Simulink® Design Optimization™

User's Guide

R2015b
How to Contact MathWorks

Latest news: www.mathworks.com

Sales and services: www.mathworks.com/sales_and_services

User community: www.mathworks.com/matlabcentral

Technical support: www.mathworks.com/support/contact_us

Phone: 508-647-7000

The MathWorks, Inc.

3 Apple Hill Drive
Natick, MA 01760-2098

Simulink® Design Optimization™ User's Guide

© COPYRIGHT 1993–2015 by The MathWorks, Inc.
The software described in this document is furnished under a license agreement. The software may be used
or copied only under the terms of the license agreement. No part of this manual may be photocopied or
reproduced in any form without prior written consent from The MathWorks, Inc.
FEDERAL ACQUISITION: This provision applies to all acquisitions of the Program and Documentation
by, for, or through the federal government of the United States. By accepting delivery of the Program
or Documentation, the government hereby agrees that this software or documentation qualifies as
commercial computer software or commercial computer software documentation as such terms are used
or defined in FAR 12.212, DFARS Part 227.72, and DFARS 252.227-7014. Accordingly, the terms and
conditions of this Agreement and only those rights specified in this Agreement, shall pertain to and
govern the use, modification, reproduction, release, performance, display, and disclosure of the Program
and Documentation by the federal government (or other entity acquiring for or through the federal
government) and shall supersede any conflicting contractual terms or conditions. If this License fails
to meet the government's needs or is inconsistent in any respect with federal procurement law, the
government agrees to return the Program and Documentation, unused, to The MathWorks, Inc.
Trademarks
MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See
www.mathworks.com/trademarks for a list of additional trademarks. Other product or brand
names may be trademarks or registered trademarks of their respective holders.
Patents
MathWorks products are protected by one or more U.S. patents. Please see
www.mathworks.com/patents for more information.
Revision History
March 2009 Online only New for Version 1 (Release 2009a)
September 2009 Online only Revised for Version 1.1 (Release 2009b)
March 2010 Online only Revised for Version 1.1.1 (Release 2010a)
September 2010 Online only Revised for Version 1.2 (Release 2010b)
April 2011 Online only Revised for Version 1.2.1 (Release 2011a)
September 2011 Online only Revised for Version 2.0 (Release 2011b)
March 2012 Online only Revised for Version 2.1 (Release 2012a)
September 2012 Online only Revised for Version 2.2 (Release 2012b)
March 2013 Online only Revised for Version 2.3 (Release 2013a)
September 2013 Online only Revised for Version 2.4 (Release 2013b)
March 2014 Online only Revised for Version 2.5 (Release 2014a)
October 2014 Online only Revised for Version 2.6 (Release 2014b)
March 2015 Online only Revised for Version 2.7 (Release 2015a)
September 2015 Online only Revised for Version 2.8 (Release 2015b)
 Contents

Data Analysis and Processing

1
Model Requirements for Importing Data . . . . . . . . . . . . . . . . 1-2
Select Input Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-3
Select Output Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-4

Import Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-6

Create Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-6
Time-Domain Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-8
Time-Series Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-11
Complex Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-12

Plot and Analyze Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-13

Why Plot the Data Before Parameter Estimation . . . . . . . . 1-13
Plot Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-13

Preprocessing Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-16

Ways to Preprocess Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-16
Remove Offset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-17
Scale Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-17
Extract Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-18
Filter Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-19
Resample Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-20
Replace Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-20

Add Preprocessed Data Sets to Estimation Project (GUI) . 1-23

Overwriting an Existing Data Set . . . . . . . . . . . . . . . . . . . . 1-23
Creating a New Data Set . . . . . . . . . . . . . . . . . . . . . . . . . . 1-24

Export Prepared Data to the MATLAB Workspace . . . . . . . 1-26

v
 Parameter Estimation
2
What Is an Experiment? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-3

Specify Estimation Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-4

Create Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-4
Specify Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-6

Specify Parameters for Estimation . . . . . . . . . . . . . . . . . . . . . 2-8

Choosing Which Parameters to Estimate First . . . . . . . . . . . 2-8
Add Model Parameters as Variables for Estimation . . . . . . . 2-8
Specify Parameters for Estimation . . . . . . . . . . . . . . . . . . . 2-11

Specify Known Initial States . . . . . . . . . . . . . . . . . . . . . . . . . 2-17

When to Specify Initial States Versus Estimate Initial States 2-17
Specify Model Initial States . . . . . . . . . . . . . . . . . . . . . . . . 2-17

Edit Experiment Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-21

Specify Experiments for Estimation . . . . . . . . . . . . . . . . . . . 2-23

Progress Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-24

Types of Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-24
Basic Steps for Creating Plots . . . . . . . . . . . . . . . . . . . . . . . 2-24

Estimation Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-30

Access Estimation Options . . . . . . . . . . . . . . . . . . . . . . . . . 2-30
General Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-30
Optimization Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-32
Specify Goodness of Fit Criteria (Cost Function) . . . . . . . . . 2-35

Progress Display Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-36

Run Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-38

Model Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-45

Load and Import Validation Data . . . . . . . . . . . . . . . . . . . . . 2-46

Specify Experiments for Validation . . . . . . . . . . . . . . . . . . . 2-47

vi Contents
Select Results for Validation . . . . . . . . . . . . . . . . . . . . . . . . . 2-49

Select Plots and Run Validation . . . . . . . . . . . . . . . . . . . . . . 2-51

Compare Measured and Simulated Responses . . . . . . . . . . 2-53

Experiment Plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-53
Residuals Plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-54

Speed Up Parameter Estimation Using Parallel Computing 2-56

When to Use Parallel Computing for Parameter Estimation 2-56
How Parallel Computing Speeds Up Estimation . . . . . . . . . 2-56

How to Use Parallel Computing for Parameter Estimation 2-60

Configure Your System for Parallel Computing . . . . . . . . . . 2-60
Model Dependencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-60
Estimate Parameters Using Parallel Computing in the
Parameter Estimation Tool . . . . . . . . . . . . . . . . . . . . . . . 2-61
Estimate Parameters Using Parallel Computing (Code) . . . . 2-64
Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-65

Use Fast Restart Mode During Parameter Estimation . . . . 2-67

Parameter Estimation Tool Workflow for Fast Restart . . . . 2-67
Command-Line Workflow for Fast Restart . . . . . . . . . . . . . 2-68
Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-69

Estimating Initial Conditions for Blocks with External Initial

Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-70

Estimation Sessions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-71

Structure of an Estimation Session . . . . . . . . . . . . . . . . . . . 2-71
Save Parameter Estimation Tool Sessions . . . . . . . . . . . . . . 2-71
Load Parameter Estimation Tool Sessions . . . . . . . . . . . . . . 2-72
Load Legacy Projects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-72

How the Software Formulates Parameter Estimation as an

Optimization Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-73
Overview of Parameter Estimation as an Optimization
Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-73
Cost Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-73
Bounds and Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-75
Optimization Methods and Problem Formulations . . . . . . . . 2-76

vii
 Write a Cost Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-83
Cost Function Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-83
Convenience Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-84
Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-85
Evaluate Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-86
Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-87

Gradient Computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-90

Estimate Model Parameter Values (Code) . . . . . . . . . . . . . . 2-92

Estimate Model Parameters and Initial States (Code) . . . 2-104

Estimate Model Parameters using Multiple Experiments

(Code) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-116

Estimate Model Parameters Per Experiment (Code) . . . . . 2-128

Estimate Model Parameters with Parameter Constraints

(Code) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-142

Estimate Model Parameter Values (GUI) . . . . . . . . . . . . . . 2-153

Estimate Model Parameters Per Experiment (GUI) . . . . . 2-166

Estimate Model Parameters and Initial States (GUI) . . . . 2-181

Generate MATLAB Code for Parameter Estimation Problems

(GUI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-192

Improving Optimization Performance using Fast Restart

(GUI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-198

Improving Optimization Performance using Fast Restart

(Code) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-206

viii Contents
 Response Optimization
3
How the Optimization Algorithm Formulates Minimization
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-3
Feasibility Problem and Constraint Formulation . . . . . . . . . . 3-3
Tracking Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-6
Gradient Descent Method Problem Formulations . . . . . . . . . 3-7
Simplex Search Method Problem Formulations . . . . . . . . . . . 3-8
Pattern Search Method Problem Formulations . . . . . . . . . . . 3-9
Gradient Computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-10

Specify Signals to Log . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-12

Specifying Step Response Characteristics . . . . . . . . . . . . . . 3-13

Specify Step Response Characteristics . . . . . . . . . . . . . . . . . 3-13

Specifying Custom Requirements . . . . . . . . . . . . . . . . . . . . . 3-17

Move Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-20

Move Constraints Graphically . . . . . . . . . . . . . . . . . . . . . . . 3-20
Position Constraints Exactly . . . . . . . . . . . . . . . . . . . . . . . . 3-21

Specify Time-Domain Design Requirements . . . . . . . . . . . . 3-23

Specify Piecewise-Linear Lower and Upper Bounds . . . . . . . 3-23
Specify Signal Property Requirements . . . . . . . . . . . . . . . . 3-24
Specify Step Response Characteristics . . . . . . . . . . . . . . . . . 3-13
Track Reference Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-30
Specify Custom Requirements . . . . . . . . . . . . . . . . . . . . . . . 3-32
Edit Design Requirements . . . . . . . . . . . . . . . . . . . . . . . . . 3-35

Edit Design Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-37

Edit Design Requirement Dialog Box Parameters . . . . . . . . 3-37

Specify Frequency-Domain Design Requirements . . . . . . . 3-39

Specify Lower Bounds on Gain and Phase Margin . . . . . . . . 3-39
Specify Piecewise-Linear Lower and Upper Bounds on
Frequency Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-41
Specify Bound on Closed-Loop Peak Gain . . . . . . . . . . . . . . 3-43
Specify Lower Bound on Damping Ratio . . . . . . . . . . . . . . . 3-45
Specify Upper and Lower Bounds on Natural Frequency . . . 3-47
Specify Upper Bound on Approximate Settling Time . . . . . . 3-49

ix
 Specify Piecewise-Linear Upper and Lower Bounds on Singular
Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-51
Specify Step Response Characteristics . . . . . . . . . . . . . . . . . 3-13
Specify Custom Requirements . . . . . . . . . . . . . . . . . . . . . . . 3-32

Specify Design Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-61

Add Model Parameters as Variables for Optimization . . . . . 3-61
Specify Design Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-64

Update Model with Design Variables Set . . . . . . . . . . . . . . . 3-66

General Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-68

Accessing General Options . . . . . . . . . . . . . . . . . . . . . . . . . 3-68
Progress Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-68
Result Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-69

Optimization Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-72

Accessing Optimization Options . . . . . . . . . . . . . . . . . . . . . 3-72
Selecting Optimization Methods . . . . . . . . . . . . . . . . . . . . . 3-73
Selecting Optimization Termination Options . . . . . . . . . . . . 3-74
Selecting Additional Optimization Options . . . . . . . . . . . . . 3-75

Create Linearization I/O Sets . . . . . . . . . . . . . . . . . . . . . . . . . 3-77

Create Linearization I/O Set . . . . . . . . . . . . . . . . . . . . . . . . 3-77

Linearization Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-79

Accessing Linearization Options . . . . . . . . . . . . . . . . . . . . . 3-79
Configuring Linearization Options . . . . . . . . . . . . . . . . . . . 3-79

Plots in the Response Optimization Tool . . . . . . . . . . . . . . . 3-82

Adding Plots in Response Optimization Tool . . . . . . . . . . . . 3-82
Plotting Current Response . . . . . . . . . . . . . . . . . . . . . . . . . 3-82
Plotting Intermediate Steps . . . . . . . . . . . . . . . . . . . . . . . . 3-82
Modifying Plot Properties . . . . . . . . . . . . . . . . . . . . . . . . . . 3-82
Plot Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-84
Export Design Variables and Requirement Values for an
Iteration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-87

Compare Requirements and Design Variables Using Spider

Plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-88

Export Design Variable Values for Specific Iteration . . . . . 3-91

x Contents
Design Optimization to Meet Time- and Frequency-Domain
Requirements (GUI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-93

Design Optimization to Meet a Custom Objective (GUI) . . 3-108

Design Optimization to Meet a Custom Objective (Code) . 3-126

Design Optimization to Meet Custom Signal Requirements

(GUI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-135

Design Optimization to Meet Frequency-Domain

Requirements (GUI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-141

Specify Custom Signal Objective with Uncertain Variable

(GUI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-160

Design Optimization with Uncertain Variables (Code) . . . 3-171

Generate MATLAB Code for Design Optimization Problems

(GUI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-181

Skip Model Simulation Based on Parameter Constraint

Violation (GUI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-188

Optimizing Parameters for Robustness . . . . . . . . . . . . . . . 3-201

What Is Robustness? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-201
Sampling Methods for Uncertain Parameters . . . . . . . . . . 3-202
Optimize Parameters for Robustness (GUI) . . . . . . . . . . . . 3-206

How to Use Accelerator Mode During Simulations . . . . . . 3-213

About Accelerating Optimization . . . . . . . . . . . . . . . . . . . . 3-213
Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-213
Setting Accelerator Mode . . . . . . . . . . . . . . . . . . . . . . . . . 3-213

Speed Up Response Optimization Using Parallel

Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-215
When to Use Parallel Computing for Response Optimization 3-215
How Parallel Computing Speeds Up Optimization . . . . . . . 3-215

How to Use Parallel Computing for Response Optimization 3-219

Configure Your System for Parallel Computing . . . . . . . . . 3-219
Model Dependencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-219
Optimize Design Using Parallel Computing (GUI) . . . . . . . 3-220

xi
 Optimize Design Using Parallel Computing (Code) . . . . . . 3-224
Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-225

Use Fast Restart Mode During Response Optimization . . 3-227

Response Optimization Tool Workflow for Fast Restart . . . 3-227
Command-Line Workflow for Fast Restart . . . . . . . . . . . . 3-228
Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-229

Optimization Does Not Make Progress . . . . . . . . . . . . . . . . 3-230

Should I worry about the scale of my responses and how
constraints and design requirements are discretized? . . 3-230
Why don't the responses and parameter values change at
all? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-230
Why does the optimization stall? . . . . . . . . . . . . . . . . . . . . 3-230

Optimization Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . 3-232

What to do if the optimization does not get close to an
acceptable solution? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-232
Why does the optimization terminate before exceeding the
maximum number of iterations, with a solution that does not
satisfy all the constraints or design requirements? . . . . 3-233
What to do if the optimization takes a long time to converge
even though it is close to a solution? . . . . . . . . . . . . . . . 3-233
What to do if the response becomes unstable and does not
recover? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-234

Optimization Speed and Parallel Computing . . . . . . . . . . . 3-235

How can I speed up the optimization? . . . . . . . . . . . . . . . . 3-235
Why are the optimization results with and without using
parallel computing different? . . . . . . . . . . . . . . . . . . . . . 3-236
Why do I not see the optimization speedup I expected using
parallel computing? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-236
Why does the optimization using parallel computing not make
any progress? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-236
Why does the optimization using parallel computing not stop
when I click the Stop optimization button? . . . . . . . . . . 3-237

Undesirable Parameter Values . . . . . . . . . . . . . . . . . . . . . . 3-238

What to do if the optimization drives the tuned compensator
elements and parameters to undesirable values? . . . . . . 3-238
What to do if the optimization violates bounds on parameter
values? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-238

xii Contents
 Reverting to Initial Parameter Values . . . . . . . . . . . . . . . . 3-240
How do I quit an optimization and revert to my initial
parameter values? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-240

Manage Response Optimization Tool Session . . . . . . . . . . 3-241

Save a Session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-241
Load a Session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-241

Optimizing Time-Domain Response of Simulink® Models

Using Parallel Computing . . . . . . . . . . . . . . . . . . . . . . . . 3-243

Design Optimization to Meet Frequency-Domain

Requirements (Code) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-252

Sensitivity Analysis
4
What Is Sensitivity Analysis? . . . . . . . . . . . . . . . . . . . . . . . . . . 4-2

Sampling Parameters for Sensitivity Analysis . . . . . . . . . . . . 4-4

Probability Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-4
Bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
Number of Samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
Method of Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
Custom Sample Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-7

Sensitivity Analysis Methods . . . . . . . . . . . . . . . . . . . . . . . . . 4-11

Visual Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-11
Quantitative Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-11

How to Use Parallel Computing for Sensitivity Analysis . . 4-14

Configure Your System for Parallel Computing . . . . . . . . . . 4-14
Model Dependencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-14
Perform Sensitivity Analysis Using Parallel Computing
(Code) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-15
Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-16

Use Fast Restart Mode During Sensitivity Analysis . . . . . . 4-17

Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-18

xiii
 Design Exploration using Parameter Sampling (Code) . . . 4-20

Identify Key Parameters for Estimation (Code) . . . . . . . . . 4-36

Optimization-Based Control Design

5
Overview of Optimization-Based Compensator Design . . . . 5-2

Time-Domain Design Requirements in Simulink . . . . . . . . . 5-4

Specify Piecewise-Linear Lower and Upper Bounds . . . . . . . 5-23
Specify Step Response Characteristics . . . . . . . . . . . . . . . . . 5-13
Track Reference Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-30
Specify Custom Requirements . . . . . . . . . . . . . . . . . . . . . . . 5-32
Edit Design Requirements . . . . . . . . . . . . . . . . . . . . . . . . . 5-35

Frequency-Domain Design Requirements in Simulink . . . . 5-16

Specify Lower Bounds on Gain and Phase Margin . . . . . . . . 5-39
Specify Piecewise-Linear Lower and Upper Bounds on
Frequency Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-41
Specify Bound on Closed-Loop Peak Gain . . . . . . . . . . . . . . 5-43
Specify Lower Bound on Damping Ratio . . . . . . . . . . . . . . . 5-45
Specify Upper and Lower Bounds on Natural Frequency . . . 5-47
Specify Upper Bound on Approximate Settling Time . . . . . . 5-49
Specify Piecewise-Linear Upper and Lower Bounds on Singular
Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-51
Specify Step Response Characteristics . . . . . . . . . . . . . . . . . 5-13
Specify Custom Requirements . . . . . . . . . . . . . . . . . . . . . . . 5-32

Time- and Frequency-Domain Requirements in SISO Design

Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-38
Root Locus Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-38
Open-Loop and Prefilter Bode Diagrams . . . . . . . . . . . . . . . 5-40
Open-Loop Nichols Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-40
Step/Impulse Response Plots . . . . . . . . . . . . . . . . . . . . . . . . 5-41

Time-Domain Simulations in SISO Design Tool . . . . . . . . . . 5-42

How to Design Optimization-Based Controllers for LTI

Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-43

xiv Contents
 Optimize LTI System to Meet Frequency-Domain
Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-44
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-44
Design Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-44
Creating an LTI Plant Model . . . . . . . . . . . . . . . . . . . . . . . 5-45
Creating Design and Analysis Plots . . . . . . . . . . . . . . . . . . 5-46
Creating a Response Optimization Task . . . . . . . . . . . . . . . 5-48
Selecting Tunable Compensator Elements . . . . . . . . . . . . . . 5-50
Adding Design Requirements . . . . . . . . . . . . . . . . . . . . . . . 5-51
Optimizing the System's Response . . . . . . . . . . . . . . . . . . . 5-59
Creating and Displaying the Closed-Loop System . . . . . . . . 5-62

Designing Linear Controllers for Simulink Models . . . . . . . 5-64

Lookup Tables
6
What are Adaptive Lookup Tables? . . . . . . . . . . . . . . . . . . . . . 6-2
Lookup Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-2
Adaptive Lookup Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-2

How to Estimate Lookup Table Values . . . . . . . . . . . . . . . . . . 6-5

Estimate Constrained Values of a Lookup Table . . . . . . . . . . 6-6

Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-6
About the Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-6
Lookup Table Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-6
Estimate the Monotonically Increasing Table Values Using
Default Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-8
Validate the Estimation Results . . . . . . . . . . . . . . . . . . . . . 6-17

Estimate Lookup Table Values from Data . . . . . . . . . . . . . . 6-23

Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-23
About the Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-23
Open a Parameter Estimation Session . . . . . . . . . . . . . . . . 6-23
Estimate the Table Values Using Default Settings . . . . . . . 6-25
Validate the Estimation Results . . . . . . . . . . . . . . . . . . . . . 6-33

Building Models Using Adaptive Lookup Table Blocks . . . 6-39

xv
 Selecting an Adaptation Method . . . . . . . . . . . . . . . . . . . . . . 6-43
Sample Mean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-43
Sample Mean with Forgetting . . . . . . . . . . . . . . . . . . . . . . . 6-44

Model Engine Using n-D Adaptive Lookup Table . . . . . . . . 6-45

Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-45
About the Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-45
Building a Model Using Adaptive Lookup Table Blocks . . . . 6-46
Adapting the Lookup Table Values Using Time-Varying I/O
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-55

Using Adaptive Lookup Tables in Real-Time Environment 6-59

xvi Contents
 1

Data Analysis and Processing

• “Model Requirements for Importing Data” on page 1-2

• “Import Data” on page 1-6
• “Plot and Analyze Data” on page 1-13
• “Preprocessing Data” on page 1-16
• “Add Preprocessed Data Sets to Estimation Project (GUI)” on page 1-23
• “Export Prepared Data to the MATLAB Workspace” on page 1-26
1 Data Analysis and Processing

Model Requirements for Importing Data

In this section...
“Select Input Signals” on page 1-3
“Select Output Signals” on page 1-4

Before you can analyze and preprocess the estimation data, you must assign the data to
the model ports or signals. In order to assign the data, the Simulink® model must contain
one of the following elements:

• Top-level Inport block

Note: You do not need an Inport block if your model already contains a fixed input
block, such as a Step block.
• Top-level Outport block
• Logged signal, which can be a top-level signal in the model or a signal in a model
subsystem

To enable signal logging for a signal, in the Simulink Editor, select the signal, click

the Simulation Data Inspector button arrow and click Log Selected Signals to
Workspace. For more information, see “Export Signal Data Using Signal Logging” in
the Simulink documentation.

When you create an experiment, as described in “Create Experiment” on page 2-4,

the top level input and output ports as well as logged signals are selected by default.
You can add or remove the input and output signals using the experiment editor. In
the experiment editor, the rows in the Inputs panel correspond to the model's top-level
Inport blocks.

1-2
 Model Requirements for Importing Data

Similarly, the rows in the Outputs panel correspond to either the top-level Outport
blocks or logged signals in the model.

Select Input Signals

You can add the Inport block in the experiment editor by clicking the Select Inputs
button in the Inputs panel to launch the Select Inputs dialog box. You can select the
Inport block you want by selecting the check box corresponding to it and clicking OK.
There is only one Inport block for the engine_idle_speed model.

Using the dialog box, you can import the input data by typing, for example,
[time,iodata(:,1)] in the Inputs panel. To learn more about importing data, see
Import Data.

1-3
1 Data Analysis and Processing

Select Output Signals

You can add the Output block in the experiment editor by clicking Select Measured
Output Signals in the Outputs panel to launch the Select Outputs dialog box. You
can select the Outport block you want by clicking the check box corresponding to it, and
clicking OK. There is only one Outport block for the engine_idle_speed model.

Using the dialog box, you can import the output data by typing, for example,
[time,iodata(:,2)] in the Inputs panel. To learn more about importing data, see
Import Data.

Related Examples
• “Import Data” on page 1-6

1-4
 Model Requirements for Importing Data

More About
• “What Is an Experiment?” on page 2-3

1-5
1 Data Analysis and Processing

Import Data
In this section...
“Create Experiment” on page 1-6
“Time-Domain Data” on page 1-8
“Time-Series Data” on page 1-11
“Complex Data” on page 1-12

Create Experiment
Before you begin data import, create an experiment. Simulink Design Optimization™
software provides a tool for setting up the estimation session.

To create an estimation session:

1 At the MATLAB® prompt, open the nonlinear idle speed model of an automotive
engine by typing :
engine_idle_speed

The model contains the Inport block BPAV and Outport block Engine Speed
for importing input and output data, respectively. To learn more, see “Model
Requirements for Importing Data” on page 1-2.
2 In the Simulink model window, open the Parameter Estimation tool by selecting
Analysis > Parameter Estimation.

1-6
 Import Data

Parameter Estimation Tool

You can organize the estimation and validation tasks inside Experiments under
Data Browser panel on the left. You can assign each experiment to an estimation
task or validation task.

To create an experiment, click the New Experiment button.

1-7
1 Data Analysis and Processing

This creates an experiment called Exp under Experiments. To change the name of
the experiment, right-click and select Rename. Call it NewData.

Note: The Simulink model must remain open to perform parameter estimation tasks.

Time-Domain Data
Experiments are collections of signal data, specifically input and output signal data.
After you create an experiment, as described in “Create Experiment” on page 1-6,
you can import data into your experiment from various sources including MATLAB®
variables, MAT-files, Excel® files, or comma-separated-value files.

To import data into your experiment right-click and select Edit.... This will launch the
experiment editor. In the experiment editor, you can define the signals contained in the
experiment.

1-8
 Import Data

For example, the rows in the Inputs panel of the editor correspond to Inport block BPAV
in the engine_idle_speed model.

The rows in the Outputs panel correspond to Outport block Engine Speed. You can
import signal data from files or MATLAB workspace.

Note: The Simulink model must contain an Inport or Outport block or logged signals to
enable importing data. For more information, see “Model Requirements for Importing
Data” on page 1-2. To select more output signals to specify data for, click Select
Measured Output Signals in the Outputs panel.

The idle-speed model of an automotive engine contains the measured data stored in the
iodata array in the workspace. The array contains two columns: the first for input data,
and the second for output data. The time data is in the time array in the workspace. You
can import the input data by typing [time,iodata(:,1)] in the Inputs panel.

1-9
1 Data Analysis and Processing

You can import the output data by typing [time,iodata(:,2)] in the Outputs panel.
You can view the data by clicking . The input data should look like this:

Output data should look like this:

1-10
 Import Data

After importing the data for NewData experiment, the experiment editor looks like this:

To import data from a file, click the button.

To learn more about the Edit Experiment: dialog box, see “Edit Experiment Data” on
page 2-21.

Time-Series Data
Time-series data is stored in time-series objects. For more information, see “Time Series
Objects” in the MATLAB documentation.

1-11
1 Data Analysis and Processing

When you import input data from a time-series object, t, for parameter estimation, you
must specify the time vector and data as [t.time,t.inputdata] in the Inport signal
dialog box. Similarly, to import output data, you must specify the time vector and data
as [t.time,t.outputdata] in the Outport signal dialog box. For more information on
how to import data into the experiment, see “Time-Domain Data” on page 1-8.

Complex Data
Complex-valued data is data whose value is a complex number. For example, a signal
with the value 1+2j is complex. You can use complex data to estimate parameters of
electrical systems, such as the magnitude and phase.

Note: You must sample the real and imaginary parts of the data as a function of the
same time vector.

To use complex data for parameter estimation:

1 Split the data into two data sets that contain the real and imaginary parts. To split
the data, use the MATLAB functions real, and imag.
2 Create two signals, one for the real part and one for the imaginary part for the
Inport or Outport block.
3 Select both signals in the experiment editor.
4 Import the data to the corresponding signal as described in “Time-Domain Data” on
page 1-8.

Related Examples
• “Plot and Analyze Data” on page 1-13
• “Preprocessing Data” on page 1-16

More About
• “Model Requirements for Importing Data” on page 1-2

1-12
 Plot and Analyze Data

Plot and Analyze Data

In this section...
“Why Plot the Data Before Parameter Estimation” on page 1-13
“Plot Data” on page 1-13

Why Plot the Data Before Parameter Estimation

After you import the estimation data as described in “Import Data” on page 1-6, you can
remove outliers, smooth, detrend, or otherwise treat the data to prepare for analysis and
estimation. To view and analyze data characteristics, plot the data on a time plot.

Plot Data
Use an experiment plot to visualize experiment data. First, create an experiment and
import data as described in “Import Data” on page 1-6. To create an experiment plot, in
the Parameter Estimation tool, on the Parameter Estimation tab, click Add Plot, and
select NewData under Experiment Plots.

This creates plots of the input signal for the Inport block BPAV and output signal for
the Outport block Engine Speed for the engine_idle_speed model (see “Create
Experiment” on page 1-6).

1-13
1 Data Analysis and Processing

You can also plot the experiment data by right-clicking NewData and selecting Plot
measured experiment data from the list.

Using the time plot, you can examine the data characteristics such as noise, outliers and
portions of the data to use for estimating parameters. After you analyze the data, you can
preprocess it as described in “Preprocessing Data” on page 1-16.

Related Examples
• “Import Data” on page 1-6
• “Preprocessing Data” on page 1-16

1-14
 Plot and Analyze Data

More About
• “Model Requirements for Importing Data” on page 1-2

1-15
1 Data Analysis and Processing

Preprocessing Data
In this section...
“Ways to Preprocess Data” on page 1-16
“Remove Offset” on page 1-17
“Scale Data” on page 1-17
“Extract Data” on page 1-18
“Filter Data” on page 1-19
“Resample Data” on page 1-20
“Replace Data” on page 1-20

Ways to Preprocess Data

In the Parameter Estimation tool, you can preprocess imported data before you use it for
parameter estimation. After plotting the measured data as shown in “Plot and Analyze
Data” on page 1-13), you have access to the Experiment Plot tab.

After importing the estimation data as described in “Import Data” on page 1-6, you can
perform the following preprocessing operations:

• “Remove Offset” on page 1-17 — Remove mean values, a constant value, or an

initial value from the data.
• “Scale Data” on page 1-17 — Scale data by a constant value, signal maximum
value, or signal initial value.
• “Extract Data” on page 1-18 — Select a subset of the data to use in the estimation.
You can graphically select the data to extract, or enter start and end times in the text
boxes.
• “Filter Data” on page 1-19 — Smooth data using a low-pass, high-pass, or band-
pass filter.

1-16
 Preprocessing Data

• “Resample Data” on page 1-20 –– Resample data using zero-order hold or linear
interpolation.
• “Replace Data” on page 1-20 –– Replace data with a constant value, region initial
value, region final value, or a line. You can use this functionality to replace outliers.

You can perform as many preprocessing operations on your data as are required for your
application. For instance, you can both filter the data and remove an offset.

Remove Offset
On the Experiment Plot tab, click Remove Offset.

It is important for good estimation results to remove data offsets. In the Remove Offset
tab, you can remove offset from all signals at once or select a particular signal using
the Remove offset from signal drop down list. Specify the value to remove using the
Offset to remove drop down list. The options are:

• A constant value. Enter the value in the box. (Default: 0)

• Mean of the data, to create zero-mean data.
• Signal initial value.

As you change the offset value, the modified data is shown in preview in the plot.

After making choices, update the existing data with the preprocessed data by clicking

Or, to save the modified data values in a new experiment, click and select

Save As: Create a new experiment from the modified data.

Scale Data
On the Experiment Plot tab, click Scale Data.

1-17
1 Data Analysis and Processing

In the Scale Data tab, you can choose to scale all signals or specify a signal to scale.
Select the scaling value from the Scale to use drop-down list. The options are:

• A constant value. Enter the value in the box. (Default: 1)

• Signal maximum value.
• Signal initial value.

As you change the scaling, the modified data is shown in preview in the plot.

After making choices, update the existing data with the preprocessed data by clicking

Or, to save the modified data values in a new experiment, click and select

Save As: Create a new experiment from the modified data.

Extract Data
To extract a portion of your data to use in the estimation process, on the Experiment
Plot tab, click Extract Data.

1-18
 Preprocessing Data

Select a subset of data to use for estimation in Extract Data tab. You can extract data
graphically or by specifying start time and end time. To extract data graphically, click
and drag the vertical bars to select a region of the data to use.

After you choose the data to extract, you can save in a new experiment by clicking Save
As.

Filter Data
You can filter your data using a low-pass, high-pass, or band-pass filter. A low-pass
filter blocks high frequency signals, a high-pass filter blocks low frequency signals, and
a band-pass filter combines the properties of both low- and high-pass filters. On the
Experiment Plot tab click one of the Low-Pass Filter, High-Pass Filter, or Band-
Pass Filter to open a new tab. For example, the low-pass filter tab appears as shown:

On the Low-Pass Filter, High-Pass Filter, or Band-Pass Filter tab, you can choose
to filter all signals or specify a particular signal. For the low-pass and high-pass filtering,
you can specify the normalized cutoff frequency of the signal. For the band-pass filter,
you can specify the normalized start and end frequencies. Specify the frequencies by
either entering the value in the associated field on the tab. Alternatively, you can specify
filter frequencies graphically, by dragging the vertical bars in the frequency-domain plot
of your data.

Click Options to specify the filter order, and select zero-phase shift filter.

After making choices, update the existing data with the preprocessed data by clicking

Or, to save the modified data values in a new experiment, click and select

Save As: Create a new experiment from the modified data.

1-19
1 Data Analysis and Processing

Resample Data
On the Experiment Plot tab, click the Resample Data button.

In the Resample Data tab, specify the sampling period using the Resample
with sample period: field. You can resample your data using one of the following
interpolation methods:

• Zero-order hold — Fill the missing data sample with the data value immediately
preceding it.
• Linear interpolation — Fill the missing data using a line that connects the two
data points.

By default, the resampling method is set to zero-order hold. You can select the
linear interpolation method from the Resample Using drop-down list.

The modified data is shown in preview in the plot.

After making choices, update the existing data with the preprocessed data by clicking

Or, to save the modified data values in a new experiment, click and select

Save As: Create a new experiment from the modified data.

Replace Data
On the Experiment Plot tab click the Replace Data button.

1-20
 Preprocessing Data

In the Replace Data tab, select data to replace by dragging across a region in the plot.
Once you select data, choose how to replace it using the Replace selected data drop-
down list. You can replace the data you select with one of these options:

• A constant value
• Region initial value
• Region final value
• A line

The replaced preview data changes color and the replacement data appears on the plot.
At any time before updating, click Clear preview to clear the data you replaced and
start over.

After making choices, update the existing data with the preprocessed data by clicking

Or, to save the modified data values in a new experiment, click and select

Save As: Create a new experiment from the modified data.

Replace Data can be useful, for example, to replace outliers. Outliers are data values
that deviate from the mean by more than three standard deviations. When estimating
parameters from data containing outliers, the results may not be accurate. Hence, you
might choose to replace the outliers in the data before you estimate the parameters.

Related Examples
• “Import Data” on page 1-6

1-21
1 Data Analysis and Processing

More About
• “Model Requirements for Importing Data” on page 1-2

1-22
 Add Preprocessed Data Sets to Estimation Project (GUI)

Add Preprocessed Data Sets to Estimation Project (GUI)

After you preprocess the data using the techniques described in “Preprocessing Data”
on page 1-16, you can add the data set to an estimation project either by overwriting an
existing data set or creating a new data set.

In this section...
“Overwriting an Existing Data Set” on page 1-23
“Creating a New Data Set” on page 1-24

Overwriting an Existing Data Set

To overwrite an existing data set with the preprocessed data:

1 In the Write results to area of the Data Preprocessing Tool GUI, select the
existing dataset option.
2 Choose the data set you want to overwrite from the drop-down list.

3 Click Add.

This action overwrites the selected data set with the modified data in the Control
and Estimation Tools Manager GUI.

1-23
1 Data Analysis and Processing

Tip You can export the preprocessed data to the MATLAB Workspace, as described in
“Export Prepared Data to the MATLAB Workspace” on page 1-26.

Creating a New Data Set

If you do not want to overwrite an existing data set with the preprocessed data, as
described in “Overwriting an Existing Data Set” on page 1-23, you can create a new
data set for the preprocessed data:

1 In the Write results to area of the Data Preprocessing Tool GUI, select the new
dataset option.
2 Specify the name of the data set in the adjacent field.

1-24
 Add Preprocessed Data Sets to Estimation Project (GUI)

3 Click Add.

This action adds a new data node in the Control and Estimation Tools Manager GUI
containing the modified data.

Tip You can export the preprocessed data to the MATLAB Workspace, as described in
“Export Prepared Data to the MATLAB Workspace” on page 1-26.

1-25
1 Data Analysis and Processing

Export Prepared Data to the MATLAB Workspace

After you add the preprocessed data to an estimation project, as described in “Add
Preprocessed Data Sets to Estimation Project (GUI)” on page 1-23, you can export the
data set to the MATLAB Workspace. You can use the data to further prepare it or
estimate parameters using the data.

1 In the Transient Data node of the Control and Estimation Tools Manager GUI,
select the node containing the prepared data set.
2 Right-click the table Data cell containing the data that you want to export, and
select Export.

The Export to Workspace dialog box opens.

3 Specify the MATLAB variable names for the prepared data and the corresponding
time vector in the Data and Time fields, respectively.

4 Click OK.

The resulting MATLAB variables data and time4 appear in the MATLAB
Workspace browser.

1-26
 2

Parameter Estimation

• “What Is an Experiment?” on page 2-3

• “Specify Estimation Data” on page 2-4
• “Specify Parameters for Estimation” on page 2-8
• “Specify Known Initial States” on page 2-17
• “Edit Experiment Data” on page 2-21
• “Specify Experiments for Estimation” on page 2-23
• “Progress Plots” on page 2-24
• “Estimation Options” on page 2-30
• “Progress Display Options” on page 2-36
• “Run Estimation” on page 2-38
• “Model Validation” on page 2-45
• “Load and Import Validation Data” on page 2-46
• “Specify Experiments for Validation” on page 2-47
• “Select Results for Validation” on page 2-49
• “Select Plots and Run Validation” on page 2-51
• “Compare Measured and Simulated Responses” on page 2-53
• “Speed Up Parameter Estimation Using Parallel Computing” on page 2-56
• “How to Use Parallel Computing for Parameter Estimation” on page 2-60
• “Use Fast Restart Mode During Parameter Estimation” on page 2-67
• “Estimating Initial Conditions for Blocks with External Initial Conditions” on page
2-70
• “Estimation Sessions” on page 2-71
• “How the Software Formulates Parameter Estimation as an Optimization Problem”
on page 2-73
• “Write a Cost Function” on page 2-83
2 Parameter Estimation

• “Gradient Computations” on page 2-90

• “Estimate Model Parameter Values (Code)” on page 2-92
• “Estimate Model Parameters and Initial States (Code)” on page 2-104
• “Estimate Model Parameters using Multiple Experiments (Code)” on page 2-116
• “Estimate Model Parameters Per Experiment (Code)” on page 2-128
• “Estimate Model Parameters with Parameter Constraints (Code)” on page 2-142
• “Estimate Model Parameter Values (GUI)” on page 2-153
• “Estimate Model Parameters Per Experiment (GUI)” on page 2-166
• “Estimate Model Parameters and Initial States (GUI)” on page 2-181
• “Generate MATLAB Code for Parameter Estimation Problems (GUI)” on page
2-192
• “Improving Optimization Performance using Fast Restart (GUI)” on page 2-198
• “Improving Optimization Performance using Fast Restart (Code)” on page 2-206

2-2
 What Is an Experiment?

What Is an Experiment?
To estimate unknown parameter values of a Simulink model, first create an experiment.
An experiment specifies measured input and output data. During estimation, the
experiment input data is used to simulate the model and the model output is compared
with the measured experiment output data. For more information about creating
experiments and importing data, see “Specify Estimation Data” on page 2-4.

In an experiment, you can specify initial-state values. To do so, specify the model initial
states for each experiment. You can optionally specify an initial guess for the initial state
values for any experiment. For more information, see “Specify Known Initial States” on
page 2-17.

To estimate a model parameter on a per-experiment basis, specify the model parameter

for each experiment. You can specify the initial values and limits for the parameter
value for any of the experiments. Alternatively, you can specify a parameter value as
a known quantity, not to be estimated. For more information, see “Specify Parameters
for Estimation” on page 2-8. You can choose to update experiments with estimated
model initial states and parameter values, or save the results in a new experiment. For
more information, see “Estimation Options” on page 2-30.

To use experiments for validating the estimated parameter values, see “Model
Validation” on page 2-45.

Related Examples
• “Specify Estimation Data” on page 2-4
• “Specify Parameters for Estimation” on page 2-8
• “Specify Known Initial States” on page 2-17
• “Specify Experiments for Estimation” on page 2-23

More About
• “Edit Experiment Data” on page 2-21

2-3
2 Parameter Estimation

Specify Estimation Data

In this section...
“Create Experiment” on page 2-4
“Specify Data” on page 2-6

Create Experiment
Before you specify estimation data, create an experiment. At the MATLAB prompt, open
the nonlinear idle speed model of an automotive engine by typing

engine_idle_speed
In the Simulink model window, open the Parameter Estimation tool by selecting
Analysis > Parameter Estimation.

2-4
 Specify Estimation Data

In the Parameter Estimation tool, on the Parameter Estimation tab, click New
Experiment.

This action creates an experiment called Exp in the Experiments list in the Data
Browser panel and opens the experiment editor. To change the name of the experiment,
right-click Exp and select Rename. If you rename it NewData, the Experiments list
now looks like this:

2-5
2 Parameter Estimation

To learn more about how to further modify an experiment using the experiment editor,
see “Edit Experiment Data” on page 2-21.

Specify Data
After you create the experiment, you can import the estimation data into the experiment.
Launch the experiment editor by right-clicking on NewData and selecting Edit....

The rows in the Inputs panel of the editor correspond to Inport block BPAV in the
engine_idle_speed model. See “Import Data” on page 1-6.

The rows in the Outputs panel correspond to Outport block Engine Speed. You can
import signal data from files or MATLAB workspace.

The idle-speed model of an automotive engine contains the measured data stored in the
iodata array in the workspace. The array contains two columns: the first for input data,
and the second for output data. The time data is in the time array in the workspace.
You can import the input data by typing [time,iodata(:,1)] in the dialog box in the
Inputs panel.

You can import the output data by typing [time,iodata(:,2)] in the dialog box in the
Outputs panel.

2-6
 Specify Estimation Data

Note: You can have more than one input or output signal, but you can have only one data
set for a signal. If you have multiple data sets, create multiple experiments.

Related Examples
• “Specify Parameters for Estimation” on page 2-8
• “Specify Known Initial States” on page 2-17

More About
• “What Is an Experiment?” on page 2-3
• “Edit Experiment Data” on page 2-21

2-7
2 Parameter Estimation

Specify Parameters for Estimation

In this section...
“Choosing Which Parameters to Estimate First” on page 2-8
“Add Model Parameters as Variables for Estimation” on page 2-8
“Specify Parameters for Estimation” on page 2-11

Choosing Which Parameters to Estimate First

Simulink Design Optimization software lets you estimate scalar, vector, and matrix
parameters. You can take an iterative approach to estimating model parameters. For
example, if you have a large number of parameters to estimate, start by estimating those
that most influence the output. After you estimate a subset of parameters and validate
the estimated parameters, you can select the remaining parameters for estimation. You
can also use sensitivity analysis for selecting the parameters that most influence the
output.

Add Model Parameters as Variables for Estimation

The software can only estimate variables that are in use by the model. Create variables
for estimation in the MATLAB or model workspace, and specify your Simulink model or
block parameters using these variables.

In this figure, the Numerator coefficients parameter of a Transfer Fcn block is

specified as a numerical value.

2-8
 Specify Parameters for Estimation

To estimate the Numerator coefficients parameter, specify it as variable gain1:

1 Create the variable gain1 in one of the following ways:

• Add the variables to the model workspace, and specify initial values.

2-9
2 Parameter Estimation

• Write initialization code in the PreloadFcn callback of the model. For more
information, see “Model Callbacks”.

gain1 = 100
2 Specify the block parameter as variable gain1 in the Transfer Fcn block dialog box.

2-10
 Specify Parameters for Estimation

You can now select gain1 for estimation. See, “Specify Parameters for Estimation” on
page 2-11.

Specify Independent Parameters for Estimation

You can also specify independent parameters that do not appear explicitly in the model
as variables for estimation. However, you cannot use this workflow with Simulink fast
restart.

Suppose that a model parameter Kint is related to independent parameters x and y such
that Kint = x+y. To estimate x and y instead of Kint:

• Create the independent variables x and y by adding them to the model workspace and
specifying initial values. This ensures that the variables are used by the model.
• Write code in the InitFcn callback of the model that defines the relationship between
Kint, x, and y. You must first use the get_param function to get the variables x and
y from the model workspace before you can use them to define Kint.

wks = get_param(gcs,'ModelWorkspace')
x = evalin(wks,'x')
y = evalin(wks,'y')
Kint = x+y;

You can now select x and y for estimation. Do not estimate the independent and
dependent parameters simultaneously. Doing so can lead to incorrect results. For
example, do not estimate Kint, x and y together.

Specify Parameters for Estimation

You can specify the parameters for estimation experiments using the Estimated
Parameters editor. In the Parameter Estimation tool, on the Parameter Estimation
tab, click Select Parameters.

To select parameters for all experiments, click Select Parameters in the Parameters
Tuned for all Experiments panel. This opens the Select model variables dialog.

2-11
2 Parameter Estimation

Here you can select the parameters you want to estimate by clicking the check box next
to it or specifying an expression. For more information see “Select Parameters Using
Select Model Variables Dialog Box” on page 2-14.

The editor looks like

For example, in the engine_idle_speed model, select freq1, freq2, freq3, gain1,
gain2, gain3 and mean_speed for estimation. You do not need to estimate the
parameters all at once. You can first select all the parameters you are interested in,
and then later select a subset to estimate. By default, all the parameters are selected
for estimation. To deselect the ones you do not want to estimate, clear the Estimate
check box for a parameter. For this example, only estimate gain1, gain2, gain3 and

2-12
 Specify Parameters for Estimation

mean_speed. Set their initial values 10, 100, 50, and 500, respectively, and then click
OK. The Edit: Estimated Parameters dialog box looks like

To learn how to specify initial values and upper and lower bounds of the parameters, see
“Specifying Initial Guesses and Upper/Lower Bounds” on page 2-15.

Select Parameters to Estimate for a Specific Experiment

To select the parameters to estimate in a specific experiment, first, select the experiment
for estimation as described in “Specify Experiments for Estimation” on page 2-23.
Then, you can use the Edit:Estimated Parameters dialog to select parameters to
estimate for that experiment. Select the experiment name from the Experiment: combo

2-13
2 Parameter Estimation

box in the Parameters and Initial States Tuned per Experiment panel. Then click
Edit experiment to launch the experiment editor for the experiment you select.

Alternatively, you can right click the experiment name in the Experiments list
and select Edit.... In the experiment editor, click the Select parameters button in
the Parameters panel. In the Select model variables dialog, you can select the
parameters you want to estimate in this experiment by checking the box next to it or
specifying an expression. For more information see “Select Parameters Using Select
Model Variables Dialog Box” on page 2-14.

Select Parameters Using Select Model Variables Dialog Box

Use this dialog box to specify parameters to estimate. The table lists all the variables in
the model workspace and the MATLAB workspace that the model uses, whether tunable
or not. Select the parameters to estimate by clicking the check box next to each variable.
If your model contains a large number of variables, filter the list by typing in the Filter
by variable name field.

The Used By column lists the blocks in the model where each variable is used. When a
parameter is used in more than one block, all blocks are listed. To locate blocks in which
a parameter is used, click the block name to highlight the block in the model.

If the parameter is not a scalar variable, click Specify expression indexing if

necessary and enter an expression that resolves to a numeric scalar value.

2-14
 Specify Parameters for Estimation

Non-scalar variables include:

• Simulink software parameter object

Example: For a Simulink parameter object k, enter k.Value.

• Structure

Example: For a structure S, enter S.fieldname, where fieldname is the name of

the field that contains the parameter.
• Cell array

Example: Type C{1} to select the first element of a cell array C.

• MATLAB array

Example: Type a(2) to select the second element of an array a.

You cannot use mathematical expressions such as a + b. Sometimes, models have

parameters that are not explicitly defined in the model itself. For example, a gain k could
be defined in the MATLAB workspace as k = a + b, where a and b are not defined in
the model but k is used. To add these independent parameters, such as k, to the Select
Parameters dialog box, see “Add Model Parameters as Variables for Estimation” on page
2-8.

Specifying Initial Guesses and Upper/Lower Bounds

After you select parameters, you can specify

• Initial guess — The value the estimation uses to start the process.
• Minimum — The smallest allowable parameter value. The default is -Inf.
• Maximum — The largest allowable parameter value. The default is +Inf.

You can enter the initial value in the dialog box below the parameter name. You can
specify the minimum and maximum value fields by clicking the arrow . The default
minimum and maximum values are -Inf and +Inf, respectively, but you can select
any range you want.

2-15
2 Parameter Estimation

If you believe a parameter lies within a finite range, it is best not to use the default
minimum and maximum values. Often, there are computational advantages in
specifying finite bounds. It can be very important to specify lower and upper bounds.
For example, if a parameter specifies the weight of a part, be sure to specify 0 as the
absolute lower bound if better information is unavailable.

Note: When you specify the minimum and maximum values for the parameters, it
does not affect your settings in the Parameters list under Data Browser pane. You
make these choices for each experiment.
• Scale — Scale is used for normalization, in situations, for example, when parameters
are in different units.

Related Examples
• “Specify Known Initial States” on page 2-17

More About
• “What Is an Experiment?” on page 2-3

2-16
 Specify Known Initial States

Specify Known Initial States

In this section...
“When to Specify Initial States Versus Estimate Initial States” on page 2-17
“Specify Model Initial States” on page 2-17

When to Specify Initial States Versus Estimate Initial States

Sets of measured data are often collected at various times and under different initial
conditions. When you estimate model parameters using one data set and subsequently
run another estimation with a second data set, your parameter values may not match.

You can use the Parameter Estimation tool to estimate the initial conditions using
procedures that are similar to those you use to estimate parameters. You can then use
these initial condition estimates as a basis for estimating parameters for your Simulink
model.

Specify Model Initial States

After you select parameters for estimation, as described in “Specify Parameters for
Estimation” on page 2-8, you can specify initial conditions of states in your model. By
default, the estimation uses initial conditions specified in the Simulink model. If you
want to specify initial conditions other than the defaults, use the Initial States panel
in the experiment editor dialog. For this example, right click NewData and select Edit...
from the list to open the experiment editor. Then, click Select Initial States button.

The Select Model States dialog for the engine_idle_speed model looks like

2-17
2 Parameter Estimation

Click the check box next to the state you would like to modify. For example, if you select
engine_idle_speed/Transfer Fcn and enter the initial values -0.2 and 0, the Initial
States panel now looks like

Click Select Parameters in the Parameter Estimation tab. After you also select the
parameters as described in “Specify Parameters for Estimation” on page 2-8, the Edit:
Estimated Parameters dialog looks like the following figure.

2-18
 Specify Known Initial States

Related Examples
• “Specify Estimation Data” on page 2-4
• “Specify Parameters for Estimation” on page 2-8

2-19
2 Parameter Estimation

More About
• “Edit Experiment Data” on page 2-21

2-20
 Edit Experiment Data

Edit Experiment Data

After creating an experiment as in “Create Experiment” on page 2-4, you can launch the
experiment editor by right-clicking on the experiment name and selecting Edit... from
the list. The experiment editor resembles the following figure.

The experiment editor has four panels. You can select output signals and import output
data in the Outputs panel. You can select input signals and import input data in the
Inputs panel. You can specify model initial states in the Initial States panel. And you
can specify parameters to estimate in the Parameters panel.

2-21
2 Parameter Estimation

Related Examples
• “Specify Estimation Data” on page 2-4
• “Specify Parameters for Estimation” on page 2-8
• “Specify Known Initial States” on page 2-17
• “Specify Experiments for Estimation” on page 2-23

More About
• “What Is an Experiment?” on page 2-3

2-22
 Specify Experiments for Estimation

Specify Experiments for Estimation

After specifying the parameters for estimation, select the experiment for the estimation
task. In the Parameter Estimation tool, on the Parameter Estimation tab, click Select
Experiments. In the Select Experiments dialog box, you can select the experiment
to use for estimation by clicking the check box corresponding to the experiment in the
Estimation column. For this example, there is only one experiment, NewData.

For more information, see “Specify Experiments for Validation” on page 2-47

Related Examples
• “Specify Estimation Data” on page 2-4
• “Specify Parameters for Estimation” on page 2-8
• “Specify Known Initial States” on page 2-17
• “Specify Experiments for Validation” on page 2-47

More About
• “What Is an Experiment?” on page 2-3

2-23
2 Parameter Estimation

Progress Plots

In this section...
“Types of Plots” on page 2-24
“Basic Steps for Creating Plots” on page 2-24

Types of Plots
The following types of plots are available for viewing and evaluating the estimation:

• Cost function — Plot the cost function value as it changes during estimation.
• Measured and simulated — Plot empirical data against simulated data.
• Parameter trajectory — Plot the parameter values as they change.
• Residuals — Plot the error between the experimental data and the simulated
output.

Basic Steps for Creating Plots

When you run estimation, Parameter Estimation tool automatically displays a parameter
trajectory plot that shows the change in the parameter values by iteration. However,
you can create other plots for viewing the progress of the estimation before you begin
estimating the parameters.

Note: An experiment must be created and selected for estimation before creating views.
To learn more, see “Create Experiment” on page 2-4 and “Specify Experiments for
Estimation” on page 2-23.

You can plot the measured data by clicking Add Plot on the Parameter Estimation
tab and selecting the experiment to use for estimation under Experiment Plots of the
drop down list. You can then add the simulated response on top of the measured data
plot by clicking Plot Model Response on the Parameter Estimation tab. Another way
of plotting the measured and simulated data is to right-click the experiment name, for
this example, NewData, and select Plot measured & simulated data from the list. The
plot looks like this:

2-24
 Progress Plots

You can see that the measured and simulated data do not match. You can edit the labels,
adjust the limits, change the units, and the font style of the plot in the Property Editor.
To launch the editor, right-click the experiment plot and select Properties from the list.

2-25
2 Parameter Estimation

You can also add a plot for the expected estimation cost by clicking Add Plot on the
Parameter Estimation tab and selecting Estimation Cost from the list. You can close
the message “There is no data for ExpCost, run the optimization to
update the plot”, by clicking the left arrow toggle to the left of the message. The plot
looks like this:

2-26
 Progress Plots

When you right-click the estimated parameters or expected cost plots, you can choose to
add the scaled values or save iteration data from the list.

2-27
2 Parameter Estimation

Add the progress plots by clicking the Add Plot button on the Parameter Estimation
tab and selecting the plot in the list. Selecting Parameter Trajectory adds the
following plot to the tool.

Use the View tab of the Parameter Estimation tool to arrange the layout of the plots, so
that Experiment plot: NewData, Iteration plot 1, and Iteration plot 2 are
all visible.

2-28
 Progress Plots

When you perform the estimation, all plots update automatically.

2-29
2 Parameter Estimation

Estimation Options

In this section...
“Access Estimation Options” on page 2-30
“General Options” on page 2-30
“Optimization Options” on page 2-32
“Specify Goodness of Fit Criteria (Cost Function)” on page 2-35

Access Estimation Options

In the Parameter Estimation tool, you can choose the cost function from the Cost
Function combo box.

You can access other estimation options by clicking the More Options button.

Note: Parallel Options (see “Estimate Parameters Using Parallel Computing in the
Parameter Estimation Tool” on page 2-61).

General Options
You can set optimization progress and result options for your estimation task in the
Estimation Options dialog box.

2-30
 Estimation Options

Progress Options
Show estimation progress window during estimation option opens an estimation
progress window. The window displays iteration information, such as the cost function,
and the termination information, such as whether the optimization converged.

Create a parameter trajectory plot during estimation option creates a plot that
shows how the parameter values change by iteration.

Update plots during optimization option updates the progress plots such a cost
function and parameter trajectory plot and estimation plots during estimation.

Result Options
Update model at the end of estimation option updates the parameter values in the
Simulink model after the estimation terminates.

2-31
2 Parameter Estimation

Overwrite estimation result with new estimated values option overwrites any
estimation results or experiments with the new estimated values.

Save estimated values as new estimation result option creates a new variable that
contains the estimated parameter values.

Optimization Options

• “Supported Estimation Methods” on page 2-32

• “Optimization Termination Options” on page 2-34
• “Additional Optimization Options” on page 2-34

Supported Estimation Methods

Both the Method and Algorithm options define the optimization method.

2-32
 Estimation Options

For the Method option

• Nonlinear least squares (default) — Uses the Optimization Toolbox™ nonlinear

least squares function lsqnonlin.
• Gradient descent — Uses the Optimization Toolbox function fmincon.
• Pattern search — Uses the pattern search method patternsearch. This option
requires Global Optimization Toolbox.
• Simplex search — Uses the Optimization Toolbox function fminsearch, which is
a direct search method. Simplex search is most useful for simple problems and is
sometimes faster than fmincon for models that contain discontinuities.

For the Nonlinear least squares and Gradient descent estimation methods

Method Algorithm Option Learn More

Nonlinear least • Trust-Region-Reflective In the Optimization
squares (default) Toolbox documentation,
• Levenberg-Marquardt see:

• “Trust-Region-
Reflective Least
Squares Algorithm”
• “Levenberg-
Marquardt Method”
Gradient descent • Trust-Region-Reflective In the Optimization
(default) Toolbox documentation,
• Interior-Point see:

• Active-Set • “fmincon Active Set

• Sequential Quadratic Algorithm”
Programming • “fmincon Interior
Point Algorithm”
• “fmincon Trust
Region Reflective
Algorithm”
• “fmincon SQP
Algorithm”

2-33
2 Parameter Estimation

Optimization Termination Options

Several options define when the optimization terminates:

• Parameter tolerance — Optimization terminates when successive parameter values

change by less than this number.
• Maximum iterations — Maximum number of iterations allowed. The optimization
terminates when the number of iterations exceeds this value.
• Function tolerance — Optimization terminates when successive function values are
less than this value.
• Use robust cost — Make the optimizer use a robust cost function instead of the
default least-squares cost. This is useful if the experimental data has many outliers,
or if your data is noisy.

By varying these parameters, you can force the optimization to continue searching for a
solution or to continue searching for a more accurate solution.

Additional Optimization Options

• Display level — Specify the form of the output that appears in the MATLAB
Command Window. The options are:

• Iteration — Display information after each iteration.

• None — Turn off all output.
• Notify — Display output only if the function does not converge.
• Final — Display only the final output.

Refer to the Optimization Toolbox documentation for more information on what type
of iterative output each method displays.

2-34
 Estimation Options

• Restarts — Indicate the number of times you want to restart, to automatically

restart the optimization. In some optimizations, the Hessian may become ill-
conditioned and the optimization does not converge. In these cases, it is sometimes
useful to restart the optimization after it stops, using the endpoint of the previous
optimization as the starting point for the next one.

Specify Goodness of Fit Criteria (Cost Function)

The cost function is a function that estimation methods minimize. You can specify
the cost function by selecting from the list in the Cost Function combo box on the
Parameter Estimation tab.

The default cost function is Sum Squared Error (sum of squared errors), which uses
a least-squares approach. You can also use Sum-Absolute Error, the sum of absolute
errors.

You can select the robust cost option from the Optimization Options tab of the
Estimation Options editor. To launch the dialog box click More Options... on the
Parameter Estimation tool. Use robust cost option makes the optimizer use a robust
cost function instead of the default least-squares cost. This is useful if the experimental
data has many outliers, or if your data is noisy.

2-35
2 Parameter Estimation

Progress Display Options

You can specify the display options in the Parameter Estimation tool. On the Parameter
Estimation tab, in the Estimate section, click the arrow , and select Open Estimation
Report.

The report displays the iteration number (Iteration), the number of times the objective
function is calculated (F-count), and cost function value (f(x)) by default. You can
change by clicking Display Options.

2-36
 Progress Display Options

Clear a check box to remove it from the display table. To learn more about the display
table, see “Iterative Display” in the Optimization Toolbox documentation.

2-37
2 Parameter Estimation

Run Estimation
Before you begin estimating the parameters, you must have configured the estimation
data, selected parameters, and specified estimation options, as described in “Specify
Estimation Data” on page 2-4, “Specify Parameters for Estimation” on page 2-8, and
“Estimation Options” on page 2-30, respectively.

To start the estimation, in the Parameter Estimation tool, on the Parameter

Estimation tab, click the Estimate button .

When starting the estimation, a progress window displays. At the end of the estimation,
the Estimation Progress Report window should resemble the following:

The estimation results are saved in EstimatedParams in the Results list on the Browse
Data pane.

2-38
 Run Estimation

Right-click EstimatedParams and select Open... from the menu. The window looks like
the following figure.

The EstimatedParams includes the values of the parameters, the cost function value,
and information about the stopping criteria for the estimation. The optimization stops
because the successive function values are less than the specified value 1e-3.

The Estimation Progress Report includes the change in the cost function in the
column titled NewData(Minimize). To see a plot of the change in the cost function

2-39
2 Parameter Estimation

during estimation, add the cost function plot by clicking the Add Plot button on the
Parameter Estimation tab and selecting Estimation Cost from the list. After the
estimation process completes, the cost function minimization plot appears as shown in
the following figure.

Usually, a lower cost function value indicates a successful estimation, meaning that the
experimental data matches the model simulation with the estimated parameters. If the
optimization went well, you should see your cost function converge on a minimum value.
The lower the cost, the more successful is the estimation.

For information on types of problems you may encounter using optimization solvers, see
the following topics in the Optimization Toolbox documentation:

2-40
 Run Estimation

• “When the Solver Fails ”

• “When the Solver Might Have Succeeded ”
• “When the Solver Succeeds ”

The estimated parameters graph shows the change in the estimated value of the
parameters by iteration.

The values of the parameters are recorded with the estimated values.

2-41
2 Parameter Estimation

The values of the estimated parameters are also updated in the MATLAB workspace.

You can also examine the measured versus simulated data plot to see how closely
the simulated data matches the measured estimation data. The next figure shows
the measured versus simulated data plot generated by running the estimation of the
engine_idle_speed model (for engine_idle_speed model, see “Create Experiment”
on page 2-4). Now, the simulated values match the measured output signal better.

2-42
 Run Estimation

Related Examples
• “Specify Estimation Data” on page 2-4
• “Specify Parameters for Estimation” on page 2-8
• “Specify Known Initial States” on page 2-17
• “Specify Experiments for Estimation” on page 2-23
• “Progress Plots” on page 2-24
• “Estimation Options” on page 2-30

2-43
2 Parameter Estimation

More About
• “What Is an Experiment?” on page 2-3
• “Model Validation” on page 2-45

2-44
 Model Validation

Model Validation
After you complete estimating parameters as described in “Run Estimation” on page
2-38, validate the results against another set of data.

To validate a model using the Parameter Estimation tool

1 Add a new experiment in the Experiments list in the Data Browser pane.
2 Import the validation data set to the experiment you want to use for validation. To
do this right-click the experiment name and select Edit....
3 Select the experiment for validation.
4 Select the results to use in the validation.
5 Select the plots to display at the end of validation.
6 Run validation and compare the measured validation data against the model output
simulated with the estimated parameter to see if they match.

You can run validation after the estimation is complete. Validations can use other
validation data sets for comparison with the model response. You must set up all
estimation plots before an estimation, and you can watch the views update in real time.
Validations appear after you have completed an estimation and do not update.

You can validate data by comparing measured and simulated data for your estimation
and validation experiments.

Also, it is useful to compare residuals of measured and simulated data in the same way.

Related Examples
• “Load and Import Validation Data” on page 2-46
• “Specify Experiments for Validation” on page 2-47
• “Select Results for Validation” on page 2-49
• “Compare Measured and Simulated Responses” on page 2-53

More About
• “What Is an Experiment?” on page 2-3

2-45
2 Parameter Estimation

Load and Import Validation Data

To validate the estimated parameters computed in “Run Estimation” on page 2-38,
import the data into the experiment you want to use for validation in the Parameter
Estimation tool.

