An experimental study for an iterative learning control scheme with adaptive filtering

This paper presents experimental investigations into a novel approach which circumvents a classic tradeoff between robustness and performance in Iterative Learning Control. The tradeoff is usually performed by designing an appropriate filtering mechanism to compensate for non-perfect resetting and noise. Time-Frequency analysis is used to examine the frequency content of error signals over the compact time support of the reference signal. This then allows the filter to effectively change its bandwidth as a function of time thereby allowing high frequency system dynamics to enter into the learning process at the appropriate time instants. At the same time, robustness is maintained by reducing the bandwidth of the filter when the system dynamics do not exhibit high frequency characteristics so as to attenuate the effect of noise on the learning stability. This adaptive Q-filter Iterative Learning Controller is needed for systems that have non-smooth nonlinearities and the advantages are demonstrated here in the simulations and experiments. A specially designed experimental study shows the fundamental limits that can affect its effectiveness.