Data-driven sparse discovery of hysteresis models for piezoelectric actuators

This paper proposes a new approach to model hysteresis in piezoelectric actuators based on recent advances in sparsity promoting machine learning methods. While sparse regression has successfully modelled several phenomena in sci-ence and engineering, its performance in modelling the nonlinear hysteresis behaviour in piezoelectric materials is still unexplored. This study applies the sequential threshold least-squares (STLSQ) algorithm to discover the governing hysteresis expressed in the form of a dynamical system. It is found that a parsimonious model governing the dynamics is extracted from a large candidate library, which predicts the hysteresis efficiently with less than le-3 relative percent error, demonstrating the algorithm's efficiency.