A unified approach to iterative learning control using neural network and integral control with anti-windup

The performance of control systems with repetitive motion can be improved by employing iterative learning controllers. Integral learning is the simplest algorithm and anti-windup is often indispensable in helping the controller to recover from the dilemma of saturation and to improve the transient performance. Neural network learning control, with the advantage of identifying system inverse dynamics, can quickly reduce output error in the first few iterations; however, the error converges very slowly thereafter. A unified hybrid controller employing a neural network and integral learning is proposed to pursue fast convergence of tracking error and good transient performance.