Introduction to Optimization in Collimator

How to launch and use the Model Optimizer tool

Built in to the core of Collimator's simulation engine is the JAX library, which accelerates machine learning tasks through auto differentiation. This unlocks the potential to efficiently compute global or local optimaβ€”in other words, optimization.

The Model Optimizer tool is one of the tools we offer to optimize the parameters of your model (i.e., initial conditions). You can optimize in three ways:

  1. Optimize model parameters to minimize a cost signal

  2. Optimize model parameters to best track reference data

  3. Optimize PID parameters for one or more PID blocks

We also offer a number of algorithms for the above, including both global and local optimization methods. A common way to approach algorithm selectionβ€”if you're not sure where to startβ€”is to run an optimization pass with a global algorithm, then follow up with a local one if needed.

Launching the Model Optimizer

From the Model Editor toolbar, you can launch the Model Optimizer tool:

This will open the Model Optimizer dialog:

Once you have filled out the options as you wish, click the "Optimize" button to start the process. The simulations will be launched, and you can track their progress in the dialog as it goes. You can cancel the process at any time.

When the optimization is finished, you will see the optimal parameter values and a plot of the cost function across simulation runs.

To apply the optimal values of all the design parameters, click "Apply and close." To apply the optimal values only for certain parameters, click the arrow button next to that parameter. Any applied values will be reflected in the model properties panel on the right of the model editor screen.

Last updated