At the MATLAB prompt, load the validation data into the MATLAB workspace by typing
load iodataval

Import this data into the validation experiment in the Parameter Estimation tool.

1 Add a new experiment in the Experiments list in the Data Browser pane. You can
rename the experiment by right-clicking and selecting Rename from the list.
2 Right click the experiment name and select Edit...
3 Type [time,iodataval(:,1)] in the dialog box in the Inputs panel.
4 Type [time,iodataval(:,2)] in the dialog box in the Outputs panel.

Related Examples
• “Import Data” on page 1-6
• “Specify Experiments for Validation” on page 2-47
• “Select Results for Validation” on page 2-49
• “Compare Measured and Simulated Responses” on page 2-53

More About
• “Model Validation” on page 2-45

2-46
 Specify Experiments for Validation

Specify Experiments for Validation

After importing data into an experiment that you want to use for validation, select the
experiment for the validation task. In the Parameter Estimation tool, on the Validation
tab, click Select Experiments.

When you create an experiment, it is by default selected for estimation. To select an

experiment for validation in the Select Experiments dialog box, deselect the check box
under the Estimation column and select the check box under the Validation column for
the corresponding experiment.

Related Examples
• “Load and Import Validation Data” on page 2-46
• “Select Results for Validation” on page 2-49
• “Compare Measured and Simulated Responses” on page 2-53

2-47
2 Parameter Estimation

More About
• “Model Validation” on page 2-45

2-48
 Select Results for Validation

Select Results for Validation

After you run the estimation as described in “Run Estimation” on page 2-38, the results
are recorded in EstimatedParams by default. After you import the validation data,
as described in “Load and Import Validation Data” on page 2-46, you must make sure
the validation uses the tuned parameter values in EstimatedParams in validation.
Validation Tab lets you do this. In the Parameter Estimation tool, on the Validation tab,
click Select Results to Validate.

In Select Results dialog box, validate the estimated parameters by selecting

EstimatedParams and deselecting Use current parameter values.

Related Examples
• “Load and Import Validation Data” on page 2-46
• “Specify Experiments for Validation” on page 2-47

2-49
2 Parameter Estimation

• “Compare Measured and Simulated Responses” on page 2-53

More About
• “Model Validation” on page 2-45

2-50
 Select Plots and Run Validation

Select Plots and Run Validation

Parameter Estimation tool displays the measured and simulated responses and the
residuals plot at the end of validation. Select the plots to display by checking the
corresponding box on the Validation Tab.

To perform the validation, on the Validation tab, click Validate. The Validation
Progress Report shows the status of the validation.

2-51
2 Parameter Estimation

Related Examples
• “Load and Import Validation Data” on page 2-46
• “Specify Experiments for Validation” on page 2-47
• “Select Results for Validation” on page 2-49
• “Compare Measured and Simulated Responses” on page 2-53

More About
• “What Is an Experiment?” on page 2-3
• “Model Validation” on page 2-45

2-52
 Compare Measured and Simulated Responses

Compare Measured and Simulated Responses

Compare the measured and simulated responses using the experiment plot and residuals
plot. To create these plots, select the corresponding check boxes on the Validation Tab
of the Parameter Estimation tool before you start validation.

Experiment Plot
The Parameter Estimation tool by default displays the experiment plot for each
experiment selected for validation. Each experiment plot shows the measured data,
as well as data from simulation using each set of results selected. For example, the
following figure shows the experiment plot for the ValidationExp data.

2-53
2 Parameter Estimation

Residuals Plot
After comparing the measured and simulated responses for an estimation, examine the
residuals. The residuals plot shows the difference between the simulated response and
measured data. To indicate a good fit between the simulated output and measured data,
the residuals should:

• Lie within a small percent of the maximum output variation.

• Not display any systematic patterns.

For example, you can see from the following figure that the residuals for
ValidationExp data satisfy both criteria.

2-54
 Compare Measured and Simulated Responses

Related Examples
• “Load and Import Validation Data” on page 2-46
• “Specify Experiments for Validation” on page 2-47
• “Select Results for Validation” on page 2-49

More About
• “Model Validation” on page 2-45

2-55
2 Parameter Estimation

Speed Up Parameter Estimation Using Parallel Computing

In this section...
“When to Use Parallel Computing for Parameter Estimation” on page 2-56
“How Parallel Computing Speeds Up Estimation” on page 2-56

When to Use Parallel Computing for Parameter Estimation

You can use Simulink Design Optimization software with Parallel Computing Toolbox™
software to speed up parameter estimation of Simulink models. Using parallel computing
may reduce the estimation time in the following cases:

• The model contains a large number parameters to estimate, and the estimation
method is specified as either Nonlinear least squares or Gradient descent.
• The Pattern search method is selected as the estimation method.
• The model is complex and takes a long time to simulate.

When you use parallel computing, the software distributes independent simulations
to run them in parallel on multiple MATLAB sessions, also known as workers. The
time required to simulate the model dominates the total estimation time. Therefore,
distributing the simulations significantly reduces the estimation time.

For information on how the software distributes the simulations and the expected
speedup, see “How Parallel Computing Speeds Up Estimation” on page 2-56.

For information on configuring your system and using parallel computing, see “How to
Use Parallel Computing for Parameter Estimation” on page 2-60.

How Parallel Computing Speeds Up Estimation

You can enable parallel computing with the Nonlinear least squares, Gradient
descent and Pattern search estimation methods. The following sections describes
the potential speedup using parallel computing for estimation:

• “Parallel Computing with Nonlinear least squares and Gradient descent Methods” on
page 2-57
• “Parallel Computing with the Pattern search Method” on page 2-58

2-56
 Speed Up Parameter Estimation Using Parallel Computing

Parallel Computing with Nonlinear least squares and Gradient descent Methods

When you select Gradient descent as the estimation method, the model is simulated
during the following computations:

• Objective value computation — One simulation per iteration

• Objective gradient computations — Two simulations for every tuned parameter per
iteration
• Line search computations — Multiple simulations per iteration

The total time, Ttotal , taken per iteration to perform these simulations is given by the
following equation:

Ttotal = T + ( Np ¥ 2 ¥ T ) + ( Nls ¥ T ) = T ¥ (1 + (2 ¥ Np) + Nls)

where T is the time taken to simulate the model and is assumed to be equal for all
simulations, Np is the number of parameters to estimate, and Nls is the number of line
searches. Nls is difficult to estimate and you generally assume it to be equal to one, two,
or three.

When you use parallel computing, the software distributes the simulations required
for objective gradient computations. The simulation time taken per iteration when the
gradient computations are performed in parallel, TtotalP , is approximately given by the
following equation:

Ê Np ˆ Ê Np ˆ
TtotalP =T + ( ceil Á ˜ ¥ 2 ¥ T ) + ( Nls ¥ T ) = T ¥ (1 + 2 ¥ ceil Á ˜ + Nls)
Ë Nw ¯ Ë Nw ¯

where Nw is the number of MATLAB workers.

Note: The equation does not include the time overheads associated with configuring the
system for parallel computing and loading Simulink software on the remote MATLAB
workers.

The expected reduction of the total estimation time is given by the following equation:

2-57
2 Parameter Estimation

Ê Np ˆ
1 + 2 ¥ ceil Á ˜ + Nls
TtotalP
= Ë Nw ¯
Ttotal 1 + ( 2 ¥ Np) + Nls

For example, for a model with Np=3, Nw=4, and Nls=3, the expected reduction of the total
Ê3ˆ
1 + 2 ¥ ceil Á ˜ + 3
estimation time equals Ë4¯ = 0.6 .
1 + (2 ¥ 3) + 3

Parallel Computing with the Pattern search Method

The Pattern search method uses search and poll sets to create and compute a set of
candidate solutions at each estimation iteration.

The total time, Ttotal , taken per iteration to perform these simulations, is given by the
following equation:

Ttotal = (T ¥ Np ¥ Nss) + ( T ¥ Np ¥ Nps) = T ¥ Np ¥ ( Nss + Nps)

where T is the time taken to simulate the model and is assumed to be equal for all
simulations, Np is the number of parameters to estimate, Nss is a factor for the search
set size, and Nps is a factor for the poll set size. Nss and Nps are typically proportional
to Np .

When you use parallel computing, Simulink Design Optimization software distributes
the simulations required for the search and poll set computations, which are evaluated in
separate parfor loops. The simulation time taken per iteration when the search and poll
sets are computed in parallel, TtotalP , is given by the following equation:

Nss Nps
TtotalP = ( T ¥ ceil( Np ¥ )) + ( T ¥ ceil( Np ¥ ))
Nw Nw
Nss Nps
= T ¥ ( ceil( Np ¥ ) + ceil( Np ¥ ))
Nw Nw

where Nw is the number of MATLAB workers.

2-58
 Speed Up Parameter Estimation Using Parallel Computing

Note: The equation does not include the time overheads associated with configuring the
system for parallel computing and loading Simulink software on the remote MATLAB
workers.

The expected speed up for the total estimation time is given by the following equation:

Nss Nps
ceil( Np ¥ ) + ceil( Np ¥ )
TtotalP Nw Nw
=
Ttotal Np ¥ ( Nss + Nps)

For example, for a model with Np=3, Nw=4, Nss=15, and Nps=2, the expected speedup
15 2
ceil(3 ¥ ) + ceil(3 ¥ )
equals 4 4 = 0 .27 .
3 ¥ (15 + 2)

Using the Pattern search method with parallel computing may not speed up the
estimation time. When you do not use parallel computing, the method stops searching
for a candidate solution at each iteration as soon as it finds a solution better than the
current solution. When you use parallel computing, the candidate solution search is more
comprehensive. Although the number of iterations may be larger, the estimation without
using parallel computing may be faster.

Related Examples
• “How to Use Parallel Computing for Parameter Estimation” on page 2-60

2-59
2 Parameter Estimation

How to Use Parallel Computing for Parameter Estimation

In this section...
“Configure Your System for Parallel Computing” on page 2-60
“Model Dependencies” on page 2-60
“Estimate Parameters Using Parallel Computing in the Parameter Estimation Tool” on
page 2-61
“Estimate Parameters Using Parallel Computing (Code)” on page 2-64
“Troubleshooting” on page 2-65

Configure Your System for Parallel Computing

You can speed up parameter estimation using parallel computing on multicore processors
or multiprocessor networks. Use parallel computing with the Parameter Estimation
tool and sdo.optimize to estimate parameters using the fmincon, lsqonlin, and
patternsearch methods. Parallel computing is not supported for the fminsearch
(Simplex search) method.

When you estimate model parameters using parallel computing, the software uses the
available parallel pool. If none is available, and you select Automatically create a
parallel pool in your Parallel Computing Toolbox preferences, the software starts a
parallel pool using the settings in those preferences. To open a parallel pool that uses a
specific cluster profile, use:

parpool(MyProfile);

MyProfile is the name of a cluster profile.

For information regarding creating a cluster profile, see “Create and Modify Cluster
Profiles” in the Parallel Computing Toolbox documentation.

Model Dependencies
Model dependencies are any referenced models, data such as model variables, S-
functions, and additional files necessary to run the model. Before starting the
optimization, verify that the model dependencies are complete. Otherwise, you may get
unexpected results.

2-60
 How to Use Parallel Computing for Parameter Estimation

Making Model Dependencies Accessible to Remote Workers

When you use parallel computing, the Simulink Design Optimization software helps you
identify model dependencies. To do so, the software uses the Simulink Manifest Tools.
The dependency analysis may not find all the files required by your model. To learn
more, see “Scope of Dependency Analysis” in the Simulink documentation. If your model
has dependencies that are undetected or inaccessible by the parallel pool workers, then
add them to the list of model dependencies.

The dependencies are made accessible to the parallel pool workers by specifying one of
the following:

• File dependencies: the model dependency files are copied to the parallel pool workers.
• Path dependencies: the paths to the model dependencies are specified to the parallel
pool workers. If you are working in a multi-platform scenario, ensure that the paths
are compatible across platforms.

Using file dependencies is recommended, however, in some cases it can be better to

choose path dependencies. For example, if parallel computing is set up on a local multi-
core computer, using path dependencies is preferred as using file dependencies creates
multiple copies of the dependency files on the local computer.

For more information, see:

• “Estimate Parameters Using Parallel Computing in the Parameter Estimation Tool”

on page 2-61
• “Estimate Parameters Using Parallel Computing (Code)” on page 2-64

Estimate Parameters Using Parallel Computing in the Parameter

Estimation Tool
To estimate model parameters using parallel computing in the Parameter Estimation
tool:

1 Ensure that the software can access parallel pool workers that use the appropriate
cluster profile.

For more information, see “Configure Your System for Parallel Computing” on page
2-60.
2 Open the Parameter Estimation tool for the Simulink model.

2-61
2 Parameter Estimation

3 Configure the estimation data, estimation parameters and states, and, optionally,
estimation settings.

For more information, see “Specify Estimation Data” on page 2-4, “Specify
Parameters for Estimation” on page 2-8, and “Optimization Options” on page 2-32.
4
On the Parameter Estimation tab, click More Options to open the
Estimation Options dialog box.
5 Select the Parallel Options tab.

6 Select the Use the parallel pool during optimization check box.

This option checks for dependencies in your Simulink model. The file dependencies
are displayed in the Model file dependencies list box, and corresponding
path to the files in Model path dependencies. The files listed in Model file
dependencies are copied to the remote workers.

Note: The automatic dependencies check may not detect all the dependencies in your
model.

2-62
 How to Use Parallel Computing for Parameter Estimation

For more information, see “Model Dependencies” on page 2-60. In this case, add
the undetected dependencies manually.
7 Add any file dependencies that the automatic check does not detect.

Specify the files in the Model file dependencies list box separated by semicolons
or on separate lines.

Alternatively, click Add file dependency to open a dialog box, and select the file to
add.

Note: If you do not want to copy the files to the remote workers, delete all entries in
the Model file dependencies list box. Populate the Model path dependencies
list box by clicking the Sync path dependencies from model, and add any
undetected path dependencies. In addition, in the list box, update the paths on local
drives to make them accessible to remote workers. For example, change C:\ to \\\
\hostname\\C$\\.
8 If you modify the Simulink model, resync the dependencies to ensure that any new
dependencies are detected. Click Sync file dependencies from model in the
Parallel Options tab to rerun the automatic dependency check for your model.

2-63
2 Parameter Estimation

This action updates the Model file dependencies list box with any new file
dependency found in the model.
9 Click OK.
10 In the Parameter Estimation tab, click Estimate to estimate the model
parameters using parallel computing.

For information on troubleshooting problems related to estimation using parallel

computing, see “Troubleshooting” on page 2-65.

Estimate Parameters Using Parallel Computing (Code)

To use parallel computing for parameter estimation at the command line:

1 Ensure that the software can access parallel pool workers that use the appropriate
cluster profile.

For more information, see “Configure Your System for Parallel Computing” on page
2-60.
2 Open the model.
3 Configure an estimation experiment. For example, see “Estimate Model Parameter
Values (Code)” on page 2-92.
4 Enable parallel computing using an optimization option set, opt.

opt = sdo.OptimizeOptions;
opt.UseParallel = 'always';
5 Find the model dependencies.

[dirs,files] = sdo.getModelDependencies(modelname)

Note: sdo.getModelDependencies may not detect all the dependencies in your

model. For more information, see “Model Dependencies” on page 2-60. In this
case, add the undetected dependencies manually.
6 Modify files to include any file dependencies that sdo.getModelDependencies
does not detect.

files = vertcat(files,'C:\matlab\work\filename.m')

2-64
 How to Use Parallel Computing for Parameter Estimation

Note: If you do not want to copy the files to the remote workers, use the path
dependencies. Add any undetected path dependencies to dirs and update
the paths on local drives to make them accessible to remote workers. See
sdo.getModelDependencies for more details.
7 Add the file dependencies for optimization.

opt.ParallelFileDependencies = files;
8 Run the optimization.

[pOpt,opt_info] = sdo.optimize(opt_fcn,param,opt);

For information on troubleshooting problems related to estimation using parallel

computing, see “Troubleshooting” on page 2-65.

Troubleshooting
• “Why Are the Estimation Results With and Without Parallel Computing Different?”
on page 2-65
• “Why Didn’t the Estimation Speed up Using Parallel Computing?” on page 2-65
• “Why Doesn’t the Estimation Using Parallel Computing Make Any Progress?” on page
2-66
• “Why Does the Estimation Using Parallel Computing Continue When I Click Stop?”
on page 2-66

Why Are the Estimation Results With and Without Parallel Computing Different?

• Different numerical precision on the client and worker machines can produce
marginally different simulation results. Thus, the optimization method can take a
different solution path and produce a different result.
• When you use parallel computing with the Pattern search method, the search
is more comprehensive and can result in a different solution. To learn more, see
“Parallel Computing with the Pattern search Method” on page 2-58.

Why Didn’t the Estimation Speed up Using Parallel Computing?

• When you estimate a few parameters or when the model does not take long to
simulate, you do not see a speedup in the estimation time. In such cases, the overhead

2-65
2 Parameter Estimation

associated with creating and distributing the parallel tasks outweighs the benefits of
running the estimation in parallel.
• Using the Pattern search method with parallel computing might not speed up the
optimization time. Without parallel computing, the method stops the search at each
iteration as soon as it finds a solution better than the current solution. The candidate
solution search is more comprehensive when you use parallel computing. Although
the number of iterations might be larger, the optimization without using parallel
computing might be faster.

To learn more about the expected speedup, see “Parallel Computing with the Pattern
search Method” on page 2-58.

Why Doesn’t the Estimation Using Parallel Computing Make Any Progress?

To troubleshoot the problem:

1 Run the optimization for a few iterations without parallel computing to see if the
optimization progresses.
2 Check whether the remote workers have access to all model dependencies. Model
dependencies include data variables and files required by the model to run.

To learn more, see “Model Dependencies” on page 2-60.

Why Does the Estimation Using Parallel Computing Continue When I Click Stop?

When you use parallel computing with the Pattern search method, the software must
wait until the current optimization iteration completes before it notifies the workers
to stop. The optimization does not terminate immediately when you click Stop, and,
instead, appears to continue running.

See Also
sdo.OptimizeOptions | parpool | sdo.getModelDependencies | sdo.optimize

More About
• “Speed Up Parameter Estimation Using Parallel Computing” on page 2-56
• “Ways to Speed Up Design Optimization Tasks”

2-66
 Use Fast Restart Mode During Parameter Estimation

Use Fast Restart Mode During Parameter Estimation

In this section...
“Parameter Estimation Tool Workflow for Fast Restart” on page 2-67
“Command-Line Workflow for Fast Restart” on page 2-68
“Troubleshooting” on page 2-69

This topic shows how to speed up parameter estimation using Simulink fast restart. You
can use the fast restart feature to speed up parameter estimation of tunable parameters
of a model.

Fast restart enables you to perform iterative simulations without compiling a model or
terminating the simulation each time. Using fast restart, you compile a model only once.
You can then tune parameters and simulate the model again without spending time on
compiling. Fast restart associates multiple simulation phases with a single compile phase
to make iterative simulations more efficient. You see a speedup of design optimization
tasks using fast restart in models that have a long compilation phase. See “How Fast
Restart Improves Iterative Simulations” in the Simulink documentation.

When you enable fast restart, you can only change tunable properties of the model during
simulation. For more information about the limitations, see “Factors Affecting Fast
Restart”.

You can perform parameter estimation using fast restart in the Parameter Estimation
tool or at the command line.

Parameter Estimation Tool Workflow for Fast Restart

To estimate model parameters using fast restart in the Parameter Estimation tool:

1 Open the Simulink model.

2 Enable fast restart in the model.

Click Fast Restart in the model window.

3 Open the Parameter Estimation tool for the model.
4 Configure the estimation data, estimation parameters and states, and, optionally,
estimation settings.

2-67
2 Parameter Estimation

For more information, see “Specify Estimation Data” on page 2-4, “Specify
Parameters for Estimation” on page 2-8, and “Optimization Options” on page 2-32.
5 In the Parameter Estimation tab, click Estimate to estimate the model
parameters in fast restart mode.
6 Disable fast restart.

In the model window, click Fast Restart .

Command-Line Workflow for Fast Restart

To estimate model parameters using fast restart at the command line:

1 Open the Simulink model.

2 Specify the model parameter values, params, to estimate. For an example, see
“Estimate Model Parameter Values (Code)” on page 2-92.
3 Create an experiment object, Exp.
Exp = sdo.Experiment('model');

Store the measured input/output data in Exp. For example, see “Estimate Model
Parameter Values (Code)” on page 2-92.
4 Create a model simulator from the experiment.
Simulator = createSimulator(Exp);

Simulator is an sdo.SimulationTest object.

Note: You must create a simulation scenario with logging information before
configuring the model for fast restart. You cannot modify logging information once
the model has been compiled for fast restart.
5 Configure the simulator and model for fast restart.
Simulator = fastRestart(Simulator,'on');
6 Create an estimation cost function, myCostfcn, and pass Simulator to the cost
function as an input. For more information, see “Write a Cost Function” on page
2-83. In the cost function, the simulator configured for fast restart is used to
update the model parameters, simulate the model, and log signals.

2-68
 Use Fast Restart Mode During Parameter Estimation

Use an anonymous function with one argument that calls myCostfcn.

estimfcn = @(param) myCostfcn(param,Simulator,Exp);

7 Run the estimation.
[param_opt,opt_info] = sdo.optimize(estimfcn,param);
8 Restore the simulator fast restart settings.
Simulator = fastRestart(Simulator,'off');

Troubleshooting
Why Don’t I See the Estimation Speedup I Expected Using Fast Restart?

You see a speedup of design optimization tasks using fast restart in models that have a
long compilation phase. If the compilation phase of your model is not long, you do not see
a significant change in estimation speed.

See Also
fastRestart | sdo.Experiment | sdo.optimize

Related Examples
• “Improving Optimization Performance using Fast Restart (GUI)” on page 2-198
• “Improving Optimization Performance using Fast Restart (Code)” on page 2-206

More About
• “Ways to Speed Up Design Optimization Tasks”

2-69
2 Parameter Estimation

Estimating Initial Conditions for Blocks with External Initial

Conditions
When an integrator block uses an initial-condition port, which you specify by an IC block,
you cannot estimate the initial conditions (ICs) of the integrator using Simulink Design
Optimization software. Estimation is not possible because external ICs have priority over
the ICs of a specific block to maintain the integrity of the model.

To tune the ICs of an integrator block with external ICs, you must modify the model to
make the external signal into a tunable parameter. For example, you can set the IC block
that feeds into the integrator to be a tunable variable and estimate it.

2-70
 Estimation Sessions

Estimation Sessions
In this section...
“Structure of an Estimation Session” on page 2-71
“Save Parameter Estimation Tool Sessions” on page 2-71
“Load Parameter Estimation Tool Sessions” on page 2-72
“Load Legacy Projects” on page 2-72

Structure of an Estimation Session

The Parameter Estimation tool, which is a tool for performing parameter estimation and
validation, stores and organizes data from a given Simulink model inside a session. To
open the Parameter Estimation tool, select Analysis > Parameter Estimation in the
Simulink model window.

When using the Parameter Estimation tool, you can:

• Manage estimation sessions

• Select parameters and initial conditions to configure the estimation
• Specify cost functions
• Import signal data (to be matched by the input and output of your Simulink model)
• Specify the initial conditions of your model

Each estimation session can include:

• One or more estimation or validation experiments

• Parameter information
• Different settings or configurations for each experiment

The default session name is the same as the Simulink model name. The session name is
shown on the title pane of Parameter Estimation tool.

Save Parameter Estimation Tool Sessions

To save your session as a MAT-file, click Save Session drop down list on the
Parameter Estimation tab.

2-71
2 Parameter Estimation

You have the options to browse to the location where you want to save the session using
the Save to file option, or save to model workspace or MATLAB workspace.

Load Parameter Estimation Tool Sessions

To open previously saved sessions, click the Open Session drop-down menu.

You have the option to browse to where the session file is, or open a session from model
workspace or the MATLAB workspace. All sessions are MAT-files.

Load Legacy Projects

You can open legacy projects that are in MAT-files by selecting Open from file
from the Open Session drop-down list. The Parameter Estimation tool recognizes and
converts them into the new session format.

2-72
 How the Software Formulates Parameter Estimation as an Optimization Problem

How the Software Formulates Parameter Estimation as an

Optimization Problem
In this section...
“Overview of Parameter Estimation as an Optimization Problem” on page 2-73
“Cost Function” on page 2-73
“Bounds and Constraints” on page 2-75
“Optimization Methods and Problem Formulations” on page 2-76

Overview of Parameter Estimation as an Optimization Problem

When you perform parameter estimation, the software formulates an optimization
problem. The optimization problem solution is the estimated parameter values set. This
optimization problem consists of:

• x — Design variables. The model parameters and initial states to be estimated.

• F(x) — Objective function. A function that calculates a measure of the difference
between the simulated and measured responses. Also called cost function or
estimation error.
•
(Optional) — Bounds. Limits on the estimated parameter values.
x£ x£x

• (Optional) C(x) — Constraint function. A function that specifies restrictions on the

design variables.

The optimization solver tunes the values of the design variables to satisfy the specified
objectives and constraints. The exact formulation of the optimization depends on the
optimization method that you use.

Cost Function
• “Types” on page 2-74
• “Time Base” on page 2-74

The software tunes the model parameters to obtain a simulated response (ysim) that
tracks the measured response or reference signal (yref). To do so, the solver minimizes the
cost function or estimation error, a measure of the difference between the simulated and
measured responses. The cost function, F(x), is the objective function of the optimization
problem.

2-73
2 Parameter Estimation

Types
The raw estimation error, e(t), is defined as:

e( t) = yref ( t) - ysim (t)

e(t) is also referred to as the error residuals or, simply, residuals.

Simulink Design Optimization software provides you the following cost functions to
process e(t):
Cost Function Formulation Option Name in GUI or
Command Line
Sum squared error (default) tN 'SSE'
F( x) = Â e(t) ¥ e(t)
t =0

N is the number of samples.

Sum absolute error tN 'SAE'
F( x) = Â |e(t)|
t =0

N is the number of samples.

Raw error È e(0) ˘ 'Residuals'
F( x) = ÍÍ M ˙˙ This option is available only
ÍÎ e( N ) ˙˚ at the command line.

N is the number of samples.

Custom function N/A This option is available only
at the command line.

Time Base
The software evaluates the cost function for a specific time interval. This interval is
dependent on the measured signal time base and the simulated signal time base.

• The measured signal time base consists of all the time points for which the measured
signal is specified. In case of multiple measured signals, this time base is the union of
the time points of all the measured signals.

2-74
 How the Software Formulates Parameter Estimation as an Optimization Problem

• The simulated signal time base consists of all the time points for which the model is
simulated.

If the model uses a variable-step solver, then the simulated signal time base can change
from one optimization iteration to another. The simulated and measured signal time
bases can be different. The software evaluates the cost function for only the time interval
that is common to both. By default, the software uses only the time points specified by
the measured signal in the common time interval.

• In the GUI, you can specify the simulation start and stop times in the Simulation
time area of the Simulation Options dialog box.
• At the command line, the software specifies the simulation stop time as the last
point of the measured signal time base. For example, the following code simulates
the model until the end time of the longest running output signal of exp, an
sdo.Experiment object:
sim_obj = createSimulator(exp);
sim_obj = sim(sim_obj);

sim_obj contains the simulated response for the model associated with exp.

Bounds and Constraints

You can specify bounds for the design variables (estimated model parameters), based on
your knowledge of the system. Bounds are expressed as:

x£ x£x

x and x are the lower and upper bounds for the design variables.

For example, in a battery discharging experiment, the estimated battery initial charge
must be greater than zero and less than Inf. These bounds are expressed as:

0< x<•

For an example of how to specify these types of bounds, see “Estimate Model Parameters
and Initial States (Code)” on page 2-104.

You can also specify other constraints, C(x), on the design variables at the command line.
C(x) can be linear or nonlinear and can describe equalities or inequalities. C(x) can also
specify multiparameter constraints. For example, for a simple friction model, C(x) can

2-75
2 Parameter Estimation

specify that the static friction coefficient must be greater than or equal to the dynamic
friction coefficient. One way of expressing this constraint is:

C( x) : x1 - x2
C( x) £ 0

x1 and x2 are the dynamic and static friction coefficients, respectively.

For an example of how to specify a constraint, see “Estimate Model Parameters with
Parameter Constraints (Code)” on page 2-142.

Optimization Methods and Problem Formulations

An optimization problem can be one of the following types:

• Minimization problem — Minimizes an objective function, F(x). You specify the

measured signal that you want the model output to track. You can optionally specify
bounds for the estimated parameters.
• Mixed minimization and feasibility problem — Minimizes an objective function, F(x),
subject to specified bounds and constraints,C(x). You specify the measured signal
that you want the model to track and bounds and constraints for the estimated
parameters.
• Feasibility problem — Finds a solution that satisfies the specified constraints, C(x).
You specify only bounds and constraints for the estimated parameters. This type of
problem is not common in parameter estimation.

The optimization method that you specify determines the formulation of the estimation
problem. The software provides the following optimization methods:

Optimization Method Description Optimization Problem Formulation

Name
• User interface: Minimizes the squares Minimization Problem
Nonlinear Least of the residuals,
Squares recommended 2
(
min F( x) 2 = min f1 ( x) 2 + f2 ( x)2 + …+f n ( x) 2 )
• Command line: method for parameter x x
'lsqnonlin' estimation. s.t. x £ x £ x

This method requires

f1(x), f2(x),...,fn(x) represent residuals. n is the
a vector of error
number of samples.

2-76
 How the Software Formulates Parameter Estimation as an Optimization Problem

Optimization Method Description Optimization Problem Formulation

Name
residuals, computed Mixed Minimization and Feasibility Problem
using a fixed time
base. Do not use this Not supported.
approach if you have a
scalar cost function or Feasibility Problem
if the number of error
Not supported.
residuals can change
from one iteration to
another.

This method uses the

Optimization Toolbox
function, lsqnonlin.

2-77
2 Parameter Estimation

Optimization Method Description Optimization Problem Formulation

Name
• User interface: General nonlinear Minimization Problem
Gradient Descent solver, uses the cost
• Command line: function gradient. min F( x)
x
'fmincon'
Use this approach if s.t. x £ x £ x
you want to specify one
or any combination of Mixed Minimization and Feasibility Problem
the following:
min F( x)
• Custom cost x
functions s.t. C( x) £ 0
• Parameter-based x£x£x
constraints
• Signal-based
constraints Note: When tracking a reference signal, the
software ignores the maximally feasible solution
This method uses the option.
Optimization Toolbox
function, fmincon.
Feasibility Problem
For information on
how the gradient • If you select the maximally feasible solution
is computed, option (i.e., the optimization continues
see “Gradient after an initial feasible solution is found),
Computations” on page the software uses the following problem
2-90. formulation:

min g
[ x,g ]
s.t. C( x) £ g
x£ x£ x
g £0

γ is a slack variable that permits a feasible

solution with C(x) ≤ γ rather than C(x) ≤ 0.

2-78
 How the Software Formulates Parameter Estimation as an Optimization Problem

Optimization Method Description Optimization Problem Formulation

Name
• If you do not select the maximally feasible
solution option (i.e., the optimization
terminates as soon as a feasible solution
is found), the software uses the following
problem formulation:

min 0
x
s.t. C( x) £ 0
x£ x£ x

2-79
2 Parameter Estimation

Optimization Method Description Optimization Problem Formulation

Name
• User interface: Based on the Nelder- Minimization Problem
Simplex Search Mead algorithm, this
• Command line: approach does not min F( x)
'fminsearch' use the cost function x
gradient.
Mixed Minimization and Feasibility Problem
Use this approach
if your cost function The software formulates the problem in two
or constraints are steps:
not continuous or
1 Finds a feasible solution.
differentiable.

This method uses the min max ( C ( x ) )

Optimization Toolbox x

functions, fminsearch 2 Minimizes the objective. The software uses

and fminbnd. fminbnd the results from step 1 as initial guesses.
is used if one scalar It maintains feasibility by introducing a
parameter is being discontinuous barrier in the optimization
optimized. Otherwise, objective.
fminsearch is used.
You cannot specify min G ( x)
parameter bounds, x
x £ x £ x , with where
fminsearch. Ï• if max ( C( x) ) > 0
G( x) = Ì
Ó F( x) otherwise.

Feasibility Problem

min max ( C ( x ) )
x

2-80
 How the Software Formulates Parameter Estimation as an Optimization Problem

Optimization Method Description Optimization Problem Formulation

Name
• User interface: Direct search Minimization Problem
Pattern Search method, based on the
• Command line: generalized pattern min F( x)
'patternsearch' search algorithm,
x
this method does not s.t. x£ x£ x
use the cost function
gradient. Mixed Minimization and Feasibility Problem

Use this approach The software formulates the problem in two

if your cost function steps:
or constraints are
not continuous or 1 Finds a feasible solution.
differentiable.
min max ( C ( x ) )
This method uses the x
Global Optimization s.t. x£ x£ x
Toolbox function,
2 Minimizes the objective. The software uses
patternsearch.
the results from step 1 as initial guesses.
It maintains feasibility by introducing a
discontinuous barrier in the optimization
objective.

min G( x)
x
s.t. x £ x £ x
where
Ï• if max ( C( x) ) > 0
G( x) = Ì
Ó F ( x) otherwise.

Feasibility Problem

min max ( C ( x ) )
x
s.t. x£ x£ x

2-81
2 Parameter Estimation

See Also
fminbnd | fmincon | fminsearch | lsqnonlin | patternsearch | sdo.Experiment
| sdo.requirements.SignalTracking | sdo.requirements.SignalTracking |
sdo.SimulationTest

Related Examples
• “Estimate Model Parameter Values (Code)” on page 2-92
• “Estimate Model Parameters with Parameter Constraints (Code)” on page 2-142
• “Estimate Parameters from Measured Data”

More About
• “Write a Cost Function” on page 2-83

2-82
 Write a Cost Function

Write a Cost Function

In this section...
“Cost Function Overview” on page 2-83
“Convenience Objects” on page 2-84
“Inputs” on page 2-85
“Evaluate Requirements” on page 2-86
“Outputs” on page 2-87

Cost Function Overview

When you use sdo.optimize to optimize model parameters (design variables), you must
provide a MATLAB function as an input to sdo.optimize. This function, also called
a cost function, must evaluate the cost and constraint values for the design variable
values for an iteration. (The cost and constraint functions are collectively referred to as
requirements.) sdo.optimize calls this function at every optimization iteration and uses
the function output to decide the optimization direction.

The cost function can also be used for global sensitivity analysis. You generate
samples of the model parameters and evaluate the cost function for each sample using
sdo.evaluate.

The cost function must have:

• Input — params , a vector of the design variables (param.Continuous objects) to be

optimized.
• Output:

• (Required) vals , a structure with one or more fields that specify the values of the
cost and constraint violations.
• (Optional) derivs, a structure with one or more fields that specify the values of
the gradients of the cost and constraint violations.

You perform the following tasks within the cost function:

• Extract the current design variable values from params.

• If the simulated response is required for evaluating the requirements, simulate the
model using the current design variable values.

2-83
2 Parameter Estimation

• Evaluate the requirements.

• Specify the requirement values as fields of vals.

To use a cost function with sdo.optimize, enter:

[param_opt,opt_info] = sdo.optimize(@myCostFunc,param)

Here, myCostFunc is the name of the MATLAB function and param is a vector of the
design variables.

Similarly, to use a cost function with sdo.evaluate, enter:

[y,info] = sdo.evaluate(@myCostFunc,param)

Convenience Objects
The software provides you with the following convenience objects that can you can use in
the cost function:

Class Name Description

sdo.SimulationTest Use an sdo.SimulationTest object, also
referred to as a simulator, to simulate a model.
The simulator allows you to simulate the model
using alternative inputs, model parameter
and initial-state values, without modifying the
model.

You configure the simulator to log the signals

needed to evaluate requirements and use the
sim method to simulate the model. Then, you
extract the model response from the object and
evaluate the requirements.
Requirements objects: Use these requirements objects to specify time-
and frequency-domain costs or constraints on
Time-domain requirements: the design variables.
• sdo.requirements.SignalBound You configure the properties of the object and
• sdo.requirements.StepResponseEnvelope then use the object’s evalRequirement method
• sdo.requirements.SignalTracking to evaluate how closely the current design
variables satisfy your design requirement.
Frequency-domain requirements:

2-84
 Write a Cost Function

Class Name Description

• sdo.requirements.GainPhaseMargin
• sdo.requirements.BodeMagnitude
• sdo.requirements.ClosedLoopPeakGain
• sdo.requirements.PZDampingRatio
• sdo.requirements.PZNaturalFrequency
• sdo.requirements.PZSettlingTime
• sdo.requirements.SignalTracking
• sdo.requirements.StepResponseEnvelope
• sdo.requirements.OpenLoopGainPhase
sdo.Experiment Use an sdo.Experiment object, also referred
to as simply an experiment, to specify the input/
output data, model parameter and initial-state
values for parameter estimation.

You update the design variable values

associated with the experiment using the
setEstimatedValues method. Then, you
create a simulator, using the createSimulator
method, to simulate the model using the
updated model configuration.

Inputs
• “Model Parameters and States” on page 2-85
• “Multiple Inputs” on page 2-86

Model Parameters and States

The function must take as input a vector of model parameter objects

(param.Continuous objects) and, optionally, initial-state objects (param.State
objects). These objects represent the design variables of the optimization problem.
You obtain these objects by using the sdo.getParameterFromModel and
sdo.getStateFromModel commands.

To access a design variable value, use:

2-85
2 Parameter Estimation

param_val = p(1).Value;

Here, p is a vector of param.Continuous objects and p(1) is either a model parameter

or an initial-state object.

Multiple Inputs

sdo.optimize requires that the cost function accept only one input argument, params.
However, you might want your cost function to use additional inputs. For instance,
you could make the model name an input argument and configure the function to
be used for multiple models. To call sdo.optimize and use a function that accepts
more than one input argument, you use an anonymous function. For example, suppose
myCostFunc_mult_inputs is a cost function that takes param, arg1, and arg2 as
inputs. Then, assuming that all input arguments are variables in the workspace, you
enter:

myCostFunc = @(param) myCostFunc_mult_inputs(param,arg1,arg2);

[param_opt,opt_info] = sdo.optimize(@myCostFunc,param);

Additional inputs can also help reduce code redundancy and computation cost, given that
the function is called repeatedly by sdo.optimize during optimization. If you use a
convenience object in your function, you can create it once, before calling sdo.optimize.
For example, you can create a simulator (sdo.SimulationTest object) to simulate your
model, and pass it to your cost function.

simulator = sdo.SimulationTest(model)
myCostFunc = @(param) myCostFunc_mult_inputs(param,arg1,arg2,simulator);
[param_opt,opt_info] = sdo.optimize(@myCostFunc,param);

Note: To perform estimation, optimization, or evaluation using Simulink fast restart at

the command line , it is necessary to create the simulator before the cost function, and
then pass the simulator to the cost function.

Evaluate Requirements
The core of the cost function is where you evaluate how well the current design variables
satisfy the design requirements. You can use MATLAB functions to do so. You can also
use the requirements objects that the Simulink Design Optimization software provides.
These objects enable you to specify requirements such as step-response characteristics,
gain/phase margin bounds, Bode magnitude bounds, etc.

2-86
 Write a Cost Function

• Parameter-only requirements — Extract the design variable values and compute the
requirement values.

For example, you can minimize the cylinder cross-sectional area, a design variable, in
a hydraulic cylinder. See “Design Optimization to Meet a Custom Objective (Code)” on
page 3-126.
• Model signal-based requirements — Simulate the model using the current design
variable values, extract the model response, and compute the requirement values.

There are multiple ways to simulate the model, including:

• Using an sdo.SimulationTest object. You update the model parameter values using
the simulator’s Parameters property. Then, you use the sim method to simulate
the model and extract the logged signals from the simulator that are of interest.
For an example, see “Design Optimization to Meet a Custom Objective (Code)” on
page 3-126.

In parameter estimation, you can use the createSimulator method of the

sdo.Experiment object to create the simulator. Before creating the simulator,
you update the experiment with the current design variable values using the
setEstimatedValues method. For an example, see “Estimate Model Parameters
Per Experiment (Code)” on page 2-128
• Using sdo.setValueInModel to update the model and then calling sim to
simulate the model.
• Linear model-based requirements — Update the model with the current design
variable values, linearize the model, and compute the requirement values.

Use sdo.setValueInModel to update the model. To linearize the model use

functions such as linmod, linearize (requires a Simulink Control Design™), or the
SystemLoggingInfo property of sdo.SimulationTest.

Note: In fast restart mode, you cannot use the linearize command. Use the
SystemLoggingInfo property of the sdo.SimulationTest object, to specify linear
systems to log when simulating the model. For an example, see “Design Optimization
to Meet Frequency-Domain Requirements (Code)” on page 3-252.

Outputs
• “Cost and Constraint Values” on page 2-88

2-87
2 Parameter Estimation

• “Multiple Objectives” on page 2-88

Cost and Constraint Values

Your cost function must return a structure containing the cost and constraint values
for the current design variables. This structure must have one or more of the following
fields, as required by your optimization problem:

• F — Cost value.
• Cleq, Ceq — Nonlinear constraint values. The solver satisfies Cleq ≤ 0 and Ceq = 0.
• leq, eq — Linear constraint values. The solver satisfies leq ≤ 0 and eq = 0.

If you have multiple constraints of one type, concatenate the values into a vector,
and specify this vector as the corresponding field value. For instance, if you have a
hydraulic cylinder, you can specify nonlinear inequality constraints on the piston position
(Cleq1) and cylinder pressure (Cleq2). In this case, specify the Cleq field of the output
structure, vals, as:

vals.Cleq = [Cleq1; Cleq2];

For an example, see “Design Optimization to Meet a Custom Objective (Code)” on page
3-126.

By default, the software computes the cost and constraint gradients using numeric
perturbation. However, you can specify the gradients and return them as an additional
output. This output must be a structure with one or more of the following fields, as
required by your optimization problem:

• F — Cost derivatives.
• Cleq — Nonlinear inequality constraints derivatives.
• Ceq — Nonlinear equality constraints derivatives.

You must also set the GradFcn property of the optimization option set to 'on'.

Multiple Objectives

Simulink Design Optimization does not support multi-objective optimization. However,

you can return the cost value (F) as a vector, representing the multiple objective values.
Using this approach does not halt the optimization. Instead, the software sums the
elements of the vector and minimizes this sum. The exception to this behavior is if you
are using the nonlinear least squares (lsqnonlin) optimization method. The nonlinear

2-88
 Write a Cost Function

least squares method, used for parameter estimation, requires that you return the error
residuals as a vector. In this case, the software minimizes the sum square of this vector.

If you are tracking multiple signals and using lsqnonlin, then you must concatenate
the error residuals for the different signals into one vector. Specify this vector as the F
field value.

For an example of single objective optimization using the gradient descent method, see
“Design Optimization to Meet a Custom Objective (Code)” on page 3-126.

For an example of multiple objective optimization using the nonlinear least squares
method, see “Estimate Model Parameters Per Experiment (Code)” on page 2-128.

See Also
sdo.OptimizeOptions | param.Continuous | sdo.SimulationTest | sdo.Experiment |
sdo.evaluate | sdo.optimize | sdo.setValueInModel

Related Examples
• “Design Optimization to Meet a Custom Objective (Code)” on page 3-126
• “Estimate Model Parameter Values (Code)” on page 2-92

More About
• “How the Optimization Algorithm Formulates Minimization Problems” on page
3-3
• “How the Software Formulates Parameter Estimation as an Optimization Problem”
on page 2-73

2-89
2 Parameter Estimation

Gradient Computations
For the Gradient descent (fmincon) optimization solver, the gradients are computed
using numerical perturbation:

Ê 1 ˆ
dx = 3 eps ¥ max Á| x|, xtypical ˜
Ë 10 ¯
dL = max ( x - dx, xmin )
dR = min ( x + d x, xmax )
FL = opt _ fcn(dL)
FR = opt _ fcn(dR)
dF ( FL - FR )
=
dx ( dL - dR )

• x is a scalar design variable.

• xmin is the lower bound of x.
• xmax is the upper bound of x.
• xtypical is the scaled value of x.
• opt_fcn is the objective function.

dx is relatively large to accommodate simulation solver tolerances.

If you want to compute the gradients in any other way, you can do so in the cost function
you write for performing design optimization programmatically. See sdo.optimize and
GradFcn of sdo.OptimizeOptions for more information.

See Also
fmincon

More About
• “How the Software Formulates Parameter Estimation as an Optimization Problem”
on page 2-73

2-90
 Gradient Computations

• “How the Optimization Algorithm Formulates Minimization Problems” on page

3-3

2-91
2 Parameter Estimation

Estimate Model Parameter Values (Code)

This example shows how to use experimental data to estimate model parameter values.

Aircraft Model

The Simulink model, sdoAircraftEstimation, models the longitudinal flight control

system of an aircraft.

open_system('sdoAircraftEstimation')

Estimation Problem

You use measured data to estimate the aircraft model parameters and states.

2-92
 Estimate Model Parameter Values (Code)

Measured output data:

• Pilot G force, output of the Pilot G-force calculation block

• Angle of attack, fourth output of the Aircraft Dynamics Model block

Parameters:

• Actuator time constant, Ta, used by the Actuator Model block

• Vertical velocity, Zd, used by the Aircraft Dynamics Model block
• Pitch rate gains, Md, used by the Aircraft Dynamics Model block

State:

• Initial state of the first-order actuator model, sdoAircraftEstimation/Actuator

Model

Define the Estimation Experiment

Get the measured data.

[time,iodata] = sdoAircraftEstimation_Experiment;

The sdoAircraftEstimation_Experiment function returns the measured output

data, iodata, and the corresponding time vector. The first column of iodata is the pilot
G force and the second column is the angle of attack.

To see the code for this function, type edit sdoAircraftEstimation_Experiment.

Create an experiment object to store the measured input/output data.

Exp = sdo.Experiment('sdoAircraftEstimation');

Create an object to store the measured pilot G-Force output.

PilotG = Simulink.SimulationData.Signal;
PilotG.Name = 'PilotG';
PilotG.BlockPath = 'sdoAircraftEstimation/Pilot G-force calculation';
PilotG.PortType = 'outport';
PilotG.PortIndex = 1;
PilotG.Values = timeseries(iodata(:,2),time);

Create an object to store the measured angle of attack (alpha) output.

2-93
2 Parameter Estimation

AoA = Simulink.SimulationData.Signal;
AoA.Name = 'AngleOfAttack';
AoA.BlockPath = 'sdoAircraftEstimation/Aircraft Dynamics Model';
AoA.PortType = 'outport';
AoA.PortIndex = 4;
AoA.Values = timeseries(iodata(:,1),time);

Add the measured pilot G-Force and angle of attack data to the experiment as the
expected output data.

Exp.OutputData = [...
PilotG; ...
AoA];

Add the initial state for the Actuator Model block to the experiment. Set its Free field
to true so that it is estimated.

Exp.InitialStates = sdo.getStateFromModel('sdoAircraftEstimation','Actuator Model');

Exp.InitialStates.Minimum = 0;
Exp.InitialStates.Free = true;

Compare the Measured Output and the Initial Simulated Output

Create a simulation scenario using the experiment and obtain the simulated output.

Simulator = createSimulator(Exp);
Simulator = sim(Simulator);

Search for the pilot G-Force and angle of attack signals in the logged simulation data.

SimLog = find(Simulator.LoggedData,get_param('sdoAircraftEstimation','SignalLoggi
PilotGSignal = find(SimLog,'PilotG');
AoASignal = find(SimLog,'AngleOfAttack');

Plot the measured and simulated data.

As expected, the model response does not match the experimental output data.

plot(time, iodata, ...

AoASignal.Values.Time,AoASignal.Values.Data,'--', ...
PilotGSignal.Values.Time,PilotGSignal.Values.Data,'-.');
title('Simulated and Measured Responses Before Estimation')
legend('Measured angle of attack', 'Measured pilot g force', ...
'Simulated angle of attack', 'Simulated pilot g force');

2-94
 Estimate Model Parameter Values (Code)

Specify the Parameters to Estimate

Select the model parameters that describe the flight control actuation system. Specify
bounds for the estimated parameter values based on our understanding of the actuation
system.

p = sdo.getParameterFromModel('sdoAircraftEstimation',{'Ta','Md','Zd'});
p(1).Minimum = 0.01; %Ta
p(1).Maximum = 1;
p(2).Minimum = -10; %Md
p(2).Maximum = 0;
p(3).Minimum = -100; %Zd
p(3).Maximum = 0;

2-95
2 Parameter Estimation

Get the actuator initial state value that is to be estimated from the experiment.

s = getValuesToEstimate(Exp);

Group the model parameters and initial states to be estimated together.

v = [p;s]

v(1,1) =

Name: 'Ta'
Value: 0.5000
Minimum: 0.0100
Maximum: 1
Free: 1
Scale: 0.5000
Info: [1x1 struct]

v(2,1) =

Name: 'Md'
Value: -1
Minimum: -10
Maximum: 0
Free: 1
Scale: 1
Info: [1x1 struct]

v(3,1) =

Name: 'Zd'
Value: -80
Minimum: -100
Maximum: 0
Free: 1
Scale: 128
Info: [1x1 struct]

v(4,1) =

Name: 'sdoAircraftEstimation/Actuator...'

2-96
 Estimate Model Parameter Values (Code)

Value: 0
Minimum: 0
Maximum: Inf
Free: 1
Scale: 1
dxValue: 0
dxFree: 1
Info: [1x1 struct]

4x1 param.Continuous

Define the Estimation Objective Function

Create an estimation objective function to evaluate how closely the simulation output,
generated using the estimated parameter values, matches the measured data.

Use an anonymous function with one input argument that calls the
sdoAircraftEstimation_Objective function. We pass the anonymous function to
sdo.optimize, which evaluates the function at each optimization iteration.
estFcn = @(v) sdoAircraftEstimation_Objective(v,Simulator,Exp);

The sdoAircraftEstimation_Objective function:

• Has one input argument that specifies the actuator parameter values and the
actuator initial state.

• Has one input argument that specifies the experiment object containing the measured
data.

• Returns a vector of errors between simulated and experimental outputs.

The sdoAircraftEstimation_Objective function requires two inputs, but

sdo.optimize requires a function with one input argument. To work around
this, estFcn is an anonymous function with one input argument, v, but it calls
sdoAircraftEstimation_Objective using two input arguments, v and Exp.

For more information regarding anonymous functions, see "Anonymous Functions".

The sdo.optimize command minimizes the return argument of the

anonymous function estFcn, that is, the residual errors returned by
sdoAircraftEstimation_Objective. For more details on how to write an

2-97
2 Parameter Estimation

objective/constraint function to use with the sdo.optimize command, type help

sdoExampleCostFunction at the MATLAB command prompt.

To examine the estimation objective function in more detail, type edit

sdoAircraftEstimation_Objective at the MATLAB command prompt.
type sdoAircraftEstimation_Objective

function vals = sdoAircraftEstimation_Objective(v,Simulator,Exp)

%SDOAIRCRAFTESTIMATION_OBJECTIVE
%
% The sdoAircraftEstimation_Objective function is used to compare model
% outputs against experimental data.
%
% vals = sdoAircraftEstimation_Objective(v,Exp)
%
% The |v| input argument is a vector of estimated model parameter values
% and initial states.
%
% The |Simulator| input argument is a simulation object used
% simulate the model with the estimated parameter values.
%
% The |Exp| input argument contains the estimation experiment data.
%
% The |vals| return argument contains information about how well the
% model simulation results match the experimental data and is used by
% the |sdo.optimize| function to estimate the model parameters.
%
% See also sdo.optimize, sdoExampleCostFunction,
% sdoAircraftEstimation_cmddemo
%

% Copyright 2012-2015 The MathWorks, Inc.

%%
% Define a signal tracking requirement to compute how well the model output
% matches the experiment data. Configure the tracking requirement so that
% it returns the tracking error residuals (rather than the
% sum-squared-error) and does not normalize the errors.
%
r = sdo.requirements.SignalTracking;
r.Type = '==';
r.Method = 'Residuals';
r.Normalize = 'off';

2-98
 Estimate Model Parameter Values (Code)

%%
% Update the experiments with the estimated parameter values.
%
Exp = setEstimatedValues(Exp,v);

%%
% Simulate the model and compare model outputs with measured experiment
% data.
%
Simulator = createSimulator(Exp,Simulator);
Simulator = sim(Simulator);

SimLog = find(Simulator.LoggedData,get_param('sdoAircraftEstimation','SignalLoggi
PilotGSignal = find(SimLog,'PilotG');
AoASignal = find(SimLog,'AngleOfAttack');

PilotGError = evalRequirement(r,PilotGSignal.Values,Exp.OutputData(1).Values);
AoAError = evalRequirement(r,AoASignal.Values,Exp.OutputData(2).Values);

%%
% Return the residual errors to the optimization solver.
%
vals.F = [PilotGError(:); AoAError(:)];
end

Estimate the Parameters

Use the sdo.optimize function to estimate the actuator parameter values and initial
state.

Specify the optimization options. The estimation function

sdoAircraftEstimation_Objective returns the error residuals between simulated
and experimental data and does not include any constraints, making this problem ideal
for the 'lsqnonlin' solver.
opt = sdo.OptimizeOptions;
opt.Method = 'lsqnonlin';

Estimate the parameters.

vOpt = sdo.optimize(estFcn,v,opt)

Optimization started 31-Jul-2015 06:07:12

2-99
2 Parameter Estimation

Step-size First-order
Iter F-count f(x) optimality
0 8 27972.2 1
1 17 10124.8 0.4744 5.69e+04
2 26 3127.92 0.3854 1.24e+04
3 35 872.751 0.4286 2.81e+03
4 44 238.66 0.5147 618
5 53 71.9182 0.493 147
6 62 17.2194 0.4163 44.9
7 71 1.82827 0.3069 11.4
8 80 0.0440753 0.1321 1.38
9 89 0.0020167 0.03129 0.0871
10 98 0.000244535 0.008145 0.119
11 107 5.74974e-05 0.005686 0.00522
Local minimum possible.

lsqnonlin stopped because the final change in the sum of squares relative to
its initial value is less than the selected value of the function tolerance.

vOpt(1,1) =

Name: 'Ta'
Value: 0.0500
Minimum: 0.0100
Maximum: 1
Free: 1
Scale: 0.5000
Info: [1x1 struct]

vOpt(2,1) =

Name: 'Md'
Value: -6.8848
Minimum: -10
Maximum: 0
Free: 1
Scale: 1
Info: [1x1 struct]

vOpt(3,1) =

2-100
 Estimate Model Parameter Values (Code)

Name: 'Zd'
Value: -63.9983
Minimum: -100
Maximum: 0
Free: 1
Scale: 128
Info: [1x1 struct]

vOpt(4,1) =

Name: 'sdoAircraftEstimation/Actuator...'
Value: 6.3839e-05
Minimum: 0
Maximum: Inf
Free: 1
Scale: 1
dxValue: 0
dxFree: 1
Info: [1x1 struct]

4x1 param.Continuous

Compare the Measured Output and the Final Simulated Output

Update the experiments with the estimated parameter values.

Exp = setEstimatedValues(Exp,vOpt);

Simulate the model using the updated experiment and compare the simulated output
with the experimental data.

The model response using the estimated parameter values closely matches the
experiment output data.

Simulator = createSimulator(Exp,Simulator);
Simulator = sim(Simulator);
SimLog = find(Simulator.LoggedData,get_param('sdoAircraftEstimation','SignalLoggi
PilotGSignal = find(SimLog,'PilotG');
AoASignal = find(SimLog,'AngleOfAttack');

plot(time, iodata, ...

2-101
2 Parameter Estimation

AoASignal.Values.Time,AoASignal.Values.Data,'-.', ...
PilotGSignal.Values.Time,PilotGSignal.Values.Data,'--')
title('Simulated and Measured Responses After Estimation')
legend('Measured angle of attack', 'Measured pilot g force', ...
'Simulated angle of attack', 'Simulated pilot g force');

Update the Model Parameter Values

Update the model with the estimated actuator parameter values. Do not update
the model actuator initial state (fourth element of vOpt) as it is dependent on the
experiment.

sdo.setValueInModel('sdoAircraftEstimation',vOpt(1:3));

2-102
 Estimate Model Parameter Values (Code)

Related Examples

To learn how to estimate model parameters using the Parameter Estimation Tool, see
"Estimate Model Parameter Values (GUI)".

Close the model.

bdclose('sdoAircraftEstimation')

2-103
2 Parameter Estimation

Estimate Model Parameters and Initial States (Code)

This example shows how to estimate the initial state and parameters of a model.

This example requires Simscape®.

RC Circuit Model

The Simulink model, sdoRCCircuit, models a simple resistor-capacitor (RC) circuit.

open_system('sdoRCCircuit');

Estimation Problem

You use the measured data to estimate the RC model parameter and state values.

Measured output data:

• Capacitor voltage, output of the PS-Simulink Converter block

Parameter:

• Capacitance, C1, used by the C1 block

State:

2-104
 Estimate Model Parameters and Initial States (Code)

• Initial voltage of the capacitor, C1

Define the Estimation Experiment

Get the measured data.

load sdoRCCircuit_ExperimentData

The variables time and data are loaded into the workspace, where data is the
measured capacitor voltage for times time.

Create an experiment object to store the experimental voltage data.

Exp = sdo.Experiment('sdoRCCircuit');

Create an object to store the measured capacitor voltage output.

Voltage = Simulink.SimulationData.Signal;
Voltage.Name = 'Voltage';
Voltage.BlockPath = 'sdoRCCircuit/PS-Simulink Converter';
Voltage.PortType = 'outport';
Voltage.PortIndex = 1;
Voltage.Values = timeseries(data,time);

Add the measured capacitor data to the experiment as the expected output data.
Exp.OutputData = Voltage;

Compare the Measured Output and the Initial Simulated Output

Create a simulation scenario using the experiment and obtain the simulated output.
Simulator = createSimulator(Exp);
Simulator = sim(Simulator);

Search for the voltage signal in the logged simulation data.

SimLog = find(Simulator.LoggedData,get_param('sdoRCCircuit','SignalLoggingName'))
Voltage = find(SimLog,'Voltage');

Plot the measured and simulated data.

The model response does not match the experimental output data.
plot(time,data,'ro',Voltage.Values.Time,Voltage.Values.Data,'b')
title('Simulated and Measured Responses Before Estimation')

2-105
2 Parameter Estimation

legend('Measured Voltage','Simulated Voltage')

Specify the Parameters to Estimate

Select the capacitance parameter from the model. Specify an initial guess for the
capacitance value (460 uF) and a minimum bound (0 F).
p = sdo.getParameterFromModel('sdoRCCircuit','C1');
p.Value = 460e-6;
p.Minimum = 0;

Define the Estimation Objective Function

Create an estimation objective function to evaluate how closely the simulation output,
generated using the estimated parameter value, matches the measured data.

2-106
 Estimate Model Parameters and Initial States (Code)

Use an anonymous function with one input argument that calls the
sdoRCCircuit_Objective function. We pass the anonymous function to
sdo.optimize, which evaluates the function at each optimization iteration.
estFcn = @(v) sdoRCCircuit_Objective(v,Simulator,Exp);

The sdoRCCircuit_Objective function:

• Has one input argument that specifies the estimated circuit capacitance value.

• Has one input argument that specifies the experiment object containing the measured
data.

• Returns a vector of errors between simulated and experimental outputs.

The sdoRCCircuit_Objective function requires two inputs, but sdo.optimize

requires a function with one input argument. To work around this, estFcn is an
anonymous function with one input argument, v, but it calls sdoRCCircuit_Objective
using two input arguments, v and Exp.

For more information regarding anonymous functions, see "Anonymous Functions".

The optimization solver minimizes the residual errors. For more details on how to write
an objective/constraint function to use with the sdo.optimize command, type help
sdoExampleCostFunction at the MATLAB command prompt.

To examine the estimation object function in more detail, type edit

sdoRCCircuit_Objective at the MATLAB command prompt.
type sdoRCCircuit_Objective

function vals = sdoRCCircuit_Objective(v,Simulator,Exp)

%SDORCCIRCUIT_OBJECTIVE
%
% The sdoRCCircuit_Objective function is used to compare model
% outputs against experimental data.
%
% vals = sdoRCCircuit_Objective(v,Exp)
%
% The |v| input argument is a vector of estimated model parameter values
% and initial states.
%
% The |Simulator| input argument is a simulation object used
% simulate the model with the estimated parameter values.

2-107
2 Parameter Estimation

%
% The |Exp| input argument contains the estimation experiment data.
%
% The |vals| return argument contains information about how well the
% model simulation results match the experimental data and is used by
% the |sdo.optimize| function to estimate the model parameters.
%
% See also sdo.optimize, sdoExampleCostFunction, sdoRCCircuit_cmddemo
%

% Copyright 2012-2015 The MathWorks, Inc.

%%
% Define a signal tracking requirement to compute how well the model output
% matches the experiment data. Configure the tracking requirement so that
% it returns the tracking error residuals (rather than the
% sum-squared-error) and does not normalize the errors.
%
r = sdo.requirements.SignalTracking;
r.Type = '==';
r.Method = 'Residuals';
r.Normalize = 'off';

%%
% Update the experiments with the estimated parameter values.
%
Exp = setEstimatedValues(Exp,v);

%%
% Simulate the model and compare model outputs with measured experiment
% data.
%
Simulator = createSimulator(Exp,Simulator);
Simulator = sim(Simulator);

SimLog = find(Simulator.LoggedData,get_param('sdoRCCircuit','SignalLoggingName'));
Voltage = find(SimLog,'Voltage');

VoltageError = evalRequirement(r,Voltage.Values,Exp.OutputData(1).Values);

%%
% Return the residual errors to the optimization solver.
%
vals.F = VoltageError(:);

2-108
 Estimate Model Parameters and Initial States (Code)

end

Estimate the Parameters

Use the sdo.optimize function to estimate the capacitance value.

Specify the optimization options. The estimation function sdoRCCircuit_Objective

returns the error residuals between simulated and experimental data and does not
include any constraints, making this problem ideal for the 'lsqnonlin' solver.

opt = sdo.OptimizeOptions;
opt.Method = 'lsqnonlin';

Estimate the parameters.

pOpt = sdo.optimize(estFcn,p,opt)

Optimization started 31-Jul-2015 06:13:01

Step-size First-order
Iter F-count f(x) optimality
0 3 55.0041 1
1 6 21.0161 0.2124 17.2
2 9 11.5085 0.1272 6.08
3 12 9.56468 0.06553 1.99
4 15 9.27666 0.02744 0.442
5 18 9.27666 0.00717 0.442
6 21 9.27131 0.001793 0.356
Local minimum possible.

lsqnonlin stopped because the final change in the sum of squares relative to
its initial value is less than the selected value of the function tolerance.

pOpt =

Name: 'C1'
Value: 1.1600e-04
Minimum: 0
Maximum: Inf
Free: 1
Scale: 0.0020
Info: [1x1 struct]

2-109
2 Parameter Estimation

1x1 param.Continuous

Compare the Measured Output and the Simulated Output

Update the experiment with the estimated capacitance value.

Exp = setEstimatedValues(Exp,pOpt);

Create a simulation scenario using the experiment and obtain the simulated output.

