Discrete-Time Iterative Learning Controller Using Approximate Inverse Model

Abstract A discrete-time iterative learning controller which utilizes an error correction algorithm with a given order is discussed in this paper. We analyze the behavior of the tracking error sequence and show that large increase of the error in transient trial stages can occur even if the ultimate convergence of the error is guaranteed. A design using a simple performance index is proposed to avoid such increase of the error. This design method is in essence equivalent to construct an approximate inverse model of a plant. However, it is pointed out that the high order error correction algorithms designed on erroneous data deteriorate the performance. To mitigate this drawback, we propose a modified design on the assumption that the uncertainty in the impulse response sequence is represented by a known probability distribution.