What Is the Reduced Order Modeler App?

Reduced Order Modeler app lets you interactively create AI-based reduced order models or ROMs of subsystems modeled in Simulink, including full-order high-fidelity third-party simulation models brought into Simulink as Functional Mockup Units. You can use ROMs for system-level desktop simulation, control design, virtual sensor modeling, hardware-in-the-loop testing and digital twin generation.

Here’s the typical workflow for creating a ROM. The Reduced Order Modeler app lets you execute these steps interactively and create ROMs without being an AI expert:

• The app lets you configure experiments by either replacing or perturbing Simulink model signals using built-in excitation signal types, or by setting block parameters to desired values or sampling them using different methods based on distributions such as Uniform and Normal. It also lets you specify which Simulink signals and block parameters to consider as ROM inputs and which signals to consider as ROM outputs. The app allows to select excitation signals that are different than ROM input signals. This provides flexibility to configure design of experiments to work for various scenarios, for example when the subsystem targeted for reduced order modeling is in a closed-loop control and shouldn’t be excited directly.
• Once experiments are set, you can visualize coverage of the design space, then proceed with running simulations with the full order model to collect training data. You can choose to run simulations in parallel to accelerate data collection and view output coverage and simulation results once they are complete.
• In the next step, you can train AI-based reduced order models such as LSTM, nonlinear ARX or neural state space model using the collected I/O data. You can interactively specify hyperparameters and evaluate them sequentially or in parallel with the Parallel Computing Toolbox. Post-training, you can compare and evaluate model performance using the accuracy metrics.
• In the last step, you can export and bring the trained ROM into Simulink for system-level simulation, control design, virtual sensor modeling, HIL testing, or digital twin generation. You can deploy the trained ROM to embedded hardware using automatic code generation or export it as an FMU with the Simulink Compiler for external use.

In summary, you can use the Reduced Order Modeler app to interactively create AI-based ROMs of subsystems modeled in Simulink, including full-order high-fidelity third-party simulation models brought into Simulink as Functional Mockup Units. Download the Reduced Order Modeler app to get started with creating AI-based ROMs.