Neural Network Based on Direct Inverse Control for Electro−Hydraulic Servo Drive

Abstract

This  paper  describes  the  use of  the  nonlinear  autoregressive with exogenous inputs neural network for the control of electro−hydraulic servo drive. The direct inverse controller was trained on a real object in an online way with the use of a programmable logic controller before starting work on a real object, and appropriate tests were performed in the MATLAB/Simulink environment in order to select the right structure of the neural network. Also, the paper includes various network learning algorithms:  gradient  descent,  resilient backpropagation, and  adaptive  moment  estimation, which belong to backpropagation algorithms.  In  order  to  compare  the suitability  of  the direct inverse controller,  the  controller  was implemented on a real electro−hudraulic test stand, and its performance was compared with the performance of a proportional−integral−derivative controller.