An inverse model control method based on signal filtering and BP neural network to improve performance of electric servo cylinder system

Abstract

The electric servo cylinder (ESC) serves as the actuator for the vibration tables and load simulators. However, nonlinear factors inherent to the ESC system, including friction and clearance, consistently constrain the performance of the system and introduce errors in experimental results. To improve the high frequency response performance of this system under the influence of nonlinear factors, a real-time inverse model control method is designed in this paper. Firstly, the least squares method (LSM) is introduced to adaptively merge the wavelet filtered signal with the original signal, and the discrete wavelet transform is utilized to zero high-frequency detail components to eliminate noise in order to increase signal accuracy. Besides, an adaptive variable step size inverse model control law based on BP neural networks (WLSM-VSS-BPLC) is developed due to the nonlinearity and uncertainty present in the ESC system model. A BP neural network is used by the controller to identify the inverse model in real-time. Meanwhile, an adaptive variable step size strategy is proposed to achieve rapid identification of the inverse model. In conclusion, theoretical analysis and experimental results demonstrate that the WLSM filter effectively reduces noise across diverse types of noise signals. And the WLSM-VSS-BPLC controller enhances both the tracking accuracy and cut-off frequency of the ESC system.