July 10, 2014

Tutorial - Optimizing Models With modeFRONTIER

Introduction
This tutorial will demonstrate how to run Hopsan simulations from modeFRONTIER1 , a multi-objective
optimization environment. This makes it possible to use powerful optimization algorithms and data
analysis tools on a Hopsan model. It is also possible to connect other programs to modeFRONTIER
simultaneously and run multi-disciplinary optimizations. This is, however, not covered by this tutorial.

Requirements
In this tutorial modeFRONTIER version 4.5.3 and Hopsan version 0.6.8 on a Windows operating
system are used. It should, however, work on Linux systems with some minor modifications. Some
basic knowledge of Hopsan and design optimization is also required. It is recommended to do the
"Getting Started", "Advanced Usage" and "Optimization" tutorials before this one.

Optimize using final values only

An optimization basically consists of three parts: A simulation model, one or more objective function(s),
and an optimization algorithm. In this case Hopsan will be responsible for the simulation, while
modeFRONTIER will take care of the algorithm. The evaluation of the objective function(s) can,
however, be placed in either one of the programs. We will demonstrate both methods.
Calculating objective function values requires access to the resulting data variables from the simulation. Letting Hopsan do the calculation can therefore be easier, since the amount of data that must
be send between the programs can be reduced. We will first show how to run an optimization by only
sending the final simulation values from Hopsan.
1. Preparations
For this example we will use the position servo example model. The objective is to optimize the
parameters of the PI-controller by analyzing a step response. First we must modify the model
so that it also calculates the objective function value.
Open the position servo example
Add an absolute value operaetor and an integrator
Connect them after the error signal and name the integrator "totalError"
Create a new empty folder
Save the model as "PositionServo.hmf" in the folder you created
It should now look similar to the picture below:
1 modeFRONTIER is the property of ESTECO SpA. The Hopsan developers are not affiliated with modeFRONTIER
or ESTECO SpA.

July 10, 2014

Hopsan and modeFRONTIER will communicate by using comma-separated files (.csv). Before
we create the modeFRONTIER project we will need to extract two files from Hopsan to use as
templates.
Open a command prompt in Windows
Go to the folder you created
Execute HopsanCLI with the following command (change the Hopsan installation path to
the one you have):
c :\" Program Files ( x86 ) "\ hopsan \ bin \ HopsanCLI -- hmf PositionServo . hmf
-- simulate hmf -- re su l ts Fi na l CS V resultsFinal . csv
-- p ar am et e rE xp or t parameters . csv

This will open the model, simulate it and generate .csv files for parameters and final values of
the data variables.
2. Build the modeFRONTIER project
Now it is time to build the project in modeFRONTIER.
Open modeFRONTIER
Create a new project and save it in the folder you created
Add the following process nodes:
Scheduler
DOS Batch Script
Logic End
Add the following nodes for input data:
2x Input Variable
Input File
Support File
Add the following nodes for output data:
Output File
Output Variable
Design Objective
Connect the nodes as shown in the picture below
2

July 10, 2014

3. Specify input variables

We will use two input variables, one for the proportional gain (Kp) and one for the integrating
gain (Ki) in the PI controller.
Double-click on one of the input variable nodes
Input Variable
Change "Name" to "Kp"
Set "Lower Bound" to 0.0 and "Upper Bound" to 0.01
Double-click on the second input variable node
Change name to "Ki", lower bound to 0.0 and upper bound to 0.005
4. Configure the input variable file
Input variables shall be written to the "parameters.csv" file by modeFRONTIER, which is in
turn loaded by Hopsan.
Double-click on the input file node
Input File
Set "Input File Node Name" to "parameters_csv"
Click on "Edit Input File"
Browse to the your folder and select "parameters.csv"
Select "Kp" in the list at the bottom
Find the row that begins with "PositionServo$GainP#k#Value"
Select the numerical value at the end of the row
Right-click and choose "Insert Variable"
Select "Ki" in the list at the bottom
Find the row that begins with "PositionServo$GainI#k#Value"
Select the numerical value at the end of the row
Right-click and choose "Insert Variable"
Click "Ok" twice to close the dialogs

July 10, 2014

5. Configure the Hopsan model file

We must also specify the model file used by Hopsan.
Double-click on the support file node
Support File
Set "Support File Node Name" to "PositionServo_hmf"
Click on "Add File" and select "PositionServo.hmf"
Click "Ok"
6. Specify the output variable
The output variable consist of the integrated error signal from the Hopsan model.
Double-click on the output variable node
Output Variable
Set "Name" to "totalError"
Click "Ok"
7. Configure the output file
The output value is read by modeFRONTIER from "resultsFinal.csv", which contains the
final values exported from Hopsan.
Double-click on the output file node
Output File
Set "Output File Node Name" to "resultsFinal_csv"
Click on "Open Output File"
Select "resultsFinal.csv"
Find the row that starts with "PositionServo$totalError#out#Value"
Select "totalError" at the bottom
Select the text "PositionServo$totalError#out#Value"
Right-click and choose "Select Relative"
Click "Ok" twice
8. Define the objective function
The objective function in modeFRONTIER will be to simply minimize the value obtained from
Hopsan.
Double-click on the design objective node
Design Objective
Change from "Maximize" to "Minimize"
Click "Ok"

