Are there any example networks, explanatory documentation, training material, or sample models available for the Flownex ROM Builder and Machine Learning Module? Additionally, can the ROM Builder be used for transient simulations, or is it limited to steady-state applications?
Yes, several resources are available to help users understand and apply the Flownex ROM Builder.
The primary reference is Section 23 of the Flownex User Manual, which explains the ROM Builder workflow and functionality. In addition, a dedicated release video demonstrates the ROM Builder process and provides a practical overview of how reduced-order models are created and used. The example network used in the release video is also available and can be used as a reference for learning and testing the functionality.
The ROM Builder is designed to create reduced-order models using neural networks trained on data generated from Flownex sensitivity analyses. These models can significantly reduce simulation times while maintaining acceptable accuracy within the trained operating range.
Regarding transient simulations, the ROM Builder is only supported for steady-state solutions. Training a neural network directly on transient simulations is generally not practical because the solution depends not only on the current inputs but also on the previous state of the system and dynamic effects such as inertia. This means that multiple valid outputs may exist for the same input conditions, making reliable neural network training difficult.
However, if a system can be reasonably approximated as quasi-steady-state, where changes occur instantaneously and inertia effects are negligible, a ROM generated from steady-state data may still be used within transient analyses to approximate system behavior.