Simulator = createSimulator(Exp,Simulator);
Simulator = sim(Simulator);

Search for the voltage signal in the logged simulation data.

SimLog = find(Simulator.LoggedData,get_param('sdoRCCircuit','SignalLoggingName'));
Voltage = find(SimLog,'Voltage');

Plot the measured and simulated data.

The simulated and measured signals match well, except for near time zero. This
mismatch is because the capacitor initial voltage defined in the model does not match the
initial voltage from the experiment.

plot(time,data,'ro',Voltage.Values.Time,Voltage.Values.Data,'b')
title('Simulated and Measured Responses After Estimation')
legend('Measured Voltage','Simulated Voltage')

2-110
 Estimate Model Parameters and Initial States (Code)

Estimate the Initial State

Add the capacitor initial voltage for the C1 block to the experiment. Set its initial guess
value to 1 V.

Exp.InitialStates = sdo.getStateFromModel('sdoRCCircuit','C1');
Exp.InitialStates.Value = 1;

Recreate the estimation function to use the experiment with initial state estimation

estFcn = @(v) sdoRCCircuit_Objective(v,Simulator,Exp);

Get the initial state and capacitance value that is to be estimated from the experiment.

2-111
2 Parameter Estimation

v = getValuesToEstimate(Exp);

Estimate the parameters.

vOpt = sdo.optimize(estFcn,v,opt)

Optimization started 31-Jul-2015 06:13:12

Step-size First-order
Iter F-count f(x) optimality
0 5 4.66337 1
1 10 2.01883 1.533 21
2 15 1.34889 0.1257 0.0803
3 20 1.34365 0.0525 0.12
4 25 1.34363 0.001294 0.000711
Local minimum found.

Optimization completed because the size of the gradient is less than

the selected value of the function tolerance.

vOpt(1,1) =

Name: 'sdoRCCircuit/C1:sdoRCCircuit.C1.vc'
Value: 2.3596
Minimum: -Inf
Maximum: Inf
Free: 1
Scale: 1
dxValue: 0
dxFree: 1
Info: [1x1 struct]

vOpt(2,1) =

Name: 'C1'
Value: 2.2638e-04
Minimum: 0
Maximum: Inf
Free: 1
Scale: 0.0020
Info: [1x1 struct]

2-112
 Estimate Model Parameters and Initial States (Code)

2x1 param.Continuous

Compare the Measured Output and the Final Simulated Output

Update the experiment with the estimated capacitance and capacitor initial voltage
values.

Exp = setEstimatedValues(Exp,vOpt);

Simulate the model with the estimated initial-state and parameter values and compare
the simulated output with the experiment data.

Simulator = createSimulator(Exp,Simulator);
Simulator = sim(Simulator);
SimLog = find(Simulator.LoggedData,get_param('sdoRCCircuit','SignalLoggingName'));
Voltage = find(SimLog,'Voltage');

plot(time,data,'ro',Voltage.Values.Time,Voltage.Values.Data,'b')
title('Simulated and Measured Responses After Initial State and Model Parameter Estimat
legend('Measured Voltage','Simulated Voltage')

2-113
2 Parameter Estimation

Update the Model Parameter Values

Update the model with the estimated capacitance value. Do not update the model
capacitor initial voltage (first element of vOpt) as it is dependent on the experiment.

sdo.setValueInModel('sdoRCCircuit',vOpt(2));

Related Examples

To learn how to estimate model parameters using the sdo.optimize command, see
"Estimate Model Parameters and Initial States (GUI)".

Close the model

2-114
 Estimate Model Parameters and Initial States (Code)

bdclose('sdoRCCircuit')

2-115
2 Parameter Estimation

Estimate Model Parameters using Multiple Experiments (Code)

This example shows how to estimate model parameters from multiple sets of
experimental data. You estimate the parameters of a mass-spring-damper system.

Open the Model and Get Experimental Data

This example uses the sdoMassSpringDamper model. The model includes two
integrators to model the velocity and position of a mass in a mass-spring-damper system.

open_system('sdoMassSpringDamper');

Load the experiment data.

load sdoMassSpringDamper_ExperimentData

The variables texp1, yexp1, texp2, and yexp2 are loaded into the workspace. yexp1
and yexp2 describe the mass position for times texp1 and texp2 respectively.

2-116
 Estimate Model Parameters using Multiple Experiments (Code)

Define the Estimation Experiments

Create a 2-element array of experiment objects to store the measured data for the two
experiments.

Create an experiment object for the first experiment.

Exp = sdo.Experiment('sdoMassSpringDamper');

Create an object to store the measured mass position output.

MeasuredPos = Simulink.SimulationData.Signal;
MeasuredPos.Values = timeseries(yexp1,texp1);
MeasuredPos.BlockPath = 'sdoMassSpringDamper/Position';
MeasuredPos.PortType = 'outport';
MeasuredPos.PortIndex = 1;
MeasuredPos.Name = 'Position';

Add the measured mass position data to the experiment as the expected output data.

Exp.OutputData = MeasuredPos;

Create an object to specify the initial state for the Velocity block. The initial velocity of
the mass is 0 m/s.

sVel = sdo.getStateFromModel('sdoMassSpringDamper','Velocity');
sVel.Value = 0;
sVel.Free = false;

sVel.Free is set to false because the initial velocity is known and does not need to be
estimated.

Create an object to specify the initial state for the Position block. Specify a guess for
the initial mass position. Set the Free field of the initial position object to true so that it
is estimated.

sPos = sdo.getStateFromModel('sdoMassSpringDamper','Position');
sPos.Free = true;
sPos.Value = -0.1;

Add the initial states to the experiment.

Exp.InitialStates = [sVel;sPos];

2-117
2 Parameter Estimation

Create a 2-element array of experiments. As the two experiments are identical except for
the expected output data, copy the first experiment twice.
Exp = [Exp; Exp];

Modify the expected output data of the second experiment object in Exp.
Exp(2).OutputData.Values = timeseries(yexp2,texp2);

Compare the Measured Output and the Initial Simulated Output

Create a simulation scenario using the first experiment and obtain the simulated output.
Simulator = createSimulator(Exp(1));
Simulator = sim(Simulator);

Search for the position signal in the logged simulation data.

SimLog = find(Simulator.LoggedData,get_param('sdoMassSpringDamper','SignalLoggingName
Position = find(SimLog,'Position');

Obtain the simulated position signal for the second experiment.

Simulator = createSimulator(Exp(2),Simulator);
Simulator = sim(Simulator);
SimLog = find(Simulator.LoggedData,get_param('sdoMassSpringDamper','SignalLoggingN
Position(2) = find(SimLog,'Position');

Plot the measured and simulated data.

The model response does not match the experimental output data.
subplot(211)
plot(...
Position(1).Values.Time,Position(1).Values.Data, ...
Exp(1).OutputData.Values.Time, Exp(1).OutputData.Values.Data,'--')
title('Experiment 1: Simulated and Measured Responses Before Estimation')
ylabel('Position')
legend('Measured Position','Simulated Position','Location','SouthEast')
subplot(212)
plot(...
Position(2).Values.Time,Position(2).Values.Data, ...
Exp(2).OutputData.Values.Time, Exp(2).OutputData.Values.Data,'--')
title('Experiment 2: Simulated and Measured Responses Before Estimation')
xlabel('Time (seconds)')
ylabel('Position')

2-118
 Estimate Model Parameters using Multiple Experiments (Code)

legend('Measured Position','Simulated Position','Location','SouthEast')

Specify Parameters to Estimate

Select the mass m, spring constant k, and damping coefficient b parameters from the
model. Specify that the estimated values for these parameters must be positive.
p = sdo.getParameterFromModel('sdoMassSpringDamper', {'b', 'k', 'm'});
p(1).Minimum = 0;
p(2).Minimum = 0;
p(3).Minimum = 0;

Get the position initial state values to be estimated from the experiment.
s = getValuesToEstimate(Exp);

2-119
2 Parameter Estimation

s contains two initial state objects, both for the Position block. Each object corresponds
to an experiment in Exp.

Group the model parameters and initial states to be estimated together.

v = [p;s]

v(1,1) =

Name: 'b'
Value: 100
Minimum: 0
Maximum: Inf
Free: 1
Scale: 128
Info: [1x1 struct]

v(2,1) =

Name: 'k'
Value: 500
Minimum: 0
Maximum: Inf
Free: 1
Scale: 512
Info: [1x1 struct]

v(3,1) =

Name: 'm'
Value: 8
Minimum: 0
Maximum: Inf
Free: 1
Scale: 8
Info: [1x1 struct]

v(4,1) =

Name: 'sdoMassSpringDamper/Position'
Value: -0.1000

2-120
 Estimate Model Parameters using Multiple Experiments (Code)

Minimum: -Inf
Maximum: Inf
Free: 1
Scale: 0.1250
dxValue: 0
dxFree: 1
Info: [1x1 struct]

v(5,1) =

Name: 'sdoMassSpringDamper/Position'
Value: -0.1000
Minimum: -Inf
Maximum: Inf
Free: 1
Scale: 0.1250
dxValue: 0
dxFree: 1
Info: [1x1 struct]

5x1 param.Continuous

Define the Estimation Objective

Create an estimation objective function to evaluate how closely the simulation output,
generated using the estimated parameter values, matches the measured data.

Use an anonymous function with one input argument that calls the
sdoMassSpringDamper_Objective function. We pass the anonymous function to
sdo.optimize, which evaluates the function at each optimization iteration.
estFcn = @(v) sdoMassSpringDamper_Objective(v,Simulator,Exp);

The sdoMassSpringDamper_Objective function:

• Has one input argument that specifies the mass, spring constant and damper values
as well as the initial mass position.

• Has one input argument that specifies the experiment object containing the measured
data.

• Returns a vector of errors between simulated and experimental outputs.

2-121
2 Parameter Estimation

The sdoMassSpringDamper_Objective function requires two inputs, but

sdo.optimize requires a function with one input argument. To work around
this, estFcn is an anonymous function with one input argument, v, but it calls
sdoMassSpringDamper_Objective using two input arguments, v and Exp.

For more information regarding anonymous functions, see "Anonymous Functions".

The sdo.optimize command minimizes the return argument of the

anonymous function estFcn, that is, the residual errors returned by
sdoMassSpringDamper_Objective. For more details on how to write an
objective/constraint function to use with the sdo.optimize command, type help
sdoExampleCostFunction at the MATLAB command prompt.

To examine the estimation objective function in more detail, type edit

sdoMassSpringDamper_Objective at the MATLAB command prompt.
type sdoMassSpringDamper_Objective

function vals = sdoMassSpringDamper_Objective(v,Simulator,Exp)

%SDOMASSSPRINGDAMPER_OBJECTIVE
%
% The sdoMassSpringDamper_Objective function is used to compare model
% outputs against experimental data.
%
% vals = sdoMassSpringDamper_Objective(v,Exp)
%
% The |v| input argument is a vector of estimated model parameter values
% and initial states.
%
% The |Simulator| input argument is a simulation object used
% simulate the model with the estimated parameter values.
%
% The |Exp| input argument contains the estimation experiment data.
%
% The |vals| return argument contains information about how well the
% model simulation results match the experimental data and is used by
% the |sdo.optimize| function to estimate the model parameters.
%
% see also sdo.optimize, sdoExampleCostFunction
%

% Copyright 2012-2015 The MathWorks, Inc.

2-122
 Estimate Model Parameters using Multiple Experiments (Code)

%%
% Define a signal tracking requirement to compute how well the model output
% matches the experiment data. Configure the tracking requirement so that
% it returns the tracking error residuals (rather than the
% sum-squared-error) and does not normalize the errors.
%
r = sdo.requirements.SignalTracking;
r.Type = '==';
r.Method = 'Residuals';
r.Normalize = 'off';

%%
% Update the experiments with the estimated parameter values.
%
Exp = setEstimatedValues(Exp,v);

%%
% Simulate the model and compare model outputs with measured experiment
% data.
%
Error = [];
for ct=1:numel(Exp)

Simulator = createSimulator(Exp(ct),Simulator);
Simulator = sim(Simulator);

SimLog = find(Simulator.LoggedData,get_param('sdoMassSpringDamper','SignalLoggingN
Position = find(SimLog,'Position');

PositionError = evalRequirement(r,Position.Values,Exp(ct).OutputData.Values);

Error = [Error; PositionError(:)];

end

%%
% Return the residual errors to the optimization solver.
%
vals.F = Error(:);
end

Estimate the Parameters

Use the sdo.optimize function to estimate the actuator parameter values and initial
state.

2-123
2 Parameter Estimation

Specify the optimization options. The estimation function

sdoMassSpringDamper_Objective returns the error residuals between simulated and
experimental data and does not include any constraints, making this problem ideal for
the 'lsqnonlin' solver.
opt = sdo.OptimizeOptions;
opt.Method = 'lsqnonlin';

Estimate the parameters. Notice that the initial mass position is estimated twice, once
for each experiment.
vOpt = sdo.optimize(estFcn,v,opt)

Optimization started 31-Jul-2015 06:12:03

Step-size First-order
Iter F-count f(x) optimality
0 11 0.777696 1
1 22 0.00413099 3.696 0.00648
2 33 0.00118327 0.3194 0.00243
3 44 0.0011106 0.06718 5.09e-05
Local minimum found.

Optimization completed because the size of the gradient is less than

the selected value of the function tolerance.

vOpt(1,1) =

Name: 'b'
Value: 58.1959
Minimum: 0
Maximum: Inf
Free: 1
Scale: 128
Info: [1x1 struct]

vOpt(2,1) =

Name: 'k'
Value: 399.9452
Minimum: 0
Maximum: Inf
Free: 1

2-124
 Estimate Model Parameters using Multiple Experiments (Code)

Scale: 512
Info: [1x1 struct]

vOpt(3,1) =

Name: 'm'
Value: 9.7225
Minimum: 0
Maximum: Inf
Free: 1
Scale: 8
Info: [1x1 struct]

vOpt(4,1) =

Name: 'sdoMassSpringDamper/Position'
Value: 0.2995
Minimum: -Inf
Maximum: Inf
Free: 1
Scale: 0.1250
dxValue: 0
dxFree: 1
Info: [1x1 struct]

vOpt(5,1) =

Name: 'sdoMassSpringDamper/Position'
Value: 0.0994
Minimum: -Inf
Maximum: Inf
Free: 1
Scale: 0.1250
dxValue: 0
dxFree: 1
Info: [1x1 struct]

5x1 param.Continuous

2-125
2 Parameter Estimation

Compare the Measured Output and the Final Simulated Output

Update the experiments with the estimated parameter values.

Exp = setEstimatedValues(Exp,vOpt);

Obtain the simulated output for the first experiment.

Simulator = createSimulator(Exp(1),Simulator);
Simulator = sim(Simulator);
SimLog = find(Simulator.LoggedData,get_param('sdoMassSpringDamper','SignalLoggingN
Position(1) = find(SimLog,'Position');

Obtain the simulated output for the second experiment.

Simulator = createSimulator(Exp(2),Simulator);
Simulator = sim(Simulator);
SimLog = find(Simulator.LoggedData,get_param('sdoMassSpringDamper','SignalLoggingN
Position(2) = find(SimLog,'Position');

Plot the measured and simulated data.

The model response using the estimated parameter values nicely matches the output
data for the experiments.

subplot(211)
plot(...
Position(1).Values.Time,Position(1).Values.Data, ...
Exp(1).OutputData.Values.Time, Exp(1).OutputData.Values.Data,'--')
title('Experiment 1: Simulated and Measured Responses After Estimation')
ylabel('Position')
legend('Measured Position','Simulated Position','Location','NorthEast')
subplot(212)
plot(...
Position(2).Values.Time,Position(2).Values.Data, ...
Exp(2).OutputData.Values.Time, Exp(2).OutputData.Values.Data,'--')
title('Experiment 2: Simulated and Measured Responses After Estimation')
xlabel('Time (seconds)')
ylabel('Voltage')
legend('Measured Position','Simulated Position','Location','SouthEast')

2-126
 Estimate Model Parameters using Multiple Experiments (Code)

Update the Model Parameter Values

Update the model m, k, and b parameter values. Do not update the model initial position
value as this is dependent on the experiment.
sdo.setValueInModel('sdoMassSpringDamper',vOpt(1:3));

Close the model

bdclose('sdoMassSpringDamper')

2-127
2 Parameter Estimation

Estimate Model Parameters Per Experiment (Code)

This example shows how to use multiple experiments to estimate a mix of model
parameter values; some that are estimated using all the experiments and others that
are estimated using individual experiments. The example also shows how to configure
estimation experiments with experiment dependent parameter values.

You estimate the parameters of a rechargeable battery based on data collected in

experiments that discharge and charge the battery.

Open the Model and Get Experimental Data

This example estimates parameters of a simple, rechargeable battery model,

sdoBattery. The model input is the battery current and the model output, the battery
terminal voltage, is computed from the battery state-of-charge.

open_system('sdoBattery');

The model is based on the equation

2-128
 Estimate Model Parameters Per Experiment (Code)

Where:

• is the battery terminal voltage in Volts.

• is the battery constant voltage in Volts.

• is the battery polarization resistance in Ohms.

• is the maximum battery capacity in Ampere-Hour.

• is the battery charge state, with 1 being fully charged and 0 zero charge.The battery
state-of-charge is computed from the integral of the battery current with a +ve
current indicating discharge and a -ve current indicating charging. The battery initial
state-of-charge is specified by in Ampere-Hour.

• is the voltage drop when charging, expressed as a fraction of the battery

constant voltage. When the battery is discharging this value is zero.

V, K, Qmax, Q0, and Loss are variables defined in the model workspace.

Load the experiment data. A 1.2V (6500mAh) battery was subjected to a discharge
experiment and a charging experiment.

load sdoBattery_ExperimentData

The variables Charge_Data and DCharge_Data are loaded into the workspace. The
first column of Charge_Data contains time data. The second and third columns of
Charge_Data describe the current and voltage during a battery charging experiment.
DCharge_Data is similarly structured and contains data for a battery discharging
experiment.

Plot the Experiment Data

subplot(221),
plot(DCharge_Data(:,1)/3600,DCharge_Data(:,2))
title('Experiment: Discharge')
xlabel('Time (hours)')
ylabel('Current (A)')
subplot(223)
plot(DCharge_Data(:,1)/3600,DCharge_Data(:,3))
xlabel('Time (hours)')
ylabel('Voltage (V)')
subplot(222),

2-129
2 Parameter Estimation

plot(Charge_Data(:,1)/3600,Charge_Data(:,2))
title('Experiment: Charge')
xlabel('Time (hours)')
ylabel('Current (A)')
subplot(224)
plot(Charge_Data(:,1)/3600,Charge_Data(:,3))
xlabel('Time (hours)')
ylabel('Voltage (V)')

Define the Estimation Experiments

Create a 2-element array of experiment objects to specify the measured data for the two
experiments.

2-130
 Estimate Model Parameters Per Experiment (Code)

Create an experiment object for the battery discharge experiment. The measured current
data is specified as a timeseries in the experiment object.
DCharge_Exp = sdo.Experiment('sdoBattery');

Specify the input data (current) as a timeseries object.

DCharge_Exp.InputData = timeseries(DCharge_Data(:,2),DCharge_Data(:,1));

Create an object to specify the measured voltage output data.

VoltageSig = Simulink.SimulationData.Signal;
VoltageSig.Name = 'Voltage';
VoltageSig.BlockPath = 'sdoBattery/SOC -> Voltage';
VoltageSig.PortType = 'outport';
VoltageSig.PortIndex = 1;
VoltageSig.Values = timeseries(DCharge_Data(:,3),DCharge_Data(:,1));

Add the voltage signal to the discharge experiment as the expected output data.
DCharge_Exp.OutputData = VoltageSig;

Specify the battery initial charge state for the experiment. The battery charge state
is modeled by the Q (Ah) block and it's initial value is specified by the variable Q0.
Create a parameter for the Q0 variable and add the parameter to the experiment. Q0 is
experiment dependent and assumes different values in the discharging and charging
experiments.
Q0 = sdo.getParameterFromModel('sdoBattery','Q0');
Q0.Value = 6.5;
Q0.Free = false;

Q0.Free is set to false because the initial battery charge is known and does not need to
be estimated.

Add the Q0 parameter to the experiment.

DCharge_Exp.Parameters = Q0;

Create an experiment object to store the charging experiment data. Add the measured
current input and measured voltage output data to the object.
Charge_Exp = sdo.Experiment('sdoBattery');
Charge_Exp.InputData = timeseries(Charge_Data(:,2),Charge_Data(:,1));
VoltageSig.Values = timeseries(Charge_Data(:,3),Charge_Data(:,1));

2-131
2 Parameter Estimation

Charge_Exp.OutputData = VoltageSig;

Add the battery initial charge and charging loss fraction parameters to the experiment.
For this experiment, the initial charge (Q0) is known (0 Ah), but the value of the charging
loss fraction (Loss) is not known.
Q0.Value = 0;

Loss = sdo.getParameterFromModel('sdoBattery','Loss');
Loss.Free = true;
Loss.Minimum = 0;
Loss.Maximum = 0.5;

Charge_Exp.Parameters = [Q0;Loss];

Loss.Free is set to true so that the value of Loss is estimated.

Collect both experiments into one vector.

Exp = [DCharge_Exp; Charge_Exp];

Compare the Measured Output and the Initial Simulated Output

Create a simulation scenario using the first (discharging) experiment and obtain the
simulated output.
Simulator = createSimulator(Exp(1));
Simulator = sim(Simulator);

Search for the voltage signal in the logged simulation data.

SimLog = find(Simulator.LoggedData,get_param('sdoBattery','SignalLoggingName'));
Voltage(1) = find(SimLog,'Voltage');

Obtain the simulated voltage signal for the second (charging) experiment.
Simulator = createSimulator(Exp(2),Simulator);
Simulator = sim(Simulator);
SimLog = find(Simulator.LoggedData,get_param('sdoBattery','SignalLoggingName'));
Voltage(2) = find(SimLog,'Voltage');

Plot the measured and simulated data.

The model response does not match the experimental output data.
subplot(211)

2-132
 Estimate Model Parameters Per Experiment (Code)

plot(...
Voltage(1).Values.Time/3600,Voltage(1).Values.Data, ...
Exp(1).OutputData.Values.Time/3600, Exp(1).OutputData.Values.Data,'-.')
title('Discharging Experiment: Simulated and Measured Responses Before Estimation')
ylabel('Voltage (V)')
legend('Simulated Voltage','Measured Voltage','Location','SouthWest')
subplot(212)
plot(...
Voltage(2).Values.Time/3600,Voltage(2).Values.Data, ...
Exp(2).OutputData.Values.Time/3600, Exp(2).OutputData.Values.Data,'-.')
title('Charging Experiment: Simulated and Measured Responses Before Estimation')
xlabel('Time (hours)')
ylabel('Voltage (V)')
legend('Simulated Voltage','Measured Voltage','Location','SouthEast')

2-133
2 Parameter Estimation

Specify Parameters to Estimate

Estimate the values of the battery voltage V, the battery polarization resistance K, and
the charging loss fraction Loss. The V and K parameters are estimated using all the
experiment data while the Loss parameter is estimated using only the charging data.

Select the battery voltage V and the battery polarization resistance K parameters from
the model. Specify minimum and maximum bounds for these parameters.

p = sdo.getParameterFromModel('sdoBattery',{'V','K'});

p(1).Minimum = 0;
p(1).Maximum = 2;

p(2).Minimum = 1e-6;
p(2).Maximum = 1e-1;

Get the experiment-specific Loss parameter from the experiment.

s = getValuesToEstimate(Exp);

Group all the parameters to be estimated.

v = [p;s]

v(1,1) =

Name: 'V'
Value: 1.2000
Minimum: 0
Maximum: 2
Free: 1
Scale: 2
Info: [1x1 struct]

v(2,1) =

Name: 'K'
Value: 1.0000e-03
Minimum: 1.0000e-06
Maximum: 0.1000
Free: 1

2-134
 Estimate Model Parameters Per Experiment (Code)

Scale: 0.0020
Info: [1x1 struct]

v(3,1) =

Name: 'Loss'
Value: 0.0100
Minimum: 0
Maximum: 0.5000
Free: 1
Scale: 0.0156
Info: [1x1 struct]

3x1 param.Continuous

Define the Estimation Objective

Create an estimation objective function to evaluate how closely the simulation output,
generated using the estimated parameter values, matches the measured data.

Use an anonymous function with one input argument that calls the
sdoBattery_Objective function. We pass the anonymous function to sdo.optimize,
which evaluates the function at each optimization iteration.
estFcn = @(v) sdoBattery_Objective(v,Simulator,Exp);

The sdoBattery_Objective function:

• Has one input argument that specifies the estimated battery parameter values.

• Has one input argument that specifies the experiment object containing the measured
data.

• Returns a vector of errors between simulated and experimental outputs.

The sdoBattery_Objective function requires two inputs, but sdo.optimize requires

a function with one input argument. To work around this, estFcn is an anonymous
function with one input argument, v, but it calls sdoBattery_Objective using two
input arguments, v and Exp.

For more information regarding anonymous functions, see "Anonymous Functions".

2-135
2 Parameter Estimation

The sdo.optimize command minimizes the return argument of the anonymous

function estFcn, that is, the residual errors returned by sdoBattery_Objective.
For more details on how to write an objective/constraint function to use with the
sdo.optimize command, type help sdoExampleCostFunction at the MATLAB
command prompt.

To examine the estimation objective function in more detail, type edit

sdoBattery_Objective at the MATLAB command prompt.

type sdoBattery_Objective

function vals = sdoBattery_Objective(v,Simulator,Exp)

%SDOBATTERY_OBJECTIVE
%
% The sdoBattery_Objective function is used to compare model
% outputs against experimental data.
%
% vals = sdoBattery_Objective(v,Exp)
%
% The |v| input argument is a vector of estimated model parameter values
% and initial states.
%
% The |Simulator| input argument is a simulation object used
% simulate the model with the estimated parameter values.
%
% The |Exp| input argument contains the estimation experiment data.
%
% The |vals| return argument contains information about how well the
% model simulation results match the experimental data and is used by
% the |sdo.optimize| function to estimate the model parameters.
%
% See also sdo.optimize, sdoExampleCostFunction, sdoBattery_cmddemo
%

% Copyright 2012-2015 The MathWorks, Inc.

%%
% Define a signal tracking requirement to compute how well the model output
% matches the experiment data. Configure the tracking requirement so that
% it returns the tracking error residuals (rather than the
% sum-squared-error) and does not normalize the errors.
%
r = sdo.requirements.SignalTracking;

2-136
 Estimate Model Parameters Per Experiment (Code)

r.Type = '==';
r.Method = 'Residuals';
r.Normalize = 'off';

%%
% Update the experiments with the estimated parameter values.
%
Exp = setEstimatedValues(Exp,v);

%%
% Simulate the model and compare model outputs with measured experiment
% data.
%
Error = [];
for ct=1:numel(Exp)

Simulator = createSimulator(Exp(ct),Simulator);
Simulator = sim(Simulator);

SimLog = find(Simulator.LoggedData,get_param('sdoBattery','SignalLoggingName'));
Voltage = find(SimLog,'Voltage');

VoltageError = evalRequirement(r,Voltage.Values,Exp(ct).OutputData(1).Values);

Error = [Error; VoltageError(:)];

end

%%
% Return the residual errors to the optimization solver.
%
vals.F = Error(:);
end

Estimate the Parameters

Use the sdo.optimize function to estimate the battery parameter values.

Specify the optimization options. The estimation function sdoBattery_Objective

returns the error residuals between simulated and experimental data and does not
include any constraints, making this problem ideal for the 'lsqnonlin' solver.
opt = sdo.OptimizeOptions;
opt.Method = 'lsqnonlin';

Estimate the parameters.

2-137
2 Parameter Estimation

vOpt = sdo.optimize(estFcn,v,opt)

Optimization started 31-Jul-2015 06:06:08

Step-size First-order
Iter F-count f(x) optimality
0 7 3272.22 1
1 14 619.356 0.1634 3.15e+05
2 21 411.131 0.2175 28.7
3 28 405.529 0.3838 2.16e+03
4 35 403.727 0.2767 15.2
5 42 403.379 0.1645 1.14e+03
Local minimum possible.

lsqnonlin stopped because the final change in the sum of squares relative to
its initial value is less than the selected value of the function tolerance.

vOpt(1,1) =

Name: 'V'
Value: 1.3083
Minimum: 0
Maximum: 2
Free: 1
Scale: 2
Info: [1x1 struct]

vOpt(2,1) =

Name: 'K'
Value: 0.0010
Minimum: 1.0000e-06
Maximum: 0.1000
Free: 1
Scale: 0.0020
Info: [1x1 struct]

vOpt(3,1) =

Name: 'Loss'
Value: 5.1801e-05
Minimum: 0

2-138
 Estimate Model Parameters Per Experiment (Code)

Maximum: 0.5000
Free: 1
Scale: 0.0156
Info: [1x1 struct]

3x1 param.Continuous

Compare the Measured Output and the Final Simulated Output

Update the experiments with the estimated parameter values.

Exp = setEstimatedValues(Exp,vOpt);

Obtain the simulated output for the first (discharging) experiment.

Simulator = createSimulator(Exp(1),Simulator);
Simulator = sim(Simulator);
SimLog = find(Simulator.LoggedData,get_param('sdoBattery','SignalLoggingName'));
Voltage(1) = find(SimLog,'Voltage');

Obtain the simulated output for the second (charging) experiment.

Simulator = createSimulator(Exp(2),Simulator);
Simulator = sim(Simulator);
SimLog = find(Simulator.LoggedData,get_param('sdoBattery','SignalLoggingName'));
Voltage(2) = find(SimLog,'Voltage');

Plot the measured and simulated data.

The simulation results match the experimental data well except in the regions when
the battery is fully charged. This is not unexpected as the simple battery model does not
model the exponential voltage drop when the battery is fully charged.
subplot(211)
plot(...
Voltage(1).Values.Time/3600,Voltage(1).Values.Data, ...
Exp(1).OutputData.Values.Time/3600, Exp(1).OutputData.Values.Data,'-.')
title('Discharging Experiment: Simulated and Measured Responses After Estimation')
ylabel('Voltage (V)')
legend('Simulated Voltage','Measured Voltage','Location','SouthWest')
subplot(212)
plot(...
Voltage(2).Values.Time/3600,Voltage(2).Values.Data, ...
Exp(2).OutputData.Values.Time/3600, Exp(2).OutputData.Values.Data,'-.')

2-139
2 Parameter Estimation

title('Charging Experiment: Simulated and Measured Responses After Estimation')

xlabel('Time (hours)')
ylabel('Voltage (V)')
legend('Simulated Voltage','Measured Voltage','Location','SouthEast')

Update the Model Parameter Values

Update the model V, K, and Loss parameter values.

sdo.setValueInModel('sdoBattery',vOpt);

Related Examples

To learn how to estimate the battery parameters using the Parameter Estimation Tool,
see "Estimate Model Parameters Per Experiment (GUI)".

2-140
 Estimate Model Parameters Per Experiment (Code)

Close the model

bdclose('sdoBattery')

2-141
2 Parameter Estimation

Estimate Model Parameters with Parameter Constraints (Code)

This example shows how to estimate model parameters while imposing constraints on
the parameter values.

You estimate dynamic and static friction coefficients of a simple friction system.

Open the Model and Get Experimental Data

This example estimates parameters for a simple friction system, sdoFriction. The
model input is the force applied to a mass and the model outputs are the mass position
and velocity.

open_system('sdoFriction');

The model is based on a mass sliding on a surface. The mass is subject to a static friction
that must be overcome before the mass moves and a dynamic friction once the mass
moves. The static friction, u_static, is a fraction of the mass normal force; similarly the
dynamic friction, u_dynamic, is a fraction of the mass normal force.

2-142
 Estimate Model Parameters with Parameter Constraints (Code)

Load the experiment data. The mass was subjected to an applied force and its position
recorded.

load sdoFriction_ExperimentData

The variables AppliedForce, Position, and Velocity are loaded into the workspace.
The first column of each of these variables represents time and the second column
represents the measured data. Because velocity is the first derivative of position, we only
use the position measurements for this example.

Plot the Experiment Data

subplot(211),
plot(AppliedForce(:,1),AppliedForce(:,2))
title('Measured Applied Force Input for Simple Friction System');
ylabel('Applied Force (N)')
subplot(212)
plot(Position(:,1),Position(:,2))
title('Measured Mass Position for Simple Friction System');
xlabel('Time (seconds)')
ylabel('Position (m)')

2-143
2 Parameter Estimation

Define the Estimation Experiment

Create an experiment object to specify the experiment data.

Exp = sdo.Experiment('sdoFriction');

Specify the input data (applied force) as a timeseries object.

Exp.InputData = timeseries(AppliedForce(:,2),AppliedForce(:,1));

Create an object to specify the measured mass position output.

PositionSig = Simulink.SimulationData.Signal;

2-144
 Estimate Model Parameters with Parameter Constraints (Code)

PositionSig.Name = 'Position';
PositionSig.BlockPath = 'sdoFriction/x';
PositionSig.PortType = 'outport';
PositionSig.PortIndex = 1;
PositionSig.Values = timeseries(Position(:,2),Position(:,1));

Add the measured mass position data to the experiment as the expected output data.

Exp.OutputData = PositionSig;

Compare the Measured Output and the Initial Simulated Output

Create a simulation scenario using the experiment and obtain the simulated output.

Simulator = createSimulator(Exp);
Simulator = sim(Simulator);

Search for the position signal in the logged simulation data.

SimLog = find(Simulator.LoggedData,get_param('sdoFriction','SignalLoggingName'));
Position = find(SimLog,'Position');

Plot the measured and simulated data.

As expected, the model response does not match the experimental output data.

figure
plot(...
Position.Values.Time,Position.Values.Data, ...
Exp.OutputData.Values.Time, Exp.OutputData.Values.Data,'-.')
title('Simulated and Measured Responses Before Estimation')
ylabel('Position (m)')
xlabel('Time (seconds)')
legend('Simulated Position','Measured Position','Location','NorthWest')

2-145
2 Parameter Estimation

Specify Parameters to Estimate

Estimate the u_static and u_dynamic friction coefficients using the experiment data.
These coefficients are used as gains in the Static Friction and Dynamic Friction
blocks, respectively. Physics indicates that friction coefficients should be constrained
so that u_static u_dynamic; this parameter constraint is implemented in the
estimation objective function.

Select the u_static and u_dynamic model parameters. Specify bounds for the
estimated parameter values. Both coefficients are limited to the range [0 1].

p = sdo.getParameterFromModel('sdoFriction',{'u_static','u_dynamic'});

2-146
 Estimate Model Parameters with Parameter Constraints (Code)

p(1).Minimum = 0;
p(1).Maximum = 1;

p(2).Minimum = 0;
p(2).Maximum = 1;

Define the Estimation Objective

Create an estimation objective function to evaluate how closely the simulation output,
generated using the estimated parameter values, matches the measured data.

Use an anonymous function with one input argument that calls the
sdoFriction_Objective function. We pass the anonymous function to
sdo.optimize, which evaluates the function at each optimization iteration.

estFcn = @(v) sdoFriction_Objective(v,Simulator,Exp);

The sdoFriction_Objective function:

• Has one input argument that specifies the estimated friction coefficients.

• Has one input argument that specifies the experiment object containing the measured
data.

• Returns the sum-squared-error errors between simulated and experimental outputs,

and returns the parameter constraint.

The sdoFriction_Objective function requires two inputs, but sdo.optimize

requires a function with one input argument. To work around this, estFcn is an
anonymous function with one input argument, v, but it calls sdoFriction_Objective
using two input arguments, v and Exp.

For more information regarding anonymous functions, see "Anonymous Functions".

The sdo.optimize command minimizes the return argument of the anonymous

function estFcn, that is, the residual errors returned by sdoFriction_Objective.
For more details on how to write an objective/constraint function to use with the
sdo.optimize command, type help sdoExampleCostFunction at the MATLAB
command prompt.

To examine the estimation objective function in more detail, type edit

sdoFriction_Objective at the MATLAB command prompt.

2-147
2 Parameter Estimation

type sdoFriction_Objective

function vals = sdoFriction_Objective(p,Simulator,Exp)

%SDOFRICTION_OBJECTIVE
%
% The sdoFriction_Objective function is used to compare model
% outputs against experimental data and measure how well constraints are
% satisfied.
%
% vals = sdoFriction_Objective(p,Exp)
%
% The |p| input argument is a vector of estimated model parameter values.
%
% The |Simulator| input argument is a simulation object used
% simulate the model with the estimated parameter values.
%
% The |Exp| input argument contains the estimation experiment data.
%
% The |vals| return argument contains information about how well the
% model simulation results match the experimental data and how well
% constraints are satisfied. The |vals| argument is used by the
% |sdo.optimize| function to estimate the model parameters.
%
% See also sdo.optimize, sdoExampleCostFunction, sdoFriction_cmddemo
%

% Copyright 2012-2015 The MathWorks, Inc.

%%
% Define a signal tracking requirement to compute how well the model output
% matches the experiment data. Configure the tracking requirement so that
% it returns the sum-squared-error.
%
r = sdo.requirements.SignalTracking;
r.Type = '==';
r.Method = 'SSE';

%%
% Update the experiments with the estimated parameter values.
%
Exp = setEstimatedValues(Exp,p);

%%
% Simulate the model and compare model outputs with measured experiment

2-148
 Estimate Model Parameters with Parameter Constraints (Code)

% data.
%
Simulator = createSimulator(Exp,Simulator);
Simulator = sim(Simulator);

SimLog = find(Simulator.LoggedData,get_param('sdoFriction','SignalLoggingName'));
Position = find(SimLog,'Position');

PositionError = evalRequirement(r,Position.Values,Exp.OutputData(1).Values);

%%
% Measure how well the parameters satisfy the friction coefficient constraint,
% |u_static| >= |u_dynamic|. Note that constraints are returned to the
% optimizer in a c <=0 format. The friction coefficient constraint is
% rewritten accordingly.
PConstr = p(2).Value - p(1).Value; % u_dynamic - u_static <= 0

%%
% Return the sum-squared-error and constraint violation to the optimization
% solver.
%
vals.F = PositionError(:);
vals.Cleq = PConstr;
end

The friction coefficient constraint, u_static u_dynamic, is implemented in the

sdoFriction_Objective function as u_dynamic - u_static 0. This is because the
optimizer requires constraint values in a format. For more information, type help
sdo.optimize at the MATLAB command prompt.

Estimate the Parameters

Use the sdo.optimize function to estimate the friction model parameter values.

Specify the optimization options. The estimation function sdoFriction_Objective

returns the sum-squared-error between simulated and experimental data and includes a
parameter constraint. The default 'fmincon' solver is ideal for this type of problem.

Estimate the parameters.

pOpt = sdo.optimize(estFcn,p)

Optimization started 31-Jul-2015 06:09:07

2-149
2 Parameter Estimation

max Step-size First-order

Iter F-count f(x) constraint optimality
0 5 27.7267 0
1 11 22.5643 0 2.21 72.9
2 15 17.4771 0 0.51 16
3 22 0.762174 0 1.33 10.7
4 29 0.40765 0 0.263 3.15
5 34 0.0254254 0 0.0897 1.22
6 39 0.00522001 0 0.0296 0.276
7 44 0.00398126 0 0.0209 0.185
8 49 0.00120167 0 0.111 0.17
9 60 0.00118106 0 0.0212 0.173
10 72 0.00110164 0 0.0262 0.165
11 91 0.00110097 0 0.0031 0.174
12 108 0.00110097 0 0.00165 0.174
Local minimum possible. Constraints satisfied.

fmincon stopped because the size of the current step is less than
the selected value of the step size tolerance and constraints are
satisfied to within the selected value of the constraint tolerance.

pOpt(1,1) =

Name: 'u_static'
Value: 0.7973
Minimum: 0
Maximum: 1
Free: 1
Scale: 0.5000
Info: [1x1 struct]

pOpt(2,1) =

Name: 'u_dynamic'
Value: 0.4021
Minimum: 0
Maximum: 1
Free: 1
Scale: 0.2500
Info: [1x1 struct]

2-150
 Estimate Model Parameters with Parameter Constraints (Code)

2x1 param.Continuous

Compare the Measured Output and the Final Simulated Output

Update the experiments with the estimated parameter values.

Exp = setEstimatedValues(Exp,pOpt);

Obtain the simulated output for the experiment.

Simulator = createSimulator(Exp,Simulator);
Simulator = sim(Simulator);
SimLog = find(Simulator.LoggedData,get_param('sdoFriction','SignalLoggingName'));
Position = find(SimLog,'Position');

Plot the measured and simulated data.

It can be seen that the model response using the estimated parameter values nicely
matches the experiment output data.

plot(...
Position.Values.Time,Position.Values.Data, ...
Exp.OutputData.Values.Time, Exp.OutputData.Values.Data,'-.')
title('Simulated and Measured Responses After Model Parameter Estimation')
ylabel('Position (m)')
xlabel('Time (seconds)')
legend('Simulated Position','Measured Position','Location','NorthWest')

2-151
2 Parameter Estimation

Update the Model Parameter Values

Update the model u_static and u_dynamic parameter values.

sdo.setValueInModel('sdoFriction',pOpt);

Close the model

bdclose('sdoFriction')

2-152
 Estimate Model Parameter Values (GUI)

Estimate Model Parameter Values (GUI)

This example shows how to use experiment data to estimate model parameters. You
estimate the parameters of an engine throttle system.

Simulink® Model of the Engine Throttle System

The Simulink® model for the engine throttle system, spe_engine_throttle, is shown
below.

Throttle Model Description

The throttle controls the air mass flow into the intake manifold of an engine. The
throttle body contains a butterfly valve that opens when the driver presses down on the
accelerator pedal. This lets more air enter the cylinders and causes the engine to produce
more torque.

A DC motor controls the opening angle of the butterfly valve. There is also a spring
attached to the valve to return it to its closed position when the DC motor is de-

2-153
2 Parameter Estimation

energized. The amount of rotation of the valve is limited to approximately 90 degrees.

Therefore, if a large command input is applied to the motor, the valve hits the hard stops
preventing it from rotating further.

The motor is modeled as a torque gain and a time-delay input with parameters Kt
and input_delay. The butterfly valve is modeled as a mass-spring-damper system with
parameters J, c and k. This system is augmented with hard stops to limit the valve
opening to 90 degrees. We know the model components, however, the parameter values of
the system are not known accurately.

Estimation Experiment Data

Double-click the Parameter Estimation GUI with preloaded data block in the
model to open a pre-configured estimation GUI session.

The saved estimation project defines three experiments; the EstimationData

experiment will be used for parameter estimation, while ValidationData1,
ValidationData2 are used for validating the estimated parameters. The
EstimateData experiment is plotted.

The signal data for the experiments can be imported from various sources including
MATLAB® variables, MAT files, Excel® files, or comma-separated-value files. See
"Importing and Preprocessing Experiment Data (GUI)" for more information.

2-154
 Estimate Model Parameter Values (GUI)

The experiment plot is also used to see how well the measured data matches the current
model. Click Plot Model Response to display simulated signal data on the experiment
plots.

2-155
2 Parameter Estimation

The simulation results show that the model does not match the measured data and that
model parameters need to be estimated.

Estimated Parameters

The next step is to define the parameter to estimate. Click Select Parameters to open
a dialog to select model parameters to estimate. In this example we have preselected the
four unknown parameters; the butterfly valve inertia, J; the damping coefficient, c; the
return spring constant, k; and the time lag in motor response, input_delay.

2-156
 Estimate Model Parameter Values (GUI)

Since we know from physical insight that all of these parameters have positive
values, we set their lower limits to zero. We also put an upper bound of 0.1 sec on the
input_delay parameter. We can also select an initial value for the parameters. These
may come from some quick calculations of some formulas that determine the parameters.

Click the right arrow toggle button to modify the parameter minimum and maximum
bounds.

2-157
2 Parameter Estimation

The Estimation Task

With the parameters for estimation selected we select experiments to use for estimation.
Click Select Experiments and select EstimationData for estimation.

2-158
 Estimate Model Parameter Values (GUI)

We are now almost ready to start our estimation but first create plots to monitor the
estimation progress. Click Add Plot and select Parameter Trajectory. This creates
a plot that shows how the estimated parameter values change during estimation. Click
the View tab to layout the plots so that the Experiment plot:EstimationData and
Iteration plot 1 are both visible.

2-159
2 Parameter Estimation

Click the Estimate button to start the estimation. You can modify estimation options by
setting the Cost Function combobox and clicking More Options....

While the estimation is running the plots update and a dialog showing estimation
progress appears. The progress dialog shows the estimation iterations, the number of
times the model has been evaluated (F-count), and the estimation cost at each iteration.

2-160
Estimate Model Parameter Values (GUI)

2-161
2 Parameter Estimation

After a number of iterations the estimation converges and terminates. The model is
updated with the estimated parameters and the estimation results are saved in the
data browser. Right click EstimatedParams and select Open... to see details of the
estimation result.

2-162
 Estimate Model Parameter Values (GUI)

Validation

It is important to validate the estimation results against other experiments. A successful

estimation will not only match the experimental data that was used for estimation but
also other independent measured data that were collected in experiments.

Click the Validation tab and click Select Experiments to select experiments for
validation. Select both ValidationData1 and ValidationData2 for validation.

2-163
2 Parameter Estimation

Click Select Results to select the estimation result(s) to use for validation. Select
EstimatedParams and deselect Use current parameter values.

Click Validate to validate the estimation result against the validation experiments.
Validation simulates the model using the estimated parameters and selected
experiments and creates plots showing the measured and simulation data. Use the
View tab to layout the plots so that the Experiment plot:ValidationData1 and
Experiment plot:ValidationData2 are both visible.

2-164
 Estimate Model Parameter Values (GUI)

The valdiation plots confirm that our estimation was successful, showing that the
estimated parameters are robust enough to handle a variety of inputs.

Related Examples

To learn how to estimate model parameters using the sdo.optimize command, see
"Estimate Model Parameter Values (Code)".

Close the model.

2-165
2 Parameter Estimation

Estimate Model Parameters Per Experiment (GUI)

This example shows how to use multiple experiments to estimate a mix of model
parameter values; some that are estimated using all the experiments and others that
are estimated using individual experiments. The example also shows how to configure
estimation experiments with experiment dependent parameter values.

You estimate the parameters of a rechargeable battery based on data collected in

experiments that discharge and charge the battery.

Open the Model and Get Experimental Data

This example estimates parameters of a simple, rechargeable battery model,

sdoBattery. The model input is the battery current and the model output, the battery
terminal voltage, is computed from the battery state-of-charge.

open_system('sdoBattery');

The model is based on the equation

2-166
 Estimate Model Parameters Per Experiment (GUI)

Where:

• is the battery terminal voltage in Volts.

• is the battery constant voltage in Volts.

• is the battery polarization resistance in Ohms.

• is the maximum battery capacity in Ampere-Hour.

• is the battery charge state, with 1 being fully charged and 0 zero charge.The battery
state-of-charge is computed from the integral of the battery current with a +ve
current indicating discharge and a -ve current indicating charging. The battery initial
state-of-charge is specified by in Ampere-Hour.

• is the voltage drop when charging, expressed as a fraction of the battery

constant voltage. When the battery is discharging this value is zero.

V, K, Qmax, Q0, and Loss are variables defined in the model workspace.

Estimation Experiment Data

A 1.2V (6500mAh) battery was subjected to a discharge and a charging experiment. This
experiment data has been loaded into a preconfigured estimation tool session.

From the sdoBattery model click Analysis and select the Parameter Estimation...
menu item to launch the Parameter Estimation tool. From the Parameter Estimation
tool click Open Session and select Open from model workspace and open the
sdoBattery_sdosession session. The measured charge and discharge experiment data
are loaded and plotted.

Click the View tab to layout the plots so that the Experiment plot:Charge_Exp and
Experiment plot:DCharge_Exp are both visible. Click Plot Model Response to see
how well the model simulation matches the measured experiment data.

2-167
2 Parameter Estimation

The plots show that the battery initial charge Q0 is not set correctly for the Charge_Exp
experiment and that the model V, K, and Loss parameters need to be estimated.

Setting Experiment Parameter Values

The previous plot indicates that the Charge_Exp battery initial charge, Q0, is not set
correctly. Add the initial charge to both experiments.

Right click Charge_Exp and select Edit.... A dialog to edit the experiment opens.

2-168
 Estimate Model Parameters Per Experiment (GUI)

Click Select Parameters to open a dialog to add model parameters to the experiment.
Select Loss and Q0 to add to the experiment. Select Loss as we need to estimate this
parameter using only the Charge_Exp experiment. Click Ok to add the Q0 and Loss
parameters to the experiment.

2-169
2 Parameter Estimation

Set the battery initial charge Q0 in the Charge_Exp to 0, i.e. there is no initial charge.

2-170
 Estimate Model Parameters Per Experiment (GUI)

Similarly add the battery initial charge Q0 to the DCharge_Exp experiment and set the
initial charge to 6.5., i.e. for this experiment there is an initial charge.

2-171
2 Parameter Estimation

Now that the experiements are updated with the correct initial battery charge click Plot
Model Response to simulate the model and compare measured and simuated data.

2-172
 Estimate Model Parameters Per Experiment (GUI)

The experiment plots show that the experiment initial conditions match but the battery
response does not. The next step is to estimate the K and V model parameters.

Select Estimation Parameters

The previous plot showed that the model response does not match the measured data and
we need to etimate the model V and K parameters.

Click Select Parameters to open a dialog to select model parameters.

2-173
2 Parameter Estimation

The upper portion of the select parameters dialog has a section for parameters that
are tuned using all experiments. Click Select Parameters and add the V and K model
parameters to the estimated parameters. Set the V minimum to 0 and the maximumn to
2, similarly set the K minimum to 1e-6 and maximum to 0.1.

2-174
Estimate Model Parameters Per Experiment (GUI)

2-175
2 Parameter Estimation

The lower section of the dialog has a section for initial states and parameters that are
tuned using individual experiments.

For the Charge_Exp we tune the Loss parameter and set its minimum to 0 maximum
to 0.5. The battery initial charge Q0 is fixed to 0 and should not be estimated; uncheck
Estimate.

2-176
 Estimate Model Parameters Per Experiment (GUI)

Select DCharge_Exp from the Experiment combobox to view the parameter settings for
the DCharge_Exp experiment. The battery initial charge Q0 is fixed to 6.5 and should
not be estimated; uncheck Estimate

Estimate Parameter Values

The experiments and estimated parameters are configured and we a ready to run the
estimation. First create a plot to monitor the estimation progress. Click Add Plot
and select Parameter Trajectory. This creates a plot that shows how the estimated
parameter values change during estimation. Click the View tab to layout the plots so
that the experiment and parameter trajectory plots are all visible.

2-177
2 Parameter Estimation

Click the Estimate button to start the estimation. You can modify estimation options by
setting the Cost Function combobox and clicking More Options....

While the estimation is running the plots update and a dialog showing estimation
progress appears. The progress dialog shows the estimation iterations, the number of
times the model has been evaluated (F-count), and the estimation cost at each iteration.

2-178
Estimate Model Parameters Per Experiment (GUI)

2-179
2 Parameter Estimation

After a number of iterations the estimation converges and terminates. The experiement
plots show the measured and simlation data matching well. The EstimatedParams plot
shows the V, K, and Loss parameters changing during the estimation; the scale of V and
K, Loss are different, right click on the plot and select Show scaled values to see how
all the parameters changed from their original values.

Related Examples

To learn how to estimate parameters per expriment using the sdo.optimize command,
see "Estimate Model Parameters Per Experiment (GUI)".

Close the model

bdclose('sdoBattery')

2-180
 Estimate Model Parameters and Initial States (GUI)

Estimate Model Parameters and Initial States (GUI)

This example shows how to estimate the physical parameters - mass (m), spring constant
(k) and damping (b) of a simple mass-spring-damper model. This example illustrates the
significance of initial state estimation.

Simulink® Model of the Mass Spring Damper System

The Simulink model for the mass-spring-damper system, msd_system, is shown below.

The model's output is the displacement response (position) of the mass in a mass-spring-
damper system, subject to a constant force (F), and an initial displacement (x0). x0 is the
initial condition of the Position integrator block. Run the simulation once to observe the
response of the model to a nominal set of parameter values.

2-181
2 Parameter Estimation

Experimental Data Sets

For estimation of the model parameters (m, b and k), two sets of experimental data are
used. These data sets were obtained using two different initial positions (0.1 and 0.3),
and contain additive noise. A plot of these data sets is shown below (orange and cyan
curves), along with the simulated response (yellow curve) of the Simulink model for
x0=-0.1 and a nominal set of parameter values (m=8, k=500, b=100).

Estimation of Model Parameters

The model has three parameters (k, b, m) that appear in the Gain blocks of the Simulink
model msd_system. We estimate these parameters using Parameter Estimation.

Double-click the Parameter Estimation GUI with preloaded data block in

the model to open a pre-configured estimation GUI session. The experimental data
sets are already loaded in the project (data_exp1 and data_exp2). Click the View
tab to layout the plots so that the Experiment plot:data_exp1 and Experiment
plot:data_exp2 are both visible. Click Plot Model Response to simulate the model
for the two experiments. The plots show that the model simulation does not match the
experiment data.

2-182
 Estimate Model Parameters and Initial States (GUI)

Parameter Estimation with No State Estimation

The tool has been configured to estimate the model parameters using both data_exp1
and data_exp2 experiments, click Select Parameters to see the selected parameters
and Select Experiments to see the experiments selected for estimation.

Click Estimate to start the estimation. You can modify estimation options by setting the
Cost Function combobox and clicking More Options....

While the estimation is running the plots update and a dialog showing estimation
progress appears. The progress dialog shows the estimation iterations, the numberof
times the model has been evaluated (F-count), and the estimation cost at each iteration.

2-183
2 Parameter Estimation

2-184
 Estimate Model Parameters and Initial States (GUI)

After a number of iterations the estimation converges and terminates. The model is
updated with the estimated parameters and the estimation results are saved in the data
browser.

The data_exp1 and data_exp2 experiment plots show that the model parameters
have been tuned to match the measured experiment data as closely as possible. The
simulated measured signals match well from the 2 second mark onward but don't match
well before 2 seconds. The simulation results for both experiments start at -0.1. This is
the initial condition of the model which was not estimated; these plots show that the
initial condition should also be estimated.

Parameter Estimation with Initial State Estimation

2-185
2 Parameter Estimation

The data_exp1 and data_exp2 experiments specify the measured output data but as
seen above must also specify the model initial state. We now add the initial states to the
experiments and estimate them.

Right click data_exp1 and select Edit... to open a dialog to configure the experiment.

Click Select Initial States and select the position state. Click OK to close the state
selector and add the selected state to the experiment.

2-186
Estimate Model Parameters and Initial States (GUI)

2-187
2 Parameter Estimation

Right click data_exp2 and select Edit.. and add the position state to the experiment.

The experiments are now configured to include initial states that can be estimated. Click
Select Parameters.

2-188
 Estimate Model Parameters and Initial States (GUI)

The upper portion of the select parameters dialog has a section for parameters that are
tuned using all experiments selected for estimation. The lower section of the dialog has
a combo-box to select an experiment and widgets to specify initial states and parameters
that are tuned using only the selected experiment. For this problem the data_exp1 and
data_exp2 experiments estimate the model initial state for each experiment.

Now we are ready to start our estimation but first create plots to monitor the estimation
progress. Click Add Plot and select Parameter Trajectory, right click the plot and
select Show scaled values. This creates a plot that shows how the estimated parameter
values change during estimation. Click the View tab to layout the plots so that the

2-189
2 Parameter Estimation

Experiment plot:data_exp1, Experiment plot:data_exp2, and Iteration

plot 1 are both visible.

Click the Estimate button to start the estimation.

After a number of iterations the estimation converges and terminates. The data_exp1
and data_exp2 experiment plots show how estimating the initial value inproves the
estimation fit. The EstimatedParams plot shows the estimated initial state for the two
experiments, the plot also shows that the estimated k value did not change while b and m
changed slightly. You can confirm this by clicking EstimatedParams and examining the
preview pane and then clicking EstimatedParams1 and examining the preview pane.

2-190
 Estimate Model Parameters and Initial States (GUI)

Alternatively right click EstimatedParams and select Open... to open a dialog to view
the results.

This example shows that it is important to independently estimate initial states for each
experiment in order to obtain the correct estimates of the model parameters.

Related Examples

To learn how to estimate model parameters and initial states using the sdo.optimize
command, see "Estimate Model Parameters and Initial States (Code)".

Close the model

2-191
2 Parameter Estimation

Generate MATLAB Code for Parameter Estimation Problems (GUI)

This example shows how to automatically generate a MATLAB function to solve a
Parameter Estimation problem. You use the Parameter Estimation tool to define an
estimation problem for a mass-spring-damper and generate MATLAB code to solve this
estimation problem.

Mass-Spring-Damper Estimation Problem

The "Estimate Model Parameters and Initial States" example shows how to use the
Parameter Estimation tool to estimate parameters of a mass-spring-damper model. In
this example we load a pre-configured Parameter Estimation tool session based on that
example.
load sdoMassSpringDamper_sdosession
spetool(SDOSessionData)

2-192
Generate MATLAB Code for Parameter Estimation Problems (GUI)

2-193
2 Parameter Estimation

Generate MATLAB Code

From the Estimate list, select Generate MATLAB Code.

2-194
 Generate MATLAB Code for Parameter Estimation Problems (GUI)

The generated code is added to the MATLAB editor as an unsaved MATLAB function.

Examine the generated code. Significant code portions are:

2-195
2 Parameter Estimation

• Specify Model Parameters to Estimate - Definition of the model parameters

being estimated.

• Define the Estimation Experiments - Definition of the measured and expected

signal dat to use for estimation.

• Create Estimation Objective Function - Creation of an anonymous function

that calls the subfunction msd_system_optFcn, which evaluates the model using
each experiment and compares simulation and measured experiment outputs. This
anonymous function is called by sdo.optimize at each iteration of the optimiztion
problem to solve the estimation problem.

• Estimate the Parameters - Solve the estimation prioblem using the

sdo.optimize command.

Select Save from the MATLAB editor to save the generated function.

Run Generated Code

Run the generated function.

The first output argument, pOpt, contains the optimized parameter values and the
second output argument, optInfo, contains optimization information.

Modify the Generated Code

You can:

2-196
 Generate MATLAB Code for Parameter Estimation Problems (GUI)

• Modify the generated spe_msd_system function to include or exclude new

experiments or change estimation options.

• Call the generated spe_msd_system function with a different set of parameters to

estimate.

For details on how to write an objective/constraint function to use with the

sdo.optimize command, type help sdoExampleCostFunction at the MATLAB
command prompt.

Close the model

delete(sdotool('msd_system','ParameterEstimation'))
bdclose('msd_system')

2-197
2 Parameter Estimation

Improving Optimization Performance using Fast Restart (GUI)

This example shows how to use the Fast Restart feature of Simulink® to speed up
optimization of a model. You use Fast Restart to estimate the parameters of an engine
throttle model.

How Fast Restart speeds up the optimization

Simulation of Simulink models requires that the model be compiled before it is

simulated. In this context compilation of a model means analyzing and formatting the
model so that it can be simulated. The idea of Fast Restart is to perform the model
compilation once and reuse the compiled information for subsequent simulations, see
"How Fast Restart Improves Iterative Simulations" in the Simulink documentation for a
description of Fast Restart.

During optimization the model is repeatedly simulated (often tens or hundreds of

times) Fast Restart means that the model is only compiled once for these simulation in
comparison to non-fast restart where the model is recompiled each time.

Models where compilation is a significant portion of overall simulation time benefit the
most from Fast Restart. Further once a model is compiled not all model parameters can
be changed, specifically only tunable parameters can be changed, see "Factors Affecting
Fast Restart" in the Simulink documentation for more information.

Open model and Estimation Tool

Load the model and click the "Parameter Estimation with preloaded data" block to load
a preconfigured parameter estimation problem. The goal is to tune the parameters of an
engine throttle model to match measured data. For details on the problem setup see the
"Estimate Model Parameter Values (GUI)" example.

open_system('spe_engine_throttle')

2-198
Improving Optimization Performance using Fast Restart (GUI)

2-199
2 Parameter Estimation

Estimate Without Using Fast Restart

To compare the estimation with and without fast restart, change the estimation options
in the tool to not update the model with estimated values.

Click Model Options... in the Parameter Estimation tool and clear Update model at
end of estimation, and select Save estimated values as new estimation result.

2-200
 Improving Optimization Performance using Fast Restart (GUI)

Click Estimate in the tool to estimate the model parameter values. The estimation
progress report shows the estimation start and end time.

2-201
2 Parameter Estimation

Estimate Using Fast Restart

To configure the model to use Fast Restart during simulation, click Enable Fast
Restart in the Simulink model.

2-202
 Improving Optimization Performance using Fast Restart (GUI)

Click Estimate in the Parameter Estimation tool. The estimation progress report shows
the estimation start and end time. Note the reduction in total estimation time compared
to the estimation without using fast restart, in this case around 28 seconds or 45% of the
original estimation time.

2-203
2 Parameter Estimation

Related Examples

The Generate MATLAB Code feature of the Parameter Estimation and Response
Optimization tool will generate the MATLAB code to configure the model for fast restart
if the tool is configured to use fast restart.

To learn how to use Fast Restart at the command line see "Improving Optimization
Performance using Fast Restart (Code)".

Close the model.

2-204
 Improving Optimization Performance using Fast Restart (GUI)

bdclose('spe_engine_throttle')

Related Examples
• “Improving Optimization Performance using Fast Restart (Code)” on page 2-206
• “Use Fast Restart Mode During Response Optimization” on page 3-227
• “Use Fast Restart Mode During Parameter Estimation” on page 2-67
• “Use Fast Restart Mode During Sensitivity Analysis” on page 4-17

More About
• “Ways to Speed Up Design Optimization Tasks”

2-205
2 Parameter Estimation

Improving Optimization Performance using Fast Restart (Code)

This example shows how to use the Fast Restart feature of Simulink® to speed up
optimization of a model. You use Fast Restart to estimate the parameters of an engine
throttle model.

How Fast Restart speeds up the optimization

Simulation of Simulink models requires that the model be compiled before it is

simulated. In this context compilation of a model means analyzing and formatting the
model so that it can be simulated. The idea of Fast Restart is to perform the model
compilation once and reuse the compiled information for subsequent simulations, see
"How Fast Restart Improves Iterative Simulations" in the Simulink documentation for a
description of Fast Restart.

During optimization the model is repeatedly simulated (often tens or hundreds of

times) Fast Restart means that the model is only compiled once for these simulation in
comparison to non-fast restart where the model is recompiled each time.

Models where compilation is a significant portion of overall simulation time benefit the
most from Fast Restart. Further once a model is compiled not all model parameters can
be changed, specifically only tunable parameters can be changed, see "Factors Affecting
Fast Restart" in the Simulink documentation for more information.

Open Model

Load the engine throttle model. The goal is to tune the parameters of the model to match
measured data. For details on the problem setup see the "Estimate Model Parameter
Values (GUI)" example.

open_system('spe_engine_throttle')

2-206
 Improving Optimization Performance using Fast Restart (Code)

Define the Estimation Problem Data

This examples focusses on the command line interface for using Fast Restart during
estimation. For a detailed description of the estimation command line interface see
"Estimate Model Parameter Values (Code)".

