Maxwell & optiSLang cosimulation optimization

For specific operation details, you can follow the ANSYS MAXWELL Base Camp official account and look at the tutorial cases below. Maxwell & optiSLang cosimulation optimization.

1. Maxwell model parameterization

I used an RMxprt example that comes with the Maxwell software to generate a Maxwell2D model with one click. I randomly selected three variables and made parameterization.

2. Import workbench

First, you have to set it up through a file, otherwise you won’t find maxwell in the workbench.
The quantities you want to optimize are cogging torque, torque ripple and average torque. Import two projects into the workbench. Double-click Geometry respectively, and click DesignXplorersetup under Optimetrics (it will only appear if you click it from the workbench. If you make some changes to the model at this time, it seems to be equivalent to local variables in programming and only works in this project) to set it up. After setting it up, you will have parameters.

3. Sensitivity analysis

Add the sensitivity analysis module. When installing optislang, there will be an option to install the workbench plug-in. In the figure below, Criteria can be understood as the objective function. Since there is no optimization involved here, it is ignored.

• DOE: Designs of Experiments, a scan performed for global sensitivity analysis.
• MOP: Metamodel of Optimal Prognosis, which removes unimportant variables from the model and improves the prediction quality of the model.

The results are as follows: (The results are different when you view them just after the simulation and when you click on the pop-up result graph at the end of the optimization. Analyze them separately)

The project file name of this picture is called Sensitivity. It was obtained just after the sensitivity analysis. The meanings of the following pictures are introduced separately (based on the guesswork in the help file)

1. Upper left corner: correlation matrix, green is low correlation, red and blue are high correlation (positive and negative), the correlation coefficient of the pole arc coefficient to the average torque is 1, because other factors have no effect on the torque
2. Upper right corner: Histogram, the horizontal axis is the input/output value, the vertical axis is the PDF probability density function. Click the arrow below to see the average value, maximum value, etc. I guess it is the distribution diagram of these 100 sampling points, but I don’t know what it is used for.
3. Lower left corner: correlation coefficient, the correlation coefficient of a certain input quantity to three output quantities. Clicking different blocks of the matrix above will result in different graphs.
4. Lower right corner: Anthill diagram, which is a scatter plot, and these points are the sampling points.

The project file name is Sensitivity_MOP. After the optimization is completed, click on the sensitivity analysis result to get the following meanings of the following figures:

1. The figure in the upper left corner is a 3D response surface diagram. The curved surface is the response surface, and the black dots are sampling points. The quantities on the XYZ axes can be changed, and XY can also belong to the same quantity.
2. The figure in the upper right corner is the residual graph, which shows the error between the model predicted value and the actual value.
3. The Cop in the lower left corner is the Coefficient of Prognosis, which is the prediction coefficient of the model. The higher the value, the more accurate the model is. This figure shows the contribution of these two quantities to the Cop coefficient.
4. The Cop matrix is ​​in the lower right corner, showing the cop of each block. The vertical axis is the slot torque and torque, and the horizontal axis is the pole arc coefficient, slot width and total. Because the calculated rotor inner diameter has no effect on these optimization quantities (Cop=0?), it is not included. When you click on each matrix block, the other three graphs will also change.

Supplement : In the above sensitivity analysis, the Sampling method was selected when setting up. If the cop is relatively low, you can drag another sensitivity analysis module and select AMOP (Adaptive Optimal Prognosis Metamodel).

4. Optimization

Theoretically, you can skip the third step and optimize directly, but this process requires running a lot of data points, and you don’t even know how many there are. According to the tutorial, there are more than 700 data points. If you do a sensitivity analysis (100 points) first, and then do the optimization (1 minute), the speed will be much faster. Some people also call it response surface-based optimization.

When setting, the average torque objective function is MAX. At this time, the value displayed on the right is negative. I guess it is because the software takes the minimum value as the benchmark, so the maximum value can only be negative. Pay attention to the difference in the following result graphs. It is positive when looking at the actual result graph, and it is negative when looking at the target value.
Next, take a look at the result graph analysis:

1. Upper left corner: Pareto chart, divided into 2D and 3D. There are many black points and red lines in the chart, but no red points. This chart is dense and looks like a surface. Click the red part, and the design on the right will become the best design. The optimal solution is not unique and cannot be found in 2D. You have to find it in the 3D chart. If you set Limit in the previous optimization settings, there will be other information here.
2. Upper right corner: Design parameter diagram, parameters of a certain point at the moment (not considering the rotor inner diameter)
3. Lower left corner: Response data graph. The percentage of the horizontal axis is the percentage of the design value in the range defined when the optimization was originally set. The vertical axis is the response value of the torque and cogging torque.
4. Lower right corner: standard value (translation is like this), which is actually the optimization target value. The average torque is still negative, and everything else is the same as the left picture.