Hysteresis modeling of piezoelectric actuators based on neural network considering load and environmental stiffness

Abstract Piezoelectric actuators (PA) are widely used in the field of aircraft wind tunnel testing due to their high displacement resolution and fast response speed. However, the displacement exhibits hysteresis and the characteristics shift under time-varying loads, making it challenging to enhance precision, ultimately leading to distorted test results. To solve this problem, a novel modeling method of PA hysteresis based on NARXNN (Nonlinear Auto-Regressive model with Exogenous Inputs Neural Network) considering load and environmental stiffness is proposed in this paper. In this method, hysteresis modeling of PA based on NARXNN is established considering time-varying load and environmental stiffness as the extended inputs. The hyperparameters of the neural network were optimized based on the particle swarm optimization (PSO). Experimental results show that the RMSE of the hysteresis model considering load and environmental stiffness is reduced by 74.86%-99.69% compared with the classic PI (Prandtl-Ishlinskii) model, which verifies the accuracy and adaptability of the NARXNN modeling method.