Optimal Tuning of PD-Type Iterative Learning Control for DC Gear Motors Using Bayesian Neural Networks

Abstract

This paper presents an efficient data-driven method to obtain the optimal parameters of a proportional-derivative (PD)-type iterative learning control (ILC) system for direct current gear motors. The proposed method aims at improving the convergence speed, tracking performance, robustness and reliability of the control system under different working conditions. Bayesian neural networks (BNN) embed probabilistic inference in the learning process, in contrast to conventional neural networks, which produce only deterministic outputs. This enables the proposed controller to better cope with uncertainties, noise and variations in the system dynamics. First, an approximate mathematical model of the DC gear motor is derived from the electrical and mechanical features of the motor-drive system. The model is then used to generate a large data set under different operating conditions with variations in system parameters and control responses. Then, the BNN is trained with these data to estimate the optimal proportional and derivative learning gains accurately to minimise the settling time and overshoot in the iterative learning process. The experimental results show that the proposed BNN-based ILC (BNN-ILC) controller significantly outperforms the conventional proportional-integral (PI) controller.