Integrated Set-Point Learning for Iterative Learning Control

Iterative Learning Control (ILC) has risen to prominence in applications where a control operation is performed repeatedly. ILC capitalizes on the repetitive nature of the operation and aims at closely tracking the user defined set-point by exploiting information from preceding trials. This information is then used to update the control input for the upcoming one. However, in systems that employ ILC, the set-point value required to achieve the desired output is not always known. In this paper we propose Integrated Set-Point Learning on top of a Linear Quadratic Direct ILC(LQ-ILC) to determine the optimal set-point profile. This is done by iteratively updating the set-point profile using gradient based algorithms upon completion of an entire control sequence. The approach is demonstrated on two systems taken from different engineering domains. In the first example of the Constant Velocity Differential Drive Robot (CVDDR) the method optimizes the robot's set-point trajectory iteratively whilst also improving tracking over the course of runs. In the second example the method is implemented on the Cott-Batch Reactor (CBR) to achieve user desired end product quality. The inter run stability of the system is investigated numerically and simulation results obtained demonstrate the efficacy of the method.