Flexible iterative learning control using a library based interpolation scheme

While learning frameworks, such as iterative learning control (ILC), offer excellent performance improvements for systems that execute a task repetitively; algorithm constraints have often limited the effectiveness of this approach for applications with non-repetitive system dynamics or trajectories. Systems with slowly varying parameters often behave like time-invariant systems for long periods. However, at a certain point, the drift state will shift enough to impact system performance and render the learned control input invalid. For these systems, the learning process must be reinitiated, leading to production losses in terms of time, money, and part fidelity while the learning controller reoptimizes the input signal. This paper presents a novel learning control strategy that utilizes an interpolated input signal derived from a library of converged signals to reinitiate the learning process for new drift states. In this manner, the controller uses training from a few critical states to inform the optimized initial inputs that span the space of the drift state. The proposed ILC framework is presented here, along with simulation results that illustrate savings in the convergence rate without a loss of performance.