Adaptive Controller with Neural Signal Predictor Applied for Two-Mass System

This article aims to present design, implementation and analysis of results prepared for adaptive control structure based on two neural networks. The first of them is main controller with reconfigurable parameters, the second is applied in training algorithm. The task of additional model is prediction of information about measured signal. Then obtained values are used in optimization process of weights. That construction of the controller can improve (accelerate) reaction of main neural networks against changes of state variables. Significant part of described work is focused on implementation of proposed model as speed controller of electrical drive. Firstly, presented results were obtained in simulations. Next, the whole control structure was applied in dSAPCE card in order to prepare laboratory experiment. Final results present high quality of control, even in presence of drive parameter changes, and also confirm initial assumptions about correct work of the predictor and better quality of control obtained using modified neural speed controller.