Can a model from the machine learning app be turned into an FMU?

I'm starting to learn to use the machine learning app and wondering if it's possible to turn a model created by that app into a FMU for use in ADAMS. Anyone have experience with this?

Answer 1

[Hiro Yoshino]

I do not have any experiences but it sounds quite feasible.

Step1. train ML model

Step2. call the model from Simulink

https://jp.mathworks.com/help/stats/predict-class-labels-using-matlab-function-block.html

Step3. Simulink --> FMU

https://jp.mathworks.com/help/simulink/ug/export-model-as-tool-coupling-fmu.html

You should note that not all of the ML models are compatible with Simulink. Looks like SVM can be used in Simulink.

[Damon Delorenzis]

I haven't tried running a ML model in Simulink yet. I'll give that a try. The best model for me so far seems to the an Optimizable GPR.

[sam l]

Hello I am working on similar project to create FMU from ML code.

But since the Mdl I created was Neural network - NARX and LSTM . How do I use above to export as FMU

[Hiro Yoshino]

@sam l - May I know your MATLAB version?

As the error message said, the saved model is incompatible with the saveLearnerForCoder function.

The object you created is "network object" from shallow network. (https://www.mathworks.com/help/deeplearning/ref/network.train.html)

The simulink block that takes a LSTM neural network model is Stateful predict block. Unfortunately this does not take the one you created. You should trian LSTM models as seriesnetwork, dagnetwork, dlnetwork.