Physics-informed learning and control of ionic polymer-metal composite-actuated flexible structure

This paper tackles the challenge of controlling ionic polymer–metal composite-actuated flexible structures (IPMC-FS), which are complex systems characterized by nonlinear, dissipative, and electromechanically coupled dynamics. The primary problem is how to accurately model and regulate these dynamics from data despite unknown or varying system parameters. Our proposed approach is motivated by the need for physically consistent models that preserve electromechanical coupling and remain computationally tractable for real-time, constraint-aware control. Existing black-box approaches lack interpretability and offer unreliable constraint handling, while nonlinear model predictive control (MPC), though powerful, can be computationally prohibitive in practice. We propose a physics-informed model learning and control framework, in which an extended Lagrangian is introduced to derive the nonlinear model via Euler–Lagrange constraints, eliminating the need for canonical coordinates. This model is embedded in a quasi-linear parameter-varying (qLPV) form to handle nonlinearities efficiently. A Physics-Informed Neural Network (PINN) is used for data-driven identification of parameters and prediction of dynamics when system parameters are unknown, enforcing Euler–Lagrange residuals to maintain physical plausibility. The learned qLPV model is integrated with a qLPV-based MPC to solve the regulation problem, reducing online computational overhead by transforming it into a sequence of quadratic programs. Simulation results demonstrate significant improvements in control precision and computational efficiency, confirming the effectiveness of the approach and its suitability for real-time, constraint-aware control of IPMC-FS.