Decentralized Iterative Learning Control for Distributed Parameter Systems

This article considers the design of a decentralized iterative learning control (ILC) scheme for spatiotemporal systems modeled by a partial differential equation of advection-diffusion type. In particular, the optimal tracking problem is formulated using a sensor-actuator network operating over an m-dimensional spatial domain. Then, a decentralized ILC design is developed to reduce complexity. The main idea is to decompose the whole system into several smaller interacting subsystems and build the control update rule for each local controller based on the respective subsystem measurements and optimal system response prediction from the other subsystems. This approach simplifies the control system and significantly reduces the learning effort while providing a control quality comparable to a fully centralized control scheme. A nontrivial simulation example on friction welding of multiple aluminum plates illustrates the effectiveness and performance of this design.