Optimized neural network-based modeling of dynamic hysteresis in piezoelectric actuators

Abstract

Piezoelectric actuators exhibit dynamic hysteresis phenomena between output displacement and input voltage during voltage control, which significantly impacts control precision. This paper proposes an optimized neural network model (ONN) that combines convolutional neural network with long short term memory network. Additionally, the input space is expanded by introducing rate-related components. The optimization of ONN model’s hyperparameters is achieved by improved sparrow search algorithm, incorporating the sine-cosine algorithm and the Cauchy mutation mechanism. Compared to traditional phenomenological models, the ONN model more accurately characterizes the amplitude-dependent and rate-dependent dynamic hysteresis of piezoelectric actuators, while also demonstrating a certain predictive capability. The model proposed in this paper is of paramount significance for enhancing the control precision of piezoelectric actuators and designing relevant controllers.