Case-study: Simulator-based control system

tuning and testing using an estimated process

model
Finn Haugen
Telemark University College

Abstract
A mathematical model of a process to be controlled can be used for off-line tuning and testing of
control systems for the process. The tuning and testing can be performed using analytical methods
or using simulations. The benefit of simulations in this respect is that it is straight-forward to
include non-linearities in the model. This presentation demonstrates the application of
commercially available functions for system identification and simulation, and control design.
Focus is on functions available in National Instrument's LabVIEW, but similar functions are
available in other tools, as MATLAB. The presentation shows a practical application, namely a
speed servo for a DC motor.

1 Introduction

1.1 Outline of the presentation

This presentation will demonstrate how to use LabVIEW toolkits for analysis, design, and
simulation of a speed control system for a DC motor. The control system is finally implemented
on a National Instruments Compact FieldPoint system for I/O and embedded control. The motor
and the Compact FieldPoint system are described briefly below.

The following LabVIEW tools are used:

• System Identification Toolkit [5]

• Simulation Module [2]
• Control Design Toolkit [3]

An HTML version of this document, including linked files, is available at http://techteach.no.

1.2 The physical process - a DC motor

In this presentation a DC motor is modelled, simulated and controlled:

Figure 1: DC motor
The motor is produced by Faulhaber. The control signal is in the range of ±10V, and the
tachometer voltage is in the range of approximately ±10V. A load inertia has been added to the
motor. The time constant of the motor including load and tachometer is approximately 0.3s. A
load torque can be applied to the motor by simply braking the motor (the load) by hand.

1.3 The Compact FieldPoint I/O and embedded control system

The Compact Fieldpoint system used in this application consists of the following modules:

• cFP-2000 controller module running LabVIEW RT 7.1

• cFP-AIO-610 analog I/O module

Figure 2 shows the Compact FieldPoint System, and Figure 3 shows how the FieldPoint system
appears in the Measurement and Automation Explorer (MAX) utility where I/O channels are
defined and tested before being used in a LabVIEW program.

Figure 2: Compact FieldPoint

Figure 3: Measurement and

Automation Explorer (MAX) showing
the Compact FieldPoint system

1.4 A procedure for analysis, design and implementation of a control

system

Figure 4 shows a procedure for analysis, design and implementation of a control system.
 Figure 4: Procedure for analysis, design and implementation of a control system

In the present application separate files implements the steps shown above. The logged data are
stored in a spreadsheet file. The estimated model is also stored in a file.

The figure below shows a block diagram of the process with controller and estimator, and blocks
containing functions for analysis and design of the control system.

Figure 5: Block diagram of the process with controller and estimator, and blocks containing
functions for analysis and design of the control system.

2 Black-box modelling of the process

2.1 Introduction

When a black-box model of a process is to be developed the process must be excited by a

sufficiently "rich" signal (we can of course not expect to derive a dynamics model from constant
signals). The excitation can be made in two ways, refer to the figure shown above):

• Closed loop excitation. This means that the process is under feedback control.
The excitation is via the setpoint.
• Open loop excitation. This means that the process is not under feedback control.
The excitation is via the control signal.

In this application open loop excitation is used.

2.2 Exciting the process and measuring the response

excite_and_logg.vi shown below saves time t, input signal u and measured response y on the
spreadsheet file logfile1. t, u, and y are columns in this file. The sampling time is h = 0.02s which
is used both for analog output (control signal) and analog input (measurement signal) throughout
this application.

Figure 6: Front panel of excite_and_logg.vi Figure 7: Block diagram of

excite_and_logg.vi

Here is the log file from one specific experiment: logfile1. The columns in the file are t, u, y.

2.3 Estimating a mathematical model from logged data

A mathematical model of the motor in the form of a discrete-time transfer function, Hp(z) is
estimated. This transfer function is actually calculated from an estimated discrete-time state space
model,

x(k+1) = Ax(k) + Bu(k)

y(k) = Cx(k) + Du(k)
using the Convert to Transfer Function Model of CDT (Control Design Toolkit) function of
the Control Design Toolkit.

The state-space model is estimated using the SI Estimate State Space Model function. This
function implements a Subspace-method which is an efficient and generally applicable method
for estimating black-box (canonical) state-space models. The function is based on the Subspace-
method reported in [4]. The SI Estimate State Space Model function is on the Parametric
Modeling palette of the System Identification function palette, see the figure below.

Figure 8: Functions palette of System Identification Toolkit

The program system_ident.vi shown below estimates a discrete-time transfer function model
compatible with the Control Design Toolkit, and saves the model on a file (for later use).
Important features of the program are as follows:

• The SI Estimate Orders of System Model function indicates the best (optimal) order
based on a calculation of the number of significant singular values of the system matrix
of the underlying state-space model used by the subspace estimation method. The user
can still freely select the order of the estimated order.
• The user can choose to apply linear detrending of the input and output data before the
data is used for system identification. The detrending is made by the SI Remove Trend
function.
• The quality of the estimated model is indicated by plotting together the measured system
output and the simulated model output which is generated using the SI Simulate Model
function.