Specify the model parameter values to estimate and any parameter bounds .

p = sdo.getParameterFromModel('spe_engine_throttle',{'J','c','input_delay','k'});
p(1).Minimum = 0;
p(2).Minimum = 0;
p(3).Minimum = 0;
p(3).Maximum = 0.1;
p(4).Minimum = 0;

Define the estimation experiment. The measured experiment data is loaded from the
sdoFastRestart_ExperimentData MATLAB file. The MATLAB file contains a
Input_SignalData and Output_SignalData variable specifying the experiment input
and output signal data.

load spe_engine_throttle_ExperimentData

2-207
2 Parameter Estimation

Exp = sdo.Experiment('spe_engine_throttle');
Input = Simulink.SimulationData.Signal;
Input.Values = Input_SignalData;
Input.BlockPath = 'spe_engine_throttle/Input';
Input.PortType = 'inport';
Input.PortIndex = 1;
Input.Name = 'spe_engine_throttle/Input:1';
Exp.InputData = Input;
Output = Simulink.SimulationData.Signal;
Output.Values = Output_SignalData;
Output.BlockPath = 'spe_engine_throttle/Throttle';
Output.PortType = 'outport';
Output.PortIndex = 1;
Output.Name = 'spe_engine_throttle/Throttle:1';
Exp.OutputData = Output;

Create a model simulator from the experiment

Simulator = createSimulator(Exp);

Configure the Simulator for Fast Restart

The simulator controls whether the model is simulated using fast restart or not. The
fastRestart command is used to configure the simulator to use Fast Restart.

The spe_engine_throttle model uses a variable-step solver, and may not output
values at the times in the measured experiment data. To output values at the times of
the measured data, use the set_param command to specify the model logging output
times as a workspace variable. In the estimation objective function, the variable is then
used to specify output times to be the same as the measured experiment data. The model
OutputTimes is set before configuring the simualtor for fast restart, as once the model is
configured for fast restart, the model logging configuration can not change.
set_param('spe_engine_throttle','OutputOption','SpecifiedOutputTimes','OutputTimes','Ou
Simulator = fastRestart(Simulator,'on');

The simulator can now be used during estimation, and the model will be simulated using
fast restart.

Run the Estimation

Create an estimation objective function to evaluate how closely the simulation

output, generated using the estimated parameter values, matches the
measured data. Use an anonymous function with one argument that calls the

2-208
 Improving Optimization Performance using Fast Restart (Code)

spe_engine_throtle_Objective function. The spe_engine_throttle_Objective

function includes the Simulator argument that has been configured to use fast restart.
optimfcn = @(P) spe_engine_throttle_Objective(P,Simulator,Exp);

Set the optimization options, and run the optimization.

Options = sdo.OptimizeOptions;
Options.Method = 'lsqnonlin';
[pOpt,Info] = sdo.optimize(optimfcn,p,Options);

Optimization started 31-Jul-2015 06:08:42

Step-size First-order
Iter F-count f(x) optimality
0 9 32.048 1
1 18 12.24 0.6495 18
2 27 3.59416 0.3919 8.65
3 36 1.11975 0.1879 3.11
4 45 0.649091 0.1966 1.25
5 54 0.287581 1.297 1.15
6 63 0.147631 0.2294 0.403
7 72 0.0874265 0.5347 0.1
8 81 0.0677538 0.3657 0.266
9 90 0.0677538 10 0.266
10 99 0.0677538 2.293 0.266
11 108 0.0673433 0.1971 0.185
Local minimum possible.

lsqnonlin stopped because the final change in the sum of squares relative to
its initial value is less than the selected value of the function tolerance.

Restore the Model

Restore the simulator fast restart settings. This clears the logging and other settings
used for the optimization problem.
Simulator = fastRestart(Simulator,'off');
set_param('spe_engine_throttle','OutputOption','RefineOutputTimes','OutputTimes','[]');

Related Examples

You can also generate code to configure you model for fast restart in the Parameter
Estimation and Response Optimization tools. Configure the model for fast restart as

2-209
2 Parameter Estimation

described in "Improving Optimization Performance using Fast Restart (GUI)". Then use
the Generate MATLAB Code feature of the tool.

Close the model.

bdclose('spe_engine_throttle')

See Also
sdo.SimulationTest | fastRestart

Related Examples
• “Improving Optimization Performance using Fast Restart (GUI)” on page 2-198
• “Use Fast Restart Mode During Response Optimization” on page 3-227
• “Use Fast Restart Mode During Parameter Estimation” on page 2-67
• “Use Fast Restart Mode During Sensitivity Analysis” on page 4-17

More About
• “Ways to Speed Up Design Optimization Tasks”

2-210
 3

Response Optimization

• “How the Optimization Algorithm Formulates Minimization Problems” on page

3-3
• “Specify Signals to Log” on page 3-12
• “Specifying Step Response Characteristics” on page 3-13
• “Specifying Custom Requirements” on page 3-17
• “Move Constraints” on page 3-20
• “Specify Time-Domain Design Requirements” on page 3-23
• “Edit Design Requirements” on page 3-37
• “Specify Frequency-Domain Design Requirements” on page 3-39
• “Specify Design Variables” on page 3-61
• “Update Model with Design Variables Set” on page 3-66
• “General Options” on page 3-68
• “Optimization Options” on page 3-72
• “Create Linearization I/O Sets” on page 3-77
• “Linearization Options” on page 3-79
• “Plots in the Response Optimization Tool” on page 3-82
• “Compare Requirements and Design Variables Using Spider Plot” on page 3-88
• “Export Design Variable Values for Specific Iteration” on page 3-91
• “Design Optimization to Meet Time- and Frequency-Domain Requirements (GUI)” on
page 3-93
• “Design Optimization to Meet a Custom Objective (GUI)” on page 3-108
• “Design Optimization to Meet a Custom Objective (Code)” on page 3-126
• “Design Optimization to Meet Custom Signal Requirements (GUI)” on page 3-135
• “Design Optimization to Meet Frequency-Domain Requirements (GUI)” on page
3-141
• “Specify Custom Signal Objective with Uncertain Variable (GUI)” on page 3-160
3 Response Optimization

• “Design Optimization with Uncertain Variables (Code)” on page 3-171

• “Generate MATLAB Code for Design Optimization Problems (GUI)” on page 3-181
• “Skip Model Simulation Based on Parameter Constraint Violation (GUI)” on page
3-188
• “Optimizing Parameters for Robustness” on page 3-201
• “How to Use Accelerator Mode During Simulations” on page 3-213
• “Speed Up Response Optimization Using Parallel Computing” on page 3-215
• “How to Use Parallel Computing for Response Optimization” on page 3-219
• “Use Fast Restart Mode During Response Optimization” on page 3-227
• “Optimization Does Not Make Progress” on page 3-230
• “Optimization Convergence” on page 3-232
• “Optimization Speed and Parallel Computing” on page 3-235
• “Undesirable Parameter Values” on page 3-238
• “Reverting to Initial Parameter Values” on page 3-240
• “Manage Response Optimization Tool Session” on page 3-241
• “Optimizing Time-Domain Response of Simulink® Models Using Parallel Computing”
on page 3-243
• “Design Optimization to Meet Frequency-Domain Requirements (Code)” on page
3-252

3-2
 How the Optimization Algorithm Formulates Minimization Problems

How the Optimization Algorithm Formulates Minimization

Problems
When you optimize parameters of a Simulink model to meet design requirements,
Simulink Design Optimization software automatically converts the requirements into
a constrained optimization problem and then solves the problem using optimization
techniques. The constrained optimization problem iteratively simulates the Simulink
model, compares the results of the simulations with the constraint objectives, and uses
optimization methods to adjust tuned parameters to better meet the objectives.

This topic describes how the software formulates the constrained optimization problem
used by the optimization algorithms. For each optimization algorithm, the software
formulates one of the following types of minimization problems:

• Feasibility
• Tracking
• Mixed feasibility and tracking

For more information on how each optimization algorithm formulates these problems,
see:

• “Gradient Descent Method Problem Formulations” on page 3-7

• “Simplex Search Method Problem Formulations” on page 3-8
• “Pattern Search Method Problem Formulations” on page 3-9
• “Gradient Computations” on page 3-10

Feasibility Problem and Constraint Formulation

Feasibility means that the optimization algorithm finds parameter values that satisfy
all constraints to within specified tolerances but does not minimize any objective or cost
function in doing so.

In the following figure, x1, x3, and xn represent a combination of parameter values P1 and
P2 and are feasible solutions because they do not violate the lower bound constraint.

3-3
3 Response Optimization

P2

x2 x3
x1
xn
Violation Lower
Bound

P1

In a Simulink model, you constrain a signal by specifying lower and upper bounds in a
Check block (Check Step Response Characteristics, ...) or a requirement object
(sdo.requirements.StepResponseEnvelope, ...), as shown in the following figure.

Upper Bounds

Lower Bounds

These constraints are piecewise linear bounds. A piecewise linear bound ybnd with n edges
can be represented as:

Ï y1 ( t) t1 £ t £ t2
Ô y (t) t £ t £ t
Ô
ybnd ( t) = Ì 2 2 3
,
Ô M M
ÔÓ yn (t) tn £ t £ tn+1

The software computes the signed distance between the simulated response and the edge.
The signed distance for lower bounds is:

3-4
 How the Optimization Algorithm Formulates Minimization Problems

È max ybnd - ysim ˘

Ít1 £t£t2 ˙
c = Í max ybnd - ysim ˙ ,
Ít2 £t£t3 ˙
Í ˙
Í max ybnd - ysim ˙
În
t £ t £ tn +1 ˚

where ysim is the simulated response and is a function of the parameters being optimized.

The signed distance for upper bounds is:

È max ysim - ybnd ˘

Í t1 £t£t2 ˙
c = Í max ysim - ybnd ˙ .
Í t2 £t£ t3 ˙
Í ˙
Í max ysim - ybnd ˙
Îtn £t£ tn +1 ˚

At the command line, opt_fcn supplies c directly from the Cleq field of vals.

If all the constraints are met (c ≤ 0) for some combination of parameter values, then that
solution is said to be feasible. In the following figure, x1 and x3 are feasible solutions.

P2

Violation
x2
x3
x1
xn

P1

When your model has multiple requirements or vector signals feeding a requirement, the
constraint vector is extended with the constraint violations for each signal and bound:

C = [ c1 ; c2 ;L ; cn ] .

3-5
3 Response Optimization

Tracking Problem
In addition to lower and upper bounds, you can specify a reference signal in a Check
Against Reference block or sdo.requirements.SignalTracking object, which the
Simulink model output can track. The tracking objective is a sum-squared-error tracking
objective.

You specify the reference signal as a sequence of time-amplitude pairs:

yref ( tref ), tref Π{Tref 0 , Tref 1 ,L , TrefN }.

The software computes the simulated response as a sequence of time-amplitude pairs:

ysim ( tsim ), tsim Π{Tsim 0 , Tsim1 ,L , TsimN },

where some values of tsim may match the values of tref.

A new time base, tnew, is formed from the union of the elements of tref and tsim. Elements
that are not within the minimum-maximum range of both tref and tsim are omitted:

tnew = {t : tsim » tref }

Using linear interpolation, the software computes the values of yref and ysim at the time
points in tnew and then computes the scaled error:

e( tnew ) =
( ysim (tnew ) - yref (tnew )) .
max yref
tnew

Finally, the software computes the weighted, integral square error:

f = w ( t ) e( t) 2 dt.
Ú

Note: The weight w(t) is 1 by default. You can specify a different value of weight only at
the command line.

3-6
 How the Optimization Algorithm Formulates Minimization Problems

When your model has requirements or vector signals feeding a requirement, the tracking
objective equals the sum of the individual tracking integral errors for each signal:

F= Â fi .

Gradient Descent Method Problem Formulations

The Gradient Descent method uses the Optimization Toolbox function fmincon to
optimize model parameters to meet design requirements.

Problem Type Problem Formulation

Feasibility Problem The software formulates the constraint C(x) as described in
“Feasibility Problem and Constraint Formulation” on page 3-3.

• If you select the maximally feasible solution option (i.e., the

optimization continues after an initial feasible solution is found),
the software uses the following problem formulation:

min g
[ x,g ]
s.t. C( x) £ g
x£ x£ x
g £0

γ is a slack variable that permits a feasible solution with C(x) ≤ γ

rather than C(x) ≤ 0.
• If you do not select the maximally feasible solution option (i.e., the
optimization terminates as soon as a feasible solution is found),
the software uses the following problem formulation:

min 0
x
s.t. C( x) £ 0
x£ x£ x

Tracking Problem The software formulates the tracking objective F(x) as described in
“Tracking Problem” on page 3-6 and minimizes the tracking
objective:

3-7
3 Response Optimization

Problem Type Problem Formulation

min F( x)
x
s.t. x £ x £ x

Mixed Feasibility and The software minimizes following problem formulation:

Tracking Problem
min F( x)
x
s.t. C( x) £ 0
x£x£x

Note: When tracking a reference signal, the software ignores the

maximally feasible solution option.

Simplex Search Method Problem Formulations

The Simplex Search method uses the Optimization Toolbox function fminsearch and
fminbnd to optimize model parameters to meet design requirements. fminbnd is used if
one scalar parameter is being optimized, otherwise fminsearch is used. You cannot use
parameter bounds x £ x £ x with fminsearch.

Problem Type Problem Formulation

Feasibility Problem The software formulates the constraint C(x) as described in
“Feasibility Problem and Constraint Formulation” on page 3-3
and then minimizes the maximum constraint violation:

min max ( C ( x ) )
x

Tracking Problem The software formulates the tracking objective F(x) as described in
“Tracking Problem” on page 3-6 and then minimizes the tracking
objective:

min F( x)
x

Mixed Feasibility and The software formulates the problem in two steps:
Tracking Problem

3-8
 How the Optimization Algorithm Formulates Minimization Problems

Problem Type Problem Formulation

1 Finds a feasible solution.

min max ( C ( x ) )
x

2 Minimizes the tracking objective. The software uses the results

from step 1 as initial guesses and maintains feasibility by
introducing a discontinuous barrier in the optimization objective.

min G ( x)
x
where
Ï• if max ( C( x) ) > 0
G( x) = Ì
Ó F ( x) otherwise.

Pattern Search Method Problem Formulations

The Pattern Search method uses the Global Optimization Toolbox function
patternsearch to optimize model parameters to meet design requirements.

Problem Type Problem Formulation

Feasibility Problem The software formulates the constraint C(x) as described in
“Feasibility Problem and Constraint Formulation” on page 3-3
and then minimizes the maximum constraint violation:

min max ( C ( x ) )
x
s.t. x£ x£ x

Tracking Problem The software formulates the tracking objective F(x) as described in
“Tracking Problem” on page 3-6 and then minimizes the tracking
objective:

min F( x)
x
s.t. x£ x£ x

3-9
3 Response Optimization

Problem Type Problem Formulation

Mixed Feasibility and The software formulates the problem in two steps:
Tracking Problem
1 Finds a feasible solution.

min max ( C ( x ) )
x
s.t. x£ x£ x

2 Minimizes the tracking objective. The software uses the results

from step 1 as initial guesses and maintains feasibility by
introducing a discontinuous barrier in the optimization objective.

min G( x)
x
s.t. x £ x £ x
where
Ï• if max ( C( x) ) > 0
G( x) = Ì
Ó F( x) otherwise.

Gradient Computations
For the Gradient descent (fmincon) optimization solver, the gradients are computed
using numerical perturbation:

Ê 1 ˆ
dx = 3 eps ¥ max Á| x|, xtypical ˜
Ë 10 ¯
dL = max ( x - dx, xmin )
dR = min ( x + d x, xmax )
FL = opt _ fcn(dL)
FR = opt _ fcn(dR)
dF ( FL - FR )
=
dx ( dL - dR )

• x is a scalar design variable.

3-10
 How the Optimization Algorithm Formulates Minimization Problems

• xmin is the lower bound of x.

• xmax is the upper bound of x.
• xtypical is the scaled value of x.
• opt_fcn is the objective function.

dx is relatively large to accommodate simulation solver tolerances.

If you want to compute the gradients in any other way, you can do so in the cost function
you write for performing design optimization programmatically. See sdo.optimize and
GradFcn of sdo.OptimizeOptions for more information.

3-11
3 Response Optimization

Specify Signals to Log

Design requirements require logged model signals. During optimization, the model is
simulated using the current value of the design variables and the logged signal is used to
evaluate the design requirements.

1 In the Response Optimization tool, select Signal in the New drop-down list. A
window opens where you select a signal to log.
2 In the Simulink model window, click the signal to which you want to add a
requirement.

The window updates and displays the name of the block and the port number where
the selected signal is located.
3
Select the signal and click to add it to the signal set.
4 In the Signal set box, enter a name for the selected signal set.

Click OK. A new variable, with the specified name, appears in the Data area of the
Response Optimization tool window.

See Also

• “Design Optimization to Track Reference Signal (GUI)”

• sdo.SimulationTest

3-12
 Specifying Step Response Characteristics

Specifying Step Response Characteristics

Specify Step Response Characteristics

To specify step response characteristics:

1 You can apply this requirement to either a signal or a linearization of your model.

In the Response Optimization Tool, click New. To apply this requirement to a

signal, select the Step Response Envelope entry in the New Time Domain
Requirement section of the New list. To apply this requirement to a linearization
of your model, select the Step Response Envelope entry in the New Frequency
Domain Requirement section of the New list. The latter option requires Simulink
Control Design software.

A window opens where you specify the step response requirements on a signal, or
system.
2 Specify a requirement name in the Name box.
3 Specify the step response characteristics:

% Overshoot
% Settling
Final Value

% Rise
Amplitude

% Undershoot
Initial Value

Rise Time Settling Time Time

3-13
3 Response Optimization

• Initial value: Input level before the step occurs

• Step time: Time at which the step takes place
• Final value: Input level after the step occurs
• Rise time: The time taken for the response signal to reach a specified percentage
of the step's range. The step's range is the difference between the final and initial
values.
• % Rise: The percentage used in the rise time.
• Settling time: The time taken until the response signal settles within a specified
region around the final value. This settling region is defined as the final step
value plus or minus the specified percentage of the final value.
• % Settling: The percentage used in the settling time.
• % Overshoot: The amount by which the response signal can exceed the final
value. This amount is specified as a percentage of the step's range. The step's
range is the difference between the final and initial values.
• % Undershoot: The amount by which the response signal can undershoot the
initial value. This amount is specified as a percentage of the step's range. The
step's range is the difference between the final and initial values.
4 Specify the signals or systems to be bound.

You can apply this requirement to a model signal or to a linearization of your

Simulink model (requires Simulink Control Design software).

• Apply this requirement to a model signal:

In the Select Signals to Bound area, select a logged signal to which you will
apply the requirement.

If you have already selected a signal to log, as described in “Specify Signals to

Log” on page 3-12, it appears in the list. Select the corresponding check-box.

If you haven’t selected a signal to log:

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

3-14
 Specifying Step Response Characteristics

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

Click OK. A new variable, with the specified name, appears in the Data area
of the Response Optimization tool window.
• Apply this requirement to a linear system.

Linear systems are defined by snapshot times at which the model is linearized
and sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a
vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in

the list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open

the Create linearization I/O set dialog box.

3-15
3 Response Optimization

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A variable with the specified requirement name appears in the Data area of the tool.
A graphical display of the requirement also appears in the Response Optimization
tool window.

Alternatively, you can use the Check Step Response Characteristics block to
specify step response bounds for a signal.

See Also

“Design Optimization to Meet Step Response Requirements (GUI)”

More About
• “Specify Time-Domain Design Requirements” on page 3-23
• “Specify Frequency-Domain Design Requirements” on page 3-39

3-16
 Specifying Custom Requirements

Specifying Custom Requirements

To specify custom requirements, such as minimizing system energy:

1 In the Response Optimization tool, select Custom Requirement in the New list. A
window opens where you specify the custom requirement.
2 Specify a requirement name in the Name box.
3 Specify the requirement type using the Type list.
4 Specify the name of the function that contains the custom requirement in the
Function box. The field must be specified as a function handle using @. The function

must be on the MATLAB path. Click to review or edit the function.

If the function does not exist, clicking opens a template MATLAB file.
Use this file to implement the custom requirement. The default function name is
myCustomRequirement.
5 (Optional) If you want to prevent the solver from considering specific parameter
combinations, select the Error if constraint is violated check box. Use this option
for parameter-only constraints.

During an optimization iteration, the solver evaluates requirements with this option
selected first.

• If the constraint is violated, the solver skips evaluating any remaining

requirements and proceeds to the next iterate.
• If the constraint is not violated, the solver evaluates the remaining requirements
for the current iterate. If any of the remaining requirements bound signals or
systems, then the solver simulates the model .

For more information, see “Skip Model Simulation Based on Parameter Constraint
Violation (GUI)” on page 3-188.

Note: If you select this check box, then do not specify signals or systems to bound. If
you do specify signals or systems, then this check box is ignored.
6 (Optional) Specify the signal or system, or both, to be bound.

3-17
3 Response Optimization

You can apply this requirement to model signals, or a linearization of your Simulink
model (requires Simulink Control Design software), or both.

Click Select Signals and Systems to Bound (Optional) to view the signal and
linearization I/O selection area.

• Apply this requirement to a model signal:

In the Signal area, select a logged signal to which you will apply the
requirement.

If you have already selected a signal to log, as described in “Specify Signals to

Log” on page 3-12, it appears in the list. Select the corresponding check box.

If you have not selected a signal to log:

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

3-18
 Specifying Custom Requirements

Click OK. A new variable, with the specified name, appears in the Data area
of the Response Optimization tool window.
• Apply this requirement to a linear system.

Linear systems are defined by snapshot times at which the model is linearized
and sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a
vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in

the list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open

the Create linearization I/O set dialog box. For more information on using
this dialog box, see “Create Linearization I/O Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

7 Click OK.

A new variable, with the specified name, appears in the Data area of the Response
Optimization tool. A graphical display of the requirement also appears in the
Response Optimization tool window.

Related Examples
• “Design Optimization to Meet a Custom Objective (GUI)” on page 3-108
• “Design Optimization to Meet Custom Signal Requirements (GUI)” on page 3-135
• “Specify Time-Domain Design Requirements” on page 3-23
• “Specify Frequency-Domain Design Requirements” on page 3-39

3-19
3 Response Optimization

Move Constraints
In this section...
“Move Constraints Graphically” on page 3-20
“Position Constraints Exactly” on page 3-21

Constraint-bound edges define time-domain constraints you would like to place on a

particular signal in your model. You can position these edges, which appear as a yellow
shaded region bordered by a black line, graphically or exactly.

Move Constraints Graphically

Use the mouse to click and drag edges in the amplitude versus time plot, as shown in the
following figure.

• To move a constraint edge boundary or to change the slope of a constraint edge,

position the pointer over a constraint edge endpoint, and press and hold down the left
mouse button. The pointer should change to a hand symbol. While still holding the
button down, drag the pointer to the target location, and release the mouse button.
Note that the edges on either side of the boundary might not maintain their slopes.
• To move an entire constraint edge up, down, left, or right, position the mouse pointer
over the edge and press and hold down the left mouse button. The pointer should

3-20
 Move Constraints

change to a four-way arrow. While still holding the button down, drag the pointer to
the target location, and release the mouse button. Note that the edges on either side
of the boundary might not maintain their slopes.

To move a constraint edge to a perfectly horizontal or vertical position, hold down the
Shift key while clicking and dragging the constraint edge. This causes the constraint
edge to snap to a horizontal or vertical position.

When moving constraint bound edges, it is sometimes helpful to display gridlines on the
axes for careful alignment of the constraint bound edges. To turn the gridlines on or off,
right-click within the axes and select Grid.

Note: You can move a lower bound constraint edge above an upper bound constraint
edge, or vice versa, but this produces an error when you attempt to run the optimization.

Position Constraints Exactly

To position a constraint edge exactly:

1 Position the pointer over the edge you want to move and right-click. Select Edit to
open the Edit Design Requirement dialog box.

3-21
3 Response Optimization

2 Specify the position of each constraint edge in the Time and Amplitude columns.

More About
• “Specify Time-Domain Design Requirements” on page 3-23
• “Specify Frequency-Domain Design Requirements” on page 3-39

3-22
 Specify Time-Domain Design Requirements

Specify Time-Domain Design Requirements

In this section...
“Specify Piecewise-Linear Lower and Upper Bounds” on page 3-23
“Specify Signal Property Requirements” on page 3-24
“Specify Step Response Characteristics” on page 3-13
“Track Reference Signals” on page 3-30
“Specify Custom Requirements” on page 3-32
“Edit Design Requirements” on page 3-35

Specify Piecewise-Linear Lower and Upper Bounds

To specify upper and lower bounds on a signal:

1 In the Response Optimization tool, select Signal Bound in the New drop-down list.
A window opens where you specify upper or lower bounds on a signal.
2 Specify a requirement name in the Name box.
3 Select the requirement type using the Type list.
4 Specify the edge start and end times and corresponding amplitude in the Time (s)
and Amplitude columns.
5
Click to specify additional bound edges.

Select a row and click to delete a bound edge.

6 In the Select Signals to Bound area, select a logged signal to apply the
requirement to.

If you have already selected signals, as described in “Specify Signals to Log” on page
3-12, they appear in the list. Select the corresponding check-box.

If you haven’t selected a signal to log:

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

3-23
3 Response Optimization

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

Click OK. A new variable, with the specified name, appears in the Data area of
the Response Optimization tool window.
7 Click OK.

A variable with the specified requirement name appears in the Data area of the tool.
A graphical display of the requirement also appears in the Response Optimization
tool window.
8 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can add a Check Custom Bounds block to your model to specify
piecewise-linear bounds.

Specify Signal Property Requirements

To specify signal property requirements:

3-24
 Specify Time-Domain Design Requirements

1 In the Response Optimization tool, select Signal Property in the New drop-down
list. The Create Requirement window opens where you specify signal property
requirements.
2 In the Name box, specify a requirement name.
3 In the Specify Property area, specify a signal property requirement using the
Property and Type lists and the Bound box.

Property List

For a signal S(t0 ),º, S(tN ) you can specify one of the following properties using the
Property list:

• Signal minimum — min(S)

• Signal maximum — max(S)
• Signal final value — S(t )
N

• Signal mean — mean(S)

• Signal median — median(S)
• Signal variance — variance(S)
• Signal interquartile range — Difference between the 75th and 25th
percentiles of the signal values.
• tN
Signal sum — Â S(i)
i =t 0

• tN
Signal sum square — Â S(i)2
i =t 0

• tN
Signal sum absolute — Â S(i)
i =t 0

Custom Signal Property

You can add a custom signal property to the Property list by editing the function
sdo.requirements.signalPropertyFcns.

3-25
3 Response Optimization

a At the MATLAB command prompt, enter edit

sdo.requirements.signalPropertyFcns.
b Add your signal property function to the FcnData cell array.

Your signal property function must be on the path.

4 In the Select Signals to Bound area, select the logged signal to which you want to
apply the requirement.

If you have already selected a signal, as described in “Specify Signals to Log” on

page 3-12, the signal appears in the list. Select the corresponding check box for that
signal.

If you have not selected a signal to log:

a
Click . The Create Signal Set window opens where you specify the logged
signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

3-26
 Specify Time-Domain Design Requirements

Click OK. A new variable, with the specified name, appears in the Data area of
the Response Optimization tool window.
5 Click OK.

A variable with the specified requirement name appears in the Data area of the tool.
An iteration plot depicting the signal property for each iteration also appears in the
Response Optimization tool window.

Specify Step Response Characteristics

To specify step response characteristics:

1 You can apply this requirement to either a signal or a linearization of your model.

In the Response Optimization Tool, click New. To apply this requirement to a

signal, select the Step Response Envelope entry in the New Time Domain
Requirement section of the New list. To apply this requirement to a linearization
of your model, select the Step Response Envelope entry in the New Frequency
Domain Requirement section of the New list. The latter option requires Simulink
Control Design software.

A window opens where you specify the step response requirements on a signal, or
system.
2 Specify a requirement name in the Name box.
3 Specify the step response characteristics:

3-27
3 Response Optimization

% Overshoot
% Settling
Final Value

% Rise
Amplitude

% Undershoot
Initial Value

Rise Time Settling Time Time

• Initial value: Input level before the step occurs

• Step time: Time at which the step takes place
• Final value: Input level after the step occurs
• Rise time: The time taken for the response signal to reach a specified percentage
of the step's range. The step's range is the difference between the final and initial
values.
• % Rise: The percentage used in the rise time.
• Settling time: The time taken until the response signal settles within a specified
region around the final value. This settling region is defined as the final step
value plus or minus the specified percentage of the final value.
• % Settling: The percentage used in the settling time.
• % Overshoot: The amount by which the response signal can exceed the final
value. This amount is specified as a percentage of the step's range. The step's
range is the difference between the final and initial values.
• % Undershoot: The amount by which the response signal can undershoot the
initial value. This amount is specified as a percentage of the step's range. The
step's range is the difference between the final and initial values.

3-28
 Specify Time-Domain Design Requirements

4 Specify the signals or systems to be bound.

You can apply this requirement to a model signal or to a linearization of your

Simulink model (requires Simulink Control Design software).

• Apply this requirement to a model signal:

In the Select Signals to Bound area, select a logged signal to which you will
apply the requirement.

If you have already selected a signal to log, as described in “Specify Signals to

Log” on page 3-12, it appears in the list. Select the corresponding check-box.

If you haven’t selected a signal to log:

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

3-29
3 Response Optimization

Click OK. A new variable, with the specified name, appears in the Data area
of the Response Optimization tool window.
• Apply this requirement to a linear system.

Linear systems are defined by snapshot times at which the model is linearized
and sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a
vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in

the list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open

the Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A variable with the specified requirement name appears in the Data area of the tool.
A graphical display of the requirement also appears in the Response Optimization
tool window.

Alternatively, you can use the Check Step Response Characteristics block to
specify step response bounds for a signal.

See Also

“Design Optimization to Meet Step Response Requirements (GUI)”

Track Reference Signals

Use reference tracking to force a model signal to match a desired signal.

To track a reference signal:

3-30
 Specify Time-Domain Design Requirements

1 In the Response Optimization tool, select Signal Tracking in the New drop-down
list. A window opens where you specify the reference signal to track.
2 Specify a requirement name in the Name box.
3 Define the reference signal by entering vectors, or variables from the workspace, in
the Time vector and Amplitude fields.

Click Update reference signal data to use the new amplitude and time vector as
the reference signal.
4 Specify how the optimization solver minimizes the error between the reference and
model signals using the Tracking Method list:

• SSE — Reduces the sum of squared errors

• SAE — Reduces the sum of absolute errors
5 In the Specify Signal to Track Reference Signal area, select a logged signal to
apply the requirement to.

If you already selected a signal to log, as described in “Specify Signals to Log” on

page 3-12, they appear in the list. Select the corresponding check-box.

If you haven’t selected a signal to log:

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

3-31
3 Response Optimization

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

Click OK. A new variable, with the specified name, appears in the Data area of
the Response Optimization tool window.
e Select the check-box corresponding to the signal and click OK.

A variable with the specified requirement name appears in the Data area of the tool.
A graphical display of the signal bound also appears in the Response Optimization
tool window.

Note: When tracking a reference signal, the software ignores the maximally feasible
solution option. For more information on this option, see “Selecting Optimization
Termination Options” on page 3-74.

Alternatively, you can use the Check Against Reference block to specify a reference
signal to track.

See Also

“Design Optimization to Track Reference Signal (GUI)”

Specify Custom Requirements

To specify custom requirements, such as minimizing system energy:

1 In the Response Optimization tool, select Custom Requirement in the New list. A
window opens where you specify the custom requirement.
2 Specify a requirement name in the Name box.
3 Specify the requirement type using the Type list.
4 Specify the name of the function that contains the custom requirement in the
Function box. The field must be specified as a function handle using @. The function

must be on the MATLAB path. Click to review or edit the function.

3-32
 Specify Time-Domain Design Requirements

If the function does not exist, clicking opens a template MATLAB file.
Use this file to implement the custom requirement. The default function name is
myCustomRequirement.
5 (Optional) If you want to prevent the solver from considering specific parameter
combinations, select the Error if constraint is violated check box. Use this option
for parameter-only constraints.

During an optimization iteration, the solver evaluates requirements with this option
selected first.

• If the constraint is violated, the solver skips evaluating any remaining

requirements and proceeds to the next iterate.
• If the constraint is not violated, the solver evaluates the remaining requirements
for the current iterate. If any of the remaining requirements bound signals or
systems, then the solver simulates the model .

For more information, see “Skip Model Simulation Based on Parameter Constraint
Violation (GUI)” on page 3-188.

Note: If you select this check box, then do not specify signals or systems to bound. If
you do specify signals or systems, then this check box is ignored.
6 (Optional) Specify the signal or system, or both, to be bound.

You can apply this requirement to model signals, or a linearization of your Simulink
model (requires Simulink Control Design software), or both.

Click Select Signals and Systems to Bound (Optional) to view the signal and
linearization I/O selection area.

• Apply this requirement to a model signal:

In the Signal area, select a logged signal to which you will apply the
requirement.

If you have already selected a signal to log, as described in “Specify Signals to

Log” on page 3-12, it appears in the list. Select the corresponding check box.

If you have not selected a signal to log:

3-33
3 Response Optimization

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

Click OK. A new variable, with the specified name, appears in the Data area
of the Response Optimization tool window.
• Apply this requirement to a linear system.

Linear systems are defined by snapshot times at which the model is linearized
and sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a
vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in

the list. Select the corresponding check box.

3-34
 Specify Time-Domain Design Requirements

If you have not created a linearization input/output set, click to open

the Create linearization I/O set dialog box. For more information on using
this dialog box, see “Create Linearization I/O Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

7 Click OK.

A new variable, with the specified name, appears in the Data area of the Response
Optimization tool. A graphical display of the requirement also appears in the
Response Optimization tool window.

See Also

• “Design Optimization to Meet a Custom Objective (GUI)” on page 3-108

• “Design Optimization to Meet Custom Signal Requirements (GUI)” on page 3-135

Edit Design Requirements

The Edit Design Requirement dialog box allows you to exactly position constraint
segments and to edit other properties of these constraints. The dialog box has two main
components:

• An upper panel to specify the constraint you are editing

• A lower panel to edit the constraint parameters

The upper panel of the Edit Design Requirement dialog box resembles the image in the
following figure.

In the context of the SISO Tool in Control System Toolbox™ software, Design
requirement refers to both the particular editor within the SISO Tool that contains the
requirement and the particular requirement within that editor. To edit other constraints
within the SISO Tool, select another design requirement from the drop-down menu.

Edit Design Requirement Dialog Box Parameters

The particular parameters shown within the lower panel of the Edit Design Requirement
dialog box depend on the type of constraint/requirement. In some cases, the lower panel

3-35
3 Response Optimization

contains a grid with one row for each segment and one column for each constraint
parameter. The following table summarizes the various constraint parameters.

Edit Design Requirement Dialog Box Parameters

Parameter Found in Description

Time Upper and lower time Defines the time range of a segment
response bounds on step and within a constraint/requirement.
impulse response plots
Amplitude Upper and lower time Defines the beginning and ending
response bounds on step and amplitude of a constraint segment.
impulse response plots
Slope (1/s) Upper and lower time Defines the slope, in 1/s, of a constraint
response bounds segment. It is an alternative method
of specifying the magnitude values.
Entering a new Slope value changes
any previously defined magnitude
values.
Final value Step response bounds Defines the input level after the step
occurs.
Rise time Step response bounds Defines a constraint segment for a
particular rise time.
% Rise Step response bounds The percentage of the step's range used
to describe the rise time.
Settling time Step response bounds Defines a constraint segment for a
particular settling time.
% Settling Step response bounds The percentage of the final value that
defines the settling region used to
describe the settling time.
% Overshoot Step response bounds  
% Undershoot Step response bounds Defines the constraint segments for a
particular percent undershoot.

3-36
 Edit Design Requirements

Edit Design Requirements

The Edit Design Requirement dialog box allows you to exactly position constraint
segments and to edit other properties of these constraints. The dialog box has two main
components:

• An upper panel to specify the constraint you are editing

• A lower panel to edit the constraint parameters

The upper panel of the Edit Design Requirement dialog box resembles the image in the
following figure.

In the context of the SISO Tool in Control System Toolbox software, Design
requirement refers to both the particular editor within the SISO Tool that contains the
requirement and the particular requirement within that editor. To edit other constraints
within the SISO Tool, select another design requirement from the drop-down menu.

Edit Design Requirement Dialog Box Parameters

The particular parameters shown within the lower panel of the Edit Design Requirement
dialog box depend on the type of constraint/requirement. In some cases, the lower panel
contains a grid with one row for each segment and one column for each constraint
parameter. The following table summarizes the various constraint parameters.

Edit Design Requirement Dialog Box Parameters

Parameter Found in Description

Time Upper and lower time Defines the time range of a segment
response bounds on step and within a constraint/requirement.
impulse response plots
Amplitude Upper and lower time Defines the beginning and ending
response bounds on step and amplitude of a constraint segment.
impulse response plots
Slope (1/s) Upper and lower time Defines the slope, in 1/s, of a constraint
response bounds segment. It is an alternative method

3-37
3 Response Optimization

Parameter Found in Description

of specifying the magnitude values.
Entering a new Slope value changes
any previously defined magnitude
values.
Final value Step response bounds Defines the input level after the step
occurs.
Rise time Step response bounds Defines a constraint segment for a
particular rise time.
% Rise Step response bounds The percentage of the step's range used
to describe the rise time.
Settling time Step response bounds Defines a constraint segment for a
particular settling time.
% Settling Step response bounds The percentage of the final value that
defines the settling region used to
describe the settling time.
% Overshoot Step response bounds  
% Undershoot Step response bounds Defines the constraint segments for a
particular percent undershoot.

3-38
 Specify Frequency-Domain Design Requirements

Specify Frequency-Domain Design Requirements

In this section...
“Specify Lower Bounds on Gain and Phase Margin” on page 3-39
“Specify Piecewise-Linear Lower and Upper Bounds on Frequency Response” on page
3-41
“Specify Bound on Closed-Loop Peak Gain” on page 3-43
“Specify Lower Bound on Damping Ratio” on page 3-45
“Specify Upper and Lower Bounds on Natural Frequency” on page 3-47
“Specify Upper Bound on Approximate Settling Time” on page 3-49
“Specify Piecewise-Linear Upper and Lower Bounds on Singular Values” on page
3-51
“Specify Step Response Characteristics” on page 3-13
“Specify Custom Requirements” on page 3-32

Specify Lower Bounds on Gain and Phase Margin

To specify lower bounds on the gain and phase margin of a linear system:

1 In the Response Optimization tool, select Gain and Phase Margin in the New list.
A window opens where you specify lower bounds on the gain and phase margin of
your linear system.
2 Specify a requirement name in Name.
3 Specify bounds on the gain margin or phase margin, or both.

3-39
3 Response Optimization

• Gain margin — Amount of gain increase or decrease required to make the loop
gain unity at the frequency where the phase angle is –180°.
• Phase margin — Amount of phase increase or decrease required to make the
phase angle –180° when the loop gain is 1.0

To specify a lower bound on the gain margin or phase margin, or both, select the
corresponding check box and enter the lower bound value.
4 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

3-40
 Specify Frequency-Domain Design Requirements

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A variable with the specified requirement name appears in the Data area of the tool.
A graphical display of the requirement also appears in the Response Optimization
tool window.
6 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Gain and Phase Margins block to specify
bounds on the gain and phase margin. (Requires Simulink Control Design.)

Specify Piecewise-Linear Lower and Upper Bounds on Frequency

Response
To specify upper or lower bounds on the magnitude of a system response:

1 In the Response Optimization tool, select Bode Magnitude in the New list. A
window opens where you specify the lower or upper bounds on the magnitude of the
system response.
2 Specify a requirement name in the Name box.
3 Specify the requirement type using the Type list.
4 Specify the edge start and end frequencies and corresponding magnitude in the
Frequency and Magnitude columns.
5 Insert or delete bound edges.

Click to specify additional bound edges.

3-41
3 Response Optimization

Select a row and click to delete a bound edge.

6 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

7 Click OK.

A new variable with the specified name appears in the Data area of the Response
Optimization tool window. A graphical display of the requirement also appears in the
Response Optimization tool window.

3-42
 Specify Frequency-Domain Design Requirements

8 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Bode Characteristics block to specify bounds
on the magnitude of the system response. (Requires Simulink Control Design.)

Specify Bound on Closed-Loop Peak Gain

To specify an upper bound on the closed-loop peak response of a system:

1 In the Response Optimization tool, select Closed-Loop Peak Gain in the New list.
A window opens where you specify an upper bound on the closed-loop peak gain of
the system.
2 Specify a requirement name in the Name box.
3 Specify the upper bound on the closed-loop peak gain in the Closed-Loop peak
gain box.

3-43
3 Response Optimization

4 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A new variable with the specified name appears in the Data area of the Response
Optimization tool window. A graphical display of the requirement also appears in the
Response Optimization tool window.

3-44
 Specify Frequency-Domain Design Requirements

6 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Nichols Characteristics block to specify
bounds on the magnitude of the system response. (Requires Simulink Control Design.)

Specify Lower Bound on Damping Ratio

To specify a lower bound on the damping ratio of the system:

1 In the Response Optimization tool, select Damping Ratio in the New list. A
window opens where you specify an upper bound on the damping ratio of the system.
2 Specify a requirement name in the Name box.
3 Specify the lower bound on the damping ratio in the Damping ratio box.

3-45
3 Response Optimization

4 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A new variable with the specified name appears in the Data area of the Response
Optimization tool. A graphical display of the requirement also appears in the
Response Optimization tool window.

3-46
 Specify Frequency-Domain Design Requirements

6 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Pole-Zero Characteristics block to specify a
bound on the damping ratio. (Requires Simulink Control Design.)

Specify Upper and Lower Bounds on Natural Frequency

To specify a bound on the natural frequency of the system:

1 In the Response Optimization tool, select Natural Frequency in the New list. A
window opens where you specify a bound on the natural frequency of the system.
2 Specify a requirement name in the Name box.

3-47
3 Response Optimization

3 Specify a lower or upper bound on the natural frequency in the Natural frequency
box.
4 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A new variable with the specified name appears in the Data area of the Response
Optimization tool. A graphical display of the requirement also appears in the
Response Optimization tool window.

3-48
 Specify Frequency-Domain Design Requirements

6 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Pole-Zero Characteristics block to specify a
bound on the natural frequency. (Requires Simulink Control Design.)

Specify Upper Bound on Approximate Settling Time

To specify an upper bound on the approximate settling time of the system:

1 In the Response Optimization tool, select Settling Time in the New list. A window
opens where you specify an upper bound on the approximate settling time of the
system.
2 Specify a requirement name in the Name box.

3-49
3 Response Optimization

3 Specify the upper bound on the approximate settling time in the Settling time box.
4 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A new variable with the specified name appears in the Data area of the Response
Optimization tool. A graphical display of the requirement also appears in the
Response Optimization tool window.

3-50
 Specify Frequency-Domain Design Requirements

6 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Pole-Zero Characteristics block to specify
the approximate settling time. (Requires Simulink Control Design.)

Specify Piecewise-Linear Upper and Lower Bounds on Singular Values

To specify piecewise-linear upper and lower bounds on the singular values of a system:

1 In the Response Optimization tool, select Singular Values in the New list. A
window opens where you specify the lower or upper bounds on the singular values of
the system.
2 Specify a requirement name in the Name box.

3-51
3 Response Optimization

3 Specify the requirement type using the Type list.

4 Specify the edge start and end frequencies and corresponding magnitude in the
Frequency and Magnitude columns, respectively.
5 Insert or delete bound edges.

Click to specify additional bound edges.

Select a row and click to delete a bound edge.

6 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

7 Click OK.

A new variable with the specified name appears in the Data area of the Response
Optimization tool. A graphical display of the requirement also appears in the
Response Optimization tool window.

3-52
 Specify Frequency-Domain Design Requirements

8 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Singular Value Characteristics block to
specify bounds on the singular value. (Requires Simulink Control Design.)

Specify Step Response Characteristics

To specify step response characteristics:

1 You can apply this requirement to either a signal or a linearization of your model.

In the Response Optimization Tool, click New. To apply this requirement to a

signal, select the Step Response Envelope entry in the New Time Domain

3-53
3 Response Optimization

Requirement section of the New list. To apply this requirement to a linearization

of your model, select the Step Response Envelope entry in the New Frequency
Domain Requirement section of the New list. The latter option requires Simulink
Control Design software.

A window opens where you specify the step response requirements on a signal, or
system.
2 Specify a requirement name in the Name box.
3 Specify the step response characteristics:

% Overshoot
% Settling
Final Value

% Rise
Amplitude

% Undershoot
Initial Value

Rise Time Settling Time Time

• Initial value: Input level before the step occurs

• Step time: Time at which the step takes place
• Final value: Input level after the step occurs
• Rise time: The time taken for the response signal to reach a specified percentage
of the step's range. The step's range is the difference between the final and initial
values.
• % Rise: The percentage used in the rise time.

3-54
 Specify Frequency-Domain Design Requirements

• Settling time: The time taken until the response signal settles within a specified
region around the final value. This settling region is defined as the final step
value plus or minus the specified percentage of the final value.
• % Settling: The percentage used in the settling time.
• % Overshoot: The amount by which the response signal can exceed the final
value. This amount is specified as a percentage of the step's range. The step's
range is the difference between the final and initial values.
• % Undershoot: The amount by which the response signal can undershoot the
initial value. This amount is specified as a percentage of the step's range. The
step's range is the difference between the final and initial values.
4 Specify the signals or systems to be bound.

You can apply this requirement to a model signal or to a linearization of your

Simulink model (requires Simulink Control Design software).

• Apply this requirement to a model signal:

In the Select Signals to Bound area, select a logged signal to which you will
apply the requirement.

If you have already selected a signal to log, as described in “Specify Signals to

Log” on page 3-12, it appears in the list. Select the corresponding check-box.

If you haven’t selected a signal to log:

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

3-55
3 Response Optimization

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

Click OK. A new variable, with the specified name, appears in the Data area
of the Response Optimization tool window.
• Apply this requirement to a linear system.

Linear systems are defined by snapshot times at which the model is linearized
and sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a
vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in

the list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open

the Create linearization I/O set dialog box.

3-56
 Specify Frequency-Domain Design Requirements

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A variable with the specified requirement name appears in the Data area of the tool.
A graphical display of the requirement also appears in the Response Optimization
tool window.

Alternatively, you can use the Check Step Response Characteristics block to
specify step response bounds for a signal.

See Also

“Design Optimization to Meet Step Response Requirements (GUI)”

Specify Custom Requirements

To specify custom requirements, such as minimizing system energy:

1 In the Response Optimization tool, select Custom Requirement in the New list. A
window opens where you specify the custom requirement.
2 Specify a requirement name in the Name box.
3 Specify the requirement type using the Type list.
4 Specify the name of the function that contains the custom requirement in the
Function box. The field must be specified as a function handle using @. The function

must be on the MATLAB path. Click to review or edit the function.

If the function does not exist, clicking opens a template MATLAB file.
Use this file to implement the custom requirement. The default function name is
myCustomRequirement.
5 (Optional) If you want to prevent the solver from considering specific parameter
combinations, select the Error if constraint is violated check box. Use this option
for parameter-only constraints.

During an optimization iteration, the solver evaluates requirements with this option
selected first.

3-57
3 Response Optimization

• If the constraint is violated, the solver skips evaluating any remaining

requirements and proceeds to the next iterate.
• If the constraint is not violated, the solver evaluates the remaining requirements
for the current iterate. If any of the remaining requirements bound signals or
systems, then the solver simulates the model .

For more information, see “Skip Model Simulation Based on Parameter Constraint
Violation (GUI)” on page 3-188.

Note: If you select this check box, then do not specify signals or systems to bound. If
you do specify signals or systems, then this check box is ignored.
6 (Optional) Specify the signal or system, or both, to be bound.

You can apply this requirement to model signals, or a linearization of your Simulink
model (requires Simulink Control Design software), or both.

Click Select Signals and Systems to Bound (Optional) to view the signal and
linearization I/O selection area.

• Apply this requirement to a model signal:

In the Signal area, select a logged signal to which you will apply the
requirement.

If you have already selected a signal to log, as described in “Specify Signals to

Log” on page 3-12, it appears in the list. Select the corresponding check box.

If you have not selected a signal to log:

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

3-58
 Specify Frequency-Domain Design Requirements

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

Click OK. A new variable, with the specified name, appears in the Data area
of the Response Optimization tool window.
• Apply this requirement to a linear system.

Linear systems are defined by snapshot times at which the model is linearized
and sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a
vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in

the list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open

the Create linearization I/O set dialog box. For more information on using
this dialog box, see “Create Linearization I/O Sets” on page 3-77.

3-59
3 Response Optimization

For more information on linearization, see “What Is Linearization?”.

7 Click OK.

A new variable, with the specified name, appears in the Data area of the Response
Optimization tool. A graphical display of the requirement also appears in the
Response Optimization tool window.

See Also

• “Design Optimization to Meet a Custom Objective (GUI)” on page 3-108

• “Design Optimization to Meet Custom Signal Requirements (GUI)” on page 3-135

3-60
 Specify Design Variables

Specify Design Variables

This topic shows how to specify design variables for optimization.

Before running the optimization, you must specify the model parameters to optimize.
These parameters form the design variables set for optimization. By tuning these
parameters, Simulink Design Optimization software attempts to make the signals meet
the requirements.

Simulink Design Optimization software optimizes the response signals of the model
by varying the tuned parameters so that the response signals lie within the constraint
bound segments or closely match a specified reference signal. The design variables can be
scalar, vector, matrix, or an expression that evaluates to one of these values.

In this section...
“Add Model Parameters as Variables for Optimization” on page 3-61
“Specify Design Variables” on page 3-64

Add Model Parameters as Variables for Optimization

The software can only optimize variables that are in use by the Simulink model. Create
variables for optimization in the MATLAB or model workspace, and specify your model or
block parameters using these variables.

In this figure, the Proportional (P) and Integral (I) gain parameters of a PID
Controller block are specified as numerical values.

3-61
3 Response Optimization

To optimize the gain parameters, specify them as variables Kp and Ki:

1 Create the variables Kp and Ki in one of the following ways:

• Add the variables to the model workspace, and specify initial values.

3-62
 Specify Design Variables

• Write initialization code in the PreloadFcn callback of the model. For more
information, see “Model Callbacks”.

Kp = 1;
Ki = 1;
2 Specify the gain parameters as the variables Kp and Ki in the PID Controller block
dialog box.

3-63
3 Response Optimization

You can now select Kp and Ki for optimization. See, “Specify Design Variables” on page
3-64.

Specify Independent Parameters for Optimization

You can also specify independent parameters that do not appear explicitly in the model
as variables for optimization. However, you cannot use this workflow with Simulink fast
restart.

Suppose that a model parameter Kint is related to independent parameters x and y such
that Kint = x+y. To optimize x and y instead of Kint:

• Create the independent variables x and y by adding them to the model workspace and
specifying initial values. This ensures that the variables are used by the model.
• Write code in the InitFcn callback of the model that defines the relationship between
Kint, x, and y. You must first use the get_param function to get the variables x and
y from the model workspace before you can use them to define Kint.
wks = get_param(gcs,'ModelWorkspace')
x = evalin(wks,'x')
y = evalin(wks,'y')
Kint = x+y;

You can now select x and y for optimization. Do not optmize the independent and
dependent parameters simultaneously. Doing so can lead to incorrect results. For
example, do not optimize Kint, x and y together.

Specify Design Variables

To specify the parameters to be tuned using the Response Optimization tool:

1 In the Design Variables Set list, select New.

A window opens where you specify design variables. All parameters in use by the
model are displayed in this window.
2 Select one or more parameter names and click

to add the selected parameters to a design variables set.

3-64
 Specify Design Variables

Note: You can add the same parameter to multiple design variable sets.
3 (Optional) Specify design variable settings.

Setting Description Default

Variable The name of the parameter. Not an editable field
Value Value of the model parameter. This value is used Current value of the parameter
by the optimization method as the initial value in the model. If you edit this
and is modified during optimization. column, click Update model
variable values to update the
values in the model.
Minimum The minimum value or lower bound for the -Inf
parameter. You can edit this field to provide an
alternate minimum value.
Maximum The maximum value or upper bound for the Inf
parameter. You can edit this field to provide an
alternate maximum value.
Scale During optimization, the design variables are Next power of 2 greater than the
scaled, or normalized, by dividing their current current value of the parameter
value by a scale value. You can edit this field to
provide an alternate scaling factor.

The check-box indicates whether the parameter is selected as an design variable

in the set. Select it if you want this parameter to be tuned during the optimization.
Deselect if you do not want this parameter to be tuned during the optimization
but you would like to keep it on the list of tuned parameters (for a subsequent
optimization).

Expand Variable Detail to see the block in the model that contains this parameter.
4 Click OK to create a design variable set.

Related Examples
• “Optimize Parameters for Robustness (GUI)” on page 3-206
• “Update Model with Design Variables Set” on page 3-66
• “Export Design Variable Values for Specific Iteration” on page 3-91

3-65
3 Response Optimization

Update Model with Design Variables Set

This example shows how to update a model with a set of design variables.

Open the Simulink model and load the pre-configured Response Optimization tool
session.
load('pidtune_demo_sdosession_update_dv.mat')
sdotool(SDOSessionData)

The Response Optimization tool opens and loads the preconfigured session. In the Data
area, DesignVars1 is a set of tuned design variables.

In the Design Variables Set list, select the design variable set, DesignVars1.

Open the Edit dialog box.

Click for the Design Variable Set list.

Select the variables you want to update in the model.

3-66
 Update Model with Design Variables Set

For this example, select Kd, Ki, and Kp.

Click Update model variables.

Plot the model response.

In the Response Optimization tab, click Plot Current Response.

Related Examples
• “Export Design Variable Values for Specific Iteration” on page 3-91

3-67
3 Response Optimization

General Options
In this section...
“Accessing General Options” on page 3-68
“Progress Options” on page 3-68
“Result Options” on page 3-69

Accessing General Options

You can set optimization progress and result options. To set these options, click Options
in the Response Optimization tool. A window opens. Select the General Options tab.

Progress Options
You can specify options related to optimization progress using the options in the
Progress Options area.

3-68
 General Options

• Show optimization progress window during optimization

Opens an optimization progress window during optimization. The window displays

information such as constraint violations and cost function if Display level
is Iteration. The window is updated at the end of the optimization with the
termination information such as whether the optimization converged.
• Update plots during optimization

Updates model response and design variable plots at each optimization iteration.

Result Options
You can specify options for optimization results in the Result Options area.

• Update model at end of optimization

Updates optimized parameter values in the Simulink model after the optimization
terminates.
• Save optimization termination information

Saves termination information returned by the optimization solver as a variable

named info in the Data area. info is a structure with one or more of the following
fields:

3-69
3 Response Optimization

• F — Optimized cost (objective) value.

• Cleq — Optimized nonlinear inequality constraint violations.

The field appears if the optimization problem includes a nonlinear inequality

constraint.

The value is a mx1 vector. Positive values indicate that the constraint has not been
satisfied. Check exitflag to confirm that the optimization succeeded.
• Ceq — Optimized nonlinear equality constraint violations.

The field appears if the optimization problem includes a nonlinear equality

constraint.

The value is a double rx1 vector. Any nonzero values indicate that the constraint
has not been satisfied. Check exitflag to confirm that the optimization
succeeded.
• Gradients — Cost and constraint gradients at the optimized parameter values.
See “How the Optimization Algorithm Formulates Minimization Problems” on
page 3-3 on how the solver computes gradients.

This field appears if the solver specified in the Method property of

sdo.OptimizeOptions computes gradients.

The value is a structure.

• exitflag — Integer identifying the reason the algorithm terminated. See
fmincon, patternsearch and fminsearch for a list of the values and the
corresponding termination reasons.
• iterations — Number of optimization iterations
• SolverOutput — A structure with solver-specific output information. The fields
of this structure depends on the optimization solver specified in the Method
property of sdo.OptimizeOptions. See fmincon, patternsearch and fminsearch
for a list of solver outputs and their description.
• Stats — A structure that contains statistics collected during optimization, such as
start and end times, number of function evaluations and restarts.
• Save requirement values at the end of optimization

Saves requirement values after the optimization terminates.

3-70
 General Options

• Design variable sets

• Overwrite optimized variable values in design variable set

Overwrites the optimized model parameter values in the design variable set
variable used in the optimization. You can see the updated values in the Value
field.

• Save optimized variable values as new design variable set

Creates a new variable in the Data area of the tool that contains the optimized
parameter values. To update the model with the optimized parameter values,
select the variable in the Design Variables Set list. Open the Edit dialog box

by clicking . Select the parameters of interest and click Update model

variables.

3-71
3 Response Optimization

Optimization Options
In this section...
“Accessing Optimization Options” on page 3-72
“Selecting Optimization Methods” on page 3-73
“Selecting Optimization Termination Options” on page 3-74
“Selecting Additional Optimization Options” on page 3-75

Accessing Optimization Options

You can set several options for the optimization. These options include the optimization
methods and the tolerances the methods use.

To set optimization options, click Options in the Response Optimization tool. A window
opens. Select the Optimization Options tab.

3-72
 Optimization Options

Selecting Optimization Methods

Both the Method and Algorithm options in the Optimization method area define the
optimization method.

The choices for the Method option are:

• Gradient descent (default) — Uses the Optimization Toolbox function fmincon to

optimize the response signal subject to the constraints.

The Algorithm options for Gradient descent are:

Algorithm Option Learn More

Sequential Quadratic Programming “fmincon SQP Algorithm” in the
(default) Optimization Toolbox documentation.
Active-Set “fmincon Active Set Algorithm” in the
Optimization Toolbox documentation.
Interior-Point “fmincon Interior Point Algorithm” in the
Optimization Toolbox documentation.
Trust-Region-Reflective “fmincon Trust Region Reflective
Algorithm” in the Optimization Toolbox
documentation.
• Pattern search — Uses the Global Optimization Toolbox function
patternsearch, an advanced direct search method, to optimize the response. This
option requires the Global Optimization Toolbox.
• Simplex search — Uses the Optimization Toolbox function fminsearch, a direct
search method, to optimize the response. Simplex search is most useful for simple
problems and is sometimes faster than Gradient descent for models that contain
discontinuities.

For more information on the problem formulations for each optimization method, see
“How the Optimization Algorithm Formulates Minimization Problems” on page 3-3.

3-73
3 Response Optimization

Selecting Optimization Termination Options

Use the Optimization options panel to specify when you want the optimization to
terminate.

• Parameter tolerance: The optimization terminates when successive parameter

values change by less than this number. For more details, refer to the discussion
of the parameter TolX in the reference page for the Optimization Toolbox function
fmincon.
• Constraint tolerance: This number determines the maximum limit by which the
constraints can be violated, and still allow a successful convergence.
• Function tolerance: The optimization terminates when successive function values
are less than this value. Changing the default Function tolerance value is only
useful when you are tracking a reference signal or using the Simplex search
method. For more details, refer to the discussion of the parameter TolFun in the
reference page for the Optimization Toolbox function fmincon.
• Maximum iterations: The maximum number of iterations allowed. The optimization
terminates when the number of iterations exceeds this number.
• Look for maximally feasible solution: When selected, the optimization continues
after it has found an initial, feasible solution, until it finds a maximally feasible,
optimal solution. When this option is unselected, the optimization terminates as soon
as it finds a solution that satisfies the constraints and the resulting response signal
sometimes lies very close to the constraint segment. In contrast, a maximally feasible
solution is typically located further inside the constraint region.

Note: If selected, the software ignores this option when tracking a reference signal.

By varying these parameters you can force the optimization to continue searching for a
solution or to continue searching for a more accurate solution.

3-74
 Optimization Options

Selecting Additional Optimization Options

At the bottom of the Optimization Options panel is a group of additional optimization
options.

• “Display Level” on page 3-75

• “Restarts” on page 3-76

Display Level

The Display level option specifies the form of the output that appears in the
Optimization Progress window. The options are:

• Iterations (default) — Displays information after each iteration

• Off — Turns off all output display
• Notify — Displays output only if the function does not converge
• Final — Displays only the final output

For more information on the type of iterative output that appears for the method you
selected in Method, see the discussion of output for the corresponding function.

Method Function Output Information

Gradient fmincon fmincon section of “Function-Specific
descent Headings” in the Optimization Toolbox
documentation
Simplex search fminsearch fminsearch section of “Function-Specific
Headings” in the Optimization Toolbox
documentation
Pattern search patternsearch “Display to Command Window Options”
in the Global Optimization Toolbox
documentation

3-75
3 Response Optimization

Restarts

In some optimizations, the Hessian may become ill-conditioned and the optimization does
not converge. In these cases, it is sometimes useful to restart the optimization after it
stops, using the endpoint of the previous optimization as the starting point for the next
one. To automatically restart the optimization, indicate the number of times you want to
restart in this field.

3-76
 Create Linearization I/O Sets

Create Linearization I/O Sets

This example shows how to create a linearization input/output set using the Response
Optimization Tool.

You create a linearization input/output set using the Create Linearization I/O Set dialog
box. This dialog box may be accessed in two ways:

• In the Response Optimization tool, select Linearization I/Os in the New drop-down
list.
•
In a requirement dialog box, in the Select Systems to Bound area, click .

Create Linearization I/O Set

To create a new linearization I/O set:

1 In your Simulink model, select one or more signals that you want to define as a
linearization input or output point.

The selected signals appear in the Create linearization I/O set dialog box under
Currently selected signals.

2 In the Create linearization I/O set dialog box, click a signal name under Currently
selected signals.

3-77
3 Response Optimization

3 Click Add. The signal appears in the list of Analysis I/Os.

4 Select the linearization point type for a signal from the Configuration drop-down
list for that signal. For example:

• If you want the selected signal to be a linearization output point, select Output
Measurement.
• If you want the signal to be an open-loop output point, select Open-loop
Output.

5 Repeat steps 1–4 for any other signals you want to define as linearization I/O points.

Tip To highlight in the Simulink model the location of any signal in the current list
of analysis I/O points, select the I/O point in the list and click Highlight.
6 After you define all the signals for the I/O set, enter a name for the I/O set in the
Variable name field located at the top-left of the window.
7 Click OK.

3-78
 Linearization Options

Linearization Options
In this section...
“Accessing Linearization Options” on page 3-79
“Configuring Linearization Options” on page 3-79

Accessing Linearization Options

If you have Simulink Control Design, you can set several options for linearization. These
options include the linearization methods and the sample time of the linear systems.

To set linearization options, click Options in the Response Optimization tool. A window
opens. Select the Linearization Options tab.

Configuring Linearization Options

Models with Time Delays

Simulink Control Design lets you choose whether to linearize models using exact
representation or Pade approximation of continuous time delays. How you treat time
delays during linearization depends on your nonlinear model.

Simulink blocks that model time delays are:

3-79
3 Response Optimization

• Transport Delay block

• Variable Time Delay block
• Variable Transport Delay block
• Delay block
• Unit Delay block

By default, linearization uses Pade approximation for representing time delays in your
linear model.

Use Pade approximation to represent time delays when:

• Applying more advanced control design techniques to your linear plant, such as LQR
or H-infinity control design.
• Minimizing the time to compute a linear model.

Specify to linearize with exact time delays for:

• Minimizing errors that result from approximating time delays

• PID tuning or loop-shaping control design methods in Simulink Control Design
• Discrete-time models (to avoid introducing additional states to the model)

The software treats discrete-time delays as internal delays in the linearized model.
Such delays do not appear as additional states in the linearized model.

