Workflow
The typical workflow has four steps.
- Preprocessing: Simulation results (or test data) that are intended to be used for training might require preprocessing (e.g., filtering). This is an optional step.
- Training: Given a certain dataset, inputs, outputs, and state variables (plus possible physical constraints) are defined and the ROM is trained with a chosen architecture and some specified training parameters.
- Evaluation: The ROM is trained, and its accuracy requires verification. Some plots and visualizations (hyperplanes, loss vs epochs, preliminary time simulation) can be leveraged in this step.
- Deployment: If the accuracy check is successful, the ROM is ready to be deployed in a system integration platform.