[Demo] Explainable MPC Control

One often overlooked topic in control systems engineering is explainability. Control engineers are typically trained, through university or research, to design systems from the ground up, starting with physics-based modeling, model identification, stability analysis, control design, and then iterating with performance evaluation. This process provides engineers insight into the system, enabling intuitive for parameter calibration.

However, in industrial practice, the situation is often quite different… Control systems are frequently inherited, developed by previous colleagues, provided as black-box modules by external suppliers, or transferred from partner or other department with limited documentation or source code. In such cases, calibrating control parameters becomes a big challenge, particularly with complex systems involving a large number of parameters. It is hard to understand the operational boundary, performance and how to tuning, calibrate parameters in each specific condition or scenario.

Recently together with our research engineer Jean Pierre Allamaa and Prof. Panagiotis (Panos) Patrinos from KU Leuven, we have developed a method to gain deeper insights of control explainability. This approach combines efficient approximate MPC, physics-informed AI, and explainable AI techniques results in a quite elegant analysis.

Please check the paper draft here: https://lnkd.in/euESZ4GY

The results is demonstrated in an interesting F1 racing use case, using Siemens Simcenter Amesim for high-fidelity vehicle dynamics and Simcenter Prescan for a realistic traffic simulation of the F1 Abu Dhabi racing track.

This research is part of the EU Marie Skłodowska-Curie project ELO-X, that Siemens is pleased to be involved: https://elo-x.eu/