Specify How Delays are Treated

To specify whether the linearization should approximate delays or use them exactly, set
the Use exact delays check box appropriately.

Linearization Sampling Time

To specify the sampling time of the linearized model, use the Linear system sample
time box. By default, the software chooses the slowest applicable sampling time. Use 0 to
specify a continuous-time linear model.

Linearization Rate Conversion Method

When you linearize models with multiple sample times, such as a discrete controller with
a continuous plant, a rate conversion algorithm generates a single-rate linear model. The
rate conversion algorithm affects linearization results.

3-80
 Linearization Options

Rate Conversion Method When to Use

Zero-Order Hold Use when you need exact discretization of
continuous dynamics in the time-domain
for staircase inputs.
Tustin Use when you need good frequency-domain
matching between a continuous-time
system and the corresponding discretized
system, or between an original system and
the resampled system.
Tustin with Prewarping Use when you need good frequency
domain matching at a particular frequency
between the continuous-time system and
the corresponding discretized system,
or between an original system and the
resampled system.
Upsampling when possible (Zero-Order Upsample discrete states when possible
Hold, Tustin, and Tustin with Prewarping) to ensure gain and phase matching of
upsampled dynamics. You can only
upsample when the new sample time is an
integer multiple of the sample time of the
original system. Otherwise, the software
uses an alternate rate conversion method.

3-81
3 Response Optimization

Plots in the Response Optimization Tool

In this section...
“Adding Plots in Response Optimization Tool” on page 3-82
“Plotting Current Response” on page 3-82
“Plotting Intermediate Steps” on page 3-82
“Modifying Plot Properties” on page 3-82
“Plot Types” on page 3-84
“Export Design Variables and Requirement Values for an Iteration” on page 3-87

Adding Plots in Response Optimization Tool

To create a new plot or to add to an existing plot in the Response Optimization Tool,
choose the variable to plot in the Data to Plot list. Then select the plot type using the
Add Plot list. The Add Plot list has entries for the supported plot types for the given
plot variable.

Plotting Current Response

To display the current response, click Plot Model Response in the Response
Optimization tab of the Response Optimization tool. The current response appears as a
thick line.

Plotting Intermediate Steps

To turn on, or off, the display of the response signal at intermediate steps during the
optimization, right-click within the white space in the plot and select Responses
> Show Iteration Responses. The response at an intermediate step is based on
parameter values at that intermediate point in the optimization.

Modifying Plot Properties

Modifying Properties of Response Plots

Right-click the white space in a plot and select Axes Properties to open the Property
Editor dialog box.

This figure shows the Property Editor dialog box for a step response.

3-82
 Plots in the Response Optimization Tool

In general, you can change the following properties of response plots.

• Labels — Titles and X- and Y-labels

To specify new text for plot titles and axis labels, type the new string in the field next
to the label you want to change. The label changes immediately as you type, so you
can see how the new text looks as you are typing.
• Limits — Numerical ranges of the x- and y- axes

Default values for the axes limits make sure that the maximum and minimum x and y
values are displayed. If you want to override the default settings, change the values in
the Limits pane fields. The Auto-Scale check box automatically clears if you click a
different field. The new limits appear immediately in the response plot.

To reestablish the default values, select the Auto-Scale check box again.
• Units — Units where applicable (e.g., rad/s to Hertz). If you cannot customize units,
the Property Editor displays that no units are available for the selected plot.
• Style — Show a grid, adjust font properties, such as font size, bold and italics, and
change the axes foreground color
• Options — Change options where applicable. These include peak response, settling
time, phase and gain margins, etc. Plot options change with each plot response
type. The Property Editor displays only the options that make sense for the selected
response plot. For example, phase and gain margins are not available for step
responses.

3-83
3 Response Optimization

Plot Types
• “Response Plots” on page 3-84
• “Spider Plots” on page 3-85
• “Iteration Plots” on page 3-86

Response Plots

You can view model signals and the requirements applied to the signal using a response
plot. You can also plot the frequency response of a system (requires Simulink Control
Design).

The response plot shows the system response as it varies during optimization. You can
also view the uncertain system responses in the plot.

3-84
 Plots in the Response Optimization Tool

Tip To select the responses and systems displayed in a given plot, right-click on the plot
and use the Systems and Responses menu.

Spider Plots

You can compare the values of design variable sets or evaluated requirements using a
spider plot.

Spider plots depict multivariate data using an axis for each variable. The various axes
are arranged clockwise and have a common intersecting point, as this example shows:

Tip To view only some of the requirement values in a given plot, right-click on the plot
and select the requirements in the Show list.

3-85
3 Response Optimization

For information on using a spider plot to compare design variables sets or evaluated
requirements, see “Compare Requirements and Design Variables Using Spider Plot” on
page 3-88

Iteration Plots

You can plot the values of design variables and requirements as they vary during
optimization using an iteration plot.

Iteration plots depict the value of the plot variable(s) for each iteration. The x-axis
represents the iteration number, as this example shows:

You can export the values of a plotted variable for a given iteration. For more
information, see “Export Design Variables and Requirement Values for an Iteration” on
page 3-87.

Tip To view scaled values of the plotted variable(s), right-click on the plot and select
Show scaled values.

3-86
 Plots in the Response Optimization Tool

Export Design Variables and Requirement Values for an Iteration

To export the values of design variables or requirements:

1 Open the Export Iteration Data dialog box.

Right-click on the iteration plot and select Export.

2 Specify the plotted variable that you want to export using the Data to export list.
3 Specify the iteration for which you want to export data in the Iteration(s) to
export box.

To specify multiple iterations, use a vector of integers. For example, [0 2 5].

4 Specify the variable name for the exported data using the Export to a variable
named box.
5 Export the data to the Data area of the tool.

Click OK. The exported data variable appears in the Data area.

Note: The iteration number is added as a suffix to the exported data variable name.

3-87
3 Response Optimization

Compare Requirements and Design Variables Using Spider Plot

This example shows how to use a spider plot to compare requirement evaluations before
and after optimizing the response. You can use a similar procedure to compare the values
of sets of design variables.

Open the Simulink model and load the pre-configured Response Optimization Tool session.

For this example, which uses a distillation column model, the step response requirements
are preconfigured and loaded in the model workspace.

1 Open the distillation model.

sys = 'distillation_demo';
open_system(sys)
2 Open the Response Optimization Tool.

In the Simulink model window, select Analysis > Response Optimization.

Alternatively, click the Response Optimization GUI with preloaded data

block in the model and skip the next step.
3 Load the preconfigured Response Optimization Tool session.

Click the Response Optimization tab. In the Open Session drop-down list, select
Open from model workspace. A window opens where you select the Response
Optimization Tool session to load. Select distillation_optim and click OK.

The preconfigured step response requirements are loaded in the Response

Optimization Tool.

Evaluate the requirement before optimization.

In the Response Optimization tab, click Evaluate Requirements.

A new variable, ReqValues, containing the evaluation of the requirements appears in

the Data area.

When optimizing the model response, you create a set of requirements that it must
satisfy. If the requirements are violated, meaning that they evaluate to nonnegative
values, the design variables must be optimized. After the optimization, you can compare
the original requirement value with the requirement evaluated using the optimized
design variable values.

3-88
 Compare Requirements and Design Variables Using Spider Plot

Plot the requirement value before optimization.

1 In the Data to Plot list, select ReqValues.

2 In the Add Plot list, select Spider plot.

The plot has an axis for each edge-and-signal combination defined in the
distillation_demo/Desired Step Response check block. Points on each axis
represent the violation for that signal-edge combination and the plot connects these
points to form a closed polygon representing the initial design. Note that some points
are negative, representing satisfied constraints, and some positive, representing violated
constraints.

Optimize the model.

Click Optimize.

A new variable, ReqValues1, containing the evaluation of the requirements using the
optimized design variables appears in the Data area.

3-89
3 Response Optimization

Compare the requirement values before and after optimization.

1 In the Data to Plot list, select ReqValues1.

2 In the Add Plot list, select Spider plot 1.

The optimized requirement values, ReqValues1, are all negative or zero, indicating that
all the constraints are satisfied.

More About
• “Export Design Variable Values for Specific Iteration” on page 3-91

3-90
 Export Design Variable Values for Specific Iteration

Export Design Variable Values for Specific Iteration

This example shows how to export the design variable values for a specific optimization
iteration.

During optimization, the optimization solver simulates the model using a different set of
design variables at each iteration. After the optimization completes, you can export the
values for an iteration from the iteration plot of the design variable set.

For this example, load a preconfigured Response Optimization tool session. Optimize the
model, and export the design variable set values for the third iteration.

Open the Simulink model and load the preconfigured Response Optimization tool session.

load('distillation_demo_sdosession_export_iter_dv.mat')
sdotool(SDOSessionData)

The Response Optimization tool opens and loads the preconfigured session. Iteration
Plot 1 is configured to plot the values of DesignVars for each optimization iteration.

Click Optimize.

The optimization completes after four iterations.

Select the iteration plot of the design variable set.

Click Iteration plot 1.

Open the Export Iteration Data dialog box.

Right-click on the iteration plot, and select Export.

3-91
3 Response Optimization

Specify details regarding exporting the design variable set data:

• In the Data to export list, select DesignVars.

• In the Iteration(s) to export box, enter 3.

To specify multiple iterations, use a vector of integers. For example, [0 2 5].

• In the Export to a variable named box, enter DesignVars_iter.

Export the design variable values set to the Data area of the tool.

Click OK. The exported data variable, DesignVars_iter_3, appears in the Data.

Note: You will see the iteration number suffixed to the exported data variable name.

Related Examples
• “Update Model with Design Variables Set” on page 3-66
• “Compare Requirements and Design Variables Using Spider Plot” on page 3-88

More About
• “Iteration Plots” on page 3-86

3-92
 Design Optimization to Meet Time- and Frequency-Domain Requirements (GUI)

Design Optimization to Meet Time- and Frequency-Domain

Requirements (GUI)
This example shows how to tune a controller to satisfy time- and frequency-domain
design requirements using the Response Optimization tool.

The example requires Simulink® Control Design™.

Aircraft Longitudinal Flight Control Model

Open the Simulink model.

sys = 'sdoAircraft';
open_system(sys);

3-93
3 Response Optimization

The aircraft model is based on the Simulink slexAircraftExample model. The model
includes:

• Subsystems to model aircraft dynamics (Aircraft Dynamics Model), wind gusts

(Dryden Wind Gust Models), and pilot G-forces (Pilot G-force calculation).

• A step change applied to the aircraft joystick at 1 second into the simulation that
causes the aircraft to pitch upward.

Controller Design Problem

You tune the controller gains to meet the following time- and frequency-domain design
requirements:

• Angle-of-attack alpha response to a step change in the joystick has a rise time of less
than 1 second, less than 1% overshoot, and settles to within 1% of steady state within
less than 5 seconds

• Pitch-rate control loop has good tracking below 1 rad/s and 20 dB noise rejection
above 100 rad/s

• Closed loop response from joystick to pilot G-Force is below 0 dB above 5 rad/s.

These requirements reduce the high frequency G-forces experienced by the pilot in
response to joystick changes while still maintaining flight performance.

The model includes the following blocks (from Simulink® Design Optimization™ and
Simulink Control Design Model Verification libraries):

• Alpha Response specifies the alpha step response requirement.

3-94
 Design Optimization to Meet Time- and Frequency-Domain Requirements (GUI)

• Pitch Rate Loop specifies the pitch-rate performance requirement.

The linearization inputs/outputs are already selected in the Linearizations tab. The
pitch-rate loop starts from the input of the controller (the controller error signal) and
ends at the output of the pitch-rate sensor. The angle-of-attack loop is opened signal so
that the block only computes the pitch-rate loop response. The linear system is computed
at a simulation time of 0.

3-95
3 Response Optimization

3-96
 Design Optimization to Meet Time- and Frequency-Domain Requirements (GUI)

The Bounds tab specifies the following pitch-rate loop shape requirements:

• Greater than 20 dB over the range 0.01 rad/s to 0.1 rad/s

• Greater than 0 dB over the range 0.1 rad/s to 1 rad/s

• Less than -20 dB over the range 100 rad/s to 1000 rad/s

3-97
3 Response Optimization

• Pilot G Response specifies the G-force requirement.

The linearization inputs/outputs are already selected in the Linearizations tab. The
linear system is computed at a simulation time of 0.

3-98
Design Optimization to Meet Time- and Frequency-Domain Requirements (GUI)

3-99
3 Response Optimization

The Bounds tab specifies the G-force requirements of less than 0 dB over the range 5
rad/s 100 rad/s.

3-100
 Design Optimization to Meet Time- and Frequency-Domain Requirements (GUI)

Open the Response Optimization Tool

Open the Response Optimization tool to configure and run design optimization problems
interactively. Click Response Optimization on the Block Parameters dialog of Alpha
Response, Pitch Rate Loop or Pilot G Response block. Alternatively, type
sdotool('sdoAircraft'). To show multiple requirement plots at the same time, use
the View tab in the tool.

The tool detects the requirements specified in the Model Verification blocks and
automatically includes them as requirements to satisfy.

3-101
3 Response Optimization

Specify Design Variables

Specify the following model parameters as design variables for optimization:

• Controller gains Ki and Kf

• Pitch-rate sensor gain Kq

• Alpha sensor gain Ka

In the Design Variables Set drop-down list, select New. A dialog to select model
parameters for optimization opens.

Select Ki, Kf, Kq and Ka. Click << to add the selected parameters to the design variables
set.

Specify minimum and maximum gain values, the Ki and Kf values must remain negative
while Ka and Kq must remain positive.

Press Enter after you enter the values.

Click OK. A new variable DesignVars appears in the Response Optimization Tool
Workspace.

3-102
 Design Optimization to Meet Time- and Frequency-Domain Requirements (GUI)

Evaluate the Initial Design

Click Plot Model Response to simulate the model and check how well the initial design
satisfies the design requirements.

The plots indicate that the current design does not satisfy the pilot G-force requirement
and the alpha step response overshoot requirement is violated.

Optimize the Design

Create a plot to display how the controller variables are modified during the
optimization. In the Data To Plot drop-down list, select DesignVars, which contains

3-103
3 Response Optimization

the optimization design variables Ki, Kf, Kq and Ka. In the Add Plot drop-down list,
select Iteration plot.

Click Optimize.

3-104
Design Optimization to Meet Time- and Frequency-Domain Requirements (GUI)

3-105
3 Response Optimization

To load a pre-configured file and run the optimization, click Open in the Response
Optimization tab and select sdoAircraft_sdosession.mat. Alternatively load the
project by typing:

>> load sdoAircraft_sdosession

>> sdotool(SDOSessionData)

The optimization progress window updates at each iteration and shows that the
optimization converged after 5 iterations.

The Alpha Response and Pilot G Response plots indicate that the design
requirements are satisfied. The DesignVars plot shows that the controller gains
converged to new values.

3-106
 Design Optimization to Meet Time- and Frequency-Domain Requirements (GUI)

To view the optimized design variable values, click DesignVars in the Response
Optimization Tool Workspace. The optimized values of the design variables are
automatically updated in the Simulink model.
% Close the model
bdclose('sdoAircraft')

3-107
3 Response Optimization

Design Optimization to Meet a Custom Objective (GUI)

This example shows how to optimize a design to meet a custom objective using the
Response Optimization tool. You optimize the cylinder parameters to minimize the
cylinder geometry and satisfy design requirements.

Hydraulic Cylinder Model

The hydraulic cylinder model is based on the Simulink model sldemo_hydcyl. The
model includes:

• Pump and Cylinder Assembly subsystems. For more information on the

subsystems, see "Single Hydraulic Cylinder Simulation".

• A step change applied to the cylinder control valve orifice area that causes the
cylinder piston position to change.

Hydraulic Cylinder Design Problem

You tune the cylinder cross-sectional area and piston spring constant to meet the
following design requirements:

• Ensure that the piston position has a step response rise time of less than 0.04 seconds
and setting time of less than 0.05 seconds.

• Limit the maximum cylinder pressures to 1.75e6 N/m.

• Minimize the cylinder cross-sectional area.

Open the Response Optimization Tool

Open the Response Optimization tool to configure and run design optimization problems
interactively.

sdotool('sdoHydraulicCylinder')

3-108
Design Optimization to Meet a Custom Objective (GUI)

3-109
3 Response Optimization

Specify Design Variables

Specify the following model parameters as design variables for optimization:

• Cylinder cross-sectional area Ac

• Piston spring constant K

In the Design Variables Set drop-down list, select New. A dialog to select model
parameters for optimization opens.

3-110
 Design Optimization to Meet a Custom Objective (GUI)

Select Ac and K. Click <- to add the selected parameters to the design variables set.

Limit the cylinder cross-sectional area to circular area with radius between 1 and 2
centimeters and the piston spring constant to a range of 1e4 to 10e4 N/m. To do so,
specify the maximum and minimum for the corresponding variable in the Maximum and
Minimum columns.

Because the variable values are different orders of magnitude, scale Ac by 1e-3 and K by
1e5.

Press Enter after you specify the values.

Click OK. A new variable DesignVars appears in the Response Optimization Tool
Workspace.

3-111
3 Response Optimization

Specify Design Requirements

The design requirements require logged model signals. During optimization, the model is
simulated using the current value of the design variables and the logged signal is used to
evaluate the design requirements.

Log the cylinder pressures, which is the first output port of the Cylinder Assembly
block.

In the New drop-down list, select Signal. A dialog to select model signals to log opens.

3-112
 Design Optimization to Meet a Custom Objective (GUI)

Enter Pressures as the signal name in the Signal set field. Then, in the Simulink
model, click the first output port of the Cylinder Assembly block named Pressure.
The dialog updates to display the selected signal.

Select the signal in the dialog and click -> to add it to the signal set.

Click OK. A new variable Pressures appears in the Response Optimization Tool
Workspace.

Similarly, log the piston position, which is the second output of the Cylinder Assembly
block, in a variable named PistonPosition.

3-113
3 Response Optimization

Specify the maximum cylinder pressure requirement of less than 1.75e6 N/m.

In the New drop-down list, select Signal Bound. A dialog to create a signal bound
requirement opens.

3-114
 Design Optimization to Meet a Custom Objective (GUI)

Designate the Requirement Name as MaxPressure. In both the start and end
Amplitude columns, enter the maximum pressure requirement of 1.75e6 N/m, and set
the Edge End Time to 0.1 s. In the Select Signals to Bound area, select Pressures,
the signal on which this requirement applies.

3-115
3 Response Optimization

Click OK.

• A new MaxPressure variable appears in the Response Optimization Tool

Workspace.

• A graphical view of the maximum pressure requirement is automatically created.

3-116
 Design Optimization to Meet a Custom Objective (GUI)

Specify the piston position step response requirement of rise time of less than 0.04
seconds and a settling time of less than 0.05 seconds.

In the New drop-down list of the Response Optimization tab, select Step Response
Envelope. A dialog to create a step response requirement opens.

Specify a requirement named PistonResponse, and the required rise and settling time
bounds. Select PistonPosition as the signal to apply the step response requirement to.

3-117
3 Response Optimization

Click OK.

Specify Custom Objective

The custom objective is to minimize the cylinder cross-sectional area.

In the New drop-down list, select Custom Requirement. A dialog to create custom
requirement opens.

3-118
 Design Optimization to Meet a Custom Objective (GUI)

Specify a function to call during optimization in the Requirement Function field. At

each optimization iteration, the software calls the function and passes the current design
variable values. You can also optionally pass logged signals to the custom requirement.
Here, you use sdoHydraulicCylinder_customObjective as the custom requirement
function, which returns the value of the cylinder cross-sectional area.

In the Requirement Type drop-down list, specify whether the requirement is an

objective to minimize (min), an inequality constraint (<=), or an equality constraint (==).

3-119
3 Response Optimization

type sdoHydraulicCylinder_customObjective

function objective = sdoHydraulicCylinder_customObjective(data)

%SDOHYDRAULICCYLINDER_CUSTOMOBJECTIVE
%
% The sdoHydraulicCylinder_customObjective function is used to define a
% custom requirement that can be used in the graphical SDTOOL environment.
%
% The |data| input argument is a structure with fields containing the
% design variable values chosen by the optimizer.
%
% The |objective| return argument is the objective value to be minimized by
% the SDOTOOL optimization solver.
%

% Copyright 2011 The MathWorks, Inc.

3-120
 Design Optimization to Meet a Custom Objective (GUI)

%For the cylinder design problem we want to minimize the cylinder

%cross-sectional area so return the cylinder cross-sectional area as an
%objective value.
Ac = data.DesignVars(1);
objective = Ac.Value;
end

Evaluate the Initial Design

Click Plot Model Response to simulate the model and check how well the initial design
satisfies the design requirements. To show both requirement plots at the same time, use
the plot layout widgets in the View tab.

3-121
3 Response Optimization

From the plots, see that the maximum pressure requirement is satisfied but the piston
position step response requirement is not satisfied.

Optimize the Design

Create a plot to display how the cylinder cross-sectional area and piston spring constant
are modified during optimization.

In the Data to Plot drop-down list, select DesignVars, which contains the optimization
design variables Ac and K. In the Add Plot drop-down, create a new iteration plot to
show the design variable trajectories. For this new plot, click Show scaled values in the
Iteration Plot tab, to facilitate viewing the two trajectories on the same axes.

3-122
 Design Optimization to Meet a Custom Objective (GUI)

Click Optimize in the Response Optimization tab.

3-123
3 Response Optimization

The optimization progress window updates at each iteration and shows that the
optimization converged after 4 iterations.

The Pressures and PistonPosition plots indicate that the design requirements
are satisfied. The MinimizeAC plot shows that the cylinder cross-sectional area Ac is
minimized.

To view the optimized design variable values, click the variable name in the Response
Optimization Tool Workspace. The optimized values of the design variables are
automatically updated in the Simulink model.

3-124
 Design Optimization to Meet a Custom Objective (GUI)

Related Examples

To learn how to optimize the cylinder design using the sdo.optimize command, see
"Design Optimization to Meet Custom Objective (Code)".
% Close the model
delete(sdotool('sdoHydraulicCylinder'))
bdclose('sdoHydraulicCylinder')

3-125
3 Response Optimization

Design Optimization to Meet a Custom Objective (Code)

This example shows how to optimize a design to meet custom objective using
sdo.optimize. You optimize the cylinder parameters to minimize the cylinder geometry
and satisfy design requirements.

Hydraulic Cylinder Model

Open the Simulink model.

sys = 'sdoHydraulicCylinder';
open_system(sys);

The hydraulic cylinder model is based on the Simulink model sldemo_hydcyl. The
model includes:

• Pump and Cylinder Assembly subsystems. For more information on the

subsystems, see "Single Hydraulic Cylinder Simulation".

• A step change applied to the cylinder control valve orifice area that causes the
cylinder piston position to change.

3-126
 Design Optimization to Meet a Custom Objective (Code)

Hydraulic Cylinder Design Problem

You tune the cylinder cross-sectional area and piston spring constant to meet the
following design requirements:

• Ensure that the piston position has a step response rise time of less than 0.04 seconds
and setting time of less than 0.05 seconds.

• Limit the maximum cylinder pressures to 1.75e6 N/m.

• Minimize the cylinder cross-sectional area.

Specify Design Variables

Select the following model parameters as design variables for optimization:

• Cylinder cross-sectional area Ac

• Piston spring constant K

Ac = sdo.getParameterFromModel('sdoHydraulicCylinder','Ac');
K = sdo.getParameterFromModel('sdoHydraulicCylinder','K');

Limit the cylinder cross-sectional area to a circular area with radius between 1 and 2
centimeters.

Ac.Minimum = pi*1e-2^2; % m^2

Ac.Maximum = pi*2e-2^2; % m^2

Limit the piston spring constant to a range of 1e4 to 10e4 N/m.

K.Minimum = 1e4; % N/m

K.Maximum = 10e4; % N/m

Specify Design Requirements

The design requirements require logged model signals. During optimization, the model is
simulated using the current value of the design variables and the logged signal is used to
evaluate the design requirements.

Log the following signals:

• Cylinder pressures, available at the first output port of the Cylinder Assembly
block

3-127
3 Response Optimization

Pressures = Simulink.SimulationData.SignalLoggingInfo;
Pressures.BlockPath = 'sdoHydraulicCylinder/Cylinder Assembly';
Pressures.OutputPortIndex = 1;

• Piston position, available at the second output port of the Cylinder Assembly block

PistonPosition = Simulink.SimulationData.SignalLoggingInfo;
PistonPosition.BlockPath = 'sdoHydraulicCylinder/Cylinder Assembly';
PistonPosition.OutputPortIndex = 2;

Create an object to store the logging information and use later to simulate the model
simulator = sdo.SimulationTest('sdoHydraulicCylinder');
simulator.LoggingInfo.Signals = [PistonPosition,Pressures];

Specify the piston position step response requirement of rise time of less than 0.04
seconds and settling time less than of 0.05 seconds.
PistonResponse = sdo.requirements.StepResponseEnvelope;
set(PistonResponse, ...
'RiseTime', 0.04, ...
'FinalValue', 0.04, ...
'SettlingTime', 0.05, ...
'PercentSettling', 1);

Specify the maximum cylinder pressure requirement of less than 1.75e6 N/m.
MaxPressure = sdo.requirements.SignalBound;
set(MaxPressure, ...
'BoundTimes', [0 0.1], ...
'BoundMagnitudes', [1.75e6 1.75e6], ...
'Type', '<=');

For convenience, collect the performance requirements into a single structure to use
later.
requirements = struct(...
'PistonResponse', PistonResponse, ...
'MaxPressure', MaxPressure);

Create Objective/Constraint Function

To optimize the cylinder cross-sectional area and piston spring constant, create a
function to evaluate the cylinder design. This function is called at each optimization
iteration.

3-128
 Design Optimization to Meet a Custom Objective (Code)

Here, use an anonymous function with one argument that calls the
sdoHydraulicCylinder_design function.
evalDesign = @(p) sdoHydraulicCylinder_design(p,simulator,requirements);

The function:

• Has one input argument that specifies the cylinder cross-sectional area and piston
spring constant values.

• Returns the optimization objective value and optimization constraint violation values.

The optimization solver minimizes the objective value and attempts to

keep the optimization constraint violation values negative. Type help
sdoExampleCostFunction for more details on how to write the objective/constraint
function.

The sdoHydraulicCylinder_design function uses the simulator

and requirements objects to evaluate the design. Type edit
sdoHydraulicCylinder_design to examine the function in more detail.
type sdoHydraulicCylinder_design

function design = sdoHydraulicCylinder_design(p,simulator,requirements)

%SDOHYDRAULICCYLINDER_DESIGN
%
% The sdoHydraulicCylinder_design function is used to evaluate a cylinder
% design.
%
% The |p| input argument is the vector of cylinder design parameters.
%
% The |simulator| input argument is a sdo.SimulinkTest object used to
% simulate the |sdoHydraulicCylinder| model and log simulation signals
%
% The |requirements| input argument contains the design requirements used
% to evaluate the cylinder design
%
% The |design| return argument contains information about the design
% evaluation that can be used by the |sdo.optimize| function to optimize
% the design.
%
% see also sdo.optimize, sdoExampleCostFunction

% Copyright 2011 The MathWorks, Inc.

3-129
3 Response Optimization

%% Simulate the model

%
% Use the simulator input argument to simulate the model and log model
% signals.
%
% First ensure that we simulate the model with the parameter values chosen
% by the optimizer.
%
simulator.Parameters = p;
% Simulate the model and log signals.
%
simulator = sim(simulator);
% Get the simulation signal log, the simulation log name is defined by the
% model |SignalLoggingName| property
%
logName = get_param('sdoHydraulicCylinder','SignalLoggingName');
simLog = get(simulator.LoggedData,logName);

%% Evaluate the design requirements

%
% Use the requirements input argument to evaluate the design requirements
%
% Check the PistonPosition signal against the stepresponse requirement
%
PistonPosition = get(simLog,'PistonPosition');
cPiston = evalRequirement(requirements.PistonResponse,PistonPosition.Values);
% Check the Pressure signals against the maximum requirement
%
Pressures = find(simLog,'Pressures');
cPressure = evalRequirement(requirements.MaxPressure,Pressures.Values);
% Use the PistonResponse and MaxPressure requirements as non-linear
% constraints for optimization.
design.Cleq = [cPiston(:);cPressure(:)];
% Add design objective to minimize the Cylinder cross-sectional area
Ac = p(1); %Since we called sdo.optimize(evalDesign,[Ac;K])
design.F = Ac.Value;
end

Evaluate the Initial Design

Call the objective function with the initial cylinder cross-sectional area and initial piston
spring constant.
initDesign = evalDesign([Ac;K]);

3-130
 Design Optimization to Meet a Custom Objective (Code)

The function simulates the model and evaluates the design requirements. The scope
shows that the maximum pressure requirement is satisfied but the piston position step
response requirement is not satisfied.

initDesign is a structure with the following fields:

• Cleq shows that some of the inequality constraints are positive indicating they are
not satisfied by the initial design.

initDesign.Cleq

ans =

-0.3839
-0.1861
-0.1836
-1.0000
0.3033
0.2909
0.1671
0.2326
-0.0480
-0.0480

• F shows the optimization objective value (in this case the cylinder cross-sectional
area). The initial design cross-sectional area, as expected, has the same value as the
initial cross-sectional area parameter Ac.

initDesign.F

ans =

1.0000e-03

Optimize the Design

Pass the objective function, initial cross-sectional area and piston spring constant values
to sdo.optimize.

[pOpt,optInfo] = sdo.optimize(evalDesign,[Ac;K]);

3-131
3 Response Optimization

Optimization started 31-Jul-2015 06:22:40

max Step-size First-order

Iter F-count f(x) constraint optimality
0 5 0.001 0.3033
1 11 0.00057281 0.07293 0.48 85.4
2 17 0.000391755 0 0.128 28
3 22 0.000388463 0 0.00232 0.00409
4 27 0.000382784 0 0.00401 0.00231
5 32 0.000378554 0 0.00299 0.000545
Local minimum found that satisfies the constraints.

Optimization completed because the objective function is non-decreasing in

feasible directions, to within the selected value of the function tolerance,
and constraints are satisfied to within the selected value of the constraint tolerance.

The optimization repeatedly evaluates the cylinder design by adjusting the cross-
sectional area and piston spring constant to meet the design requirements. From the
scope, see that the maximum pressure and piston response requirements are met.

The sdo.optimize function returns:

• pOpt shows the optimized cross-sectional area and piston spring constant values.
pOpt

pOpt(1,1) =

Name: 'Ac'
Value: 3.7855e-04
Minimum: 3.1416e-04
Maximum: 0.0013
Free: 1
Scale: 0.0020
Info: [1x1 struct]

pOpt(2,1) =

Name: 'K'
Value: 1.5816e+04
Minimum: 10000
Maximum: 100000

3-132
 Design Optimization to Meet a Custom Objective (Code)

Free: 1
Scale: 65536
Info: [1x1 struct]

2x1 param.Continuous

• optInfo is a structure that contains optimization termination information such as

number of optimization iterations and the optimized design.
optInfo

optInfo =

Cleq: [10x1 double]

F: 3.7855e-04
Gradients: [1x1 struct]
exitflag: 1
iterations: 5
SolverOutput: [1x1 struct]
Stats: [1x1 struct]

For example, the Cleq field shows the optimized non-linear inequality constraints are all
non-positive to with-in optimization tolerances, indicating that the maximum pressure
and piston response requirements are satisfied.
optInfo.Cleq

ans =

-0.0968
-0.0126
-0.0126
-1.0000
-0.2067
-0.0052
-0.0074
-0.0004
-0.0476
-0.0476

3-133
3 Response Optimization

The F field contains the optimized cross-sectional area. The optimized cross-sectional
area value is nearly 50% less that the initial value.
optInfo.F

ans =

3.7855e-04

Update the Model Variable Values

By default, the model variables Ac and K are not updated at the end of optimization. Use
the setValueInModel command to update the model variable values.

sdo.setValueInModel('sdoHydraulicCylinder',pOpt)

Related Examples

To learn how to optimize the cylinder design using the Response Optimization tool, see
"Design Optimization to Meet Custom Objective (GUI)".
% Close the model
bdclose('sdoHydraulicCylinder')

3-134
 Design Optimization to Meet Custom Signal Requirements (GUI)

Design Optimization to Meet Custom Signal Requirements (GUI)

This example shows how to optimize a design to meet a custom signal requirement. You
optimize the controller parameters to minimize the plant actuation signal energy while
satisfying step response requirements.

1 Load a saved Response Optimization tool session.

load sldo_model1_custom_signal_session
sdotool(SDOSessionData);

The following Simulink model opens.

The Response Optimization tool, configured with the following settings, also opens:

• Step response characteristics, specified on the output of the Plant block, that the
model output must satisfy:

• Maximum overshoot of 5%
• Maximum rise time of 10 seconds
• Maximum settling time of 30 seconds
• Design variable set with the controller parameters Kp, Ki and Kd. These
parameters have a minimum value of 0.
• The variables for step requirements (PlantResponse), logged signal
(PlantOutput) and design variables (DesignVars) which appear in the Data
area.

3-135
3 Response Optimization

2 Specify a signal to log. You apply the custom requirement on this logged signal.

a Select New > Signal.

A window opens where you select a signal to log.

b In the Simulink model window, click the output of the Controller block.

The window updates to display the selected signal.

c
Select the signal and click to add it to the signal set.

3-136
 Design Optimization to Meet Custom Signal Requirements (GUI)

d In Signal set, enter PlantActuator.

Click OK. A new variable PlantActuator appears in the Data area.

3 Specify the custom requirement to apply to the signal.

The custom requirement calls the objective function

sldo_model1_minimize_energy which returns the energy in the PlantActuator
signal. The signal energy is minimized. This function accepts:

• An input argument data which is a structure with fields for the design variables
in the Data area. Signals are logged for the nominal and uncertain parameter
values if there are any.
• Returns the objective value to be minimized.

Tip To see the contents of this function, type edit

sldo_model1_minimize_energy .

a Select New > Custom Requirement.

A window opens where you specify the custom requirement.

b Specify MinimizeEnergy as the Name.
c Specify @sldo_model1_minimize_energy as the Function.
d Select Minimize the function output as the Type.

3-137
3 Response Optimization

4 In the Select Signals and Systems to Bound area, select the PlantActuator
check box to associate the custom requirement with that signal.

Click OK. A new variable appears in the Data area of the tool. The window also
updates to graphically display the custom signal requirement.
5 Click Optimize.

3-138
 Design Optimization to Meet Custom Signal Requirements (GUI)

After a few iterations, the optimization converges to meet both the custom signal and
step response requirements.

6 Close the model.

setOption(sdotool('sldo_model1'),'NoPromptClose',true)
bdclose('sldo_model1')

3-139
3 Response Optimization

Related Examples
• “Design Optimization Using Frequency-Domain Check Blocks (GUI)”

3-140
 Design Optimization to Meet Frequency-Domain Requirements (GUI)

Design Optimization to Meet Frequency-Domain Requirements

(GUI)
This example shows how to tune model parameters to meet frequency-domain
requirements using the Response Optimization tool.

This example requires Simulink® Control Design™.

Suspension Model

Open the Simulink Model.

open_system('sdoSimpleSuspension')

Mass-spring-damper models represent simple suspension systems and for this example
we tune the system to meet typical suspension requirements. The model implements

3-141
3 Response Optimization

the second order system representing a mass-spring-damper using Simulink blocks and
includes:

• a Mass gain block parameterized by the total suspended mass, m0+mLoad. The total
mass is the sum of a nominal mass,|m0|, and a variable load mass, mLoad.

• a Damper gain block parameterized by the damping coefficient, b.

• a Spring gain block parameterized by the spring constant, k.

• two integrator blocks to compute the mass velocity and position.

• a Band-Limited Disturbance Force block applying a disturbance force to the

mass. The disturbance force is assumed to be band-limited white noise.

Simulate the model to view the system response to the applied disturbance force.

3-142
 Design Optimization to Meet Frequency-Domain Requirements (GUI)

Design Problem

The initial system has a bandwidth that is too high. This can be seen from the
spiky position signal. You tune the spring and damper values to meet the following
requirements:

• The -3dB system bandwidth must not exceed 10 rad/s.

• The damping ratio of the system must be less than 1/sqrt(2). This ensures that no
frequencies in pass band are amplified by the system.

• Minimize the expected failure rate of the system. The expected failure rate is
described by a Weibull distribution dependent on the mass, spring, and damper
values.

• These requirements must all be satisfied as the load mass ranges from 0 to 20.

Open the Response Optimization Tool

In the Simulink model Analysis menu, select Response Optimization.

Specify Design Variables

In the Design Variables Set list, select New. Add the b and k model variables to the
design variable set.

• Specify the Minimum and Maximum values for the b variable as 100 and 10000
respectively.

• Specify the Minimum and Maximum values for the k variable as 10000 and 100000
respectively.

3-143
3 Response Optimization

Click OK. A new variable, DesignVars, appears in the Response Optimization Tool
Workspace.

In the Uncertain Variables Set list, select New. Add the mLoad variable to the
uncertain variables set.

• Specify the Uncertain Values value for the mLoad variable as [10 15 20]

Click OK. A new variable, UncVars, appears in the Response Optimization Tool
Workspace.

Specify Linear Analysis Input/Output Points

Specify the input/output points defining the linear system used to compute the
bandwidth and damping ratio.

To specify the input/output points:

• In the New list, select Linearization I/Os.

• In the Simulink model, click the signal at the output of the Band-Limited
Disturbance Force block. The Create linearization I/O set dialog box is updated
and the chosen signal appears in it.

• In the Create linearization I/O dialog box, select the signal and click Add.

• In the Configuration list for the selected signal, choose Input Perturbation to
specify it as an input signal.

3-144
 Design Optimization to Meet Frequency-Domain Requirements (GUI)

• Similarly, add the pos signal from the Simulink model. Specify this signal as an
output. In the Configuration list, select Output Measurement.

3-145
3 Response Optimization

• Click OK. A new variable, IOs, appears in the Response Optimization Tool
Workspace.

Add Bandwidth and Damping-Ratio Requirements

Tune the spring and damper values to satisfy bandwidth and damping ratio
requirements.

To specify the bandwidth requirement:

• Open a dialog to specify bounds on the Bode magnitude. In the New list, select Bode
Magnitude.

• Specify the requirement name as Bandwidth.

• Specify the edge start frequency and magnitude as 10 rad/s and -3db, respectively.

• Specify the edge end frequency and magnitude as 100 rad/s and -3db, respectively.

• Specify the input/output set to which the requirement applies by clicking Select
Systems to Bound. Select the IOs check box .

3-146
 Design Optimization to Meet Frequency-Domain Requirements (GUI)

• Click OK. A new requirement, Bandwidth, appears in the Response Optimization

Tool Workspace and a graphical view of the bandwidth requirement is
automatically created.

To specify the damping ratio requirement:

3-147
3 Response Optimization

• Open a dialog to specify bounds on the damping ratio. In the New list, select
Damping Ratio.

• Specify the damping ratio bound value as 0.7071.

• Specify the input/output set to which the requirement applies by clicking Select
Systems to Bound. Select the IOs check box .

• Click OK. A new requirement, DampingRatio, appears in the Response

Optimization Tool Workspace and a graphical view of the damping ratio
requirement is automatically created.

3-148
 Design Optimization to Meet Frequency-Domain Requirements (GUI)

Add a Reliability Requirement

Tune the spring and damper values to minimize the expected failure rate over a lifetime
of 100e3 miles. The failure rate is computed using a Weibull distribution on the damping
ratio of the system. As the damping ratio increases the failure rate is expected to
increase.

Specify the reliability requirement as a custom requirement:

• Open a dialog box to specify the custom requirements. In the New list, select Custom
Requirement.

• Specify the custom requirement name as MinFailureRate.

• In the Specify Function area, select Minimize the function output from the
Type list.

• Specify the function as @sdoSuspension_FailureRate.

3-149
3 Response Optimization

• Click OK. A new requirement, MinFailureRate, appears in the Response

Optimization Tool Workspace and a graphical view of the custom requirement is
automatically created.

The @sdoSuspension_FailureRate function returns expected failure rate for a lifetime of

100e3 miles.

type sdoSuspension_FailureRate

function pFailure = sdoSuspension_FailureRate(data)

%SDOSUSPENDION_FAILURERATE
%
% The sdoSuspension_FailureRate function is used to define a custom
% requirement that can be used in the graphical SDTOOL environment.
%
% The |data| input argument is a structure with fields containing the
% design variable values chosen by the optimizer.
%
% The |pFailure| return argument is the failure rate to be minimized by the
% SDOTOOL optimization solver. The failure rate is given by a Weibull
% distribution that is a function of the mass, spring and damper values.
% The design minimizes the failure rate for a 100e3 mile lifetime.
%

% Copyright 2012 The MathWorks, Inc.

%Get the spring and damper design values

allVarNames = {data.DesignVars.Name};
idx = strcmp(allVarNames,'k');
k = data.DesignVars(idx).Value;
idx = strcmp(allVarNames,'b');
b = data.DesignVars(idx).Value;

%Get the nominal mass from the model workspace

wksp = get_param('sdoSimpleSuspension','ModelWorkspace');
m = evalin(wksp,'m0');

%The expected failure rate is defined by the Weibull cumulative

%distribution function, 1-exp(-(x/l)^k), where k=3, l is a function of the
%mass, spring and damper values, and x the lifetime.
d = b/2/sqrt(m*k);
pFailure = 1-exp(-(100e3*d/250e3)^3);
end

3-150
 Design Optimization to Meet Frequency-Domain Requirements (GUI)

Optimize the Design

Before running the optimization be sure to have completed the earlier steps.
Alternatively, you can load the sdoSimpleSuspension_sdosession from the model
workspace into the Response Optimization tool.

To save the initial design variable values and later compare them with the optimized
values configure the optimization.

• Click Options.

• Select the Save optimized variable values as new design variable set
option.

To study how the design variable values change during optimization:

• In the Data to Plot list, select DesignVars.

• In the Add Plot list, and select Iteration Plot.

• View the design variables in an appropriately scaled manner. Right-click on the

DesignVars plot and select Show scaled values.

To evaluate the requirements at the initial design point, click Evaluate Requirements
. The requirement plots are updated and a ReqValues variable is added to the
Response Optimization Tool Workspace.

3-151
3 Response Optimization

To optimize the design, click Optimize. The plots are updated during optimization.
At the end of optimization, the optimal design values are written to the DesignVars1
variable. The requirement values for the optimized design are written to the
ReqValues1 variable.

3-152
Design Optimization to Meet Frequency-Domain Requirements (GUI)

3-153
3 Response Optimization

Analyze the Design

To compare design variables before and after optimization:

• In the Data to Plot list, select DesignVars.

• In the Add Plot list, select Spider Plot.

• To add the optimized design variables to the same plot, select DesignVars1 in
the Response Optimization Tool Workspace and drag it onto the Spider plot.
Alternatively, in the Data to Plot list, select DesignVars1. Then, in the Add Plot
list, select Spider plot 1 from the Add to Existing Plot section.

3-154
 Design Optimization to Meet Frequency-Domain Requirements (GUI)

The plot shows that the optimizer reduced both the k and b values for the optimal design.

3-155
3 Response Optimization

To compare requirements before and after optimization:

• In the Data to Plot list, select ReqValues.

• In the Add Plot list, select Spider Plot.

• To add the optimized requirement values to the same plot, select ReqValues1 in
the Response Optimization Tool Workspace and drag it onto the Spider plot.
Alternatively, in the Data to Plot list, select ReqValues1. Then, in the Add Plot
list, select Spider plot 2 from the Add to Existing Plot section.

3-156
Design Optimization to Meet Frequency-Domain Requirements (GUI)

3-157
3 Response Optimization

The plot shows that the optimal design has a lower failure rate (the MinFailureRate axis)
and better satisfies the bandwidth requirement. The value plotted on the bandwidth
axis is the difference between the bandwidth bound and the bandwidth value. The
optimization satisfies the bound by keeping this value negative; a more negative value
indicates better satisfaction of the bound.

The improved reliability and bandwidth are achieved by pushing the damping ratio
closer to the damping ratio bound. The plot has two axes for the damping ratio
requirement, one for each system pole, and the plotted values are the difference between
the damping ratio bound and the damping ratio value. The optimization satisfies the
bound by keeping this value negative.

Finally the simulated mass position is smoother than the initial position response
(indication of a lower bandwidth as required) at the expense of larger position deflection.

Related Examples

To learn how to optimize the suspension design using the sdo.optimize command, see
"Design Optimization to Meet Frequency-Domain Requirements (Code)".

3-158
 Design Optimization to Meet Frequency-Domain Requirements (GUI)

% Close the model

bdclose('sdoSimpleSuspension')

3-159
3 Response Optimization

Specify Custom Signal Objective with Uncertain Variable (GUI)

This example shows how to specify a custom objective function for a model signal.
You calculate the objective function value using a variable that models parameter
uncertainty.

Competitive Population Dynamics Model

The Simulink model sdoPopulation models a simple two-organism ecology using the
competitive Lotka-Volterra equations:

• is the population size of the n-th organism.

• is the inherent per capita growth rate of each organism.

• is the competitive delay for each organism.

• is the carrying capacity of the organism environment.

• is the proximity of the two populations and how strongly they affect each other.

The model uses normalized units.

Open the model.

open_system('sdoPopulation')

3-160
Specify Custom Signal Objective with Uncertain Variable (GUI)

3-161
3 Response Optimization

3-162
 Specify Custom Signal Objective with Uncertain Variable (GUI)

The two-dimensional signal, P, models the population sizes for P1 (first element) and P2
(second element). The model is initially configured with one organism, P1, dominating
the ecology. The Population scope shows the P1 population oscillating between high
and low values, while P2 is constant at 0.1. The Population Phase Portrait block
shows the population sizes of the two organisms in relation to each other.

Population Stabilization Design Problem

Tune the , , and values to meet the following design requirements:

• Minimize the population range, that is, the maximum difference between P1 and P2.

• Stabilize P1 and P2, that is, ensure that neither organism population dies off or grows
extremely large.

You must tune the parameters for different values of the carrying capacity, . This
ensures robustness to environment carrying-capacity uncertainty.

Open Response Optimization Tool

Double-click the Open Optimization Tool block in the model to open a pre configured
Response Optimization tool session. The session specifies the following variables:

3-163
3 Response Optimization

• DesignVars - Design variables set for the , and model parameters.

• K_unc - Uncertain parameter modeling the carrying capacity of the organism

environment ( ). K_unc specifies the nominal value and two sample values.

• P1 and P2 - Logged signals representing the populations of the two organisms.

Specify Custom Signal Objective Function

Specify a custom requirement to minimize the maximum difference between the two
population sizes. Apply this requirement to the P1 and P2 model signals.

1 Open the Create Requirement dialog box. In the New list, select Custom
Requirement.
2 Specify the following in the Create Requirement dialog box:

• Name - Enter PopulationRange.

• Type - Select Minimize the function output from the list.
• Function - Enter @sdoPopulation_PopRange. For more information about this
function, see Custom Signal Objective Function Details.
• Select Signals and Systems to Bound (Optional) - Select the P1 and P2 check
boxes.

3-164
Specify Custom Signal Objective with Uncertain Variable (GUI)

3-165
3 Response Optimization

3. Click OK.

A new variable, PopulationRange, appears in the Response Optimization Tool

Workspace.

Custom Signal Objective Function Details

PopulationRange uses the sdoPopulation_PopRange function. This function

computes the maximum difference between the populations, across different environment
carrying capacity values. By minimizing this value, you can achieve both design goals.
The function is called by the optimizer at each iteration step.

To view the function, type edit sdoPopulation_PopRange. The following discusses

details of this function.

Input/Output

The function accepts data, a structure with the following fields:

• DesignVars - Current iteration values of , and .

• Nominal - Logged signal data, obtained by simulating the model using parameter
values specified by data.DesignVars and nominal values for all other parameters.
The Nominal field is itself a structure with fields for each logged signal. The field
names are the logged signal names. The custom requirement uses the logged signals,
P1 and P2. Therefore, data.Nominal.P1 and data.Nominal.P2 are timeseries
objects corresponding to P1 and P2.

• Uncertain - Logged signal data, obtained by simulating the model using the
sample values of the uncertain variable K_unc. The Uncertain field is a vector of N
structures, where N is the number of sample values specified for K_unc. Each element
of this vector is similar to data.Nominal and contains simulation results obtained
from a corresponding sample value specified for K_unc.

The function returns the maximum difference between the population sizes across
different carrying capacities. The following code snippet in the function performs this
action:

val = max(maxP(1)-minP(2),maxP(2)-minP(1));

Data Time Range

3-166
 Specify Custom Signal Objective with Uncertain Variable (GUI)

When computing the design goals, discard the initial population growth data to eliminate
biases from the initial-condition. The following code snippet in the function performs this
action:

%Get the population data

tMin = 5; %Ignore signal values prior to this time
iTime = data.Nominal.P1.Time > tMin;
sigData = [data.Nominal.P1.Data(iTime), data.Nominal.P2.Data(iTime)];

iTime represents the time interval of interest, and the columns of sigData contain P1
and P2 data for this interval.

Optimization for Different Values of Carrying Capacity

The function includes the effects of varying the carrying capacity by iterating through the
elements of data.Uncertain. The following code snippet in the function performs this
action:

...
for ct=1:numel(data.Uncertain)
iTime = data.Uncertain(ct).P1.Time > tMin;
sigData = [data.Uncertain(ct).P1.Data(iTime), data.Uncertain(ct).P2.Data(iTime)];

maxP = max([maxP; max(sigData)]); %Update maximum if new signals are bigger

minP = min([minP; min(sigData)]); %Update minimum if new signals are smaller
end
...

The maximum and minimal populations are obtained across all the simulations
contained in data.Uncertain.

Optimize Design

Click Optimize.

The optimization converges after a number of iterations.

3-167
3 Response Optimization

The P1,P2 plot shows the population dynamics, with the first organism population in
blue and the second organism population in red. The dotted lines indicate the population
dynamics for different environment capacity values. The PopulationRange plot shows
that the maximum difference between the two organism populations reduces over time.

3-168
 Specify Custom Signal Objective with Uncertain Variable (GUI)

The Population Phase Portrait block shows the populations initially varying, but
they eventually converge to stable population sizes.

3-169
3 Response Optimization

% Close the model

bdclose('sdoPopulation')

3-170
 Design Optimization with Uncertain Variables (Code)

Design Optimization with Uncertain Variables (Code)

This example shows how to optimize a design when there are uncertain variables. You
optimize the dimensions of a Continuously Stirred Tank Reactor (CSTR) to minimize
product concentration variation and production cost in case of varying, or uncertain, feed
stock.

Continuously Stirred Tank Reactor (CSTR) Model

Continuously Stirred Tank Reactors (CSTRs) are common in the process industry. The
Simulink model, sdoCSTR, models a jacketed diabatic (i.e., non-adiabatic) tank reactor
described in [1]. The CSTR is assumed to be perfectly mixed, with a single first-order
exothermic and irreversible reaction, . , the reactant, is converted to , the
product.

In this example, you use the following two-state CSTR model, which uses basic
accounting and energy conservation principles:

• , and - Concentrations of A in the CSTR and in the feed [kgmol/m^3]

• , , and - CSTR, feed, and coolant temperatures [K]

• and - Volumetric flow rate [m^3/h] and the density of the material in the CSTR
[1/m^3]

• and - Height [m] and heated cross-sectional area [m^2] of the CSTR.

• - Pre-exponential non-thermal factor for reaction [1/h]

3-171
3 Response Optimization

• and - Activation energy and heat of reaction for [kcal/kgmol]

• - Boltzmann's gas constant [kcal/(kgmol * K)]

• and - Heat capacity [kcal/K] and heat transfer coefficients [kcal/(m^2 * K * h)]

Open the Simulink model.

open_system('sdoCSTR');

The model includes a cascaded PID controller in the Controller subsystem. The
controller regulates the reactor temperature, , and reactor residual concentration, .

CSTR Design Problem

Assume that the CSTR is cylindrical, with the coolant applied to the base of the cylinder.
Tune the CSTR cross-sectional area, , and CSTR height, , to meet the following design
goals:

• Minimize the variation in residual concentration, . Variations in the residual

concentration negatively affect the quality of the CSTR product. Minimizing the
variations also improves CSTR profit.

3-172
 Design Optimization with Uncertain Variables (Code)

• Minimize the mean coolant temperature . Heating or cooling the jacket coolant
temperature is expensive. Minimizing the mean coolant temperature improves CSTR
profit.

The design must allow for variations in the quality of supply feed concentration, ,
and feed temperature, . The CSTR is fed with feed from different suppliers. The
quality of the feed differs from supplier to supplier and also varies within each supply
batch.

Specify Design Variables

Select the following model parameters as design variables for optimization:

• Cylinder cross-sectional area

• Cylinder height

p = sdo.getParameterFromModel('sdoCSTR',{'A','h'});

Limit the cross-sectional area to a range of [1 2] m^2.

p(1).Minimum = 1;
p(1).Maximum = 2;

Limit the height to a range of [1 3] m.

p(2).Minimum = 1;
p(2).Maximum = 3;

Specify Uncertain Variables

Select the feed concentration and feed temperature as uncertain variables. You evaluate
the design using different values of feed temperature and concentration.

pUnc = sdo.getParameterFromModel('sdoCSTR',{'FeedCon0','FeedTemp0'});

Create a parameter space for the uncertain variables. Use normal distributions for both
variables. Specify the mean as the current parameter value. Specify a variance of 5% of
the mean for the feed concentration and 1% of the mean for the temperature.

uSpace = sdo.ParameterSpace(pUnc);
uSpace = setDistribution(uSpace,'FeedCon0',makedist('normal',pUnc(1).Value,0.05*pUnc(1)
uSpace = setDistribution(uSpace,'FeedTemp0',makedist('normal',pUnc(2).Value,0.01*pUnc(2

3-173
3 Response Optimization

The feed concentration is inversely correlated with the feed temperature. Add this
information to the parameter space.
%uSpace.RankCorrelation = [1 -0.6; -0.6 1];

The rank correlation matrix has a row and column for each parameter with the (i,j) entry
specifying the correlation between the i and j parameters.

Sample the parameter space. The scatter plot shows the correlation between
concentration and temperature.
rng('default'); %For reproducibility
uSmpl = sdo.sample(uSpace,60);
sdo.scatterPlot(uSmpl)

3-174
 Design Optimization with Uncertain Variables (Code)

Ideally you want to evaluate the design for every combination of points in the design and
uncertain spaces, which implies 30*60 = 1800 simulations. Each simulation takes around
0.5 sec. You can use parallel computing to speed up the evaluation. For this example
you instead only use the samples that have maximum & minimum concentration and
temperature values, reducing the evaluation time to around 1 min.
[~,iminC] = min(uSmpl.FeedCon0);
[~,imaxC] = max(uSmpl.FeedCon0);
[~,iminT] = min(uSmpl.FeedTemp0);
[~,imaxT] = max(uSmpl.FeedTemp0);
uSmpl = uSmpl(unique([iminC,imaxC,iminT,imaxT]) ,:);

Specify Design Requirements

The design requirements require logging model signals. During optimization, the model
is simulated using the current value of the design variables. Logged signals are used to
evaluate the design requirements.

Log the following signals:

• CSTR concentration, available at the second output port of the sdoCSTR/CSTR block
Conc = Simulink.SimulationData.SignalLoggingInfo;
Conc.BlockPath = 'sdoCSTR/CSTR';
Conc.OutputPortIndex = 2;
Conc.LoggingInfo.NameMode = 1;
Conc.LoggingInfo.LoggingName = 'Concentration';

• Coolant temperature, available at the first output of the sdoCSTR/Controller block

Coolant = Simulink.SimulationData.SignalLoggingInfo;
Coolant.BlockPath = 'sdoCSTR/Controller';
Coolant.OutputPortIndex = 1;
Coolant.LoggingInfo.NameMode = 1;
Coolant.LoggingInfo.LoggingName = 'Coolant';

Create and configure a simulation test object to log the required signals.
simulator = sdo.SimulationTest('sdoCSTR');
simulator.LoggingInfo.Signals = [Conc,Coolant];

Create Objective/Constraint Function

Create a function to evaluate the CSTR design. This function is called at each
optimization iteration.

3-175
3 Response Optimization

Use an anonymous function with one argument that calls the sdoCSTR_design function.

evalDesign = @(p) sdoCSTR_design(p,simulator,pUnc,uSmpl);

The evalDesign function:

• Has one input argument that specifies the CSTR dimensions

• Returns the optimization objective value

The sdoCSTR_design function uses a for loop that iterates through the sample values
specified for the feed concentration. Within the loop, the function:

• Simulates the model using the current iterate, feed concentration, and feed
temperature values

• Calculates the residual concentration variation and coolant temperature costs

To view the objective function, type edit sdoCSTR_design.

Evaluate Initial Design

Call the evalDesign function with the initial CSTR dimensions.

dInit = evalDesign(p)

dInit =

F: 11.3358
costConc: 6.4387
costCoolant: 4.8971

Plot the model response for the initial design. Simulate the model using the sample feed
concentration values. The plot shows the variation in the residual concentration and
coolant temperature.

sdoCSTR_plotModelResponse(p,simulator,pUnc,uSmpl);

3-176
 Design Optimization with Uncertain Variables (Code)

The sdoCSTR_plotModelResponse function plots the model response. To view this

function, type edit sdoCSTR_plotModelResponse.

Optimize Design
Pass the objective function and initial CSTR dimensions to sdo.optimize.
pOpt = sdo.optimize(evalDesign,p)

Optimization started 31-Jul-2015 06:16:19

max Step-size First-order

Iter F-count f(x) constraint optimality
0 4 5.17935 0

3-177
3 Response Optimization

1 8 3.81245 0 2.01 7.81

2 12 2.65827 0 0.574 3.06
3 16 2.53697 0 0.162 0.423
4 20 2.52022 0 0.0154 0.249
5 24 2.48533 0 0.072 0.163
6 28 2.47909 0 0.0285 0.107
7 37 2.47708 0 0.0019 0.143
8 42 2.46892 0 0.0477 0.452
9 50 2.46495 0 0.00891 0.397
10 65 2.46444 0 0.00127 0.351
11 72 2.46444 0 0.000801 0.351
Local minimum possible. Constraints satisfied.

fmincon stopped because the size of the current step is less than
the selected value of the step size tolerance and constraints are
satisfied to within the selected value of the constraint tolerance.

pOpt(1,1) =

Name: 'A'
Value: 2
Minimum: 1
Maximum: 2
Free: 1
Scale: 0.5000
Info: [1x1 struct]

pOpt(2,1) =

Name: 'h'
Value: 2.2093
Minimum: 1
Maximum: 3
Free: 1
Scale: 2
Info: [1x1 struct]

2x1 param.Continuous

3-178
 Design Optimization with Uncertain Variables (Code)

Evaluate Optimized Design

Call the evalDesign function with the optimized CSTR dimensions.

dFinal = evalDesign(pOpt)

dFinal =

F: 2.4644
costConc: 1.4454
costCoolant: 1.0191

Plot the model response for the optimized design. Simulate the model using the sample
feed concentration values. The optimized design reduces the residual concentration
variation and average coolant temperature for different feed stocks.

sdoCSTR_plotModelResponse(pOpt,simulator,pUnc,uSmpl);

3-179
3 Response Optimization

Related Examples

To learn how to use sensitivity analysis to explore the CSTR design space and select
an initial design for optimization, see "Design Optimization with Uncertain Variables
(Code)".

References

[1] Bequette, B.W. Process Dynamics: Modeling, Analysis and Simulation. 1st ed. Upper
Saddle River, NJ: Prentice Hall, 1998.
% Close the model
bdclose('sdoCSTR')

3-180
 Generate MATLAB Code for Design Optimization Problems (GUI)

Generate MATLAB Code for Design Optimization Problems (GUI)

This example shows how to automatically generate a MATLAB function to solve a
Design Optimization problem. You use the Response Optimization tool to define an
optimization problem for a hydraulic cylinder design and generate MATLAB code to solve
this optimization problem.

Hydraulic Cylinder Design Problem

The "Design Optimization to Meet a Custom Objective Using the Response Optimization
Tool" example shows how to use the Response Optimization tool to optimize a cylinder
design. In this example we load a pre-configured Response Optimization tool session
based on that example.

load sdoHydraulicCylinder_sdosession
sdotool(SDOSessionData)

3-181
3 Response Optimization

3-182
 Generate MATLAB Code for Design Optimization Problems (GUI)

Generate MATLAB Code

From the Optimize list, select Generate MATLAB Code.

3-183
3 Response Optimization

The generated code is added to the MATLAB editor as an unsaved MATLAB function.

3-184
 Generate MATLAB Code for Design Optimization Problems (GUI)

Examine the generated code. Significant code portions are:

• Specify Design Variables - Definition of the model parameters being optimized.

• Specify Design Requirements - Definition of the design requirements.

• Create Optimization Objective Function - Creation of an anonymous

function that calls the subfunction sdoHydraulicCylinder_optFcn, which
evaluates the cylinder design. sdo.optimize calls the anonymous function at each
iteration.

• Evaluate custom parameter requirement functions -

Evaluates the custom requirement, MinimizeAC, that uses the
sdoHydraulicCylinder_customObjective function.

3-185
3 Response Optimization

• Optimize the Design - Optimization using the sdo.optimize command.

Select Save from the MATLAB editor to save the generated function.

Run Generated Code

Run the generated function.

The first output argument, pOpt, contains the optimized parameter values and the
second output argument, optInfo, contains optimization information.

Modify the Generated Code

You can:

• Modify the generated sdo_sdoHydraulicCylinder function to include or exclude

new design requirements or change the optimization options.

• Call the generated sdo_sdoHydraulicCylinder function with a different set of

parameters to optimize.

For details on how to write an objective/constraint function to use with the

sdo.optimize command, type help sdoExampleCostFunction at the MATLAB
command prompt.

3-186
 Generate MATLAB Code for Design Optimization Problems (GUI)

Close the model

delete(sdotool('sdoHydraulicCylinder'))
bdclose('sdoHydraulicCylinder')

3-187
3 Response Optimization

Skip Model Simulation Based on Parameter Constraint Violation

(GUI)
This example shows how to optimize a design and specify parameter-only constraints
that prevent the model from being evaluated in an invalid solution space.

During optimization, the solver may try a design variable set that results in a model
simulation error, which can be computationally expensive. If you can define a parameter-
only constraint that identifies such a design variable set, then the solver can use the
constraint to skip such sets. In other words, you can configure the optimization to be
more efficient by disallowing design variable sets that lead to simulation errors.

In this example, you optimize thermostat settings to minimize temperature set-point

deviations while satisfying some constraints. One of the constraints applies to the model
parameters that define the thermostat switch on/switch off points. If the switch-off point
is greater than the switch-on point, evaluating the model leads to a simulation error.

Thermostat Model

Open the model.

open_system('sdoThermostat');

3-188
 Skip Model Simulation Based on Parameter Constraint Violation (GUI)

The model describes a simple heater & thermostat that regulate the temperature of a
room. The room is subject to external temperature fluctuations. The room temperature is
computed using a first-order heat-flow equation:

Where:

• is the room temperature (C).

• the external temperature (C).

• the heat supplied by the heater (W).

• the room thermal capacity (J/C).

The heater is controlled by a thermostat that turns on when the difference between the
room temperature and temperature set-point exceeds a threshold. The heater turns off
when the error drops below a threshold.