Figure 9: Front panel of system_ident.vi Figure 10: Block diagram of

 system_ident.vi

2.4 Is the model accurate? Comparing process simulation and

measurements

In compare_process_simulation_and_measurements.vi shown below the estimated model is

further evaluated. The user excites both the simulated process and the real process with an
arbitrary input signal (adjusted by the user), and the simulated response and the real measurement
are compared. The simulation is based on the LabVIEW Simulation Module [2]. The functions
palette of the Simulation Module is shown below.

Figure 11: The functions palette of the Simulation Module

Since in general the linear model is valid only around an operating point, the model is excited by
the devation, du, from the operating point value, u0, of the input signal, and the simulated
response is thus the deviation, dy, from the operating point value, y0, of the output signal.

In other words:

• The input signal to the simulated process is du, while the total input signal to the physical
process is

u = u0 + du

where u0 is fixed and du is arbitrary, but small in amplitude.

• The response from the simulated process is dy_sim, while the response from the real
process is

y_meas = y0 + dy_meas

giving

dy_meas = y_meas - y0

The model is accurate if the difference between dy_meas and dy_sim is small!
 Figure 13: Block diagram of
compare_process_simulation_
Figure 12: Front panel of and_measurements.vi
compare_process_simulation_and_measurements.vi

3 Control system analysis and design

3.1 Analysis of control system using Control Design Toolkit

The function palette of the Control Design Toolkit [3] is shown below.

Figure 14: Function palette of the Control Design Toolkit

The figure below shows a block diagram of the control system, including a measurement low pass
filter (which is a second order Butterworth lowpass filter). The blocks contains discrete-time
transfer functions (i.e. z-transfer functions).

Figure 15: Block diagram of the control system

control_analysis_design.vi shown below analyses the control system with respect to the following
[1]:

• Bandwidth (loop transfer function crossover frequency)

• Gain Margin GM and Phase Margin PM
 • Simulated step response in the process output variable y (due to a unit step in the
setpoint)

In addition, PID settings are calculated from the Ziegler-Nichols' closed-loop method interpreted
in the frequency domain.

Figure 16: Front panel of

control_analysis_design.vi

Figure 17: Block diagram of

control_analysis_design.vi

3.2 Comparing simulated control system and real control system (target:
PC)

real_and_simulated_control_system.vi shown below simulates a PID control system for the

estimated process model.
 Figure 18: Front panel of
real_and_simulated_control_system.vi

Figure 19: Block diagram of

real_and_simulated_control_system.vi

4 Implementing and testing the control system on a real-time

target (Compact FieldPoint)
rt_control_system.vi shown below implements the control system to be downloaded and run on
the FieldPoint system.

Figure 20: Front panel of rt_control_system.vi Figure 21: Block diagram of

rt_control_system.vi
5 Discussion, conclusion, and future work
It has been demonstrated how functions for system identification, simulation and control analysis
and design can be applied for tuning and testing of a practical control system. These functions are
based on an estimated discrete-time transfer function of the process to be controlled. The transfer
function stems from a state-space model estimated using a subspace method. Functions available
in LabVIEW were demonstrated in this presentation, but similar functions are available in other
tools, as MATLAB.

The model order was selected by the user based on a visual comparison of the measured response
and simulated response. Based on this visual comparion, the model order was acually selected
different from the "optimal" order indicated by the Subspace-method. Although the visual
comparison may be a good basis for selecting model order in a partly manual application, it can,
of course, not be used as a validation method in a fully automated system. Future work will focus
on building a fully automated system for model estimation and control design. Alternative
validation methods and alternative estimation methods will then be considered, as the prediction
error method with validation based on e.g. the FPE (Final Prediction-Error croterion) or the AIC
(Akaike's Information Theoretic Criterion).

Finally, LabVIEW is a nice tool for developing application for system identification, simulation,
and control system analysis and design. This is due to its graphical programming environment,
easy physical I/O support, and several powerful toolkits recently added. At Telemark University
College we use LabVIEW increasingly in student labs to bring theory into practical applications.

References
[1] Haugen, F. 2005: Discrete-time signals and systems,
http://techteach.no/publications/discretetime_signals_systems/discrete.pdf

[2] Haugen, F. 2005 Introduction to LabVIEW Simulation Module,

http://techteach.no/publications/labview_simulation

[3] Haugen, F. 2005: Introduction to LabVIEW Control Design Toolkit,

http://techteach.no/publications/labview_control_design

[4] Overschee, P. V., and B. De Moor. 1993. N4SID: Subspace algorithms for the stochastic
identification problem. Automatica 29, no. 3:649–660.

[5] National Istruments. 2004. System Identification Toolkit User Manual