基于LSTM神经网络的压电执行器位移迟滞建模

Displacement Hysteresis Modeling of Piezoelectric Actuator Based on LSTM Neural Network

该文搭建迟滞测量实验平台,测量一种用于LED晶圆检测压电执行器的迟滞效应,设计了一种基于长短期记忆(LSTM)神经网络的压电迟滞模型,使用时间序列预测法对压电执行器位移迟滞效应建模。将该模型与传统的Prandtl-Ishlinskii(PI)模型进行对比。实验结果表明,神经网络模型具有较好、较广泛的迟滞建模效果,对于正弦波,位移预测精度保证小于2%;对于衰减正弦波,位移预测精度可保证小于3%。较高的模型预测精度为使用压电执行器进行LED晶圆检测提供了依据。

In this paper,a hysteresis measurement experiment platform is built to measure the hysteresis effect of a piezoelectric actuator used for LED wafer detection.A piezoelectric hysteresis model based on LSTM neural network is designed,and the hysteresis effect of piezoelectric actuator displacement is modeled by time series prediction method.Comparing the model with the traditional PI model,the experimental results show that the model has better and more extensive hysteresis modeling effect.For sine wave,the displacement prediction accuracy is guaranteed to be within 2%,and for attenuated sine wave,it is guaranteed to be within 3%,which can basically meet the precision requirements of wafer detection in industry.