July 10, 2014

9. Write a script for controlling HopsanCLI

Hopsan will be controlled through the command line interface (HopsanCLI.exe) from a batch
script in modeFRONTIER.
Double-click on the script node
DOS Batch Script
Set "Name" to "HopsanCLI"
Click on "Edit DOS Batch Script"
Enter the following command:
C :\" Program Files ( x86 ) "\ Hopsan \ bin \ HopsanCLI -- hmf PositionServo . hmf
-- simulate hmf -- re su l ts Fi na l CS V resultsFinal . csv
-- p ar am et e rI mp or t parameters . csv

This will tell HopsanCLI to load the specified model, load parameters from parameters.csv,
simulate the model and then write the final values of all data variables to resultsFinal.csv.
Click "Ok" twice
10. Choose initial distribution and algorithm
Finally, we must configure the optimization in modeFRONTIER. For this example we will use
a random initial distribution and the Simplex algorithm. It is of course possible to use other
distributions and algorithms as well.
Double-click on the DOE node
DOE
Select "Random" under "Space Fillers" to the left
Click on "Add DOE Sequence"
Click "Ok"
Double-click on the scheduler node
Scheduler
Select "Simplex" under "Heuristic Optimizers"
Click "Ok"
11. Run an optimization
Now everything is prepared for starting the optimization!
Make sure the model is saved
Go to the "Run Analysis" mode
Start the optimization by clicking on the green arrow to the top right of the workspace
Run/Stop

Optimizing using full variable export

Putting the objective function in Hopsan is not always possible (or desirable). Another method is to
export the data to modeFRONTIER instead, so that the calculation can be made there.
1. Save a copy of the model
Save the model from the previous section in the same folder but with a different name.

July 10, 2014

2. Generate output variable file

First we need to generate a new .csv file to use as a template for the data.
Open the command prompt in Windows and go to your folder
Start HopsanCLI with the following command:
c :\ Program Files ( x86 ) "\ hopsan \ bin \ HopsanCLI -- hmf PositionServo . hmf
-- simulate hmf -- re su l ts Fi na l CS V resultsFinal . csv

This will generate a file called "resultsFinal.csv", which we will use below.
3. Modify the HopsanCLI script
Now we must modify the script where modeFRONTIER calls HopsanCLI, so that the full result
file is updated after each simulation.
Double-click on the script node
DOS Batch Script
Click on "Edit DOS Batch Script"
Modify the command to the following:
C :\" Program Files ( x86 ) "\ Hopsan \ bin \ HopsanCLI -- hmf PositionServo . hmf
-- simulate hmf -- re sultsFul lCSV resultsFull . csv
-- p ar am et e rI mp or t parameters . csv

Click "Ok" twice

4. Modify the project
The project in modeFRONTIER must be modified, so that the objective function is calculated.

Remove the output file node (called "resultsFinal_csv")

Add an Output Template node and a Calculator node between the script and the exit nodes
Add a Transfer File node between the Script node and the Output Template node
Add two Output Vector nodes between the Output Template node and the Calculator node
Connect the calculator node with the existing "totalError" node

The model should now look like this:

July 10, 2014

5. Specify output variables

We will use two output variables from Hopsan, the reference piston position and the actual
position.
Double-click on one of the output vector nodes
Vector Output Variable
Set "Name" to "x_actual"
Set "Size" to 2048
Click "Ok"
Double-click on the second output vector node
Set "Name" to "x_reference"
Set "Size" to 2048
Click "Ok"
6. Configure the output varible file
The output variables must be obtained from the generated .csv file.
Double-click on the transfer file node
Transfer File
Set "Transfer File Node Name" to "resultsFull_csv"
Click on "Add File"
Browse to your folder and select "resultsFull.csv"
Click "Ok" twice
7. Configure the data mining
Now we must specify the data mining, to map each variable to a block in the .csv file.
Double-click on the output template node
Output Template
Set "Output Template Node Name" to "resultsFull_template"
Click on "Edit Output Template"
Browse to your folder and select "resultsFull.csv"
Click on "x_actual" at the bottom
Find the row that starts with "PositionServo$Position_Sensor#out#Value"
Select the text, right-click and choose "Add Rule For" "x_actual"
Select the text again, right-click and choose "Set Anchor" "x_actual"
Uncheck column 1, 2, 3 and 4 to the left
Click on "x_reference" at the bottom
Find the row that starts with "PositionServo$Step#out#Value"
Select the text, right-click and choose "Add Rule For" "x_reference"
Select the text again, right-click and choose "Set Anchor" "x_reference"
Uncheck column 1, 2, 3 and 4 to the left
Click "Ok" twice

July 10, 2014

8. Define the objective function calculation

Finally, we must also specify the equation for the objective function.
Double-click on the calculator node
Calculator
Set "Name" to "objective_function"
Click on "Edit Calculator Expression"
Enter the following expression:
t o t a l E r r o r = sum ( a b s ( s u b t r a c t ( x _ a c t u a l , x _ r e f e r e n c e ) ) )
Click "Ok" twice to close the dialog
9. Save the model and run an optimization
If everything was successful, the optimization should work and give reasonable results. This
method should, however, be less time-efficient, due to the large amount of data being exchanged
between the programs.