3-189
3 Response Optimization

The heater operation is displayed in the Heater use scope. The upper axis is the
delivered heat and the lower axis shows the times when the heater is switched on.

The room temperature is displayed in the Temperature scope.

3-190
 Skip Model Simulation Based on Parameter Constraint Violation (GUI)

Thermostat Design Problem

You tune the thermostat turn-on and turn-off temperature thresholds, and also the
heater power. The Thermostat switch block specifies the turn-on and turn-off
thresholds using the variables H_on and H_off. The Heater block specifies the heater
power using the variable Hgain.

The design requirements are:

• Minimize the difference between the room temperature and temperature set-point
over a 24 hour period.

• The heater must not turn on more than 12 times during the 24 hour period.

• The thermostat turn-on temperature must be greater than the thermostat turn-
off temperature. If this constraint is violated, the model is invalid and cannot be
simulated or evaluated.

Open the Response Optimization Tool

Open a pre-configured Response Optimization tool session.

load sdoThermostat_sdosession
sdotool(SDOSessionData)

3-191
3 Response Optimization

3-192
Skip Model Simulation Based on Parameter Constraint Violation (GUI)

3-193
3 Response Optimization

The pre-configured session specifies the following variables:

• DesignVars - Design variables set for the H_on, H_off, and Hgain model
parameters.
• Minimize_T_error - Requirement to minimize the temperature deviation from the
set-point.
• LimitH_on - Requirement to limit the number of times the thermostat is turned on.
• H_on_sig and T_error - Logged signals. H_on_sig represents when the heater is
on. T_error is the difference between the room temperature and the set-point.

3-194
 Skip Model Simulation Based on Parameter Constraint Violation (GUI)

Specify Parameter Constraint

The H_on > H_off requirement is not yet defined. Use a custom requirement to specify
this constraint and configure the requirement to error if it is not satisfied.

In the New drop-down list, select Custom Requirement. The Create Requirement
dialog opens.

In this dialog, specify the following:

• Name - SwitchConstraint.

• Type - Select Constrain the function output to be >= 0 from the Type list.

• Function - @sdoThermostat_SwitchingConstraint.

• Error if constraint is violated - Select this check box.

3-195
3 Response Optimization

The software calls the sdoThermostat_SwitchingConstraint function at each

optimization iteration with a structure containing all the design variables. The output
of the sdoThermostat_SwitchingConstraint function is the difference between
the H_on and H_off values. This difference must be positive for the requirement to be
satisfied.

The software evaluates custom requirements that test parameter-only constraints, such
as SwitchConstraint, before simulating the model and evaluating the remaining
requirements.

• If the constraint is violated while the Error if constraint is violated check box
is selected, the software does not simulate the model to evaluate the remaining
requirements. Instead, the solver assigns the cost function a NaN value for this
iteration, evaluates the terminating conditions, and continues.

• If the constraint is violated while the Error if constraint is violated check box
is cleared, the solver will attempt to simulate the model to evaluate the remaining
requirements. Simulating the model may lead to a hard error; for example, simulating
the thermostat model when SwitchConstraint is violated will lead to an error. In
this case, the solver assigns the cost function a NaN value for this iteration, evaluates
the terminating conditions, and continues.

To examine the constraint function, type edit

sdoThermostat_SwitchingConstraint. The requirement that H_on > H_off is
implemented as H_on - H_off > 0

Optimize the Design

Click Optimize.

3-196
Skip Model Simulation Based on Parameter Constraint Violation (GUI)

3-197
3 Response Optimization

The Optimization Progress window appears and updates at each iteration. The
optimization successfully minimizes the temperature error while satisfying the switching
constraints.

During this optimization, the H_on and H_off values never approach the H_on > H_off
constraint boundary. So, there is never a danger of violating the constraint. However,
changing the optimization algorithm may produce different behavior. For example,
changing the optimization algorithm from the one used here, 'Interior-Point', to 'Active-
Set' results in H_on and H_off values that are at the constraint boundary. This violation

3-198
 Skip Model Simulation Based on Parameter Constraint Violation (GUI)

triggers the SwitchConstraint requirement and prevents model simulation for the
relevant iterations.

View Optimized Model Response

Simulate the model with the optimized thermostat settings. The optimized heater
operation is displayed in the Heater use scope where the upper axis is the delivered
heat and the lower axis the heater switch on times.

3-199
3 Response Optimization

The optimized room temperature is displayed in the Temperature scope.

Close the model

delete(sdotool('sdoThermostat'))
bdclose('sdoThermostat')

3-200
 Optimizing Parameters for Robustness

Optimizing Parameters for Robustness

In this section...
“What Is Robustness?” on page 3-201
“Sampling Methods for Uncertain Parameters” on page 3-202
“Optimize Parameters for Robustness (GUI)” on page 3-206

What Is Robustness?
A design is robust when it's response does not violate design requirements under
model parameter variations. Your model may contain parameters whose values are not
precisely known. Such parameters vary over a given range of values and are defined as
uncertain parameters. You may know the nominal value and the range of values in which
these uncertain parameters vary.

You can use Simulink Design Optimization software to incorporate the parameter
uncertainty to test the robustness of your design. When you optimize parameters for
robustness, the optimization solver uses the responses computed using all the uncertain
parameter values to adjust the design variable values.

You can specify the same parameter both as a design and uncertain variable. However,
you cannot use a parameter both as a design and uncertain variable in the same
optimization run. Also, you cannot add uncertainty to controller or plant parameters
during optimization-based control design in the SISO Design Tool.

The uncertain variables can be scalar, vector, matrix or an expression.

You can test and optimize parameters for model robustness in the following ways:

• Before Optimization. Specify the parameter uncertainty before you optimize

the parameters to meet the design requirements. In this case, the optimization
method optimizes the signals based on both nominal parameter values as well as the
uncertain values. This mode requires more computational time.
• After Optimization. Specify the parameter uncertainty after you have optimized
the model parameters to meet design requirements. You can then test the effect of
the uncertain parameters by plotting the model's response. If the response violates
the design requirements, you can optimize the parameters again by including the
parameter uncertainty during the optimization.

3-201
3 Response Optimization

Related Examples

“Optimize Parameters for Robustness (GUI)” on page 3-206

More About

“Sampling Methods for Uncertain Parameters” on page 3-202

Sampling Methods for Uncertain Parameters

Sample values for uncertain parameters are a vector of numerical values. You can specify
the vector yourself or generate a vector of random numbers using the software. The
sample values you specify can be uniformly distributed or random. For example, four
sample values for two uncertain parameters a and b in the range [0 3] and [1 2.5]
may look like the following figure.

a
0 3

2.5

There are two methods to determine the number of sample values to use during
optimization:

• Only the combination of minimum and maximum values (circled)

3-202
 Optimizing Parameters for Robustness

0 a 3

2.5

• Combination of the entire set of values (all solid dots in the previous figure)

Tip Using only the minimum and maximum values during optimization increases the
computation speed when compared to using the entire set of values.

For the previous example, there are 4 combinations using the minimum and maximum
values and 16 combinations if you use all sample values.

In the Response Optimization tool, you specify the sampling method using the options as
shown in the following figure.

3-203
3 Response Optimization

3-204
 Optimizing Parameters for Robustness

Related Examples

“Optimize Parameters for Robustness (GUI)” on page 3-206

More About

• “What Is Robustness?” on page 3-201

3-205
3 Response Optimization

Optimize Parameters for Robustness (GUI)

This example shows how to optimize parameters for model robustness.

1 Load a saved Response Optimization tool session.

load sldo_model1_desreq_optim_sdosession;
sdotool(SDOSessionData);

The sdotool command opens the following Simulink model and a saved Response
Optimization tool session.

The parameters of this model, Kp, Ki and Kd, have already been optimized to meet
the following step response requirements:

• Maximum overshoot of 5%
• Maximum rise time of 10 seconds
• Maximum settling time of 30 seconds
2 Specify parameter uncertainty.

a In the Uncertain Variables Set drop-down list, select New.

A window opens where you specify uncertain variables.

3-206
 Optimizing Parameters for Robustness

b Click w0 and zeta to select them.

c
Click to add the selected parameters to an uncertain variables set.

The software displays the following parameter settings:

3-207
3 Response Optimization

• Variable — Parameter name

• Nominal Value — Nominal value of the parameters as specified in the
Simulink model
• Uncertain Values — Values that the uncertain parameter can take. By
default, the maximum and minimum values vary by 10% of the nominal
value.

The total number of sample values to use during optimization is a

combination of the maximum and minimum values of the uncertain
parameters.

The check-box indicates that the parameter is included in the uncertain variable
set. The default uncertain variable set name is UncVars.

Click OK. A new variable UncVars appears in Data area of the Response
Optimization tool.

Specify Random Values

Instead of specifying sample values, you can auto-generate random values in a

specific range. Select a parameter and click Set Uncertain Values.

A window opens where you specify the range and the number of samples.

3-208
 Optimizing Parameters for Robustness

3 Test the model robustness to the uncertain parameters.

a Click Plot Current Response.

The step response plot, displaying the requirements, updates.

3-209
3 Response Optimization

• The solid curve corresponds to the model response computed using the
optimized parameters and nominal values of the uncertain parameter.
• The four dashed curves correspond to the model response with the minimum
and maximum values of the uncertain parameters.

The dashed plot lines show that the response during the period of 10 to 20
seconds violates the design requirements.
4 Optimize the parameters for model robustness. Click Optimize.

The Optimization Progress window opens which displays the optimization iterations.

After the optimization completes, the message Optimization converged indicates

that the final model response computed by varying the uncertain parameters meets
the specified design requirements.

3-210
 Optimizing Parameters for Robustness

5 Examine the responses.

Tip To view only the final responses of the model, right-click the white area in the
plot and uncheck Responses > Show Iteration Responses.

3-211
3 Response Optimization

The final responses appear as the thick solid and dashed curves. The nominal and
uncertain responses with parameter variations now meet the design requirements.

More About

• “What Is Robustness?” on page 3-201

• “Sampling Methods for Uncertain Parameters” on page 3-202

Related Examples
• “Design Optimization with Uncertain Variables (Code)” on page 3-171

3-212
 How to Use Accelerator Mode During Simulations

How to Use Accelerator Mode During Simulations

In this section...
“About Accelerating Optimization” on page 3-213
“Limitations” on page 3-213
“Setting Accelerator Mode” on page 3-213

About Accelerating Optimization

Simulink Design Optimization software supports Normal and Accelerator simulation
modes. You can accelerate the design optimization computations by changing the
simulation mode of your Simulink model to Accelerator. For information about these
modes, see “How Acceleration Modes Work” in the Simulink documentation.

The default simulation mode is Normal. In this mode, Simulink uses interpreted code,
rather than compiled C code during simulations.

In the Accelerator mode, Simulink Design Optimization software runs simulations

during optimization with compiled C code. Using compiled C code speeds up the
simulations and reduces the time to optimize the model response signals.

Limitations
You cannot use the Accelerator mode if your model contains algebraic loops. If the
model contains MATLAB function blocks, you must either remove them or replace them
with Fcn blocks.

If the model structure changes during optimization, the model is compiled to regenerate
the C code for each iteration. In this case, using the Accelerator mode increases the
computation time. To learn more about code regeneration, see “Code Regeneration in
Accelerated Models” in the Simulink documentation.

Setting Accelerator Mode

To set the simulation mode to Accelerator, open the Simulink model window and
perform one of the following actions:

• Select under Simulation > Mode > Accelerator.

3-213
3 Response Optimization

• Choose Accelerator from the drop-down list as shown in the next figure.

Tip To obtain the maximum performance from the Accelerator mode, close all
Scope blocks in your model.

More About
• “Ways to Speed Up Design Optimization Tasks”

3-214
 Speed Up Response Optimization Using Parallel Computing

Speed Up Response Optimization Using Parallel Computing

In this section...
“When to Use Parallel Computing for Response Optimization” on page 3-215
“How Parallel Computing Speeds Up Optimization” on page 3-215

When to Use Parallel Computing for Response Optimization

You can use Simulink Design Optimization software with Parallel Computing Toolbox
software to speed up the response optimization of a Simulink model. Using parallel
computing may reduce model optimization time in the following cases:

• The model contains a large number of tuned parameters, and the Gradient
descent method is selected for optimization.
• The Pattern search method is selected for optimization.
• The model contains a large number of uncertain parameters and uncertain parameter
values.
• The model is complex and takes a long time to simulate.

When you use parallel computing, the software distributes independent simulations to
run them in parallel on multiple MATLAB sessions, also known as workers. Distributing
the simulations significantly reduces the optimization time because the time required to
simulate the model dominates the total optimization time.

For information on how the software distributes the simulations and the expected
speedup, see “How Parallel Computing Speeds Up Optimization” on page 3-215.

For information on configuring your system and using parallel computing, see “How to
Use Parallel Computing for Response Optimization” on page 3-219.

How Parallel Computing Speeds Up Optimization

You can enable parallel computing with the Gradient descent and Pattern search
optimization methods. When you enable parallel computing, the software distributes
independent simulations during optimization on multiple MATLAB sessions. The
following topics describe which simulations are distributed and the potential speedup
using parallel computing:

3-215
3 Response Optimization

• “Parallel Computing with the Gradient Descent Method” on page 3-216

• “Parallel Computing with the Pattern Search Method” on page 3-217

Parallel Computing with the Gradient Descent Method

When you select Gradient descent as the optimization method, the model is simulated
during the following computations:

• Constraint and objective value computation — One simulation per iteration

• Constraint and objective gradient computations — Two simulations for every tuned
parameter per iteration
• Line search computations — Multiple simulations per iteration

The total time, Ttotal , taken per iteration to perform these simulations is given by the
following equation:

Ttotal = T + ( Np ¥ 2 ¥ T ) + ( Nls ¥ T ) = T ¥ (1 + (2 ¥ Np) + Nls)

where T is the time taken to simulate the model and is assumed to be equal for all
simulations, Np is the number of tuned parameters, and Nls is the number of line
searches. Nls is difficult to estimate and you generally assume it to be equal to one, two,
or three.

When you use parallel computing, the software distributes the simulations required for
constraint and objective gradient computations. The simulation time taken per iteration
when the gradient computations are performed in parallel, TtotalP , is approximately
given by the following equation:

Ê Np ˆ Ê Np ˆ
TtotalP =T + ( ceil Á ˜ ¥ 2 ¥ T ) + ( Nls ¥ T ) = T ¥ (1 + 2 ¥ ceil Á ˜ + Nls)
Ë Nw ¯ Ë Nw ¯

where Nw is the number of MATLAB workers.

Note: The equation does not include the time overheads associated with configuring the
system for parallel computing and loading Simulink software on the remote MATLAB
workers.

3-216
 Speed Up Response Optimization Using Parallel Computing

The expected speedup for the total optimization time is given by the following equation:

Ê Np ˆ
1 + 2 ¥ ceil Á ˜ + Nls
TtotalP
= Ë Nw ¯
Ttotal 1 + ( 2 ¥ Np) + Nls

For example, for a model with Np=3, Nw=4, and Nls=3, the expected speedup equals
Ê3ˆ
1 + 2 ¥ ceil Á ˜ + 3
Ë4¯ = 0.6 .
1 + (2 ¥ 3) + 3

For an example of the performance improvement achieved with the Gradient descent
method, see Improving Optimization Performance Using Parallel Computing.

Parallel Computing with the Pattern Search Method

The Pattern search optimization method uses search and poll sets to create and
compute a set of candidate solutions at each optimization iteration.

The total time, Ttotal , taken per iteration to perform these simulations, is given by the
following equation:

Ttotal = (T ¥ Np ¥ Nss) + ( T ¥ Np ¥ Nps) = T ¥ Np ¥ ( Nss + Nps)

where T is the time taken to simulate the model and is assumed to be equal for all
simulations, Np is the number of tuned parameters, Nss is a factor for the search set
size, and Nps is a factor for the poll set size. Nss and Nps are typically proportional to
Np .

When you use parallel computing, Simulink Design Optimization software distributes
the simulations required for the search and poll set computations, which are evaluated in
separate parfor loops. The simulation time taken per iteration when the search and poll
sets are computed in parallel, TtotalP , is given by the following equation:

Nss Nps
TtotalP = ( T ¥ ceil( Np ¥ )) + ( T ¥ ceil( Np ¥ ))
Nw Nw
Nss Nps
= T ¥ ( ceil( Np ¥ ) + ceil( Np ¥ ))
Nw Nw

3-217
3 Response Optimization

where Nw is the number of MATLAB workers.

Note: The equation does not include the time overheads associated with configuring the
system for parallel computing and loading Simulink software on the remote MATLAB
workers.

The expected speed up for the total optimization time is given by the following equation:

Nss Nps
ceil( Np ¥ ) + ceil( Np ¥ )
TtotalP Nw Nw
=
Ttotal Np ¥ ( Nss + Nps)

For example, for a model with Np=3, Nw=4, Nss=15, and Nps=2, the expected speedup
15 2
ceil(3 ¥ ) + ceil(3 ¥ )
equals 4 4 = 0 .27 .
3 ¥ (15 + 2)

Note: Using the Pattern search method with parallel computing may not speed up
the optimization time. To learn more, see “Why do I not see the optimization speedup I
expected using parallel computing?” on page 3-236

For an example of the performance improvement achieved with the Pattern search
method, see Improving Optimization Performance Using Parallel Computing.

Related Examples
• “How to Use Parallel Computing for Response Optimization” on page 3-219

3-218
 How to Use Parallel Computing for Response Optimization

How to Use Parallel Computing for Response Optimization

In this section...
“Configure Your System for Parallel Computing” on page 3-219
“Model Dependencies” on page 3-219
“Optimize Design Using Parallel Computing (GUI)” on page 3-220
“Optimize Design Using Parallel Computing (Code)” on page 3-224
“Troubleshooting” on page 3-225

Configure Your System for Parallel Computing

You can speed up model optimization using parallel computing on multicore processors
or multiprocessor networks. Use parallel computing with the Response Optimization tool
and sdo.optimize to optimize using the fmincon, lsqonlin, and patternsearch
methods. Parallel computing is not supported for the fminsearch (Simplex search)
method.

When you optimize model parameters using parallel computing, the software uses the
available parallel pool. If none is available, and you select Automatically create a
parallel pool in your Parallel Computing Toolbox preferences, the software starts a
parallel pool using the settings in those preferences. To open a parallel pool that uses a
specific cluster profile, use:

parpool(MyProfile);

MyProfile is the name of a cluster profile.

For information regarding creating a cluster profile, see “Create and Modify Cluster
Profiles” in the Parallel Computing Toolbox documentation.

Model Dependencies
Model dependencies are any referenced models, data such as model variables, S-
functions, and additional files necessary to run the model. Before starting the
optimization, verify that the model dependencies are complete. Otherwise, you may get
unexpected results.

3-219
3 Response Optimization

Making Model Dependencies Accessible to Remote Workers

When you use parallel computing, the Simulink Design Optimization software helps you
identify model dependencies. To do so, the software uses the Simulink Manifest Tools.
The dependency analysis may not find all the files required by your model. To learn
more, see “Scope of Dependency Analysis” in the Simulink documentation. If your model
has dependencies that are undetected or inaccessible by the parallel pool workers, then
add them to the list of model dependencies.

The dependencies are made accessible to the parallel pool workers by specifying one of
the following:

• File dependencies: the model dependency files are copied to the parallel pool workers.
• Path dependencies: the paths to the model dependencies are specified to the parallel
pool workers. If you are working in a multi-platform scenario, ensure that the paths
are compatible across platforms.

Using file dependencies is recommended, however, in some cases it can be better to

choose path dependencies. For example, if parallel computing is set up on a local multi-
core computer, using path dependencies is preferred as using file dependencies creates
multiple copies of the dependency files on the local computer.

For more information, see:

• “Optimize Design Using Parallel Computing (GUI)” on page 3-220

• “Optimize Design Using Parallel Computing (Code)” on page 3-224

Optimize Design Using Parallel Computing (GUI)

To optimize a model response using parallel computing in the Response Optimization
tool:

1 Ensure that the software can access parallel pool workers that use the appropriate
cluster profile.

For more information, see “Configure Your System for Parallel Computing” on page
3-219.
2 Open the Response Optimization tool for the Simulink model.
3 Configure the design variables, design requirements, and, optionally, optimization
settings.

3-220
 How to Use Parallel Computing for Response Optimization

For more information, see “Specify Design Variables” on page 3-61, “Specify Time-
Domain Design Requirements” on page 3-23, “Specify Frequency-Domain Design
Requirements” on page 3-39, and “Optimization Options” on page 3-72.
4
On the Response Optimization tab, click Options to open the Response
Optimization Options dialog box.
5 Select the Parallel Options tab.

6 Select the Use the parallel pool during optimization check box.

This option checks for dependencies in your Simulink model. The file dependencies
are displayed in the Model file dependencies list box, and corresponding
path to the files in Model path dependencies. The files listed in Model file
dependencies are copied to the remote workers.

Note: The automatic dependencies check may not detect all the dependencies in your
model.

3-221
3 Response Optimization

For more information, see “Model Dependencies” on page 3-219. In this case, add
the undetected dependencies manually.

7 Add any file dependencies that the automatic check does not detect.

Specify the files in the Model file dependencies list box separated by semicolons
or on separate lines.

3-222
 How to Use Parallel Computing for Response Optimization

Alternatively, click Add file dependency to open a dialog box, and select the file to
add.

Note: If you do not want to copy the files to the remote workers, delete all entries in
the Model file dependencies list box. Populate the Model path dependencies
list box by clicking the Sync path dependencies from model, and add any
undetected path dependencies. In addition, in the list box, update the paths on local
drives to make them accessible to remote workers. For example, change C:\ to \\\
\hostname\\C$\\.
8 If you modify the Simulink model, resync the dependencies to ensure that any new
dependencies are detected. Click Sync file dependencies from model in the
Parallel Options tab to rerun the automatic dependency check for your model.

This action updates the Model file dependencies list box with any new file
dependency found in the model.
9 Click OK.
10 In the Response Optimization tool, click Optimize to optimize the model response
using parallel computing.

3-223
3 Response Optimization

For information on troubleshooting problems related to optimization using parallel

computing, see “Troubleshooting” on page 3-225.

Optimize Design Using Parallel Computing (Code)

To optimize a model response using parallel computing at the command line:

1 Ensure that the software can access parallel pool workers that use the appropriate
cluster profile.

For more information, see “Configure Your System for Parallel Computing” on page
3-219.
2 Open the model.
3 Specify design requirements and design variables. For example see, “Design
Optimization to Meet Step Response Requirements (Code)”.
4 Enable parallel computing using an optimization option set, opt.
opt = sdo.OptimizeOptions;
opt.UseParallel = 'always';
5 Find the model dependencies.
[dirs,files] = sdo.getModelDependencies(modelname)

Note: sdo.getModelDependencies may not detect all the dependencies in your

model. For more information, see “Model Dependencies” on page 3-219. In this
case, add the undetected dependencies manually.
6 Modify files to include any file dependencies that sdo.getModelDependencies
does not detect.
files = vertcat(files,'C:\matlab\work\filename.m')

Note: If you do not want to copy the files to the remote workers, use the path
dependencies. Add any undetected path dependencies to dirs and update
the paths on local drives to make them accessible to remote workers. See
sdo.getModelDependencies for more details.
7 Add the file dependencies for optimization.
opt.ParallelFileDependencies = files;

3-224
 How to Use Parallel Computing for Response Optimization

8 Run the optimization.

[pOpt,opt_info] = sdo.optimize(opt_fcn,param,opt);

For information on troubleshooting problems related to optimization using parallel

computing, see “Troubleshooting” on page 3-225.

Troubleshooting
• “Why Are the Optimization Results With and Without Using Parallel Computing
Different?” on page 3-225
• “Why Don’t I See the Optimization Speed up I Expected Using Parallel Computing?”
on page 3-225
• “Why Doesn’t the Optimization Using Parallel Computing Make Any Progress?” on
page 3-226
• “Why Doesn’t the Optimization Using Parallel Computing Stop When I Click the Stop
Optimization Button?” on page 3-226

Why Are the Optimization Results With and Without Using Parallel Computing Different?

• Different numerical precision on the client and worker machines can produce
marginally different simulation results. Thus, the optimization method can take a
different solution path and produce a different result.
• When you use parallel computing with the Pattern search method, the search is
more comprehensive and can result in a different solution.

To learn more, see “Parallel Computing with the Pattern Search Method” on page
3-217.

Why Don’t I See the Optimization Speed up I Expected Using Parallel Computing?

• When you optimize a model that does not have a large number of parameters or does
not take long to simulate, you might not see a speedup in the optimization time. In
such cases, the overhead associated with creating and distributing the parallel tasks
outweighs the benefits of running the optimization in parallel.
• Using the Pattern search method with parallel computing might not speed up
the optimization time. Without parallel computing, the method stops the search at
each iteration when it finds a solution better than the current solution. The candidate
solution search is more comprehensive when you use parallel computing. Although

3-225
3 Response Optimization

the number of iterations might be larger, the optimization without using parallel
computing might be faster.

To learn more about the expected speedup, see “Parallel Computing with the Pattern
Search Method” on page 3-217.

Why Doesn’t the Optimization Using Parallel Computing Make Any Progress?

To troubleshoot the problem:

1 Run the optimization for a few iterations without parallel computing to see if the
optimization progresses.
2 Check whether the remote workers have access to all model dependencies. Model
dependencies include data variables and files required by the model to run.

To learn more, see “Model Dependencies” on page 3-219.

Why Doesn’t the Optimization Using Parallel Computing Stop When I Click the Stop
Optimization Button?

When you use parallel computing with the Pattern search method, the software must
wait until the current optimization iteration completes before it notifies the workers
to stop. The optimization does not terminate immediately when you click Stop, and,
instead, appears to continue running.

See Also
sdo.OptimizeOptions | parpool | sdo.getModelDependencies | sdo.optimize

Related Examples
• “Optimizing Time-Domain Response of Simulink® Models Using Parallel
Computing” on page 3-243

More About
• “Speed Up Response Optimization Using Parallel Computing” on page 3-215
• “Ways to Speed Up Design Optimization Tasks”

3-226
 Use Fast Restart Mode During Response Optimization

Use Fast Restart Mode During Response Optimization

In this section...
“Response Optimization Tool Workflow for Fast Restart” on page 3-227
“Command-Line Workflow for Fast Restart” on page 3-228
“Troubleshooting” on page 3-229

This topic shows how to speed up response optimization using Simulink fast restart. You
can use the fast restart feature to speed up response optimization of tunable parameters
of a model.

Fast restart enables you to perform iterative simulations without compiling a model or
terminating the simulation each time. Using fast restart, you compile a model only once.
You can then tune parameters and simulate the model again without spending time on
compiling. Fast restart associates multiple simulation phases with a single compile phase
to make iterative simulations more efficient. You see a speedup of design optimization
tasks using fast restart in models that have a long compilation phase. See “How Fast
Restart Improves Iterative Simulations” in the Simulink documentation.

When you enable fast restart, you can only change tunable properties of the model during
simulation. For more information about the limitations, see “Factors Affecting Fast
Restart”.

You can optimize using fast restart in the Response Optimization tool or at the command
line.

Response Optimization Tool Workflow for Fast Restart

To optimize a model response using fast restart in the Response Optimization tool:

1 Open the Simulink model.

2 Enable fast restart in the model.

Click Fast Restart in the model window.

3 Open the Response Optimization tool for the model.
4 Configure the design variables, design requirements, and, optionally, optimization
settings.

3-227
3 Response Optimization

For more information, see “Specify Design Variables” on page 3-61, “Specify Time-
Domain Design Requirements” on page 3-23, “Specify Frequency-Domain Design
Requirements” on page 3-39, and “Optimization Options” on page 3-72.
5 Click Optimize to optimize the model response in fast restart mode.
6 Disable fast restart.

In the model window, click Fast Restart .

Command-Line Workflow for Fast Restart

To optimize a model response using fast restart at the command line:

1 Open the Simulink model.

2 Create a model simulation scenario. You must create a simulation scenario with
logging information before configuring the model for fast restart. You cannot modify
logging information once the model has been compiled for fast restart.

Simulator = sdo.SimulationTest('model');

Specify model signals to log during model simulation.

For response optimization problems that include frequency-domain

requirements, the model is linearized using Simulink Control Design. Use the
SystemLoggingInfo property of the sdo.SimulationTest object, Simulator,
to specify linear systems to log when simulating the model. For an example, see
“Design Optimization to Meet Frequency-Domain Requirements (Code)” on page
3-252.

Note: In fast restart mode, you cannot use the linearize command from Simulink
Control Design to specify and compute linear systems. Using linearize generates
an error.
3 Specify design requirements, Requirements, and design variables, param. For and
example, see “Design Optimization to Meet Step Response Requirements (Code)”.
4 Configure the model and simulation scenario for fast restart.

Simulator = fastRestart(Simulator,'on');

3-228
 Use Fast Restart Mode During Response Optimization

5 Create an optimization cost function, myCostfcn, and pass Simulator to the cost
function as an input. For more information, see “Write a Cost Function” on page
2-83. In the cost function, the simulator configured for fast restart is used to update
the model parameters, simulate the model, and log signals.

Use an anonymous function with one argument that calls myCostfcn.

optimfcn = @(param) myCostfcn(param,Simulator,Requirements);

Here, myCostfcn is a cost function that takes design variables, param, simulation
scenario, Simulator, and design requirements, Requirements, as inputs.
6 Run the optimization.
[param_opt,opt_info] = sdo.optimize(optimfcn,param);
7 Restore the simulator fast restart settings.
Simulator = fastRestart(Simulator,'off');

Troubleshooting
Why Don’t I See the Optimization Speedup I Expected Using Fast Restart?

You see a speedup of design optimization tasks using fast restart in models that have a
long compilation phase. If the compilation phase of your model is not long, you do not see
a significant change in optimization speed.

See Also
fastRestart | sdo.optimize | sdo.SimulationTest

Related Examples
• “Improving Optimization Performance using Fast Restart (GUI)” on page 2-198
• “Improving Optimization Performance using Fast Restart (Code)” on page 2-206

More About
• “Ways to Speed Up Design Optimization Tasks”

3-229
3 Response Optimization

Optimization Does Not Make Progress

In this section...
“Should I worry about the scale of my responses and how constraints and design
requirements are discretized?” on page 3-230
“Why don't the responses and parameter values change at all?” on page 3-230
“Why does the optimization stall?” on page 3-230

Should I worry about the scale of my responses and how constraints and
design requirements are discretized?
No. Simulink Design Optimization software automatically normalizes constraints, design
requirement and response data.

Why don't the responses and parameter values change at all?

The optimization problem you formulated might be nonsmooth. This means that small
parameter changes have no effect on the amount by which response signals satisfy or
violate the constraints and only large changes will make a difference. Try switching to
a search-based method such as simplex search or pattern search. Alternatively, look for
initial guesses outside of the dead zone where parameter changes have no effect. If you
are optimizing the response of a Simulink model, you could also try removing nonlinear
blocks such as Quantizer or Dead Zone.

Why does the optimization stall?

When optimizing a Simulink model, certain parameter combinations can make the
simulation stall for models with strong nonlinearities or frequent mode switching. In
these cases, the ODE solvers take smaller and smaller step sizes. Stalling can also occur
when the model's ODEs become too stiff for some parameter combinations. A symptom
of this behavior is when the Simulink model status is Running and clicking the Stop
button fails to interrupt the optimization. When this happens, you can try one of the
following solutions:

• Switch to a different ODE solver, especially one of the stiff solvers.

• Specify a minimum step size.

3-230
 Optimization Does Not Make Progress

• Disable zero crossing detection if chattering is occurring.

• Tighten the lower and upper bounds on parameters that cause simulation difficulties.
In particular, eliminate regions of the parameter space where some model
assumptions are invalid and the model behavior can become erratic.

3-231
3 Response Optimization

Optimization Convergence
In this section...
“What to do if the optimization does not get close to an acceptable solution?” on page
3-232
“Why does the optimization terminate before exceeding the maximum number
of iterations, with a solution that does not satisfy all the constraints or design
requirements?” on page 3-233
“What to do if the optimization takes a long time to converge even though it is close to a
solution?” on page 3-233
“What to do if the response becomes unstable and does not recover?” on page 3-234

What to do if the optimization does not get close to an acceptable

solution?
• If you are using pattern search, check that you have specified appropriate maximum
and minimum values for all your tuned parameters or compensator elements. The
pattern search method looks inside these bounds for a solution. When they are set to
their default values of Inf and -Inf, the method searches within ±100% of the initial
values of the parameters. In some cases this region is not large enough and changing
the maximum and minimum values can expand the search region.
• Your optimization problem might have local minima. Consider running one of the
search-based methods first to get closer to an acceptable solution.
• Reduce the number of tuned parameters and compensator elements by removing from
the design variables or from the Compensators pane (when using a SISO Design
Task) those parameters that you know only mildly influence the optimized responses.
After you identify reasonable values for the key parameters, add the fixed parameters
back to the tunable list and restart the optimization using these reasonable values as
initial guesses.
• The software may have encountered errors during the optimization. Review the errors
to determine if you can make changes to improve the optimization results. Changes
may require modifications to the model, requirements, or optimization settings.

• In the Response Optimization tool, the software creates a structure named

EvalErrors in the Data area when the optimization completes with errors.
Export this structure to the MATLAB workspace and examine its contents at the

3-232
 Optimization Convergence

command line. EvalErrors has two fields, Errors and DesignVars, containing
the errors encountered during optimization and the corresponding design variable
values. To reproduce a specific error, use sdo.setValueInModel to run the
model using the design variables that correspond to the error.
• At the command line, the second output of sdo.optimize, opt_info,
is a structure that provides information regarding the optimization.
opt_info.exitflag identifies the reason the optimization terminated. For more
information regarding exit flags, see “Exit Flags and Exit Messages”.

Why does the optimization terminate before exceeding the maximum

number of iterations, with a solution that does not satisfy all the
constraints or design requirements?
• It might not be possible to achieve your specifications. Try relaxing the constraints
or design requirements that the response signals violate the most. After you find an
acceptable solution to the relaxed problem, tighten some constraints again and restart
the optimization.
• The optimization might have converged to a local minimum that is not a feasible
solution. Restart the optimization from a different initial guess and/or use one of
the search-based methods to identify another local minimum that satisfies the
constraints.

What to do if the optimization takes a long time to converge even though

it is close to a solution?
• In a Response Optimization tool, click Stop to interrupt the optimization when you
think the current optimized response signals are acceptable.

When you use a SISO Design Task, click Stop Optimization in the Optimization
panel of the Response Optimization node in the Control and Estimation Tools
Manager, when you think the current optimized response signals are acceptable.
• If you use the gradient descent method, try restarting the optimization. Restarting
resets the Hessian estimate and might speed up convergence.
• Increase the convergence tolerances in the Optimization Options dialog to force
earlier termination.
• Relax some of the constraints or design requirements to increase the size of the
feasibility region.

3-233
3 Response Optimization

What to do if the response becomes unstable and does not recover?

While the optimization formulation has explicit safeguards against unstable or divergent
response signals, the optimization can sometimes venture into an unstable region where
simulation results become erratic and gradient methods fail to find a way back to the
stable region. In these cases, you can try one of the following solutions:

• Add or tighten the lower and upper bounds on compensator element and parameter
values. Instability often occurs when you allow some parameter values to become too
large.
• Use a search-based method to find parameter values that stabilize the response
signals and then start the gradient-based method using these initial values.
• When optimizing responses in a SISO Design Task, you can try adding additional
design requirements that achieve the same or similar goal. For example, in addition
to a settling time design requirement on a step response plot, you could add a settling
time design requirement on a root-locus plot that restricts the location of the real
parts of the poles. By adding overlapping design requirements in this way, you can
force the optimization to meet the requirements.

3-234
 Optimization Speed and Parallel Computing

Optimization Speed and Parallel Computing

In this section...
“How can I speed up the optimization?” on page 3-235
“Why are the optimization results with and without using parallel computing different?”
on page 3-236
“Why do I not see the optimization speedup I expected using parallel computing?” on
page 3-236
“Why does the optimization using parallel computing not make any progress?” on page
3-236
“Why does the optimization using parallel computing not stop when I click the Stop
optimization button?” on page 3-237

How can I speed up the optimization?

• The optimization time is dominated by the time it takes to simulate the model.
When optimizing a Simulink model, you can enable the Accelerator mode using
Simulation > Mode > Accelerator in the Simulink Editor, to dramatically reduce
the optimization time.

Note: The Rapid Accelerator mode in Simulink software is not supported for speeding
up the optimization. For more information, see “How to Use Accelerator Mode During
Simulations” on page 3-213.
• The choice of ODE solver can also significantly affect the overall optimization time.
Use a stiff solver when the simulation takes many small steps, and use a fixed-step
solver when such solvers yield accurate enough simulations for your model. (These
solvers must be accurate in the entire parameter search space.)
• Reduce the number of tuned compensator elements or parameters and constrain their
range to narrow the search space.
• When specifying parameter uncertainty (not available when optimizing responses in
a SISO Design Task), keep the number of sample values small since the number of
simulations grows exponentially with the number of samples. For example, a grid of 3
parameters with 10 sample values for each parameter requires 103=1000 simulations
per iteration.

3-235
3 Response Optimization

Why are the optimization results with and without using parallel
computing different?
• Different numerical precision on the client and worker machines can produce
marginally different simulation results. Thus, the optimization method can take a
different solution path and produce a different result.
• When you use parallel computing with the Pattern search method, the search is
more comprehensive and can result in a different solution.

To learn more, see “Parallel Computing with the Pattern Search Method” on page
3-217.

Why do I not see the optimization speedup I expected using parallel

computing?
• When you optimize a model that does not have a large number of parameters or does
not take long to simulate, you might not see a speedup in the optimization time. In
such cases, the overhead associated with creating and distributing the parallel tasks
outweighs the benefits of running the optimization in parallel.
• Using the Pattern search method with parallel computing might not speed up
the optimization time. Without parallel computing, the method stops the search at
each iteration when it finds a solution better than the current solution. The candidate
solution search is more comprehensive when you use parallel computing. Although
the number of iterations might be larger, the optimization without using parallel
computing might be faster.

To learn more about the expected speedup, see “Parallel Computing with the Pattern
Search Method” on page 3-217.

Why does the optimization using parallel computing not make any
progress?
To troubleshoot the problem:

1 Run the optimization for a few iterations without parallel computing to see if the
optimization progresses.
2 Check whether the remote workers have access to all model dependencies. Model
dependencies include data variables and files required by the model to run.

3-236
 Optimization Speed and Parallel Computing

To learn more, see “Model Dependencies” on page 3-219.

Why does the optimization using parallel computing not stop when I click
the Stop optimization button?
When you use parallel computing with the Pattern search method, the software must
wait until the current optimization iteration completes before it notifies the workers
to stop. The optimization does not terminate immediately when you click Stop, and,
instead, appears to continue running.

3-237
3 Response Optimization

Undesirable Parameter Values

In this section...
“What to do if the optimization drives the tuned compensator elements and parameters
to undesirable values?” on page 3-238
“What to do if the optimization violates bounds on parameter values?” on page 3-238

What to do if the optimization drives the tuned compensator elements

and parameters to undesirable values?
• When a tuned compensator element or parameter is positive, or when its value is
physically constrained to a given range, enter the lower and upper bounds (Minimum
and Maximum) in one of the following:

• Dialog box to select design variables (in Response optimization tool)

• Compensators pane (in a SISO Design Task)

This information helps guide the optimization method towards a reasonable solution.
• Specify initial guesses that are within the range of desirable values.
• In the Compensators pane in a SISO Design Task, verify that no integrators/
differentiators are selected for optimization. Optimizing the pole/zero location of
integrators/differentiators can result in drastic changes in the system gain and lead to
undesirable values.

What to do if the optimization violates bounds on parameter values?

The Gradient descent optimization method fmincon violates the parameter bounds
when it cannot simultaneously satisfy the signal constraints and the bounds. When this
happens, try one of the following:

• Specify a different value for the parameter bound and restart the optimization. A
guideline is to adjust the bound by 1% of the typical value.

For example, for a parameter with a typical value of 1 and lower bound of 0, change
the lower bound to 0.01.
• Relax the signal constraints and restart the optimization. This approach results in a
different solution path for the Gradient descent method.

3-238
 Undesirable Parameter Values

• Restart the optimization immediately after it terminates by clicking Optimize in the

Response Optimization tool. This approach uses the previous optimization results as
the starting point for the next optimization cycle to refine the results.
• Use the following two-step approach to perform the optimization:

1 Run an initial optimization to satisfy the signal constraints.

For example, run the optimization using the Simplex search method. This
method satisfies the signal constraints but does not support the bounds on
parameter values. The results obtained using this method provide the starting
point for the optimization performed in the next step. To learn more about this
method, see the fminsearch function reference page in the Optimization Toolbox
documentation.
2 Reconfigure the optimization by selecting a different optimization method to
satisfy both the signal constraints and the parameter bounds.

For example, change the optimization method to Gradient descent and run
the optimization again.

Tip If Global Optimization Toolbox software is installed, you can select the Pattern
search optimization method to optimize the model response.

3-239
3 Response Optimization

Reverting to Initial Parameter Values

How do I quit an optimization and revert to my initial parameter values?

• Before running an optimization, do one of the following:

• In the Response Optimization tool, click Options. Uncheck Update model at end
of optimization in the General Options tab.
• In the Response Optimization tool, click Options. Select Save optimized
variable values as new design variable set in the General Options tab.
• Make a copy of the design variable set in the Data area.

If you want to revert to the initial parameter values after the optimization terminates
or you stop the optimization by clicking Stop, select the design variable that contains

the initial values in the Design Variable Set drop-down list and click adjacent
to Design variables Set. Select the design variables in the dialog box and click
Update model variable values to revert the model parameters to their original
values.
• When using a SISO Design Task, the Start Optimization button becomes a Stop
Optimization button after the optimization has begun. To quit the optimization,
click the Stop Optimization button. To revert to the initial parameter values, select
Edit > Undo Optimize compensators from the menu in the SISO Design Tool
window.

3-240
 Manage Response Optimization Tool Session

Manage Response Optimization Tool Session

In this section...
“Save a Session” on page 3-241
“Load a Session” on page 3-241

Save a Session
Saving a session lets you reuse your settings and optimization results later. These
settings include design requirements, design and uncertain variables, plots and
optimization settings. Each Response Optimization tool session is associated with a
Simulink model.

You can save the session as a MAT-file or workspace variable:

• To save the session as a MAT-file, in the Save Session drop-down list, click Save
to file in the Response Optimization tab. A window opens where you specify the
MAT-file name.
• To save the session as a model or MATLAB workspace variable, select Save to model
workspace or Save to MATLAB workspace in the Save Session drop-down list.

Load a Session
You can load a saved MAT-file or workspace session:

3-241
3 Response Optimization

1 Open a Response Optimization tool for the model.

2 To load a MAT-file, in the Response Optimization tab click on the Open Session
drop-down list, and select Open from file . A window opens where you select the
MAT-file to load.

To load a workspace variable, select Open from model workspace or Open from
MATLAB workspace in the Open Session drop-down list.

3-242
 Optimizing Time-Domain Response of Simulink® Models Using Parallel Computing

Optimizing Time-Domain Response of Simulink® Models Using

Parallel Computing
This example shows how to use parallel computing to optimize the time-domain
response of a Simulink® model. You use Simulink® Design Optimization™ and Parallel
Computing Toolbox™ to tune the gains of a discrete PI controller of a boiler to meet the
design requirements. The example also shows how the software automatically handles
model file dependencies.

This example requires Parallel Computing Toolbox™.

Opening the Model

The Simulink model consists of a boiler model and a discrete PI controller. When using
parallel computing, Simulink Design Optimization performs a model dependency check,
which recognizes the boiler model library as an installed library.

In order to illustrate how model dependencies are handled when using parallel
computing, we copy the boiler model and library block to a temporary folder before
opening the model.

pathToLib = boilerpressure_setup; %Copies boiler model and library to a temporary fol

addpath(pathToLib);
open_system('boilerpressure_demo')

3-243
3 Response Optimization

Design Requirements

The boiler pressure is regulated by a discrete PI controller. The design requirement for
the controller is to limit the pressure variation of the boiler within +-%5 of the nominal
pressure.

The initial controller has fairly good regulation characteristics but in the presence of
additional heat disturbances, modeled by the Heat Disturbance block, we want to tune
the controller performance to provide tighter pressure regulation.

Double-click the 'Response Optimization GUI with preloaded data' block in the Simulink
model to open a pre-configured Response Optimization Tool. The Response Optimization
Tool is configured with:

1. Upper and lower bounds representing a +-5% allowable range on the drum pressure

2. A reference tracking objective to minimize the deviation of the drum pressure from
nominal

3. The PI controller gains, Kp and Ki, are selected for tuning

3-244
 Optimizing Time-Domain Response of Simulink® Models Using Parallel Computing

Click Plot Model Response to display the drum pressure variations with the initial
controller.

Configuring and Running the Optimization in the GUI Using Parallel Computing

When computing the model response with the initial controller, this complex model took
a long time to simulate. Using parallel computing can reduce the optimization time
by simulating the model in parallel. For more information on parallel computing and
optimization performance see the tutorial "Improving Optimization Performance Using
Parallel Computing".

To configure the optimization problem to use parallel computing click Options in the
Response Optimization Tool and select the Parallel Options tab. Select the "Use the

3-245
3 Response Optimization

parallel pool during optimization" option. This triggers an automated search for any
model dependencies. In this example, the Steam table library (libsteam.slx) is found
as a model dependency (in addition to the boilerpressure_demo model itself), and is
displayed in the Model file dependencies list box.

Clicking OK configures the optimization to use parallel computing.

To run the optimization click the Optimize button. A progress window opens displaying
optimization progress and the plots update to show the optimized response.

3-246
 Optimizing Time-Domain Response of Simulink® Models Using Parallel Computing

The final response shows that the optimized regulator tracks the reference pressure
much more closely and the drum pressure constraints are satisfied.

Configuring and Running the Optimization at the Command Line

You can also use the command line functions to configure the optimization to use parallel
computing and run the optimization.

Select the model variables for optimization and set lower limits

p = sdo.getParameterFromModel('boilerpressure_demo',{'Kp','Ki'});
p(1).Minimum = 0.001;
p(2).Minimum = 0.001;

3-247
3 Response Optimization

Select the model signal to bound and create a simulator to simulate the model.
nPressure = Simulink.SimulationData.SignalLoggingInfo;
nPressure.BlockPath = 'boilerpressure_demo/1//y0';
nPressure.OutputPortIndex = 1;
nPressure.LoggingInfo.NameMode = 1;
nPressure.LoggingInfo.LoggingName = 'nPressure';

simulator = sdo.SimulationTest('boilerpressure_demo');
simulator.LoggingInfo.Signals = nPressure;

Get the optimization requirements defined by the check blocks in the model so that we
can use them in the optimization problem.
bnds = getbounds('boilerpressure_demo/Drum pressure constraint');
PressureLimits = [bnds{:}];
bnds = getbounds('boilerpressure_demo/Drum pressure constraint(Reference Tracking)');
PressureRegulation = [bnds{:}];
requirements = struct(...
'PressureLimits', PressureLimits, ...
'PressureRegulation', PressureRegulation);

Define the function called during optimization. Notice that the function uses the
simulator and requirements defined earlier to evaluate the design.
evalDesign = @(p) boilerpressure_design(p,simulator,requirements);
type boilerpressure_design

function design = boilerpressure_design(p,simulator,requirements)

%BOILERPRESSURE_DESIGN
%
% The boilerpressure_design function is used to evaluate a boiler
% controller design design.
%
% The |p| input argument is the vector of controller parameters.
%
% The |simulator| input argument is a sdo.SimulinkTest object used to
% simulate the |boilerpressure_demo| model and log simulation signals.
%
% The |requirements| input argument contains the design requirements used
% to evaluate the boiler controller design.
%
% The |design| return argument contains information about the design
% evaluation that can be used by the |sdo.optimize| function to optimize

3-248
 Optimizing Time-Domain Response of Simulink® Models Using Parallel Computing

% the design.
%
% see also sdo.optimize, sdoExampleCostFunction

% Copyright 2011 The MathWorks, Inc.

%% Simulate the model

%
% Use the simulator input argument to simulate the model and log model
% signals.
%
% First ensure that we simulate the model with the parameter values chosen
% by the optimizer.
%
simulator.Parameters = p;
%%
% Simulate the model and log signals.
%
simulator = sim(simulator);
%%
% Get the simulation signal log, the simulation log name is defined by the
% model |SignalLoggingName| property
%
logName = get_param('boilerpressure_demo','SignalLoggingName');
simLog = get(simulator.LoggedData,logName);

%% Evaluate the design requirements

%
% Use the requirements input argument to evaluate the design requirements
%
% Check the Pressure signal against the |PressureLimits| requirements.
%
nPressure = get(simLog,'nPressure');
c = [...
evalRequirement(requirements.PressureLimits(1),nPressure.Values); ...
evalRequirement(requirements.PressureLimits(2),nPressure.Values)];
%%
% Use the PressureLimits requirements as non-linear constraints for
% optimization.
design.Cleq = c(:);
%%
% Check the pressure signal against the |PressureRegulation| requirement.
%
f = evalRequirement(requirements.PressureRegulation,nPressure.Values);

3-249
3 Response Optimization

%%
% Use the PressureRegulation requirement as an objective for optimization.
design.F = f;
end

Setup optimization options to use the parallel pool and specify the model and model files
dependencies.

opt = sdo.OptimizeOptions;
opt.UseParallel = 'always';
opt.OptimizedModel = 'boilerpressure_demo';
[dirs,files] = sdo.getModelDependencies('boilerpressure_demo');
opt.ParallelFileDependencies = files;

To run the optimization using the parallel pool pass the optimization options, opt, to the
sdo.optimize command.

3-250
 Optimizing Time-Domain Response of Simulink® Models Using Parallel Computing

Closing the Model

After the model is optimized, we remove the boiler model and library file from the
temporary folder.
bdclose('boilerpressure_demo')
rmpath(pathToLib)
boilerpressure_cleanup(pathToLib)

3-251
3 Response Optimization

Design Optimization to Meet Frequency-Domain Requirements

(Code)
This example shows how to tune model parameters to meet frequency-domain
requirements, using the sdo.optimize command.

This example requires Simulink® Control Design™.

Suspension Model

Open the Simulink Model.

open_system('sdoSimpleSuspension')

Mass-spring-damper models represent simple suspension systems and for this example
we tune the system to meet typical suspension requirements. The model implements

3-252
 Design Optimization to Meet Frequency-Domain Requirements (Code)

the second order system representing a mass-spring-damper using Simulink blocks and
includes:

• a Mass gain block parameterized by the total suspended mass, m0+mLoad. The total
mass is the sum of a nominal mass, m0, and a variable load mass, mLoad.

• a Damper gain block parameterized by the damping coefficient, b.

• a Spring gain block parameterized by the spring constant, k.

• two integrator blocks to compute the mass velocity and position.

• a Band-Limited Disturbance Force block applying a disturbance force to the

mass. The disturbance force is assumed to be band-limited white noise.

Simulate the model to view the system response to the applied disturbance force.

3-253
3 Response Optimization

Design Problem

The initial system has a bandwidth that is too high. This can be seen from the
spiky position signal. You tune the spring and damper values to meet the following
requirements:

• The -3dB system bandwidth must not exceed 10 rad/s.

• The damping ratio of the system must be less than 1/sqrt(2). This ensures that no
frequencies in pass band are amplified by the system.

• Minimize the expected failure rate of the system. The expected failure rate is
described by a Weibull distribution dependent on the mass, spring, and damper
values.

• These requirements must all be satisfied as the load mass ranges from 0 to 20 kg.

Specify Design Variables

DesignVars = sdo.getParameterFromModel('sdoSimpleSuspension',{'b','k'});
DesignVars(1).Minimum = 100;
DesignVars(1).Maximum = 10000;
DesignVars(2).Minimum = 10000;
DesignVars(2).Maximum = 100000;

Specify Uncertain Variables

Specify mLoad, the load mass, as an uncertian variable. This will ensure the optimal
solution is robust to variations in load mass.

UncVars = sdo.getParameterFromModel('sdoSimpleSuspension','mLoad');
UncVars_Values = {...
10; ...
20};

Specify Design Requirements

Specify design requirements to satisfy during optimization.

Requirements = struct;
Requirements.Bandwidth = sdo.requirements.BodeMagnitude(...
'BoundFrequencies', [10 100], ...
'BoundMagnitudes', [-3 -3]);
Requirements.DampingRatio = sdo.requirements.PZDampingRatio;

3-254
 Design Optimization to Meet Frequency-Domain Requirements (Code)

The reliability requirement will be specified in the optimization objective function

described below. The reliability requirement is used to tune the spring and damper
values to minimize the expected failure rate over a lifetime of 100e3 miles. The failure
rate is computed using a Weibull distribution on the damping ratio of the system. As the
damping ratio increases, the failure rate is expected to increase.

Linearization Definition

The design requirements use the bandwidth and damping ratio of the system, these
frequency domain characteristics require linearizing the model. Create a simulator
for the model and use the simulator to compute the linear systems used by the
requirements.

Simulator = sdo.SimulationTest('sdoSimpleSuspension');

Specify linear systems to compute by specifying the linearization inputs and outputs of
the system. The linear system input is the output of the Band-Limited Disturbance
Force block and the linear system output is the output of the x_dot block (the
position signal).

IOs(1) = linio('sdoSimpleSuspension/Band-Limited Disturbance Force',1,'input');

IOs(2) = linio('sdoSimpleSuspension/x_dot',1,'output');

Add the linearization IOs to a sdo.SystemLoggingInfo object with a logging name,

and linearization snaphot time. In this case, the snaphot time is set to 0, the model initial
condition.

sys1 = sdo.SystemLoggingInfo;
sys1.Source = 'IOs';
sys1.LoggingName = 'IOs';
sys1.LinearizationIOs = IOs;
sys1.SnapshotTimes = 0;

Add the system logging info to the simulator. When the simulator is used to run the
model, the linear system defined by the specifed linearization IOs is computed and added
to the simulation log with the specified logging name.

Simulator.SystemLoggingInfo = sys1;

Create Optimization Objective Function

Create a function that is called at each optimization iteration to evaluate the design
requirements.

3-255
3 Response Optimization

Use an anonymous function with one argument that calls

sdoSimpleSuspension_Design. The sdoSimpleSuspension_Design function
has arguments for the design parameters, the simulator, the design requirements, the
uncertain variables, and uncertain variable values.

optimfcn = @(P) sdoSimpleSuspension_Design(P,Simulator,Requirements,UncVars,UncVars_Val

type sdoSimpleSuspension_Design

function Vals = sdoSimpleSuspension_Design(P,Simulator,Requirements,UncVars,UncVars_Val

%SDOSIMPLESUSPENSION_DESIGN
%
% The sdoSimpleSuspension_Design function is used to evaluate a simple
% suspension system design.
%
% The |P| input argument is the vector of suspension design parameters.
%
% The |Simulator| input argument is a sdo.SimulinkTest object used to
% simulate the |sdoSimpleSuspension| model and log simulation signals.
%
% The |Requirements| input argument contains the design requirements
% used to evaluate the suspension design.
%
% The |UncVars| and |UncVars_Values| arguments specify the uncertain
% variable and uncertain variable values.
%
% The |Vals| return argument contains information about the design
% evaluation that can be used by the |sdo.optimize| function to optimize
% the design.
%
% See also sdoSimpleSuspension_cmddemo

% Copyright 2015 The MathWorks, Inc.

%% Evaluate the suspension reliability

%
%Get the spring and damper design values
allVarNames = {P.Name};
idx = strcmp(allVarNames,'k');
k = P(idx).Value;
idx = strcmp(allVarNames,'b');
b = P(idx).Value;

%Get the nominal mass from the model workspace

3-256
 Design Optimization to Meet Frequency-Domain Requirements (Code)

wksp = get_param('sdoSimpleSuspension','ModelWorkspace');
m = evalin(wksp,'m0');

%The expected failure rate is defined by the Weibull cumulative

%distribution function, 1-exp(-(x/l)^k), where k=3, l is a function of the
%mass, spring and damper values, and x the lifetime.
d = b/2/sqrt(m*k);
pFailure = 1-exp(-(100e3*d/250e3)^3);

%% Nominal Evaluation
%
% Evaluate the model and requirements with uncertain parameters set to their
% nominal values.

% Simulate the model.

Simulator.Parameters = [P(:);UncVars(:)];
Simulator = sim(Simulator);

% Retrieve logged linearizations.

Sys = find(Simulator.LoggedData,'IOs');

% Evaluate the design requirements.

Cleq_Bandwidth = evalRequirement(Requirements.Bandwidth,Sys.values);
Cleq_DampingRatio = evalRequirement(Requirements.DampingRatio,Sys.values);

%% Uncertain Evaluation
%
% Evaluate the model and requirements for all combinations of uncertain
% parameter values.
for ct=1:size(UncVars_Values,1)
UncVars(1).Value = UncVars_Values{ct,1};

% Simulate the model.

Simulator.Parameters = [P(:);UncVars(:)];
Simulator = sim(Simulator);

% Retrieve logged linearizations.

Sys = find(Simulator.LoggedData,'IOs');

% Evaluate the design requirements.

Cleq_Bandwidth_UncVars = evalRequirement(Requirements.Bandwidth,Sys.values);
Cleq_DampingRatio_UncVars = evalRequirement(Requirements.DampingRatio,Sys.values);

Cleq_Bandwidth = [Cleq_Bandwidth; Cleq_Bandwidth_UncVars]; %#ok<AGROW>

3-257
3 Response Optimization

Cleq_DampingRatio = [Cleq_DampingRatio; Cleq_DampingRatio_UncVars]; %#ok<AGROW>

end

%% Return Values.
%
% Collect the evaluated design requirement values in a structure to
% return to the optimization solver.
Vals.F = pFailure;
Vals.Cleq = [...
Cleq_Bandwidth(:); ...
Cleq_DampingRatio(:)];
end

Optimization Options

Specify optimization options.

Options = sdo.OptimizeOptions;
Options.MethodOptions.MaxFunEvals = [];
Options.MethodOptions.Algorithm = 'active-set';
Options.MethodOptions.LargeScale = 'on';
Options.MethodOptions.TolGradCon = 1e-06;

Optimize the Design

Call sdo.optimize with the objective function handle, parameters to optimize, and
options.

[Optimized_DesignVars,Info] = sdo.optimize(optimfcn,DesignVars,Options);

Optimization started 31-Jul-2015 06:19:07

max Step-size First-order

Iter F-count f(x) constraint optimality
0 5 0.0619897 0.7642
1 10 0.351266 -0.1277 0.461 1.22
2 15 0.189345 -0.03881 0.188 0.312
3 20 0.0829063 0.01312 0.51 0.425
4 24 0.0365398 0 0.308 1.55
5 28 0.0294977 0 0.0143 0.733
6 32 0.0293065 1.57e-16 0.000417 0.00142
Local minimum found that satisfies the constraints.

Optimization completed because the objective function is non-decreasing in

3-258
 Design Optimization to Meet Frequency-Domain Requirements (Code)

feasible directions, to within the selected value of the function tolerance,

and constraints are satisfied to within the selected value of the constraint tolerance.

Related Examples

To learn how to optimize the suspension system using the Response Optimization tool,
see "Design Optimization to Meet Frequency-Domain Requirements (GUI)".

Close the model.

bdclose('sdoSimpleSuspension')

3-259
 4

Sensitivity Analysis

• “What Is Sensitivity Analysis?” on page 4-2

• “Sampling Parameters for Sensitivity Analysis” on page 4-4
• “Sensitivity Analysis Methods” on page 4-11
• “How to Use Parallel Computing for Sensitivity Analysis” on page 4-14
• “Use Fast Restart Mode During Sensitivity Analysis” on page 4-17
• “Design Exploration using Parameter Sampling (Code)” on page 4-20
• “Identify Key Parameters for Estimation (Code)” on page 4-36
4 Sensitivity Analysis

What Is Sensitivity Analysis?

Generally, sensitivity analysis is defined as the study of how uncertainty in the output
of a model can be attributed to different sources of uncertainty in the model input[1]. In
the context of using Simulink Design Optimization software, sensitivity analysis refers
to understanding how the parameters and states (optimization design variables) of a
Simulink model influence the optimization cost function. Examples of using sensitivity
analysis include:

• Before optimization — Determine the influence of the parameters of a Simulink

model on the output. Use sensitivity analysis to rank parameters in order of influence
so that you can determine the most influential parameters. Optimize the model by
tuning the most influential parameters or perform experiments to better characterize
those parameters.
• After optimization — Test how robust the cost function is to small changes in the
values of optimized parameters.

One approach to sensitivity analysis is local sensitivity analysis, which is derivative

based (numerical or analytical). Mathematically, the sensitivity of the cost function with
respect to certain parameters is equal to the partial derivative of the cost function with
respect to those parameters. The term local refers to the fact that all derivatives are
taken at a single point. For simple cost functions, this approach is efficient. However,
this approach can be infeasible for complex models, where formulating the cost function
(or the partial derivatives) is nontrivial. For example, models with discontinuities do not
always have derivatives.

Local sensitivity analysis is a one-at-a-time (OAT) technique. OAT techniques analyze

the effect of one parameter on the cost function at a time, keeping the other parameters
fixed. They explore only a small fraction of the design space, especially when there are
many parameters. Also, they do not provide insight about how the interactions between
parameters influence the cost function.

Another approach to sensitivity analysis is global sensitivity analysis, often implemented

using Monte Carlo techniques. This approach uses a representative (global) set of
samples to explore the design space. Use Simulink Design Optimization software to
perform global sensitivity analysis. The workflow is as follows:

1 Sample the model parameters using experimental design principles. That is, for each
parameter, generate multiple values that the parameter can assume. Define the
parameter sample space by specifying probability distributions for each parameter.
You can also specify parameter correlations.

4-2
 What Is Sensitivity Analysis?

For information about sampling parameters, see “Sampling Parameters for

Sensitivity Analysis” on page 4-4.
2 Evaluate the optimization cost function at each sample point. You can plot the cost
function output for the samples to visually analyze trends.
3 (Optional) Formally analyze the relation between the cost function and the samples.
Analysis methods include correlation, partial correlation (requires a Statistics
and Machine Learning Toolbox™ license), and standardized regression. You can
configure each analysis method to use either raw or ranked data.

For information about the analysis methods, see “Sensitivity Analysis Methods” on
page 4-11.

References
[1] Saltelli, A., Ratto, M., Andres, T., Campolongo, F., Cariboni, J., Gatelli, D., Saisana,
M., and Tarantola, S. Global Sensitivity Analysis. The Primer, John Wiley and
Sons, 2008.

See Also
sdo.analyze | sdo.evaluate | sdo.sample | sdo.scatterPlot

Related Examples
• “Design Exploration using Parameter Sampling (Code)” on page 4-20
• “Identify Key Parameters for Estimation (Code)” on page 4-36

More About
• “Sensitivity Analysis Methods” on page 4-11
• “Sampling Parameters for Sensitivity Analysis” on page 4-4

4-3
4 Sensitivity Analysis

Sampling Parameters for Sensitivity Analysis

You can perform global sensitivity analysis using Simulink Design Optimization
software. You vary the value of the Simulink model parameters and states of interest
in a specific range. These parameters and states are the optimization design variables,
collectively referred to as parameters. Each combination of values for the different
parameters is referred to as a sample or sample point. A collection of samples is referred
to as a design space, parameter sample space, or, simply, sample space. You evaluate
the optimization cost function for each point in the sample space. Then, you analyze the
relation between the parameter value variations and the cost function value variations to
understand how the parameters influence the cost function.

