Physics-informed neural networks with data-driven in modeling and characterizing piezoelectric micro-bender

Abstract Despite the rapid development and widespread adoption of physics-informed neural networks (PINNs) in various engineering fields, their applications in microelectromechanical coupling systems (MEMS) remain relatively unexplored. In this study, we demonstrate a novel implementation of PINNs for modeling and characterizing a piezoelectric microactuator. By leveraging the beam-like structure, the governing equations for a multi-layered piezoelectric actuator are first derived and subsequently incorporated into the PINNs model to accurately predict the deformation of the piezoelectric actuator in response to a given voltage input. Furthermore, by integrating experimental deflection data obtained from Laser Doppler Vibrometer measurements into the neural network, we further demonstrate the potential of PINNs in identifying the piezoelectric material coefficient through inverse analysis. Our contribution in applying PINNs to models and characterizing piezoelectric actuators in MEMS serves as a promising starting point for the broader utilization of machine learning techniques in this field.