Data-based feedforward controller reconstruction from iterative learning control algorithm

Different from previous work discussing how to improve the performance of the ILC algorithm by assigning a well-approximated plant inversion as the learning filter, a novel way to reconstruct a feedforward filter from the ILC algorithm is presented. Assuming the ILC algorithm has already been implemented for the controlled plant, a stable FIR feedforward filter can be extracted from the converged control inputs which are learned for tracking a smoother impulse. This proposed ILC-based feedforward filter (ILCFF) is model-less and only requires minimal digital signal processing compared to existing inversion methods. The reconstructed ILCFF can serve as a feedforward controller for tracking challenging trajectories, and may be put back as the learning filter for improving the convergence rate of the original ILC algorithm.