Each model evaluation has a computational expense and can be time intensive.
Therefore, ideally, you want to use the smallest sample set that yields useful results. You
can use techniques such as design of experiments (DOE) (also referred to as experimental
design) to choose an efficient sample set for sensitivity analysis.

Use sdo.ParameterSpace to define the parameter space. This object specifies the
probability distributions and correlations for the parameters. Use this object as an input
to sdo.sample to generate samples from the specified parameter space.

Common considerations for parameter sampling include:

• “Probability Distribution” on page 4-4

• “Bounds” on page 4-5
• “Number of Samples” on page 4-5
• “Method of Sampling” on page 4-5
• “Custom Sample Sets” on page 4-7

Probability Distribution
Specify the probability distribution function that is best suited for each parameter.
Use your knowledge of the system (empirical or theoretical) to choose the probability
distribution for the parameter.

The Simulink Design Optimization software allows you to specify uniform (default) and
normal distributions. If you have a Statistics and Machine Learning Toolbox license, you
can also specify any univariate probability distribution that the toolbox provides.

4-4
 Sampling Parameters for Sensitivity Analysis

Specify the probability distribution of a parameter using the ParameterDistributions

property of an sdo.ParameterSpace object.

Bounds
Specify the upper and lower bounds of the value of a parameter. These bounds define the
sampling range for the parameter. Use your knowledge of the parameter’s range of likely
values to choose the bounds.

To specify the bounds of a parameter, use the Minimum and Maximum properties of a
param.Continuous object. Use this updated param.Continuous object when you specify
the parameter space using an sdo.ParameterSpace object.

Number of Samples
Ideally, you want to use the smallest number of samples that yield useful results,
because each sample requires a model evaluation.

As the number of parameters increases, the number of samples needed to explore the
design space generally increases. For correlation or regression analysis, consider using
10Np samples, where Np is the number of parameters.

Specify the number of samples as the second input argument of sdo.sample.

Method of Sampling
After specifying the probability distributions and bounds for the parameters, you
generate samples for the specified parameter space. You can specify the method used
to generate these samples using the Method property of an sdo.SampleOptions object.
Use this object as an input to sdo.sample to specify the sampling options. Specify the
sampling method as one of the following:

• 'random' — Random samples drawn from the probability distributions specified for
the parameters.

Suppose you specified a value for the RankCorrelation property of the

sdo.ParameterSpace object that you use for sampling. The software uses the Iman-
Conover method to impose the parameter correlations.

4-5
4 Sensitivity Analysis

• 'lhs' — Latin hypercube samples drawn from the probability distributions specified
for the parameters. Use this option for a more systematic space-filling approach
than random sampling. The following figure shows the difference between random
sampling and Latin hybercube design-based sampling.

The figure shows 8 samples for 2 parameters, drawn from a uniform distribution, in
the interval from 0 to 1. Random sampling can result in the clustering of samples (see
the top right-hand corner of the plot). Latin hypercube designs, with their stratified
approach to sampling, are better able to avoid such clustering.

Suppose you specified a value for the RankCorrelation property of the

sdo.ParameterSpace object that you use for sampling. The software uses the Iman-
Conover method to impose the parameter correlations.

This method requires a Statistics and Machine Learning Toolbox license.

• 'copula' — Random samples drawn from a copula. Use this option to impose
correlations between the parameters.

You can use either a Gaussian copula (default) or a t copula. Specify the choice of
copula using the MethodOptions property of the sdo.SampleOptions object.
Use t copulas when the probability of extreme parameter values is not negligible.
You must specify the degrees of freedom for a t copula. As you increase the degrees
of freedom, the t copula converges to the Gaussian copula, and the probability of

4-6
 Sampling Parameters for Sensitivity Analysis

extreme parameter values becomes negligible. The following figure shows 1000
samples drawn for 2 parameters, in the interval from 0 to 1, using the Gaussian and t
copulas.

In comparison to the Gaussian copula, the t copula has more samples that represent
the extreme values of the parameters.

You can specify the correlation type as either Spearman’s rank correlation or
Kendall’s rank correlation.

For the 'copula' method, you must also specify the value of the RankCorrelation
property of the sdo.ParameterSpace object that you use for sampling.

This method requires a Statistics and Machine Learning Toolbox license.

Custom Sample Sets

You can specify a custom sample set. For example, suppose you want to generate a 1000
samples of the model parameter, R, which is a resistor.

R = param.Continuous('R',10);

4-7
4 Sensitivity Analysis

Sample R in the 5% range of its nominal value. However, resistors of 1% tolerance are
removed. So, you do not need to sample R in the 1% range. You can use the following
approaches:

• “Specify Customized Probability Distribution” on page 4-8

• “Create Table of Custom Samples” on page 4-9

To visualize the samples and validate the sample space, use sdo.scatterPlot.

Specify Customized Probability Distribution

Use a customized probability distribution to configure the parameter space and generate
samples.

You can use tools provided by the Statistics and Machine Learning Toolbox software to
create customized probability distributions. For example,

x = [0.95 0.99 1.01 1.05]*R.Value;

F = [0 0.5 0.5 1];

pdR = makedist('PiecewiseLinear','x',x,'Fx',F);

x = linspace(.9*R.Value, 1.1*R.Value, 1e3);

plot(x, pdf(pdR, x));

4-8
 Sampling Parameters for Sensitivity Analysis

The call to makedist specifies a piecewise linear distribution for the resistor value, with
a “hole” in the 1% range.

Specify pdR as the probability distribution for the R parameter when you create the
sdo.ParameterSpace object to define the parameter space.
ps = sdo.ParameterSpace(R,pdR);

Generate the samples using sdo.sample.

x = sdo.sample(ps,1000);

Create Table of Custom Samples

Create a table of the custom samples. Specify one column for each parameter, and one
row for each sample. The column name must be the same as the parameter name.

For example:
Rval = R.Value;

4-9
4 Sensitivity Analysis

Ns = 1000;
x = table([linspace(.95*Rval,.99*Rval,Ns*.5) ...
linspace(1.01*Rval,1.05*Rval,Ns*.5)]','VariableNames',{'R'});

Consider another example, where you have two model parameters, A and B.

A = param.Continuous('A',1);
B = param.Continuous('B',10);

Vary A for the following values: 2,3,4. Vary B for the following values: 20,30,40. Generate
a table of samples for every combination of A and B.

Avals = [2 3 4];
Bvals = [20 30 40];
[Agrid,Bgrid] = meshgrid(Avals,Bvals);
x = table(Agrid(:),Bgrid(:),'VariableNames',{'A','B'});

See Also
sdo.SampleOptions | sdo.sample

Related Examples
• “Design Exploration using Parameter Sampling (Code)” on page 4-20

More About
• “What Is Sensitivity Analysis?” on page 4-2

4-10
 Sensitivity Analysis Methods

Sensitivity Analysis Methods

To analyze how the parameters and states (collectively referred to as parameters) of a
Simulink model influence the cost function, first generate samples of the parameters.
Then, evaluate the cost function for each sample. Finally, analyze the relationship
between the parameter variations and the cost function values. Perform this analysis in
the following ways:

• “Visual Analysis” on page 4-11

• “Quantitative Analysis” on page 4-11

Visual Analysis
Plot the cost function evaluations against the parameter samples to identify trends. This
method is informal and provides visual intuition about how the various parameters affect
the cost function.

You can use tools such as:

• sdo.scatterPlot — Scatter plot of the parameter samples against the cost function
evaluation
• surf, mesh — 3-D plot of samples of two parameters against the cost function
evaluation

Quantitative Analysis
Obtain summary statistics using sdo.analyze. This function performs linear
correlation analysis by default. You can specify other analysis method(s) using an
sdo.AnalyzeOptions options object.

Available analysis methods include:

Method Name Description

Correlation Use to analyze how a model parameter and
the cost function are correlated
Partial correlation (requires a Statistics Use to analyze how a model parameter and
and Machine Learning Toolbox license) the cost function are correlated, removing
the effects of the remaining parameters

4-11
4 Sensitivity Analysis

Method Name Description

Standardized regression Use when you expect that the model
parameters linearly influence the cost
function

For information about the formulations of these methods, see sdo.AnalyzeOptions.

Linear vs. Ranked Analysis

Each of these methods supports the following analysis types:

• Linear analysis, also referred to as Pearson analysis — Uses raw data for analysis.
Best suited when you expect a linear relation between the parameters and cost
function.
• Ranked analysis, also referred to as Spearman analysis and ranked transformation —
Uses ranks of data for analysis. Best suited when you expect a nonlinear monotonic
relation between the parameters and the cost function.

For an example of ranked analysis, suppose you had the following data set:

x1 x2 y
9 20 340
5 60 106
2.3 50.4 870.5

Here x1 and x2 are model parameters, and y is the cost function. Each row represents
a sample and the associated cost function evaluation.

The data is ranked on a per column basis. For example, when you rank the data in
column 1 (x1), which contains the entries 9, 5, and 2.3, the ranked data is equal to 3, 2,
and 1. The ranked data set for the samples of x1, x2 and y are as follows:

x1 x2 y
3 1 2
2 3 1
1 2 3

4-12
 Sensitivity Analysis Methods

The ranked data set can be used for correlation, partial correlation, or standardized
regression analysis.

Linear analysis retains information about intervals between data values, whereas
ranked analysis does not.

See Also
sdo.AnalyzeOptions | sdo.analyze | sdo.evaluate | sdo.sample

Related Examples
• “Identify Key Parameters for Estimation (Code)” on page 4-36

More About
• “What Is Sensitivity Analysis?” on page 4-2

4-13
4 Sensitivity Analysis

How to Use Parallel Computing for Sensitivity Analysis

In this section...
“Configure Your System for Parallel Computing” on page 4-14
“Model Dependencies” on page 4-14
“Perform Sensitivity Analysis Using Parallel Computing (Code)” on page 4-15
“Troubleshooting” on page 4-16

Configure Your System for Parallel Computing

To perform global sensitivity analysis, you sample the model parameters and states
(collectively referred to as parameters), and evaluate the cost function for each sample.
You use sdo.evaluate to perform the cost function evaluation (also referred to as model
evaluation). Evaluating the model for many samples can be time consuming. You can
speed up the performance of sdo.evaluate using parallel computing on multicore
processors or multiprocessor networks.

When you call sdo.evaluate with the parallel computing option enabled, the software
uses the available parallel pool. If none is available, and Automatically create a
parallel pool is selected in your Parallel Computing Toolbox preferences, the software
starts a parallel pool using the settings in those preferences. To open a parallel pool that
uses a specific cluster profile, use:

parpool(MyProfile);

MyProfile is the name of a cluster profile.

For information regarding creating a cluster profile, see “Create and Modify Cluster
Profiles” in the Parallel Computing Toolbox documentation.

Model Dependencies
Model dependencies are any referenced models, data such as model variables, S-
functions, or additional files necessary to run the model. Before starting the parallel
model evaluation, verify that the model dependencies are complete. Otherwise, you may
get unexpected results.

4-14
 How to Use Parallel Computing for Sensitivity Analysis

Making Model Dependencies Accessible to Remote Workers

When you use parallel computing, the Simulink Design Optimization software helps you
identify model dependencies. To do so, the software uses the Simulink Manifest Tools.
The dependency analysis may not find all the files required by your model. To learn
more, see “Scope of Dependency Analysis” in the Simulink documentation. If your model
has dependencies that are undetected or inaccessible by the parallel pool workers, then
add them to the list of model dependencies.

The dependencies are made accessible to the parallel pool workers by specifying one of
the following:

• File dependencies: the model dependency files are copied to the parallel pool workers.
• Path dependencies: the paths to the model dependencies are specified to the parallel
pool workers. If you are working in a multi-platform scenario, ensure that the paths
are compatible across platforms.

Using file dependencies is recommended, however, in some cases it can be better to

choose path dependencies. For example, if parallel computing is set up on a local multi-
core computer, using path dependencies is preferred as using file dependencies creates
multiple copies of the dependency files on the local computer.

Perform Sensitivity Analysis Using Parallel Computing (Code)

To evaluate a model using parallel computing:

1 Ensure that the software can access parallel pool workers that use the appropriate
cluster profile.

For more information, see “Configure Your System for Parallel Computing” on page
4-14.
2 Open the model.
3 Specify the cost function and generate parameter samples for sensitivity analysis.
For example, see “Design Exploration using Parameter Sampling (Code)” on page
4-20.
4 Enable parallel computing using an evaluation option set.
opt = sdo.EvaluateOptions;
opt.UseParallel = 'always';
5 Find the model dependencies.

4-15
4 Sensitivity Analysis

[dirs,files] = sdo.getModelDependencies(modelname)

Note: sdo.getModelDependencies may not detect all the dependencies in your

model. For more information, see “Model Dependencies” on page 4-14. In this
case, add the undetected dependencies manually.
6 Modify files to include any file dependencies that sdo.getModelDependencies
does not detect.
files = vertcat(files,'C:\matlab\work\filename.m')

Note: If you do not want to copy the files to the remote workers, add any undetected
path dependencies to dirs and update the paths on local drives to make them
accessible to remote workers. See sdo.getModelDependencies for more details.
7 Add the file dependencies for evaluation.
opt.ParallelFileDependencies = files;
8 Specify the name of the model to be evaluated in parallel.
opt.EvaluatedModel = modelname;
9 Evaluate the model.
[pOpt,opt_info] = sdo.evaluate(fcn,samples,opt);

Troubleshooting
Why Don’t I See the Evaluation Speed up I Expected Using Parallel Computing?

When you evaluate a model that does not have a large number of evaluations or does not
take long to simulate, you might not see a speedup in the evaluation time. In such cases,
the overhead associated with creating and distributing the parallel tasks outweighs the
benefits of running the evaluation in parallel.

See Also
sdo.EvaluateOptions | parpool | sdo.evaluate | sdo.getModelDependencies

More About
• “Ways to Speed Up Design Optimization Tasks”

4-16
 Use Fast Restart Mode During Sensitivity Analysis

Use Fast Restart Mode During Sensitivity Analysis

This topic shows how to speed up sensitivity analysis at the command line using
Simulink fast restart. You can use the fast restart feature to speed up sensitivity
analysis of tunable parameters of a model.

Fast restart enables you to perform iterative simulations without compiling a model or
terminating the simulation each time. Using fast restart, you compile a model only once.
You can then tune parameters and simulate the model again without spending time on
compiling. Fast restart associates multiple simulation phases with a single compile phase
to make iterative simulations more efficient. You see a speedup of design optimization
tasks using fast restart in models that have a long compilation phase. See “How Fast
Restart Improves Iterative Simulations” in the Simulink documentation.

When you enable fast restart, you can only change tunable properties of the model during
simulation. For more information about the limitations, see “Factors Affecting Fast
Restart”.

You can use sensitivity analysis to evaluate which model parameters most influence a
cost function. You can use these parameters during parameter estimation or response
optimization. Suppose that you want to use sensitivity analysis to reduce the number of
parameters that you need to estimate to fit a model.

To evaluate the model in fast restart mode using a cost function aimed at parameter
estimation:

1 Open the Simulink model.

2 Specify the model parameter values, params, to estimate and generate parameter
samples, params_samples. For an example, see “Identify Key Parameters for
Estimation (Code)” on page 4-36.
3 Create an experiment object, Exp.
Exp = sdo.Experiment('model');

Store the measured input-output data in Exp. For an example, see “Identify Key
Parameters for Estimation (Code)” on page 4-36.
4 Create a model simulator from the experiment.
Simulator = createSimulator(Exp);

Simulator is an sdo.SimulationTest object.

4-17
4 Sensitivity Analysis

Note: You must create a simulation scenario with logging information before
configuring the model for fast restart. You cannot modify logging information once
the model has been compiled for fast restart.
5 Configure the model and simulator for fast restart.
Simulator = fastRestart(Simulator,'on');
6 Create a cost function, myCostfcn, and pass Simulator to the cost function as
an input. For more information, see “Write a Cost Function” on page 2-83. In the
cost function, the simulator configured for fast restart is used to update the model
parameters, simulate the model, and log signals.

Use an anonymous function with one argument that calls myCostfcn.

evalfcn = @(param) myCostfcn(param,Simulator,Exp);
7 Evaluate the model.
[param_opt,opt_info] = sdo.evaluate(evalfcn,param,param_samples);
8 Restore the simulator fast restart settings.
Simulator = fastRestart(Simulator,'off');

The fast restart workflow is similar for sensitivity analysis that identifies design
variables using a cost function aimed at response optimization. See “Use Fast Restart
Mode During Response Optimization” on page 3-227.

Troubleshooting
Why Don’t I See the Evaluation Speedup I Expected Using Fast Restart?

You see a speedup of design optimization tasks using fast restart in models that have a
long compilation phase. If the compilation phase of your model is not long, you do not see
a significant change in estimation speed.

See Also
fastRestart | sdo.evaluate | sdo.SimulationTest

Related Examples
• “Improving Optimization Performance using Fast Restart (GUI)” on page 2-198

4-18
 Use Fast Restart Mode During Sensitivity Analysis

• “Improving Optimization Performance using Fast Restart (Code)” on page 2-206

More About
• “Ways to Speed Up Design Optimization Tasks”

4-19
4 Sensitivity Analysis

Design Exploration using Parameter Sampling (Code)

This example shows how to sample and explore a design space. You explore the design
of a Continuously Stirred Tank Reactor to minimize product concentration variation and
production cost. The design includes feed stock uncertainty.

You explore the CSTR design by characterizing design parameters using probability
distributions. You use the distributions to generate random samples in the design space
and perform Monte-Carlo evaluation of the design at these sample points. You then
create plots to visualize the design space and select the best design. You can then use the
best design as an initial guess for optimization of the design.

You can also use the sampled design space and Monte-Carlo evaluation output to analyze
the influence of design parameters on the design, see "Sensitivity Analysis for Parameter
Estimation (Code)"

Continuously Stirred Tank Reactor (CSTR) Model

Continuously Stirred Tank Reactors (CSTRs) are common in the process industry. The
Simulink model, sdoCSTR, models a jacketed diabatic (i.e., non-adiabatic) tank reactor
described in [1]. The CSTR is assumed to be perfectly mixed, with a single first-order
exothermic and irreversible reaction, . , the reactant, is converted to , the
product.

In this example, you use the following two-state CSTR model, which uses basic
accounting and energy conservation principles:

• , and - Concentrations of A in the CSTR and in the feed [kgmol/m^3]

• , , and - CSTR, feed, and coolant temperatures [K]

• and - Volumetric flow rate [m^3/h] and the density of the material in the CSTR
[1/m^3]

4-20
 Design Exploration using Parameter Sampling (Code)

• and - Height [m] and heated cross-sectional area [m^2] of the CSTR.

• - Pre-exponential non-thermal factor for reaction [1/h]

• and - Activation energy and heat of reaction for [kcal/kgmol]

• - Boltzmann's gas constant [kcal/(kgmol * K)]

• and - Heat capacity [kcal/K] and heat transfer coefficients [kcal/(m^2 * K * h)]

Open the Simulink model.

open_system('sdoCSTR');

CSTR Design Problem

Assume that the CSTR is cylindrical, with the coolant applied to the base of the cylinder.
Tune the CSTR cross-sectional area, , and CSTR height, , to meet the following design
goals:

• Minimize the variation in residual concentration, . Variations in the residual

concentration negatively affect the quality of the CSTR product. Minimizing the
variations also improves CSTR profit.

4-21
4 Sensitivity Analysis

• Minimize the mean coolant temperature . Heating or cooling the jacket coolant
temperature is expensive. Minimizing the mean coolant temperature improves CSTR
profit.

The design must allow for variations in the quality of supply feed concentration, ,
and feed temperature, . The CSTR is fed with feed from different suppliers. The
quality of the feed differs from supplier to supplier and also varies within each supply
batch.

Specify Design Variables

Select the following model parameters as design variables:

• Cylinder cross-sectional area

• Cylinder height

p = sdo.getParameterFromModel('sdoCSTR',{'A','h'});

Limit the cross-sectional area to a range of [0.2 2] m^2.

p(1).Minimum = 0.2;
p(1).Maximum = 2;

Limit the height to a range of [0.5 3] m.

p(2).Minimum = 0.5;
p(2).Maximum = 3;

Sample the Design Space

Create a parameter space for the design variables. The parameter space characterizes
the allowable parameter values and combinations of parameter values.
pSpace = sdo.ParameterSpace(p);

The parameter space uses default uniform distributions for the design variables. The
distribution lower and upper bounds are set to the design variable minimum and
maximum value respectively.

Use the sdo.sample function to generate samples from the parameter space. You use
the samples to evaluate the model and explore the design space.
rng default; % For reproducibility

4-22
 Design Exploration using Parameter Sampling (Code)

pSmpl = sdo.sample(pSpace,30);

Use the sdo.scatterPlot command to visualize the sampled parameter space. The
scatter plot shows the parameter distributions on the diagonal subplots and pairwise
parameter combinations on the off diagonal subplots.
figure, sdo.scatterPlot(pSmpl)

Specify Uncertain Variables

Select the feed concentration and feed temperature as uncertain variables. You evaluate
the design using different values of feed temperature and concentration.
pUnc = sdo.getParameterFromModel('sdoCSTR',{'FeedCon0','FeedTemp0'});

4-23
4 Sensitivity Analysis

Create a parameter space for the uncertain variables. Use normal distributions for both
variables. Specify the mean as the current parameter value. Specify a variance of 5% of
the mean for the feed concentration and 1% of the mean for the temperature.

uSpace = sdo.ParameterSpace(pUnc);
uSpace = setDistribution(uSpace,'FeedCon0',makedist('normal',pUnc(1).Value,0.05*pUnc(1)
uSpace = setDistribution(uSpace,'FeedTemp0',makedist('normal',pUnc(2).Value,0.01*pUnc(2

The feed concentration is inversely correlated with the feed temperature. Add this
information to the parameter space.

uSpace.RankCorrelation = [1 -0.6; -0.6 1];

The rank correlation matrix has a row and column for each parameter with the (i,j) entry
specifying the correlation between the i and j parameters.

Sample the parameter space. The scatter plot shows the correlation between
concentration and temperature.

uSmpl = sdo.sample(uSpace,60);
sdo.scatterPlot(uSmpl)

4-24
 Design Exploration using Parameter Sampling (Code)

Ideally you want to evaluate the design for every combination of points in the design and
uncertain spaces, which implies 30*60 = 1800 simulations. Each simulation takes around
0.5 sec. You can use parallel computing to speed up the evaluation. For this example
you instead only use the samples that have maximum & minimum concentration and
temperature values, reducing the evaluation time to around 1 min.
[~,iminC] = min(uSmpl.FeedCon0);
[~,imaxC] = max(uSmpl.FeedCon0);
[~,iminT] = min(uSmpl.FeedTemp0);
[~,imaxT] = max(uSmpl.FeedTemp0);
uSmpl = uSmpl(unique([iminC,imaxC,iminT,imaxT]) ,:)

uSmpl =

4-25
4 Sensitivity Analysis

FeedCon0 FeedTemp0
________ _________

9.4555 303.58
11.175 288.13
11.293 290.73
8.9308 294.16

Create Evaluation Function

Create a function that evaluates the design for a given sample point in the design space.
The design is evaluated on how well it minimizes the variation in residual concentration
and mean coolant temperature.

Specify Design Requirements

Evaluating a point in the design space requires logging model signals. Logged signals are
used to evaluate the design requirements.

Log the following signals:

• CSTR concentration, available at the second output port of the sdoCSTR/CSTR block
Conc = Simulink.SimulationData.SignalLoggingInfo;
Conc.BlockPath = 'sdoCSTR/CSTR';
Conc.OutputPortIndex = 2;
Conc.LoggingInfo.NameMode = 1;
Conc.LoggingInfo.LoggingName = 'Concentration';

• Coolant temperature, available at the first output of the sdoCSTR/Controller block

Coolant = Simulink.SimulationData.SignalLoggingInfo;
Coolant.BlockPath = 'sdoCSTR/Controller';
Coolant.OutputPortIndex = 1;
Coolant.LoggingInfo.NameMode = 1;
Coolant.LoggingInfo.LoggingName = 'Coolant';

Create and configure a simulation test object to log the required signals.
simulator = sdo.SimulationTest('sdoCSTR');
simulator.LoggingInfo.Signals = [Conc,Coolant];

Evaluation Function

4-26
 Design Exploration using Parameter Sampling (Code)

Use an anonymous function with one argument that calls the sdoCSTR_design function.
evalDesign = @(p) sdoCSTR_design(p,simulator,pUnc,uSmpl);

The evalDesign function:

• Has one input argument that specifies the CSTR dimensions

• Returns the optimization objective value

The sdoCSTR_design function uses a for loop that iterates through the sample values
specified for the feed concentration and temperature. Within the loop, the function:

• Simulates the model using the current design point, feed concentration, and feed
temperature values

• Calculates the residual concentration variation and coolant temperature costs

To view the objective function, type edit sdoCSTR_design.

type sdoCSTR_design

function design = sdoCSTR_design(p,simulator,pUnc,smplUnc)

%SDOCSTR_DESIGN
%
% The sdoCSTR_design function is used to evaluate a CSTR design.
%
% The |p| input argument is the vector of CSTR dimensions.
%
% The |simulator| input argument is a sdo.SimulinkTest object used to
% simulate the |sdoCSTR| model and log simulation signals.
%
% The |pUnc| input argument is a vector of parameters to specify the CSTR
% input feed concentration and feed temperature. The |smplUnc| argument is
% a table of different feed concentration and temperature values.
%
% The |design| return argument contains information about the design
% evaluation that can be used by the |sdo.optimize| function to optimize
% the design.
%
% see also sdo.optimize, sdoExampleCostFunction
%

% Copyright 2012-2013 The MathWorks, Inc.

4-27
4 Sensitivity Analysis

%% Model Simulations and Evaluations

%
% For each value in |smplUnc|, configure and simulate the model. Use
% the logged concentration and coolant signals to compute the design cost.
%
costConc = 0;
costCoolant = 0;
for ct=1:size(smplUnc,1)
%Set the feed concentration and temperature values
pUnc(1).Value = smplUnc{ct,1};
pUnc(2).Value = smplUnc{ct,2};

%Simulate model
simulator.Parameters = [p; pUnc];
simulator = sim(simulator);
logName = get_param('sdoCSTR','SignalLoggingName');
simLog = get(simulator.LoggedData,logName);

%Compute Concentration cost based on the standard deviation of the

%concentration from a nominal value.
Sig = find(simLog,'Concentration');
costConc = costConc+10*std(Sig.Values-2);

%Compute coolant cost based on the mean deviation from room

%temperature.
Sig = find(simLog,'Coolant');
costCoolant = costCoolant+abs(mean(Sig.Values - 294))/30;
end

%% Return Total Cost

%
% Compute the total cost as a sum of the concentration and coolant costs.
%
design.F = costConc + costCoolant;

%%
% Add the individual cost terms to the return argument. These are not used
% by the optimizer, but included for convenience.
design.costConc = costConc;
design.costCoolant = costCoolant;
end

4-28
 Design Exploration using Parameter Sampling (Code)

Evaluate

Use the sdo.evaluate command to evaluate the model at the sample design points.

y = sdo.evaluate(evalDesign,p,pSmpl);

Model evaluated at 30 samples.

View the results of the evaluation using a scatter plot. The scatter plot shows pairwise
plots for each design variable (A,h) and design cost. The plot include the total cost, F, as
well as coolant and concentration costs, costCoolant and costConc respectively.

sdo.scatterPlot(pSmpl,y);

4-29
4 Sensitivity Analysis

The plot shows that larger cross-sectional areas result in lower total costs. However it is
difficult to tell how the height influences the total cost.

Create a mesh plot showing the total cost as a function of A and h.

Ftotal = scatteredInterpolant(pSmpl.A,pSmpl.h,y.F);
xR = linspace(min(pSmpl.A),max(pSmpl.A),60);
yR = linspace(min(pSmpl.h),max(pSmpl.h),60);
[xx,yy] = meshgrid(xR,yR);
zz = Ftotal(xx,yy);
mesh(xx,yy,zz)
view(56,30)
title('Total cost as function of A and h')
zlabel('Ftotal')
xlabel(p(1).Name), ylabel(p(2).Name);

4-30
 Design Exploration using Parameter Sampling (Code)

The plot shows that high values of A and h result in lower costs. The best design in the
sampled space corresponds to the design with the lowest cost value.

[~,idx] = min(y.F);
pBest = [y(idx,:), pSmpl(idx,:)]

pBest =

F costConc costCoolant A h
______ ________ ___________ ______ ______

2.1052 1.5757 0.52953 1.8483 2.1158

4-31
4 Sensitivity Analysis

Refine the Design Space

The total cost surface plot shows that low cost designs are designs with A in the range
[1.5 2] and h in the range [2 3]. Modify the parameter space distributions for A and h and
resample the design space to focus on this region.
pSpace = setDistribution(pSpace,'A',makedist('uniform',1.5,2));
pSpace = setDistribution(pSpace,'h',makedist('uniform',2,3));
pSmpl = sdo.sample(pSpace,30);

Add the pBest found earlier to the new samples so that it is part of the refined design
space.
pSmpl = [pSmpl;pBest(:,4:5)];
sdo.scatterPlot(pSmpl)

4-32
 Design Exploration using Parameter Sampling (Code)

Evaluate using Refined Design Space

y = sdo.evaluate(evalDesign,p,pSmpl);

Model evaluated at 31 samples.

Create a mesh plot for this section of the design space. The surface indicates that better
designs are near the A = 1.9, h = 2.1 point.

Ftotal = scatteredInterpolant(pSmpl.A,pSmpl.h,y.F);
xR = linspace(min(pSmpl.A),max(pSmpl.A),60);
yR = linspace(min(pSmpl.h),max(pSmpl.h),60);
[xx,yy] = meshgrid(xR,yR);
zz = Ftotal(xx,yy);
mesh(xx,yy,zz)
view(56,30)
title('Total cost as function of A and h')
zlabel('Ftotal')
xlabel(p(1).Name), ylabel(p(2).Name);

4-33
4 Sensitivity Analysis

Find the best design from the new design space and compare with the best design point
found earlier.

[~,idx] = min(y.F);
pBest = [pBest; [y(idx,:), pSmpl(idx,:)]]

pBest =

F costConc costCoolant A h
______ ________ ___________ ______ ______

2.1052 1.5757 0.52953 1.8483 2.1158

1.979 1.4838 0.49528 1.9695 2.1174

4-34
 Design Exploration using Parameter Sampling (Code)

The best design in the refined design space is better than the design found earlier. This
indicates that there may be better designs in the same region and warrants refining the
design space further. Alternatively you can use the best design point as an initial guess
for optimization.

Related Examples

To learn how to optimize the CSTR design using the sdo.optimize command, see
"Design Optimization with Uncertain Variables (Code)".

To learn how to analyze the influence of design parameters on the design using the
sdo.analyze command, see "Sensitivity Analysis for Parameter Estimation (Code)"

References

[1] Bequette, B.W. Process Dynamics: Modeling, Analysis and Simulation. 1st ed. Upper
Saddle River, NJ: Prentice Hall, 1998.

Close the model

bdclose('sdoCSTR')

4-35
4 Sensitivity Analysis

Identify Key Parameters for Estimation (Code)

This example shows how to use sensitivity analysis to narrow down the number of
parameters that you need to estimate to fit a model. This example uses a model of the
vestibulo-ocular reflex, which generates compensatory eye movements.

Model Description

The vestibulo-ocular reflex (VOR) enables the eyes to move at the same speed and in the
opposite direction as the head, so that vision is not blurred when the head moves during
normal activity. For example, if the head turns in one direction, the eyes turn in the
opposite direction, with the same speed. This happens even in the dark. In fact, the VOR
is most easily characterized by measurements in the dark, to ensure that eye movements
are predominantly driven by the VOR.

The file sdoVOR_Data.mat contains uniformly sampled data of stimulation and eye
movements. If the VOR were perfectly compensatory, then a plot of eye movement data,
when flipped vertically, would overlay exactly on top of a plot of head motion data. Such
a system would be described by a gain of 1 and a phase of 180 degrees. However, when
we plot the data in the file sdoVOR_Data.mat, the eye movements are close, but not
perfectly compensatory.

load sdoVOR_Data.mat; % Column vectors: Time HeadData EyeData

figure
plot(Time, HeadData, ':b', Time, EyeData, '-g')
xlabel('Time (sec)')
ylabel('Angular Velocity (deg/sec)')
ylim([-110 110])
legend('Head Data', 'Eye Data')

4-36
 Identify Key Parameters for Estimation (Code)

The eye movement data does not perfectly overlay the head motion data, and this can be
modeled by several factors. Head rotation is sensed by organs in the inner ears, known as
semicircular canals. These detect head motion and transmit signals about head motion
to the brain, which sends motor commands to the eye muscles, so that eye movements
compensate for head motion. We would like to use this eye movement data to estimate
the parameters in the models for these various stages. The model we will use is shown
below. There are four parameters in the model: Delay, Gain, Tc, and Tp.

model_name = 'sdoVOR';
open_system(model_name)

4-37
4 Sensitivity Analysis

The Delay parameter models the fact that there is some delay in communicating the
signals from the inner ear to the brain and the eyes. This delay is due to the time needed
for chemical neurotransmitters to traverse the synaptic clefts between nerve cells. Based
on the number of synapses involved in the vestibulo-ocular reflex, this delay is expected
to be around 5 ms. For estimation purposes, we will assume it is between 2 and 9 ms.

Delay = sdo.getParameterFromModel(model_name, 'Delay');

Delay.Value = 0.005; % seconds
Delay.Minimum = 0.002;
Delay.Maximum = 0.009;

The Gain parameter models the fact that the the eyes do not move quite as much as the
head does. We will use 0.8 as our initial guess, and assume it is between 0.6 and 1.

Gain = sdo.getParameterFromModel(model_name, 'Gain');

Gain.Value = 0.8;
Gain.Minimum = 0.6;
Gain.Maximum = 1;

The Tc parameter models the dynamics associated with the semicircular canals, as well
as some additional neural processing. The canals are high-pass filters, because after a
subject has been put into rotational motion, the neurally active membranes in the canals
slowly relax back to resting position, so the canals stop sensing motion. Thus in the plot
above, after the stimulation undergoes transition edges, the eye movements tend to
depart from the stimulation over time. Based on mechanical characteristics of the canals,
combined with additional neural processing which prolongs this time constant to improve

4-38
 Identify Key Parameters for Estimation (Code)

the accuracy of the VOR, we will estimate the Tc parameter to be 15 seconds, and assume
it is between 10 and 30 seconds.

Tc = sdo.getParameterFromModel(model_name, 'Tc');
Tc.Value = 15;
Tc.Minimum = 10;
Tc.Maximum = 30;

Finally, the Tp parameter models the dynamics of the oculomotor plant, i.e. the eye and
the muscles and tissues attached to it. The plant can be modeled by two poles, however
it is believed that the pole with the larger time constant is cancelled by precompensation
in the brain, to enable the eye to make quick movements. Thus in the plot, when the
stimulation undergoes transition edges, the eye movements follow with only a little
delay. For the Tp parameter, we will use 0.01 seconds as our initial guess, and assume it
is between 0.005 and 0.05 seconds.

Tp = sdo.getParameterFromModel(model_name, 'Tp');
Tp.Value = 0.01;
Tp.Minimum = 0.005;
Tp.Maximum = 0.05;

Collect these parameters into a vector.

v = [Delay Gain Tc Tp];

Compare Measured Data to Initial Simulated Output

Create an Experiment object. Specify HeadData as input.

Exp = sdo.Experiment(model_name);
Exp.InputData = timeseries(HeadData, Time);

Associate eye movement data with model output.

EyeMotion = Simulink.SimulationData.Signal;
EyeMotion.Name = 'EyeMotion';
EyeMotion.BlockPath = [model_name '/Oculomotor Plant'];
EyeMotion.PortType = 'outport';
EyeMotion.PortIndex = 1;
EyeMotion.Values = timeseries(EyeData, Time);

Add EyeMotion to the experiment.

Exp.OutputData = EyeMotion;

4-39
4 Sensitivity Analysis

Use the data's timing characteristics in the model.

stop_time = Time(end);
set_param(gcs, 'StopTime', num2str(stop_time));
dt = Time(2) - Time(1);
set_param(gcs, 'FixedStep', num2str(dt))

Create a simulation scenario using the experiment, and obtain the simulated output.

Exp = setEstimatedValues(Exp, v); % use vector of parameters/states

Simulator = createSimulator(Exp);
Simulator = sim(Simulator);

Search for the model_residual signal in the logged simulation data.

SimLog = find(Simulator.LoggedData, ...

get_param(model_name, 'SignalLoggingName') );
EyeSignal = find(SimLog, 'EyeMotion');

The model output does not match the data very well, as shown by the residual, which we
can compute by calling the objective function.

estFcn = @(v) sdoVOR_Objective(v, Simulator, Exp, 'Residuals');

Model_Error = estFcn(v);
plot(Time, EyeData, '-g', ...
EyeSignal.Values.Time, EyeSignal.Values.Data, '--c', ...
Time, Model_Error.F, '-r');
xlabel('Time (sec)');
ylabel('Angular Velocity (deg/sec)');
legend('Eye Data', 'Model', 'Residual');

4-40
 Identify Key Parameters for Estimation (Code)

The objective function used above is defined in the file "sdoVOR_Objective.m".

type sdoVOR_Objective.m

function vals = sdoVOR_Objective(v, Simulator, Exp, Method)

% Compare model output with data
%
% Inputs:
% v - vector of parameters and/or states
% Simulator - used to simulate the model
% Exp - Experiment object
% Method - 'SSE' for scalar output, 'Residuals' for vector of residuals

4-41
4 Sensitivity Analysis

% Copyright 2014-2015 The MathWorks, Inc.

% Requirement setup
req = sdo.requirements.SignalTracking;
req.Type = '==';
req.Method = Method;

% If Residuals requested, keep on same scale as signals, for plotting

switch Method
case 'Residuals'
req.Normalize = 'off';
end

% Simulate the model

Exp = setEstimatedValues(Exp, v); % use vector of parameters/states
Simulator = createSimulator(Exp,Simulator);
Simulator = sim(Simulator);

% Compare model output with data

SimLog = find(Simulator.LoggedData, ...
get_param(Exp.ModelName, 'SignalLoggingName') );
OutputModel = find(SimLog, 'EyeMotion');
Model_Error = evalRequirement(req, OutputModel.Values, Exp.OutputData.Values);
vals.F = Model_Error;

Sensitivity Analysis

Create an object to sample the parameter space.

ps = sdo.ParameterSpace([Delay ; Gain ; Tc ; Tp]);

Generate 100 samples from the parameter space.

rng default; % for reproducibility

x = sdo.sample(ps, 100);
sdo.scatterPlot(x);

4-42
 Identify Key Parameters for Estimation (Code)

The sampling above used default options, and these are reflected in the plots above.
Parameter values were selected at random from distributions that were uniform over the
range of each parameter. Consequently, the histogram plots along the diagonal appear
approximately uniform. If Statistics and Machine Learning Toolbox™ is available, a
number of distributions may be used in addition to uniform and normal, and sampling
can be done in a Latin hypercube pattern.

The off-diagonal plots above show scatter plots between pairs of different variables.
Since we did not specify a RankCorrelation matrix in ps, the scatter plots do not indicate
correlations. However, if parameters were believed to be correlated, this can be specified
using the RankCorrelation property of ps.

4-43
4 Sensitivity Analysis

For sensitivity analysis, it is simpler to use a scalar objective, so we will specify the sum
of squared errors, "SSE":
estFcn = @(v) sdoVOR_Objective(v, Simulator, Exp, 'SSE');
y = sdo.evaluate(estFcn, ps, x);

Model evaluated at 100 samples.

Evaluation could also be sped up using parallel computing.

Obtain the standardized regression coefficients.

opts = sdo.AnalyzeOptions;
opts.Method = 'StandardizedRegression';
sensitivities = sdo.analyze(x, y, opts);

Other types of analysis include correlation and, if Statistics and Machine Learning
Toolbox is available, partial correlation.

We can view the analysis results.

disp(sensitivities)

F
_________

Delay 0.01303
Gain -0.90873
Tc -0.044395
Tp 0.19919

For standardized regression, parameters that highly influence the model output have
sensitivity magnitudes close to 1. On the other hand, less influential parameters have
smaller sensitivity magnitudes. We see that this objective function is sensitive to changes
in the Gain and Tp parameters, but much less sensitive to changes in the Delay and Tc
parameters.

You can validate sensitivity analysis results by resampling and reevaluating the
objective function for the samples. You can also use engineering intuition for a quick
analysis. For example, in this model, the time constant Tc ranges from 10 to 30 seconds.
Even the minimum value of 10 seconds is large compared to the 2-second duration
for which the head motion stimulation is held at constant velocity. Therefore, Tc is
not expected to affect the output greatly. However, even when this kind of intuition

4-44
 Identify Key Parameters for Estimation (Code)

is not readily available in other models, sensitivity analysis can help highlight which
parameters are influential.

Based on the results of sensitivity analysis, designate the Delay and Tc parameters
as fixed when optimizing. This reduction in the number of free parameters speeds up
optimization.
Delay.Free = false;
Tc.Free = false;

Optimization
We can use the minimum from sensitivity analysis as the initial guess for optimization.
[fval, idx_min] = min(y.F);
Delay.Value = x.Delay(idx_min);
Gain.Value = x.Gain(idx_min);
Tc.Value = x.Tc(idx_min);
Tp.Value = x.Tp(idx_min);
%
v = [Delay Gain Tc Tp];
opts = sdo.OptimizeOptions;
opts.Method = 'fmincon';

As was the case with model evaluations in sensitivity analysis, parallel computing could
be used to speed up the optimization.
vOpt = sdo.optimize(estFcn, v, opts);
disp(vOpt)

Optimization started 31-Jul-2015 06:11:09

max Step-size First-order

Iter F-count f(x) constraint optimality
0 5 13.4798 0
1 18 12.2052 0 0.129 305
2 30 11.1441 0 0.0648 781
3 41 10.0493 0 0.0843 289
4 46 9.23607 0 0.0758 227
5 51 8.76122 0 0.0183 10.1
6 56 8.75862 0 0.00184 0.476
7 57 8.75862 0 8.41e-05 0.476
Local minimum possible. Constraints satisfied.

fmincon stopped because the size of the current step is less than

4-45
4 Sensitivity Analysis

the selected value of the step size tolerance and constraints are
satisfied to within the selected value of the constraint tolerance.

(1,1) =

Name: 'Delay'
Value: 0.0038
Minimum: 0.0020
Maximum: 0.0090
Free: 0
Scale: 0.0078
Info: [1x1 struct]

(1,2) =

Name: 'Gain'
Value: 0.9012
Minimum: 0.6000
Maximum: 1
Free: 1
Scale: 1
Info: [1x1 struct]

(1,3) =

Name: 'Tc'
Value: 16.6833
Minimum: 10
Maximum: 30
Free: 0
Scale: 16
Info: [1x1 struct]

(1,4) =

Name: 'Tp'
Value: 0.0157
Minimum: 0.0050
Maximum: 0.0500
Free: 1
Scale: 0.0156

4-46
 Identify Key Parameters for Estimation (Code)

Info: [1x1 struct]

1x4 param.Continuous

Visualizing Result of Optimization

Obtain the model response after estimation. Search for the model_residual signal in the
logged simulation data.

Exp = setEstimatedValues(Exp, vOpt);

Simulator = createSimulator(Exp,Simulator);
Simulator = sim(Simulator);
SimLog = find(Simulator.LoggedData, ...
get_param(model_name, 'SignalLoggingName') );
EyeSignal = find(SimLog, 'EyeMotion');

Comparing the measured eye data with the optimized model response shows that the
residuals are much smaller.

estFcn = @(v) sdoVOR_Objective(v, Simulator, Exp, 'Residuals');

Model_Error = estFcn(vOpt);
plot(Time, EyeData, '-g', ...
EyeSignal.Values.Time, EyeSignal.Values.Data, '--c', ...
Time, Model_Error.F, '-r');
xlabel('Time (sec)');
ylabel('Angular Velocity (deg/sec)');
legend('Eye Data', 'Model', 'Residual');

4-47
4 Sensitivity Analysis

Close the model

bdclose(model_name)

4-48
 5

Optimization-Based Control Design

• “Overview of Optimization-Based Compensator Design” on page 5-2

• “Time-Domain Design Requirements in Simulink” on page 5-4
• “Frequency-Domain Design Requirements in Simulink” on page 5-16
• “Time- and Frequency-Domain Requirements in SISO Design Tool” on page 5-38
• “Time-Domain Simulations in SISO Design Tool” on page 5-42
• “How to Design Optimization-Based Controllers for LTI Systems” on page 5-43
• “Optimize LTI System to Meet Frequency-Domain Requirements” on page 5-44
• “Designing Linear Controllers for Simulink Models” on page 5-64
5 Optimization-Based Control Design

Overview of Optimization-Based Compensator Design

You can design optimization-based controllers for Simulink models to meet time and
frequency-domain design requirements, as described in “Design Optimization to Meet
Time- and Frequency-Domain Requirements (GUI)” on page 3-93.

If you have Control System Toolbox software installed, you can also design and optimize
control systems by tuning controller elements or parameters within a SISO Design Task
in the Control and Estimation Tools Manager. You can tune elements or parameters such
as poles, zeros, and gains within any controller in the system and optimize the open and
closed loop responses to meet time- and frequency-domain requirements.

Optimize the responses of systems in the SISO Design Task to meet both time- and
frequency-domain performance requirements by graphically constraining signals:

• Add frequency-domain design requirements to plots such as root-locus, Nichols, and

Bode in the SISO Design Task graphical tuning editor called SISO Design Tool.
• Add time-domain design requirements to plots such as step or impulse response
(when displayed within the Linear System Analyzer as part of a SISO Design Task).

You can use optimization methods in a SISO Design Task in the Control and Estimation
Tools Manager to tune both command-line LTI models as well as Simulink models:

• Create an LTI model using the Control System Toolbox command-line functions
and use the controlSystemDesigner function to create a SISO Design Task for
the model. For an example, see “Optimize LTI System to Meet Frequency-Domain
Requirements” on page 5-44.
• Use a Simulink Compensator Design task (from Simulink Control Design software) to
automatically analyze the model and then create a SISO Design Task for a linearized
version of the model. You can then use the optimization techniques in the SISO
Design Task to tune the response of the linearized Simulink model. For an example,
see “Design Optimization-Based PID Controller for Linearized Simulink Model
(GUI)”.

Note: When using response optimization within a SISO Design Task you cannot add
uncertainty to system parameters.

When using a SISO Design Task, Simulink Design Optimization software automatically
sets the model's simulation start and stop time and you cannot directly change them.

5-2
 Overview of Optimization-Based Compensator Design

By default, the simulation starts at 0 and continues until the SISO Design Task
determines that the dynamics of the model have settled out. In addition, when the design
requirements extend beyond this point, the simulation continues to the extent of the
design requirements. Although you cannot directly adjust the start or stop time of the
simulation, you can adjust the design requirements to extend further in time and thus
force the simulation to continue to a certain point.

5-3
5 Optimization-Based Control Design

Time-Domain Design Requirements in Simulink

In this section...
“Specify Piecewise-Linear Lower and Upper Bounds” on page 5-23
“Specify Step Response Characteristics” on page 5-13
“Track Reference Signals” on page 5-30
“Specify Custom Requirements” on page 5-32
“Edit Design Requirements” on page 5-35

Specify Piecewise-Linear Lower and Upper Bounds

To specify upper and lower bounds on a signal:

1 In the Response Optimization tool, select Signal Bound in the New drop-down list.
A window opens where you specify upper or lower bounds on a signal.
2 Specify a requirement name in the Name box.
3 Select the requirement type using the Type list.
4 Specify the edge start and end times and corresponding amplitude in the Time (s)
and Amplitude columns.
5
Click to specify additional bound edges.

Select a row and click to delete a bound edge.

6 In the Select Signals to Bound area, select a logged signal to apply the
requirement to.

If you have already selected signals, as described in “Specify Signals to Log” on page
3-12, they appear in the list. Select the corresponding check-box.

If you haven’t selected a signal to log:

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

5-4
 Time-Domain Design Requirements in Simulink

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

Click OK. A new variable, with the specified name, appears in the Data area of
the Response Optimization tool window.
7 Click OK.

A variable with the specified requirement name appears in the Data area of the tool.
A graphical display of the requirement also appears in the Response Optimization
tool window.
8 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can add a Check Custom Bounds block to your model to specify
piecewise-linear bounds.

Specify Step Response Characteristics

To specify step response characteristics:

5-5
5 Optimization-Based Control Design

1 You can apply this requirement to either a signal or a linearization of your model.

In the Response Optimization Tool, click New. To apply this requirement to a

signal, select the Step Response Envelope entry in the New Time Domain
Requirement section of the New list. To apply this requirement to a linearization
of your model, select the Step Response Envelope entry in the New Frequency
Domain Requirement section of the New list. The latter option requires Simulink
Control Design software.

A window opens where you specify the step response requirements on a signal, or
system.
2 Specify a requirement name in the Name box.
3 Specify the step response characteristics:

% Overshoot
% Settling
Final Value

% Rise
Amplitude

% Undershoot
Initial Value

Rise Time Settling Time Time

• Initial value: Input level before the step occurs

• Step time: Time at which the step takes place
• Final value: Input level after the step occurs

5-6
 Time-Domain Design Requirements in Simulink

• Rise time: The time taken for the response signal to reach a specified percentage
of the step's range. The step's range is the difference between the final and initial
values.
• % Rise: The percentage used in the rise time.
• Settling time: The time taken until the response signal settles within a specified
region around the final value. This settling region is defined as the final step
value plus or minus the specified percentage of the final value.
• % Settling: The percentage used in the settling time.
• % Overshoot: The amount by which the response signal can exceed the final
value. This amount is specified as a percentage of the step's range. The step's
range is the difference between the final and initial values.
• % Undershoot: The amount by which the response signal can undershoot the
initial value. This amount is specified as a percentage of the step's range. The
step's range is the difference between the final and initial values.
4 Specify the signals or systems to be bound.

You can apply this requirement to a model signal or to a linearization of your

Simulink model (requires Simulink Control Design software).

• Apply this requirement to a model signal:

In the Select Signals to Bound area, select a logged signal to which you will
apply the requirement.

If you have already selected a signal to log, as described in “Specify Signals to

Log” on page 3-12, it appears in the list. Select the corresponding check-box.

If you haven’t selected a signal to log:

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

5-7
5 Optimization-Based Control Design

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

Click OK. A new variable, with the specified name, appears in the Data area
of the Response Optimization tool window.
• Apply this requirement to a linear system.

Linear systems are defined by snapshot times at which the model is linearized
and sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a
vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in

the list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open

the Create linearization I/O set dialog box.

5-8
 Time-Domain Design Requirements in Simulink

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A variable with the specified requirement name appears in the Data area of the tool.
A graphical display of the requirement also appears in the Response Optimization
tool window.

Alternatively, you can use the Check Step Response Characteristics block to
specify step response bounds for a signal.

See Also

“Design Optimization to Meet Step Response Requirements (GUI)”

Track Reference Signals

Use reference tracking to force a model signal to match a desired signal.

To track a reference signal:

1 In the Response Optimization tool, select Signal Tracking in the New drop-down
list. A window opens where you specify the reference signal to track.
2 Specify a requirement name in the Name box.
3 Define the reference signal by entering vectors, or variables from the workspace, in
the Time vector and Amplitude fields.

Click Update reference signal data to use the new amplitude and time vector as
the reference signal.
4 Specify how the optimization solver minimizes the error between the reference and
model signals using the Tracking Method list:

• SSE — Reduces the sum of squared errors

• SAE — Reduces the sum of absolute errors
5 In the Specify Signal to Track Reference Signal area, select a logged signal to
apply the requirement to.

5-9
5 Optimization-Based Control Design

If you already selected a signal to log, as described in “Specify Signals to Log” on

page 3-12, they appear in the list. Select the corresponding check-box.

If you haven’t selected a signal to log:

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

Click OK. A new variable, with the specified name, appears in the Data area of
the Response Optimization tool window.
e Select the check-box corresponding to the signal and click OK.

A variable with the specified requirement name appears in the Data area of the tool.
A graphical display of the signal bound also appears in the Response Optimization
tool window.

5-10
 Time-Domain Design Requirements in Simulink

Note: When tracking a reference signal, the software ignores the maximally feasible
solution option. For more information on this option, see “Selecting Optimization
Termination Options” on page 3-74.

Alternatively, you can use the Check Against Reference block to specify a reference
signal to track.

See Also

“Design Optimization to Track Reference Signal (GUI)”

Specify Custom Requirements

To specify custom requirements, such as minimizing system energy:

1 In the Response Optimization tool, select Custom Requirement in the New list. A
window opens where you specify the custom requirement.
2 Specify a requirement name in the Name box.
3 Specify the requirement type using the Type list.
4 Specify the name of the function that contains the custom requirement in the
Function box. The field must be specified as a function handle using @. The function

must be on the MATLAB path. Click to review or edit the function.

If the function does not exist, clicking opens a template MATLAB file.
Use this file to implement the custom requirement. The default function name is
myCustomRequirement.
5 (Optional) If you want to prevent the solver from considering specific parameter
combinations, select the Error if constraint is violated check box. Use this option
for parameter-only constraints.

During an optimization iteration, the solver evaluates requirements with this option
selected first.

• If the constraint is violated, the solver skips evaluating any remaining

requirements and proceeds to the next iterate.

5-11
5 Optimization-Based Control Design

• If the constraint is not violated, the solver evaluates the remaining requirements
for the current iterate. If any of the remaining requirements bound signals or
systems, then the solver simulates the model .

For more information, see “Skip Model Simulation Based on Parameter Constraint
Violation (GUI)” on page 3-188.

Note: If you select this check box, then do not specify signals or systems to bound. If
you do specify signals or systems, then this check box is ignored.
6 (Optional) Specify the signal or system, or both, to be bound.

You can apply this requirement to model signals, or a linearization of your Simulink
model (requires Simulink Control Design software), or both.

Click Select Signals and Systems to Bound (Optional) to view the signal and
linearization I/O selection area.

• Apply this requirement to a model signal:

In the Signal area, select a logged signal to which you will apply the
requirement.

If you have already selected a signal to log, as described in “Specify Signals to

Log” on page 3-12, it appears in the list. Select the corresponding check box.

If you have not selected a signal to log:

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

5-12
 Time-Domain Design Requirements in Simulink

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

Click OK. A new variable, with the specified name, appears in the Data area
of the Response Optimization tool window.
• Apply this requirement to a linear system.

Linear systems are defined by snapshot times at which the model is linearized
and sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a
vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in

the list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open

the Create linearization I/O set dialog box. For more information on using
this dialog box, see “Create Linearization I/O Sets” on page 3-77.

5-13
5 Optimization-Based Control Design

For more information on linearization, see “What Is Linearization?”.

7 Click OK.

A new variable, with the specified name, appears in the Data area of the Response
Optimization tool. A graphical display of the requirement also appears in the
Response Optimization tool window.

See Also

• “Design Optimization to Meet a Custom Objective (GUI)” on page 3-108

• “Design Optimization to Meet Custom Signal Requirements (GUI)” on page 3-135

Edit Design Requirements

The Edit Design Requirement dialog box allows you to exactly position constraint
segments and to edit other properties of these constraints. The dialog box has two main
components:

• An upper panel to specify the constraint you are editing

• A lower panel to edit the constraint parameters

The upper panel of the Edit Design Requirement dialog box resembles the image in the
following figure.

In the context of the SISO Tool in Control System Toolbox software, Design
requirement refers to both the particular editor within the SISO Tool that contains the
requirement and the particular requirement within that editor. To edit other constraints
within the SISO Tool, select another design requirement from the drop-down menu.

Edit Design Requirement Dialog Box Parameters

The particular parameters shown within the lower panel of the Edit Design Requirement
dialog box depend on the type of constraint/requirement. In some cases, the lower panel
contains a grid with one row for each segment and one column for each constraint
parameter. The following table summarizes the various constraint parameters.

5-14
 Time-Domain Design Requirements in Simulink

Edit Design Requirement Dialog Box Parameters

Parameter Found in Description

Time Upper and lower time Defines the time range of a segment
response bounds on step and within a constraint/requirement.
impulse response plots
Amplitude Upper and lower time Defines the beginning and ending
response bounds on step and amplitude of a constraint segment.
impulse response plots
Slope (1/s) Upper and lower time Defines the slope, in 1/s, of a constraint
response bounds segment. It is an alternative method
of specifying the magnitude values.
Entering a new Slope value changes
any previously defined magnitude
values.
Final value Step response bounds Defines the input level after the step
occurs.
Rise time Step response bounds Defines a constraint segment for a
particular rise time.
% Rise Step response bounds The percentage of the step's range used
to describe the rise time.
Settling time Step response bounds Defines a constraint segment for a
particular settling time.
% Settling Step response bounds The percentage of the final value that
defines the settling region used to
describe the settling time.
% Overshoot Step response bounds  
% Undershoot Step response bounds Defines the constraint segments for a
particular percent undershoot.

5-15
5 Optimization-Based Control Design

Frequency-Domain Design Requirements in Simulink

In this section...
“Specify Lower Bounds on Gain and Phase Margin” on page 5-39
“Specify Piecewise-Linear Lower and Upper Bounds on Frequency Response” on page
5-41
“Specify Bound on Closed-Loop Peak Gain” on page 5-43
“Specify Lower Bound on Damping Ratio” on page 5-45
“Specify Upper and Lower Bounds on Natural Frequency” on page 5-47
“Specify Upper Bound on Approximate Settling Time” on page 5-49
“Specify Piecewise-Linear Upper and Lower Bounds on Singular Values” on page 5-51
“Specify Step Response Characteristics” on page 5-13
“Specify Custom Requirements” on page 5-32

Specify Lower Bounds on Gain and Phase Margin

To specify lower bounds on the gain and phase margin of a linear system:

1 In the Response Optimization tool, select Gain and Phase Margin in the New list.
A window opens where you specify lower bounds on the gain and phase margin of
your linear system.
2 Specify a requirement name in Name.
3 Specify bounds on the gain margin or phase margin, or both.

5-16
 Frequency-Domain Design Requirements in Simulink

• Gain margin — Amount of gain increase or decrease required to make the loop
gain unity at the frequency where the phase angle is –180°.
• Phase margin — Amount of phase increase or decrease required to make the
phase angle –180° when the loop gain is 1.0

To specify a lower bound on the gain margin or phase margin, or both, select the
corresponding check box and enter the lower bound value.
4 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

5-17
5 Optimization-Based Control Design

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A variable with the specified requirement name appears in the Data area of the tool.
A graphical display of the requirement also appears in the Response Optimization
tool window.
6 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Gain and Phase Margins block to specify
bounds on the gain and phase margin. (Requires Simulink Control Design.)

Specify Piecewise-Linear Lower and Upper Bounds on Frequency

Response
To specify upper or lower bounds on the magnitude of a system response:

1 In the Response Optimization tool, select Bode Magnitude in the New list. A
window opens where you specify the lower or upper bounds on the magnitude of the
system response.
2 Specify a requirement name in the Name box.
3 Specify the requirement type using the Type list.
4 Specify the edge start and end frequencies and corresponding magnitude in the
Frequency and Magnitude columns.
5 Insert or delete bound edges.

Click to specify additional bound edges.

5-18
 Frequency-Domain Design Requirements in Simulink

Select a row and click to delete a bound edge.

6 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

7 Click OK.

A new variable with the specified name appears in the Data area of the Response
Optimization tool window. A graphical display of the requirement also appears in the
Response Optimization tool window.

5-19
5 Optimization-Based Control Design

8 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Bode Characteristics block to specify bounds
on the magnitude of the system response. (Requires Simulink Control Design.)

Specify Bound on Closed-Loop Peak Gain

To specify an upper bound on the closed-loop peak response of a system:

1 In the Response Optimization tool, select Closed-Loop Peak Gain in the New list.
A window opens where you specify an upper bound on the closed-loop peak gain of
the system.
2 Specify a requirement name in the Name box.
3 Specify the upper bound on the closed-loop peak gain in the Closed-Loop peak
gain box.

5-20
 Frequency-Domain Design Requirements in Simulink

4 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A new variable with the specified name appears in the Data area of the Response
Optimization tool window. A graphical display of the requirement also appears in the
Response Optimization tool window.

5-21
5 Optimization-Based Control Design

6 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Nichols Characteristics block to specify
bounds on the magnitude of the system response. (Requires Simulink Control Design.)

Specify Lower Bound on Damping Ratio

To specify a lower bound on the damping ratio of the system:

1 In the Response Optimization tool, select Damping Ratio in the New list. A
window opens where you specify an upper bound on the damping ratio of the system.
2 Specify a requirement name in the Name box.
3 Specify the lower bound on the damping ratio in the Damping ratio box.

5-22
 Frequency-Domain Design Requirements in Simulink

4 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A new variable with the specified name appears in the Data area of the Response
Optimization tool. A graphical display of the requirement also appears in the
Response Optimization tool window.

5-23
5 Optimization-Based Control Design

6 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Pole-Zero Characteristics block to specify a
bound on the damping ratio. (Requires Simulink Control Design.)

Specify Upper and Lower Bounds on Natural Frequency

To specify a bound on the natural frequency of the system:

1 In the Response Optimization tool, select Natural Frequency in the New list. A
window opens where you specify a bound on the natural frequency of the system.
2 Specify a requirement name in the Name box.

5-24
 Frequency-Domain Design Requirements in Simulink

3 Specify a lower or upper bound on the natural frequency in the Natural frequency
box.
4 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A new variable with the specified name appears in the Data area of the Response
Optimization tool. A graphical display of the requirement also appears in the
Response Optimization tool window.

5-25
5 Optimization-Based Control Design

6 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Pole-Zero Characteristics block to specify a
bound on the natural frequency. (Requires Simulink Control Design.)

Specify Upper Bound on Approximate Settling Time

To specify an upper bound on the approximate settling time of the system:

1 In the Response Optimization tool, select Settling Time in the New list. A window
opens where you specify an upper bound on the approximate settling time of the
system.
2 Specify a requirement name in the Name box.

5-26
 Frequency-Domain Design Requirements in Simulink

3 Specify the upper bound on the approximate settling time in the Settling time box.
4 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A new variable with the specified name appears in the Data area of the Response
Optimization tool. A graphical display of the requirement also appears in the
Response Optimization tool window.

5-27
5 Optimization-Based Control Design

6 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Pole-Zero Characteristics block to specify
the approximate settling time. (Requires Simulink Control Design.)

Specify Piecewise-Linear Upper and Lower Bounds on Singular Values

To specify piecewise-linear upper and lower bounds on the singular values of a system:

1 In the Response Optimization tool, select Singular Values in the New list. A
window opens where you specify the lower or upper bounds on the singular values of
the system.
2 Specify a requirement name in the Name box.

5-28
 Frequency-Domain Design Requirements in Simulink

3 Specify the requirement type using the Type list.

4 Specify the edge start and end frequencies and corresponding magnitude in the
Frequency and Magnitude columns, respectively.
5 Insert or delete bound edges.

Click to specify additional bound edges.

Select a row and click to delete a bound edge.

6 In the Select Systems to Bound section, select the linear systems to which this
requirement applies.

Linear systems are defined by snapshot times at which the model is linearized and
sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in the
list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open the
Create linearization I/O set dialog box.

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

7 Click OK.

A new variable with the specified name appears in the Data area of the Response
Optimization tool. A graphical display of the requirement also appears in the
Response Optimization tool window.

5-29
5 Optimization-Based Control Design

8 (Optional) In the graphical display, you can:

• “Move Constraints Graphically” on page 3-20

• “Position Constraints Exactly” on page 3-21

Alternatively, you can use the Check Singular Value Characteristics block to
specify bounds on the singular value. (Requires Simulink Control Design.)

Specify Step Response Characteristics

To specify step response characteristics:

1 You can apply this requirement to either a signal or a linearization of your model.

In the Response Optimization Tool, click New. To apply this requirement to a

signal, select the Step Response Envelope entry in the New Time Domain

5-30
 Frequency-Domain Design Requirements in Simulink

Requirement section of the New list. To apply this requirement to a linearization

of your model, select the Step Response Envelope entry in the New Frequency
Domain Requirement section of the New list. The latter option requires Simulink
Control Design software.

A window opens where you specify the step response requirements on a signal, or
system.
2 Specify a requirement name in the Name box.
3 Specify the step response characteristics:

% Overshoot
% Settling
Final Value

% Rise
Amplitude

% Undershoot
Initial Value

Rise Time Settling Time Time

• Initial value: Input level before the step occurs

• Step time: Time at which the step takes place
• Final value: Input level after the step occurs
• Rise time: The time taken for the response signal to reach a specified percentage
of the step's range. The step's range is the difference between the final and initial
values.
• % Rise: The percentage used in the rise time.

5-31
5 Optimization-Based Control Design

• Settling time: The time taken until the response signal settles within a specified
region around the final value. This settling region is defined as the final step
value plus or minus the specified percentage of the final value.
• % Settling: The percentage used in the settling time.
• % Overshoot: The amount by which the response signal can exceed the final
value. This amount is specified as a percentage of the step's range. The step's
range is the difference between the final and initial values.
• % Undershoot: The amount by which the response signal can undershoot the
initial value. This amount is specified as a percentage of the step's range. The
step's range is the difference between the final and initial values.
4 Specify the signals or systems to be bound.

You can apply this requirement to a model signal or to a linearization of your

Simulink model (requires Simulink Control Design software).

• Apply this requirement to a model signal:

In the Select Signals to Bound area, select a logged signal to which you will
apply the requirement.

If you have already selected a signal to log, as described in “Specify Signals to

Log” on page 3-12, it appears in the list. Select the corresponding check-box.

If you haven’t selected a signal to log:

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

5-32
 Frequency-Domain Design Requirements in Simulink

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

Click OK. A new variable, with the specified name, appears in the Data area
of the Response Optimization tool window.
• Apply this requirement to a linear system.

Linear systems are defined by snapshot times at which the model is linearized
and sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a
vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in

the list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open

the Create linearization I/O set dialog box.

5-33
5 Optimization-Based Control Design

For more information on using this dialog box, see “Create Linearization I/O
Sets” on page 3-77.

For more information on linearization, see “What Is Linearization?”.

5 Click OK.

A variable with the specified requirement name appears in the Data area of the tool.
A graphical display of the requirement also appears in the Response Optimization
tool window.

Alternatively, you can use the Check Step Response Characteristics block to
specify step response bounds for a signal.

See Also

“Design Optimization to Meet Step Response Requirements (GUI)”

Specify Custom Requirements

To specify custom requirements, such as minimizing system energy:

1 In the Response Optimization tool, select Custom Requirement in the New list. A
window opens where you specify the custom requirement.
2 Specify a requirement name in the Name box.
3 Specify the requirement type using the Type list.
4 Specify the name of the function that contains the custom requirement in the
Function box. The field must be specified as a function handle using @. The function

must be on the MATLAB path. Click to review or edit the function.

If the function does not exist, clicking opens a template MATLAB file.
Use this file to implement the custom requirement. The default function name is
myCustomRequirement.
5 (Optional) If you want to prevent the solver from considering specific parameter
combinations, select the Error if constraint is violated check box. Use this option
for parameter-only constraints.

During an optimization iteration, the solver evaluates requirements with this option
selected first.

5-34
 Frequency-Domain Design Requirements in Simulink

• If the constraint is violated, the solver skips evaluating any remaining

requirements and proceeds to the next iterate.
• If the constraint is not violated, the solver evaluates the remaining requirements
for the current iterate. If any of the remaining requirements bound signals or
systems, then the solver simulates the model .

For more information, see “Skip Model Simulation Based on Parameter Constraint
Violation (GUI)” on page 3-188.

Note: If you select this check box, then do not specify signals or systems to bound. If
you do specify signals or systems, then this check box is ignored.
6 (Optional) Specify the signal or system, or both, to be bound.

You can apply this requirement to model signals, or a linearization of your Simulink
model (requires Simulink Control Design software), or both.

Click Select Signals and Systems to Bound (Optional) to view the signal and
linearization I/O selection area.

• Apply this requirement to a model signal:

In the Signal area, select a logged signal to which you will apply the
requirement.

If you have already selected a signal to log, as described in “Specify Signals to

Log” on page 3-12, it appears in the list. Select the corresponding check box.

If you have not selected a signal to log:

a
Click . A window opens where you specify the logged signal.
b In the Simulink model window, click the signal to which you want to add a
requirement.

5-35
5 Optimization-Based Control Design

The window updates and displays the name of the block and the port number
where the selected signal is located.
c
Select the signal and click to add it to the signal set.
d In the Signal set box, enter a name for the selected signal set.

Click OK. A new variable, with the specified name, appears in the Data area
of the Response Optimization tool window.
• Apply this requirement to a linear system.

Linear systems are defined by snapshot times at which the model is linearized
and sets of linearization I/O points defining the system inputs and outputs.

a Specify the simulation time at which the model is linearized using the
Snapshot Times box. For multiple simulation snapshot times, specify a
vector.
b Select the linearization input/output set from the Linearization I/O area.

If you have already created a linearization input/output set, it will appear in

the list. Select the corresponding check box.

If you have not created a linearization input/output set, click to open

the Create linearization I/O set dialog box. For more information on using
this dialog box, see “Create Linearization I/O Sets” on page 3-77.

5-36
 Frequency-Domain Design Requirements in Simulink

For more information on linearization, see “What Is Linearization?”.

7 Click OK.

A new variable, with the specified name, appears in the Data area of the Response
Optimization tool. A graphical display of the requirement also appears in the
Response Optimization tool window.

See Also

• “Design Optimization to Meet a Custom Objective (GUI)” on page 3-108

• “Design Optimization to Meet Custom Signal Requirements (GUI)” on page 3-135

5-37
5 Optimization-Based Control Design

Time- and Frequency-Domain Requirements in SISO Design Tool

In this section...
“Root Locus Diagrams” on page 5-38
“Open-Loop and Prefilter Bode Diagrams” on page 5-40
“Open-Loop Nichols Plots” on page 5-40
“Step/Impulse Response Plots” on page 5-41

Root Locus Diagrams

• “Settling Time” on page 5-38
• “Percent Overshoot” on page 5-38
• “Damping Ratio” on page 5-39
• “Natural Frequency” on page 5-39
• “Region Constraint” on page 5-39

Settling Time

If you specify a settling time in the continuous-time root locus, a vertical line appears on
the root locus plot at the pole locations associated with the value provided (using a first-
order approximation). In the discrete-time case, the constraint is a curved line.

It is required that Re{ pole} < -4 .6 / Tsettling for continuous systems and
log( abs( pole)) / Tdiscrete < -4 .6 / Tsettling for discrete systems. This is an approximation of
the settling time based on second-order dominant systems.

Percent Overshoot

Specifying percent overshoot in the continuous-time root locus causes two rays, starting
at the root locus origin, to appear. These rays are the locus of poles associated with
the percent value (using a second-order approximation). In the discrete-time case, the
constraint appears as two curves originating at (1,0) and meeting on the real axis in the
left-hand plane.

The percent overshoot p.o constraint can be expressed in terms of the damping ratio, as
in this equation:

5-38
 Time- and Frequency-Domain Requirements in SISO Design Tool

2
p.o. = 100 e-pz / 1-z

where z is the damping ratio.

Damping Ratio

Specifying a damping ratio in the continuous-time root locus causes two rays, starting
at the root locus origin, to appear. These rays are the locus of poles associated with
the damping ratio. In the discrete-time case, the constraint appears as curved lines
originating at (1,0) and meeting on the real axis in the left-hand plane.

The damping ratio defines a requirement on - Re{ pole} / abs( pole) for continuous
systems and on

r = abs( pSys)
t = angle( pSys)

c = - log( r) / (log( r)) 2 + t2

for discrete systems.

Natural Frequency

If you specify a natural frequency, a semicircle centered around the root locus origin
appears. The radius equals the natural frequency.

The natural frequency defines a requirement on abs(pole) for continuous systems and on

r = abs( pSys)
t = angle( pSys)

c = (log( r)) 2 + t2 / Tsmodel

for discrete systems.

Region Constraint

Specifies an exclusion region in the complex plane, causing a line to appear between the
two specified points with a shaded region below the line. The poles must not lie in the
shaded region.

5-39
5 Optimization-Based Control Design

Open-Loop and Prefilter Bode Diagrams

• “Gain and Phase Margins” on page 5-40
• “Upper Gain Limit” on page 5-40
• “Lower Gain Limit” on page 5-40

Gain and Phase Margins

Specify a minimum phase and or a minimum gain margin.

Upper Gain Limit

You can specify an upper gain limit, which appears as a straight line on the Bode
magnitude curve. You must select frequency limits, the upper gain limit in decibels, and
the slope in dB/decade.

Lower Gain Limit

Specify the lower gain limit in the same fashion as the upper gain limit.

Open-Loop Nichols Plots

• “Phase Margin” on page 5-40
• “Gain Margin” on page 5-40
• “Closed-Loop Peak Gain” on page 5-41
• “Gain-Phase Requirement” on page 5-41

Phase Margin

Specify a minimum phase amount.

While displayed graphically at only one location around a multiple of -180 degrees, this
requirement applies to phase margin regardless of actual phase (i.e., it is interpreted for
all multiples of -180).

Gain Margin

Specify a minimum gain margin.

While displayed graphically at only one location around a multiple of -180 degrees, this
requirement applies to gain margin regardless of actual phase (i.e., it is interpreted for
all multiples of -180).

5-40
 Time- and Frequency-Domain Requirements in SISO Design Tool

Closed-Loop Peak Gain

Specify a peak closed-loop gain at a given location. The specified value can be positive
or negative in dB. The constraint follows the curves of the Nichols plot grid, so it is
recommended that you have the grid on when using this feature.

While displayed graphically at only one location around a multiple of -180 degrees, this
requirement applies to gain margin regardless of actual phase (i.e., it is interpreted for
all multiples of -180).

Gain-Phase Requirement

Specifies an exclusion region for the response on the Nichols plot. The response must not
pass through the shaded region.

This only applies to the region (phase and gain) drawn.

Step/Impulse Response Plots

• “Upper Time Response Bound” on page 5-41
• “Lower Time Response Bound” on page 5-41

Upper Time Response Bound

You can specify an upper time response bound.

Lower Time Response Bound

You can specify a lower time response bound.

Related Examples
• “How to Design Optimization-Based Controllers for LTI Systems” on page 5-43
• “Optimize LTI System to Meet Frequency-Domain Requirements” on page 5-44
• “Design Optimization-Based PID Controller for Linearized Simulink Model (GUI)”

5-41
5 Optimization-Based Control Design

Time-Domain Simulations in SISO Design Tool

When using a SISO Design Task, Simulink Design Optimization software automatically
sets the model's simulation start and stop time and you cannot directly change them.
By default, the simulation starts at 0 and continues until the SISO Design Task
determines that the dynamics of the model have settled out. In addition, when the design
requirements extend beyond this point, the simulation continues to the extent of the
design requirements. Although you cannot directly adjust the start or stop time of the
simulation, you can adjust the design requirements to extend further in time and thus
force the simulation to continue to a certain point.

5-42
 How to Design Optimization-Based Controllers for LTI Systems

How to Design Optimization-Based Controllers for LTI Systems

To design optimization-based linear controller for an LTI model:

1 Create and import a linear model into a SISO Design Task. You can create an LTI
model at the MATLAB command line, as described in “Creating an LTI Plant Model”
on page 5-45.
2 Create a SISO Design Task with design and analysis plots, as described in “Creating
Design and Analysis Plots” on page 5-46.

To learn more about SISO Design Tool, see “Using the SISO Design Task in
the Controls & Estimation Tools Manager” in the Control System Toolbox
documentation.
3 Under Automated Tuning select Optimization based tuning as the Design
Method and then click the Optimize Compensators button to create a Response
Optimization task within the Control and Estimation Tools Manager. See
“Creating a Response Optimization Task” on page 5-48 for more information.
4 Within the Response Optimization node, select the Compensators pane to select
and configure the compensator elements you want to tune during the response
optimization. See “Selecting Tunable Compensator Elements” on page 5-50 for
more information.

Note: Compensator elements or parameters cannot have uncertainty when used with
frequency-domain based response optimization.
5 Under Design requirements in the Response Optimization node, select
the design requirements you want the system to satisfy. See “Adding Design
Requirements” on page 5-51 for more information.
6 Click the Start Optimization button within the Response Optimization node.
The optimization progress results appear under Optimization. The Compensators
pane contains the new, optimized compensator element values. See “Optimizing the
System's Response” on page 5-59 for more information.

5-43
5 Optimization-Based Control Design

Optimize LTI System to Meet Frequency-Domain Requirements

In this section...
“Introduction” on page 5-44
“Design Requirements” on page 5-44
“Creating an LTI Plant Model” on page 5-45
“Creating Design and Analysis Plots” on page 5-46
“Creating a Response Optimization Task” on page 5-48
“Selecting Tunable Compensator Elements” on page 5-50
“Adding Design Requirements” on page 5-51
“Optimizing the System's Response” on page 5-59
“Creating and Displaying the Closed-Loop System” on page 5-62

Introduction
When you have Control System Toolbox software, you can place Simulink Design
Optimization design requirements or constraints on plots in the SISO Design Tool
graphical tuning editor and analysis plots that are part of a SISO Design Task. This
allows you to include design requirements for response optimization in the frequency-
domain in addition to the time-domain. This topic guides you through an example using
frequency-domain design requirements to optimize the response of a system in the SISO
Design Task.

You can specify frequency-domain design requirements to optimize response signals for
any model that you can design within a SISO Design Task:

• Command-line LTI models created with the Control System Toolbox commands
• Simulink models that have been linearized using Simulink Control Design software

Design Requirements
In this example, you use a linearized version of the following Simulink model.

5-44
 Optimize LTI System to Meet Frequency-Domain Requirements

You use optimization methods to design a compensator so that the closed loop system
meets the following design specifications when you excite the system with a unit step
input:

• A maximum 30-second settling time

• A maximum 10% overshoot
• A maximum 10-second rise time
• A limit of ±0.7 on the actuator signal

Creating an LTI Plant Model

In the srotut1 model, the plant model is composed of a gain, a limited integrator, a
transfer function, and a transport delay.

You want to design the compensator for the open loop transfer function of the linearized
srotut1 model. The linearized srotut1 plant model is composed of the gain, an
unlimited integrator, the transfer function, and a Padé approximation to the transport
delay.

To create an open loop transfer function based on the linearized srotut1 model, enter
the following commands:
w0 = 1;
zeta = 1;
Kint = 0.5;
Tdelay = 1;
[delayNum,delayDen] = pade(Tdelay,1);
integrator = tf(Kint,[1 0]);
transfer_fcn = tf(w0^2,[1 2*w0*zeta w0^2]);
delay_block = tf(delayNum,delayDen);
open_loopTF = integrator*transfer_fcn*delay_block;

If the plant model is an array of models, the controller is designed for a nominal model
only but you can analyze the control design for the remaining models in the array. For

5-45
5 Optimization-Based Control Design

more information, see “Control Design Analysis of Multiple Models” in the Control
System Toolbox documentation.

Tip You can directly linearize the Simulink model using Simulink Control Design
software.

Creating Design and Analysis Plots

This example uses a root locus diagram to design the response of the open loop transfer
function, open_loopTF. To create a SISO Design Task, containing a root-locus plot for
the open loop transfer function, use the following command:

controlSystemDesigner('rlocus',open_loopTF)

A SISO Design Task is created within the Control and Estimation Tools Manager, as
shown in the following figure.

5-46
 Optimize LTI System to Meet Frequency-Domain Requirements

The Control and Estimation Tools Manager is a graphical environment for managing
and performing tasks such as designing SISO systems. The SISO Design Task node
contains five panels that perform actions related to designing SISO control systems. For
more information, see “Using the SISO Design Task in the Controls & Estimation Tools
Manager” in Control System Toolbox documentation.

The Architecture pane, within the SISO Design Task node, lets you choose the
architecture for the control system you are designing. This example uses the default
architecture. In this system, the plant model, G, is the open loop transfer function
open_loopTF, the prefilter, F, and the sensor, H, are set to 1, and the compensator, C, is
the compensator that will be designed using response optimization methods.

The SISO Design Task also contains a root locus diagram in the SISO Design Tool
graphical tuning editor.

5-47
5 Optimization-Based Control Design

In addition to the root-locus diagram, it is helpful to visualize the response of the system
with a step response plot. To add a step response:

1 Select the Analysis Plots pane with the SISO Design Task node of the Control
and Estimation Tool Manager.
2 Select Step for the Plot Type of Plot 1.
3 Under Contents of Plots, select the check box in column 1 for the response Closed
Loop r to y.

A step response plot appears in an Linear System Analyzer. The plot shows the response
of the closed loop system from r (input to the prefilter, F) to y (output of the plant model,
G):

Creating a Response Optimization Task

There are several possible methods for designing a SISO system; this example uses an
automated approach involving response optimization methods. After creating the design

5-48
 Optimize LTI System to Meet Frequency-Domain Requirements

and analysis plots as discussed in “Creating Design and Analysis Plots” on page 5-46,
you are ready to start a response optimization task to design the compensator.

To create a response optimization task:

1 Select the Automated Tuning pane within the SISO Design Task node in the
Control and Estimation Tools Manager.
2 In the Automated Tuning pane, select Optimization based tuning as the
Design Method.
3 Click the Optimize Compensators button to create the Response Optimization
node under the SISO Design Task node in the tree browser in the left pane of the
Control and Estimation Tools Manager.

The Response Optimization node contains four panes as shown in the next figure.

5-49
5 Optimization-Based Control Design

With the exception of the first pane, each corresponds to a step in the response
optimization process:

• Overview: A schematic diagram of the response optimization process.

• Compensators: Select and configure the compensator elements that you want to
tune. See “Selecting Tunable Compensator Elements” on page 5-50.
• Design requirements: Select the design requirements that you want the system to
meet after tuning the compensator elements. See “Adding Design Requirements” on
page 5-51.
• Optimization: Configure optimization options and view the progress of the response
optimization. See “Optimizing the System's Response” on page 5-59.

Note: When optimizing responses in a SISO Design Task, you cannot add uncertainty to
parameters or compensator elements.

Selecting Tunable Compensator Elements

You can tune elements or parameters within compensators in your system so that
the response of the system meets the design requirements you specify. To specify the
compensator elements to tune:

1 Select the Compensators pane within the Response Optimization node.

2 Within the Compensators pane, select the check boxes in the Optimize column
that correspond to the compensator elements you want to tune.

In this example, to tune the Gain in the compensator C, select the check box next to
this element, as shown in the following figure.

5-50
 Optimize LTI System to Meet Frequency-Domain Requirements

Note: Compensator elements or parameters cannot have uncertainty when used with
frequency-domain based response optimization.

Adding Design Requirements

You can use both frequency-domain and time-domain design requirements to tune
parameters in a control system. The Design requirements pane within the Response
Optimization node of the Control and Estimation Tools Manager provides an interface
to create new design requirements and select those you want to use for a response
optimization.

5-51
5 Optimization-Based Control Design

This example uses the design specifications described in “Design Requirements” on page
5-44. The following sections each create a new design requirement to meet these
specifications:

• “Settling Time Design Requirement” on page 5-52

• “Overshoot Design Requirement” on page 5-53
• “Rise Time Design Requirement” on page 5-54
• “Actuator Limit Design Requirement” on page 5-56

After you add the design requirements, you can select a subset of requirements for
controller design, as described in “Selecting the Design Requirements to Use During
Response Optimization” on page 5-59.

Settling Time Design Requirement

The first design specification for this example is to have a settling time of 30 seconds or
less. This specification can be represented on a root-locus diagram as a constraint on the
real parts of the poles of the open loop system.

To add this design requirement:

1 Select the Design requirements pane within the Response Optimization node of
the Control and Estimation Tools Manager.
2 Click the Add new design requirement button. This opens the New Design
Requirement dialog box.

Within this dialog box you can specify new design requirements and add them to a
new or existing design or analysis plot.
3 Add a design requirement to the existing root-locus diagram:

a Select Pole/zero settling time from the Design requirement type

menu.
b Select Open-Loop L from the Requirement for response menu.
c Enter 30 seconds for the Settling time.
d Click OK.

A vertical line should appear on the root-locus diagram, as shown in the following
figure.

5-52
 Optimize LTI System to Meet Frequency-Domain Requirements

Overshoot Design Requirement

The second design specification for this example is to have a percentage overshoot of
10% or less. This specification is related to the damping ratio on a root-locus diagram.
In addition to adding a design requirement with the Add new design requirement
button, you can also right-click directly on the design or analysis plots to add the
requirement, as shown next.

To add this design requirement:

1 Right-click anywhere within the white space of the root-locus diagram in the SISO
Design Tool window. Select Design Requirements > New to open the New Design
Requirement dialog box.

5-53
5 Optimization-Based Control Design

2 Select Percent overshoot as the Design requirement type and enter 10 as the
Percent overshoot.
3 Click OK to add the design requirement to the root-locus diagram. The design
requirement appears as two lines radiating at an angle from the origin, as shown in
the following figure.

Rise Time Design Requirement

The third design specification for this example is to have a rise time of 10 seconds or less.
This specification is related to a lower limit on a Bode Magnitude diagram.

To add this design requirement:

5-54
 Optimize LTI System to Meet Frequency-Domain Requirements

1 Select the Graphical Tuning pane in the SISO Design Task node of the Control
and Estimation Tools Manager.
2 For Plot 2, set Plot Type to Open-Loop Bode.
3 Right-click anywhere within the white space of the open-loop bode diagram in the
SISO Design Tool window. Select Design Requirements > New to open the New
Design Requirement dialog box.
4 Create a design requirement to represent the rise time and add it to the new Bode
plot:

a Select Lower gain limit from the Design requirement type menu.
b Enter 1e-2 to 0.17 for the Frequency range.
c Enter 0 to 0 for the Magnitude range.
d Click OK.

A Bode diagram appears within the SISO Design Tool window. The magnitude plot
of the Bode diagram includes a horizontal line representing the design requirement,
as shown in the following figure.

5-55
5 Optimization-Based Control Design

Actuator Limit Design Requirement

The fourth design specification for this example is to limit the actuator signal to within
±0.7. To add this design requirement:

1 Select the Design requirements pane in the Response Optimization node of the
Control and Estimation Tools Manager.
2 Click the Add new design requirement button to open the New Design
Requirement dialog box.
3 Create a time-domain design requirement to represent the upper limit on the
actuator signal, and add it to a new step response plot in the Linear System
Analyzer:

5-56
 Optimize LTI System to Meet Frequency-Domain Requirements

a Select Step response upper amplitude limit from the Design

requirement type menu.
b Select Closed Loop r to u from the Requirement for response menu.
c Enter 0 to 10 for the Time range.
d Enter 0.7 to 0.7 for the Amplitude range.
e Click OK. A second step response plot for the closed loop response from r to
u appears in the Linear System Analyzer. The plot contains a horizontal line
representing the upper limit on the actuator signal.
f To extend this limit for all times (to t = ∞), right click on the black edge of the
design requirement, somewhere toward the right edge, and select Extend to
inf. The diagram should now appear as shown next.

To add the corresponding design requirement for the lower limit on the actuator signal:

5-57
5 Optimization-Based Control Design

1 Select the Design requirements pane in the Response Optimization node of the
Control and Estimation Tools Manager.
2 Click the Add new design requirement button to open the New Design
Requirement dialog box.
3 Create a time-domain design requirement to represent the lower limit on the
actuator signal, and add it to the step response plot in the Linear System Analyzer:

a Select Step response lower amplitude limit from the Design

requirement type menu.
b Select Closed Loop r to u from the Requirement for response menu.
c Enter 0 to 10 for the Time range.
d Enter -0.7 to -0.7 for the Amplitude range.
e Click OK. The step response plot now contains a second horizontal line
representing the lower limit on the actuator signal.
f To extend this limit for all times (to t = ∞), right-click in the yellow shaded area
and select Extend to inf. The diagram should now appear as shown in the
following figure.

5-58
 Optimize LTI System to Meet Frequency-Domain Requirements

Selecting the Design Requirements to Use During Response Optimization

The design requirements give constraints on the dynamics of the system and the
values of response signals. The table in the Design requirements tab lists all design
requirements in the design and analysis plots. Select the check boxes next to the design
requirements you want to use in the response optimization. This example uses all the
current design requirements.

Optimizing the System's Response

After selecting the compensator elements to tune and adding design requirements for the
response signals to satisfy, you are ready to being the response optimization.

The Optimization pane within the Response Optimization node of the Control and
Estimation Tools Manager displays the progress of the response optimization. The pane

5-59
5 Optimization-Based Control Design

also contains options to configure the types of progress information displayed during the
optimization and options to configure the optimization methods and algorithms.

To optimize the response of the system in this example, click the Start Optimization
button.

The Optimization pane displays the progress of the optimization, iteration by iteration,
as shown next. Termination messages from the optimization method and suggestions for
improving convergence also appear here.

The optimized signals in the design and analysis plots appear as follows:

5-60
Optimize LTI System to Meet Frequency-Domain Requirements

5-61
5 Optimization-Based Control Design

Creating and Displaying the Closed-Loop System

After designing a compensator by optimizing the response of the system, you can export
the compensator to the MATLAB workspace, and create a model of the full closed-loop
system.

1 Within the SISO Design Tool window, select File > Export to open the SISO Tool
Export dialog box.
2 Select the compensator you designed, Compensator C, and then click the Export
to Workspace button.

At the command line, enter the following command to create the closed-loop system, CL,
from the open-loop transfer function, open_loopTF, and the compensator, C:

CL=feedback(C*open_loopTF,1)

5-62
 Optimize LTI System to Meet Frequency-Domain Requirements

This returns the following model:

Zero/pole/gain from input to output "Output":
-0.19414 (s-2)
----------------------------------------------
(s^2 + 0.409s + 0.1136) (s^2 + 3.591s + 3.418)

To create a step response plot of the closed loop system, enter the following command:
step(CL);

This produces the following figure:

5-63
5 Optimization-Based Control Design

Designing Linear Controllers for Simulink Models

When you have Control System Toolbox and Simulink Control Design software, you can
perform frequency-domain optimization of Simulink models.

You can use Simulink Control Design software to configure SISO Design Tool with
compensators, inputs, outputs, and loops computed from a Simulink model. For
more information, see “Creating a SISO Design Task” in Simulink Control Design
documentation.

After you configure the SISO Design Tool, use Simulink Design Optimization software to
optimize the controller parameters of the linearized Simulink model. For an example of
optimization-based control design for a model linearized using Simulink Control Design
software, see “Design Optimization-Based PID Controller for Linearized Simulink Model
(GUI)”.

There is only one difference when tuning compensators derived from Simulink Control
Design software: The tuning of compensators from a Simulink model is done through
the masks of the Simulink blocks representing each compensator. When selecting
parameters to optimize, users can tune the compensator in the pole, zero, or gain format,
or in a format consistent with the Simulink block mask as shown in the following figure.
Changing the compensator format is not possible when optimizing pure SISO Tool
models (those not derived using Simulink Control Design software).

5-64
 Designing Linear Controllers for Simulink Models

Mask of a Simulink compensator block

5-65
5 Optimization-Based Control Design

Response optimization compensators pane

5-66
 6

Lookup Tables

• “What are Adaptive Lookup Tables?” on page 6-2

• “How to Estimate Lookup Table Values” on page 6-5
• “Estimate Constrained Values of a Lookup Table” on page 6-6
• “Estimate Lookup Table Values from Data” on page 6-23
• “Building Models Using Adaptive Lookup Table Blocks” on page 6-39
• “Selecting an Adaptation Method” on page 6-43
• “Model Engine Using n-D Adaptive Lookup Table” on page 6-45
• “Using Adaptive Lookup Tables in Real-Time Environment” on page 6-59
6 Lookup Tables

What are Adaptive Lookup Tables?

Lookup Tables
Lookup tables store numeric data in a multidimensional array format. In the simpler
two-dimensional case, lookup tables can be represented by matrices. Each element of a
matrix is a numerical quantity, which can be precisely located in terms of two indexing
variables. At higher dimensions, lookup tables can be represented by multidimensional
matrices, whose elements are described in terms of a corresponding number of indexing
variables.

Lookup tables provide a means to capture the dynamic behavior of a physical

(mechanical, electronic, software) system. The behavior of a system with M inputs and N
outputs can be approximately described by using N lookup tables, each consisting of an
array with M dimensions.

You usually generate lookup tables by experimentally collecting or artificially creating

the input and output data of a system. In general, you need as many indexing
parameters as the number of input variables. Each indexing parameter may take a value
within a predetermined set of data points, which are called the breakpoints. The set of all
breakpoints corresponding to an indexing variable is called a grid. Thus, a system with
M inputs is gridded by M sets of breakpoints. The software uses the breakpoints to locate
the array elements, where the output data of the system are stored. For a system with
N outputs, the software locates the N array elements and then stores the corresponding
data at these locations.

After you create a lookup table using the input and output measurements as described
previously, you can use the corresponding multidimensional array of values in
applications without having to remeasure the system outputs. In fact, you need only
the input data to locate the appropriate array elements in the lookup table because the
software reads the approximate system output from the data stored at these locations.
Therefore, a lookup table provides a suitable means of capturing the input-output
mapping of a static system in the form of numeric data stored at predetermined array
locations.

Adaptive Lookup Tables

Statically defined lookup tables, as described in “Lookup Tables” on page 6-2, cannot
accommodate the time-varying behavior (characteristics) of a physical plant. Static
lookup tables establish a permanent and static mapping of input-output behavior of a

6-2
 What are Adaptive Lookup Tables?

physical system. Conversely, the behavior of actual physical systems often varies with
time due to wear, environmental conditions, and manufacturing tolerances. With such
variations, the static mapping of input-output behavior of a plant described by the lookup
table may no longer provide a valid representation of the plant characteristics.

Adaptive lookup tables incorporate the time-varying behavior of physical plants into the
lookup table generation and maintenance process while providing all of the functionality
of a regular lookup table.

The adaptive lookup table receives the input and output measurements of a plant's
behavior, which are then used to dynamically create and update the content of the
underlying lookup table. In addition to requiring the input data to create the lookup
table, the adaptive lookup table also uses the output data of the plant to recalculate the
table values. For example, you can collect the output data of the plant by placing sensors
at appropriate locations in a physical system.

The software uses the input measurements to locate the array elements by comparing
these input values with the breakpoints defined for each indexing variable. Next, it
uses the output measurements to recalculate the numeric value stored at these array
locations. However, unlike a regular table, which only stores the array data before the
actual use of the lookup table, the adaptive table continuously improves the content of
the lookup table. This continuous improvement of the table data is referred to as the
adaptation process or learning process.

The adaptation process involves statistical and signal processing algorithms to recapture
the input-output behavior of the plant. The adaptive lookup table always tries to provide
a valid representation of the plant dynamics even though the plant behavior may be time
varying. The underlying signal processing algorithms are also robust against reasonable
measurement noise and they provide appropriate filtering of noisy output measurements.

See Also
Adaptive Lookup Table (1D Stair-Fit) | Adaptive Lookup Table (2D
Stair-Fit) | Adaptive Lookup Table (nD Stair-Fit)

Related Examples
• “Model Engine Using n-D Adaptive Lookup Table” on page 6-45

More About
• “About Lookup Table Blocks”

6-3
6 Lookup Tables

• “Building Models Using Adaptive Lookup Table Blocks” on page 6-39

6-4
 How to Estimate Lookup Table Values

How to Estimate Lookup Table Values

You can use lookup table Simulink blocks to approximate a system's behavior, as
described in “About Lookup Table Blocks” in the Simulink documentation. After you
build your system using lookup tables, you can use Simulink Design Optimization
software to estimate the table values from measured I/O data.

Estimating lookup table values is an example of estimating parameters which are

matrices or multi-dimensional arrays. The workflow for estimating parameters of a
lookup table consist of the following tasks:

1 Creating a Simulink model using lookup table blocks.

2 Importing the measured input and output (I/O) data from which you want to
estimate the table values.
3 Analyzing and preparing the I/O data for estimation.
4 Estimating the lookup table values.
5 Validating the estimated table values using a validation data set.

Related Examples
• “Estimate Lookup Table Values from Data” on page 6-23
• “Estimate Constrained Values of a Lookup Table” on page 6-6

6-5
6 Lookup Tables

Estimate Constrained Values of a Lookup Table

In this section...
“Objectives” on page 6-6
“About the Data” on page 6-6
“Lookup Table Output” on page 6-6
“Estimate the Monotonically Increasing Table Values Using Default Settings” on page
6-8
“Validate the Estimation Results” on page 6-17

Objectives
This example shows how to estimate constrained values of a lookup table. Apply
monotonically increasing constraints to the lookup table output values, and use the
Parameter Estimation tool to estimate the table values.

About the Data

In this example, use lookup_increasing.mat, which contains the measured I/O data
for estimating the lookup table values. The MAT-file includes the following variables:

• xdata1 — Input data consisting of 602 uniformly sampled data points in the range
[-5,5].
• ydata1 — Output data corresponding to the input data samples.
• time1 — Time vector.

Use the I/O data to estimate monotonically increasing output values of the lookup table
in the lookup_increasing Simulink model.

Lookup Table Output

1 Open the lookup table model by typing the following command at the MATLAB
prompt:
lookup_increasing

This command opens the Simulink model, and loads the estimation data in the
MATLAB workspace.

6-6
 Estimate Constrained Values of a Lookup Table

2 View the table output values by double-clicking the Lookup Table block.

The table contains 11 output values at breakpoints [-5:5], specified in the Function
Block Parameters dialog box. To learn more about how to specify the table values,
see “Enter Breakpoints and Table Data” in the Simulink documentation.

6-7
6 Lookup Tables

The Table data field shows that the table output values are the cumulative sum
of the values stored in variable ydelta. Thus, if yn are the 11 table output values,
ydelta is (y1, y2–y1, y3–y2, ..., y11–y10). The initial ydelta values are loaded from
lookup_increasing.mat.

The initial table output values are not monotonically increasing. To ensure
monotonically increasing table output values, the difference between adjacent
table output values should be positive. To do so, estimate ydelta in the
Parameter Estimation tool using the measured I/O estimation data, and constrain
ydelta(2:end) to be positive during estimation.

Estimate the Monotonically Increasing Table Values Using Default Settings

1 Open a parameter estimation session.

In the Simulink model, select Analysis > Parameter Estimation to open a session
with the name lookup_increasing in the Parameter Estimation tool.

6-8
 Estimate Constrained Values of a Lookup Table

2 Create an experiment and import the I/O data.

On the Parameter Estimation tab, click New Experiment. Type

[time1,ydata1] in Outputs and [time1,xdata1] in Inputs of the Edit
Experiment dialog box. Click OK. A new experiment with name Exp is created in
the Experiments area of the tool. Rename the experiment EstimationData by
right-clicking the default experiment name, Exp, and selecting Rename. For more
information, see “Import Data” on page 1-6.

6-9
6 Lookup Tables

3 Run an initial simulation to view the measured data, simulated model values, and
the initial table values by typing the following commands at the MATLAB prompt.

sim('lookup_increasing')
figure(1); plot(xdata1,ydata1,'m*',xout,yout,'b^')
hold on; plot(-5:5,cumsum(ydelta),'k','LineWidth',2)
xlabel('Input Data'); ylabel('Output Data');
legend('Measured data','Initial simulated data','Initial table output')

6-10
 Estimate Constrained Values of a Lookup Table

The initial table output values and simulated data do not match the measured data.
4 Select parameter for estimation.

On the Parameter Estimation tab, click Select Parameters. The Edit: Estimated
Parameters dialog box opens. In the Parameters Tuned for all Experiments
panel, click Select parameters to open the Select Model Variables dialog box.
Check the box next to ydelta, and click OK.

6-11
6 Lookup Tables

The ydelta values are selected for estimation by default in the Edit: Estimated
Parameters dialog box.

6-12
 Estimate Constrained Values of a Lookup Table

5 Apply a monotonically increasing constraint on the table output values. For more
details about the table, see “Lookup Table Output” on page 6-6.

In the Edit: Estimated Parameters dialog box, click the arrow next to the
ydelta values. In the expanded menu, set Minimum ydelta values to [-
Inf,zeros(1,10)]. Thus, while the first value in ydelta can by anything,
subsequent values which are the difference between adjacent table output values,
must be positive.

6-13
6 Lookup Tables

6 Select EstimationData experiment for estimation.

On the Parameter Estimation tab, click Select Experiment. By default,

EstimationData is selected for estimation. If not, check the box under the
Estimation column, and click OK.

7 Estimate the table values using default settings.

On the Parameter Estimation tab, click Estimate.

The Parameter Trajectory plot shows the change in the parameter values at each
iteration.

6-14
 Estimate Constrained Values of a Lookup Table

The Estimation Progress Report shows the iteration number, number of times the
objective function is evaluated, and value of the cost function at the end of each
iteration.

6-15
6 Lookup Tables

The estimated parameters are saved in a new variable, EstimatedParams,

in the Results area of the tool. To view the estimated parameters, right-click
EstimatedParams and select Open.

6-16
 Estimate Constrained Values of a Lookup Table

The estimated ydelta(2:end) values are positive. Thus, the output of the table,
which is the cumulative sum of the values stored in ydelta, is monotonically
increasing.

Validate the Estimation Results

After you estimate the table values, as described in “Estimate the Monotonically
Increasing Table Values Using Default Settings” on page 6-8, you use another
measured data set to validate and check that you have not over-fit the model. You can
plot and examine the following plots to validate the estimation results:

• Residuals plot
• Measured and simulated data plots

1 Create an experiment to use for validation and import the validation I/O data.

6-17
6 Lookup Tables

On the Parameter Estimation tab, click New Experiment. Type

[time2,ydata2] in Outputs and [time2,xdata2] in Inputs of the Edit
Experiment dialog box. Name the experiment ValidationData by right-clicking the
default experiment name, Exp, in the Experiments area of the tool, and selecting
Rename. For more information, see “Import Data” on page 1-6.
2 Select the experiment for validation.

Click Select Experiments on the Parameter Estimation tab. The

ValidationData experiment is selected for estimation by default. Clear
Estimation and select the box for Validation.

3 Select the results to validate.

On the Validation tab, click Select Results to Validate. Clear Use current
parameter values, select EstimatedParams, and click OK.

6-18
 Estimate Constrained Values of a Lookup Table

4 Select the plots to display during validation.

The Parameter Estimation tool displays the experiment plot after validation by
default. Add the residuals plot by selecting the corresponding box on the Validation
tab.

Click Validate.
5 Examine the plots.

a The experiment plot shows the data simulated using estimated parameters
agrees with the measured validation data.

6-19
6 Lookup Tables

b To view the residuals plot, click Residual plot: ValidationData tab.

6-20
 Estimate Constrained Values of a Lookup Table

The residuals, which show the difference between the simulated and measured
data, lie within 15% of the maximum output variation. This indicates a good
match between the measured and simulated table data values.
c Plot and examine the validation data, simulated data, and estimated table
values.

sim('lookup_increasing')
figure(2); plot(xdata2,ydata2,'m*',xout,yout,'b^')
hold on; plot(-5:5,cumsum(ydelta),'k','LineWidth', 2)
xlabel('Input Data'); ylabel('Output Data');
legend('Validation data','Simulated data','Table output values');

6-21
6 Lookup Tables

The table output values match both the measured data and the simulated table
values. The table output values cover the entire range of input values, which
indicates that all the lookup table values have been estimated.

Related Examples
• “Estimate Lookup Table Values from Data” on page 6-23

6-22
 Estimate Lookup Table Values from Data

Estimate Lookup Table Values from Data

In this section...
“Objectives” on page 6-23
“About the Data” on page 6-23
“Open a Parameter Estimation Session” on page 6-23
“Estimate the Table Values Using Default Settings” on page 6-25
“Validate the Estimation Results” on page 6-33

Objectives
This example shows how to estimate lookup table values from time-domain input-output
(I/O) data.

About the Data

In this example, use the I/O data in lookup_regular.mat to estimate the values of a
lookup table. The MAT-file includes the following variables:

• xdata1 — Consists of 63 uniformly-sampled input data points in the range [0,6.5]

• ydata1 — Consists of output data corresponding to the input data samples
• time1 — Time vector

Use the I/O data to estimate the lookup table values in the lookup_regular Simulink
model. The lookup table in the model contains ten values, which are stored in the
MATLAB variable table. The initial table values comprise a vector of 0s. To learn more
about how to model a system using lookup tables, see “Guidelines for Choosing a Lookup
Table” in the Simulink documentation.

Open a Parameter Estimation Session

To estimate the lookup table values, open a Parameter Estimation session.

1 Open the lookup table model by typing the following command at the MATLAB
prompt:

6-23
6 Lookup Tables

lookup_regular

This command opens the Simulink model, and loads the estimation data into the
MATLAB workspace.

2 In the Simulink model, select Analysis > Parameter Estimation to open a new
session with name lookup_regular in the Parameter Estimation tool.

6-24
 Estimate Lookup Table Values from Data

Estimate the Table Values Using Default Settings

Use the following steps to estimate the lookup table values.

1 Create a new experiment by clicking New Experiment on the Parameter

Estimation tab. Name it EstimationData. Then import the I/O data, xdata1
and ydata1, and the time vector, time1, into the experiment. To do this open the
experiment editor by right-clicking EstimationData and selecting Edit.... Type
[time1,ydata1] in the output dialog box and [time1,xdata1] in the input dialog
box in the experiment editor. For more information, see “Import Data” on page 1-6.
After you import the data the experiment looks as follows:

6-25
6 Lookup Tables

2 Run an initial simulation to view the I/O data, simulated output, and the initial
table values. To do so, type the following commands at the MATLAB prompt:

sim('lookup_regular')
figure(1); plot(xdata1,ydata1, 'm*', xout, yout,'b^')
hold on; plot(linspace(0,6.5,10), table, 'k', 'LineWidth', 2);
legend('Measured data','Initial simulation data','Initial table values');

6-26
 Estimate Lookup Table Values from Data

The x-axis and y-axis of the figure represent the input and output data, respectively.
The figure shows the following plots:

• Measured data — Represented by the magenta stars (*).

• Initial table values — Represented by the black line.
• Initial simulation data — Represented by the blue deltas (Δ).

You can see that the initial table values and simulated data do not match with the
measured data.
3 To select the table values to estimate, on the Parameter Estimation tab, click the
Select Parameters button to open the Edit:Estimated Parameters dialog. In
the Parameters Tuned for all Experiments panel, click Select parameters to
launch the Select Model Variables dialog. Check the box next to table, and click OK.

6-27
6 Lookup Tables

The Edit:Estimated Parameters window now looks as follows. The table values
are selected for estimation by default.

6-28
 Estimate Lookup Table Values from Data

4 On the Parameter Estimation tab, click Select Experiment. EstimationData

is selected for estimation by default. If not, check the box under the Estimation
column, and click OK.

6-29
6 Lookup Tables

5 To estimate the table values using the default settings, on the Parameter
Estimation tab, click Estimate to open the Parameter Trajectory plot and
Estimation Progress Report window. The Parameter Trajectory plot shows the
change in the parameter values at each iteration.

After the estimation converges, the Parameter Trajectory plot looks like this:

6-30
 Estimate Lookup Table Values from Data

The Estimation Progress Report shows the iteration number, number of times
the objective function is evaluated, and the value of the cost function at the end of
each iteration. After the estimation converges, the Estimation Progress Report
looks like this:

6-31
6 Lookup Tables

The estimated parameters are saved in EstimatedParams in the Results

section of the Data Browser pane on the left. To view the results, right-click on
EstimatedParams and then select Open. The report resembles the following.

6-32
 Estimate Lookup Table Values from Data

This report includes the estimated parameter values, the final value of the cost
function, and other optimization results. You can see that the optimization stopped
when the size of the gradient, 1.18e-14 was less than the criteria value, 1e-3.

Validate the Estimation Results

After you estimate the table values, as described in “Estimate the Table Values Using
Default Settings” on page 6-25, you must use another data set to validate that you
have not over-fitted the model. You can plot and examine the following plots to validate
the estimation results:

• Residuals plot
• Measured and simulated data plots

To validate the estimation results:

6-33
6 Lookup Tables

1 Create a new experiment to use for validation. Name it ValidationData. Import

the validation I/O data, xdata2 and ydata2, and time vector, time2 in the
ValidationData experiment. To do this open the experiment editor by right-
clicking ValidationData and selecting Edit.... Then, type [time2,ydata2] in the
output dialog box and [time2,xdata2] in the input dialog box in the experiment
editor. For more information, see “Import Data” on page 1-6.
2 To select the experiment for validation, on the Parameter Estimation tab, click
Select Experiments. The ValidationData experiment is selected for estimation
by default. Deselect the box for estimation and check it for validation.

3 To select results to use, on the Validation tab, click Select Results to Validate.
Deselect Use current parameter values and select EstimatedParams, and
click OK.

6-34
 Estimate Lookup Table Values from Data

4 The Parameter Estimation tool, by default, displays the experiment plot after
validation. Add the residuals plot by checking the corresponding box on the
Validation tab.

To start validation, on the Validation tab, click Validate.

5 Examine the plots

a Experiment plot

6-35
6 Lookup Tables

You can see that the data simulated using the estimated parameters agrees with
the measured validation data.
b Click Residual plot: ValidationData to open the residuals plot.

6-36
 Estimate Lookup Table Values from Data

The residuals, which show the difference between the simulated and measured
data, lie in the range [-0.15,0.15]— within 15% of the maximum output
variation. This indicates a good match between the measured and the simulated
table data values.
c Plot and examine the estimated table values against the validation data set and
the simulated table values by typing the following commands at the MATLAB
prompt.

sim('lookup_regular')
figure(2); plot(xdata2,ydata2, 'm*', xout, yout,'b^')
hold on; plot(linspace(0,6.5,10), table, 'k', 'LineWidth', 2)

6-37
6 Lookup Tables

The plot shows that the table values, displayed as the black line, match both
the validation data and the simulated table values. The table data values cover
the entire range of input values, which indicates that all the lookup table values
have been estimated.

Related Examples
• “Estimate Constrained Values of a Lookup Table” on page 6-6

6-38
 Building Models Using Adaptive Lookup Table Blocks

Building Models Using Adaptive Lookup Table Blocks

Simulink Design Optimization software provides blocks for modeling systems as adaptive
lookup tables. You can use the adaptive lookup table blocks to create lookup tables from
measured or simulated data. You build a model using the adaptive lookup table blocks,
and then simulate the model to adapt the lookup table values to the time-varying I/O
data. During simulation, the software uses the input data to locate the table values, and
then uses the output data to recalculate the table values. The updated table values are
stored in the adaptive lookup table block. For more information, see “What are Adaptive
Lookup Tables?” on page 6-2.

The Adaptive Lookup Table library has the following blocks:

• Adaptive Lookup Table (1D Stair-Fit) — One-dimensional adaptive lookup

table
• Adaptive Lookup Table (2D Stair-Fit) — Two-dimensional adaptive lookup
table
• Adaptive Lookup Table (nD Stair-Fit) — Multidimensional adaptive lookup
table

Note: Use the Adaptive Lookup Table (nD Stair-Fit) block to create lookup tables of
three or more dimensions.

To access the Adaptive Lookup Tables library:

1 Open the Simulink Library Browser.

At the MATLAB prompt, enter simulink.

2 Open the Simulink Design Optimization library.

In the Libraries pane, expand the Simulink Design Optimization node.

3 In the Simulink Design Optimization library tree, click Adaptive Lookup Tables.

6-39
6 Lookup Tables

By default, the Adaptive Lookup Table blocks have two inputs and outputs. You can
display additional inputs and outputs in a block by selecting the corresponding options in
the Function Block Parameters dialog box. To learn more about the options, see the block
reference pages.

Adaptive Lookup Table Block Showing Inputs and Outputs

The 2-D Adaptive Lookup Table block has the following inputs and outputs:

• u and y — Input and output data of the system being modeled, respectively.

6-40
 Building Models Using Adaptive Lookup Table Blocks

For example, to model an engine's efficiency as a function of engine rpm and manifold
pressure, specify u as the rpm, y as the pressure, and y as the efficiency signals.
• Tin — The initial table data.
• Enable — Signal to enable, disable, or reset the adaptation process.
• Lock — Signal to update only specified cells in the table.
• y — Value of the cell currently being adapted.
• N — Number of the cell currently being adapted.
• Tout — Values of the adapted table data.

For more information on how to use adaptive lookup tables, see “Model Engine Using n-D
Adaptive Lookup Table” on page 6-45.

A typical Simulink diagram using an adaptive lookup table block is shown in the next
figure.

Simulink Diagram Using an Adaptive Lookup Table

In this figure, the Experiment Data block imports a set of experimental data into
Simulink through MATLAB workspace variables. The initial table is specified in the
block mask parameters. When the simulation runs, the initial table begins to adapt to
new data inputs and the resulting table is copied to the block's output.

See Also
Adaptive Lookup Table (1D Stair-Fit) | Adaptive Lookup Table (2D
Stair-Fit) | Adaptive Lookup Table (nD Stair-Fit)

Related Examples
• “Model Engine Using n-D Adaptive Lookup Table” on page 6-45

6-41
6 Lookup Tables

More About
• “What are Adaptive Lookup Tables?” on page 6-2
• “Selecting an Adaptation Method” on page 6-43

6-42
 Selecting an Adaptation Method

Selecting an Adaptation Method

You specify the algorithm using the Adaptation Method drop-down list in the Function
Block Parameters dialog box of an adaptive lookup table block. This section discusses the
details of these algorithms.

Sample Mean
Sample mean provides the average value of n output data samples and is defined as:

n
$y( n) = 1
Â
n i =1
y( i)

where y(i) is the ith measurement collected within a particular cell. For each input
data u, the sample mean at the corresponding cell is updated using the output data
measurement, y. Instead of accumulating n samples of data for each cell, a recursive
relation is used to calculate the sample mean. The recursive expression is obtained by the
following equation:

È n -1 ˘ n - 1 È 1 n -1 ˘ 1 n -1 $
$y( n) = 1 Í 1
Â
n ÍÎ i =1
y( i) + y(n) ˙ =
˙˚
Í Â
n ÍÎ n - 1 i=1
y(i) ˙ + y( n) =
˙˚ n n
y( n - 1) + y(n)
n

where y(n) is the nth data sample.

Defining a priori estimation error as e( n) = y(n) - $y( n - 1) , the recursive relation can be
written as:

$y( n) = $y(n - 1) + 1 e( n)
n

where n ≥ 1 and the initial estimate $y(0) is arbitrary.

In this expression, only the number of samples, n, for each cell— rather than n data
samples—is stored in memory.

6-43
6 Lookup Tables

Sample Mean with Forgetting

The adaptation method “Sample Mean” on page 6-43 has an infinite memory. The
past data samples have the same weight as the final sample in calculating the sample
mean. Sample mean (with forgetting) uses an algorithm with a forgetting factor
or Adaptation gain that puts more weight on the more recent samples. This algorithm
provides robustness against initial response transients of the plant and an adjustable
speed of adaptation. Sample mean (with forgetting) is defined as:

n
1
$y( n) =
 l n-i y(i)
 l n- i i =1
n
i =1

1 Èn -1 ˘ s( n - 1)
n -i $y(n - 1) + 1 y( n)
= Â
Í l y(i) + y(n) ˙ =
Â
n
i =1
l n- i ÍÎ i =1 ˙˚ s(n) s( n)

k
where l Œ [ 0,1 ] is the Adaptation gain and s( k) =  l n -i .
i =1

Defining a priori estimation error as e( n) = y(n) - $y( n - 1) , where n ≥ 1 and the initial

estimate $y(0) is arbitrary, the recursive relation can be written as:

$y( n) = $y(n - 1) + 1 e(n) = $y( n - 1) + 1 - l e( n)

s( n) 1 - ln

A small value of λ results in faster adaptation. A value of 0 indicates short memory (last
data becomes the table value), and a value of 1 indicates long memory (average all data
received in a cell).

6-44
 Model Engine Using n-D Adaptive Lookup Table

Model Engine Using n-D Adaptive Lookup Table

In this section...
“Objectives” on page 6-45
“About the Data” on page 6-45
“Building a Model Using Adaptive Lookup Table Blocks” on page 6-46
“Adapting the Lookup Table Values Using Time-Varying I/O Data” on page 6-55

Objectives
In this example, you learn how to capture the time-varying behavior of an engine using
an n-D adaptive lookup table. You accomplish the following tasks using the Simulink
software:

• Configure an adaptive lookup table block to model your system.

• Simulate the model to update the lookup table values dynamically.
• Export the adapted lookup table values to the MATLAB workspace.
• Lock a specific cell in the table during adaptation.
• Disable the adaptation process and use the adaptive lookup table as a static lookup
table.

About the Data

In this example, you use the data in vedata.mat which contains the following variables
measured from an engine:

• X — 10 input breakpoints for intake manifold pressure in the range [10,100]

• Y — 36 input breakpoints for engine speed in the range [0,7000]
• Z — 10x36 matrix of table data for engine volumetric efficiency

To learn more about breakpoints and table data, see “Anatomy of a Lookup Table” in
the Simulink documentation.

The output volumetric efficiency of the engine is time varying, and a function of two
inputs—intake manifold pressure and engine speed. The data in the MAT-file is used to
generate the time-varying input and output (I/O) data for the engine.

6-45
6 Lookup Tables

Building a Model Using Adaptive Lookup Table Blocks

In this portion of the tutorial, you learn how to build a model of an engine using an
Adaptive Lookup Table block.

1 Open a preconfigured Simulink model by typing the model name at the MATLAB
prompt:

enginetable1_data

The Experimental Data subsystem in the Simulink model generates time-varying I/

O data during simulation.

This command also loads the variables X, Y and Z into the MATLAB workspace. To
learn more about this data, see “About the Data” on page 6-45.
2 Add an Adaptive Lookup Table block to the Simulink model.

a Open the Simulink Library Browser.

At the MATLAB prompt, enter simulink.

6-46
 Model Engine Using n-D Adaptive Lookup Table

b Open the Simulink Design Optimization library.

In the Libraries pane, expand the Simulink Design Optimization node.

c In the Simulink Design Optimization library tree, click Adaptive Lookup
Tables.

d Drag and drop the Adaptive Lookup Table (nD Stair-Fit) block from the
Adaptive Lookup Tables library to the Simulink model window.

3 Double-click the Adaptive Lookup Table (nD Stair-Fit) block to open the Function
Block Parameters: Adaptive Lookup Table (nD Stair-Fit) dialog box.

6-47
6 Lookup Tables

4 In the Function Block Parameters dialog box:

a Specify the following block parameters:

• Table breakpoints (cell array) — Enter {[X; 110], [Y; 7200]} to

specify the range of input breakpoints.
• Table data (initial) — Enter rand(10,36) to specify random numbers as
the initial table values for the volumetric efficiency.

6-48
 Model Engine Using n-D Adaptive Lookup Table

• Table numbering data — Enter reshape(1:360,10,36) to specify a

numbering scheme for the table cells.
b Verify that Sample mean (with forgetting) is selected in the Adaptation
method drop-down list.
c Enter 0.98 in the Adaptation gain (0 to 1) field to specify the forgetting factor
for the Sample mean (with forgetting) adaptation algorithm.

An adaptation gain close to 1 indicates high robustness of the lookup table

values to input noise. To learn more about the adaptation gain, see “Sample
Mean with Forgetting” on page 6-44 in “Selecting an Adaptation Method” on
page 6-43.
d Select the Make adapted table an output check box.

This action adds a new port named Tout to the Adaptive Lookup Table block.
You use this port to plot the table values as they are being adapted.
e Select the Add adaptation enable/disable/reset port check box.

This action adds a new port named Enable to the Adaptive Lookup Table block.
You use this port to enable or disable the adaptation process.
f Select the Add cell lock enable/disable port check box.

This action adds a new port named Lock to the Adaptive Lookup Table block.
You use this port to lock a cell during the adaptation process.
g Verify that Ignore is selected in the Action for out-of-range drop-down list.

This selection specifies that the software ignores any time-varying inputs
outside the range of input breakpoints during adaptation.

Tip To learn more, see Adaptive Lookup Table (nD Stair-Fit) block
reference page.

After you configure the parameters, the block parameters dialog box looks like
the following figure.

6-49
6 Lookup Tables

h Click OK to close the Function Block Parameters dialog box.

The Simulink model now looks similar to the following figure.

6-50
 Model Engine Using n-D Adaptive Lookup Table

5 Assign the input and output data to the engine model by connecting the U and Y
ports of the Experimental Data block to the u and y ports of the Adaptive Lookup
Table block, respectively.

Tip To learn how to connect blocks in the Simulink model window, see “Block and
Signal Line Shortcuts and Actions” in the Simulink documentation.
6 Design a logic using Simulink blocks to enable or disable the adaptation process.
Connect the logic to the Adaptive Lookup Table block, as shown in the following
figure.

This logic outputs an initial value of 1 which enables the adaptation process.
7 Design a logic to lock a cell during adaptation. Connect the logic to the Adaptive
Lookup Table block, as shown in the following figure.

6-51
6 Lookup Tables

8 In the Simulink Library Browser, select the Simulink > Sinks library, and drag
Display blocks to the model window. Connect the blocks, as shown in the following
figure.

During simulation, the Display blocks show the following:

• Display block — Shows the value of the current cell being adapted.
• Display1 block — Shows the number of the current cell being adapted.
9 Write a MATLAB function to plot the lookup table values as they adapt during
simulation.

Alternatively, type enginetable at the MATLAB prompt to open a preconfigured

Simulink model. The Efficiency Surface subsystem contains a function to plot
the lookup table values, as shown in the next figure.

6-52
 Model Engine Using n-D Adaptive Lookup Table

10 Connect a To Workspace block to export the adapted table values:

a In the Simulink Library Browser, select the Simulink > Sinks library, and drag
the To Workspace block to the model window.

To learn more about this block, see the To Workspace block reference page in the
Simulink documentation.
b Double-click the To Workspace block to open the Sink Block Parameters dialog
box, and type Tout in the Variable name field.

6-53
6 Lookup Tables

c Click OK.
d Connect the To Workspace block to the adaptive lookup table output signal
Tout, as shown in the next figure.

6-54
 Model Engine Using n-D Adaptive Lookup Table

You have now built the model for updating and viewing the adaptive lookup table values.
You must now simulate the model to start the adaptation, as described in “Adapting the
Lookup Table Values Using Time-Varying I/O Data” on page 6-55.

Adapting the Lookup Table Values Using Time-Varying I/O Data

In this portion of the tutorial, you learn how to update the lookup table values to adapt to
the time-varying input and output values.

You must have already built the Simulink model, as described “Building a Model Using
Adaptive Lookup Table Blocks” on page 6-46.

To perform the adaptation:

1 In the Simulink Editor, specify the simulation time as inf.

The simulation time of infinity specifies that the adaptation process continues as
long as the input and output values of the engine change.
2 In the Simulink Editor, select Simulation > Run to start the adaptation process.

A figure window opens that shows the volumetric efficiency of the engine as a
function of the intake manifold pressure and engine speed:

• The left plot shows the measured volumetric efficiency as a function of intake
manifold pressure and engine speed.
• The right plot shows the volumetric efficiency as it adapts with the time-varying
intake manifold pressure and engine speed.

6-55
6 Lookup Tables

During simulation, the lookup table values displayed on the right plot adapt to the
variations in the I/O data. The left and the right plots resemble each other after a
few seconds, as shown in the next figure.

6-56
 Model Engine Using n-D Adaptive Lookup Table

Tip During simulation, the Cell Number and Adaptive Table Outputs blocks in
the Simulink model display the cell number, and the adapted lookup table value in
the cell, respectively.
3 Pause the simulation by selecting Simulation > Pause.

This action also exports the adapted table values Tout to the MATLAB workspace.

Note: After you pause the simulation, the adapted table values are stored in the
Adaptive Lookup Table block.
4 Examine that the left and the right plots match. This resemblance indicates that the
table values have adapted to the time-varying I/O data.
5 Lock a table cell so that only one cell adapts. You may find this feature useful if a
portion of the data is highly erratic or otherwise difficult for the algorithm to handle.

a Select Simulation > Run to restart the simulation.

b Double-click the Lock block. This action toggles the switch and feeds the output
of the ON block to theLock input port of the Adaptive Lookup Table(nD
Stair-Fit) block.

You can view the number of the locked cell in the Cell Number block in the
Simulink model.
6 After the table values adapt to the time-varying I/O data, you can continue to use the
Adaptive Lookup Table block as a static lookup table:

a In the Simulink model window, double-click the Enable block. This action
toggles the switch, and disables the adaptation.
b Select Simulation > Run to restart the simulation, if it is not already running.

During simulation, the Adaptive Lookup Table block works like a static lookup
table, and continues to estimate the output values as the input values change.
You can see the current lookup table value in the Adaptive Table Outputs
block in the Simulink model window.

6-57
6 Lookup Tables

Note: After you disable the adaptation, the Adaptive Lookup Table block does not
update the stored table values, and the figure that displays the table values does not
update.

See Also
Adaptive Lookup Table (nD Stair-Fit)

More About
• “What are Adaptive Lookup Tables?” on page 6-2

6-58
 Using Adaptive Lookup Tables in Real-Time Environment

Using Adaptive Lookup Tables in Real-Time Environment

You can use experimental data from sensor measurements collected by running various
tests on a system in real time. The measured data is then sent to the adaptive table block
to generate a lookup table describing the relation between the system inputs and output.

You can also use the Adaptive Lookup Table block in a real-time environment, where
some time-varying properties of a system need to be captured. To do so, generate C code
using Simulink Coder™ code generation software that can then be run in Simulink Real-
Time™ or dSPACE® software. Because you can start, stop, or reset the adaptation if you
want, use logic to enable the adaptation of the table data only when it is desired. The
cell number output N, and the Enable and Lock inputs facilitate this process. Use the
Enable input to start and stop the adaptation and the Lock input to update only one of
the table cells. The Lock input combined with some logic using the cell number output N
provide the means for updating only the desired table cells during a simulation run.

See Also
Adaptive Lookup Table (1D Stair-Fit) | Adaptive Lookup Table (2D
Stair-Fit) | Adaptive Lookup Table (nD Stair-Fit)

Related Examples
• “Model Engine Using n-D Adaptive Lookup Table” on page 6-45

More About
• “What are Adaptive Lookup Tables?” on page 6-2

6-59