Feedforward learning-nonlinear processes and adaptation

A feedforward learning control (LC) scheme is reported in this paper. This scheme is applicable to a class of smoothly nonlinear (repetitive) control tasks. Rapid reductions in tracking error are demonstrated using a single-input, single-output (SISO), nonlinear model of the rapid thermal processing (RTP) wafer manufacturing process. This scheme preserves the simplicity of the basic LC scheme reported in Tao, Kosut, and Aral (1994) by using the (measurable) impulse response model. Since the “impulse response” of a nonlinear process varies as the process trajectory moves along, adaptation is used to adjust the impulse response model between task repeats. Analytical justifications for the proposed adaptation are provided using a nonlinear equation solving analogy. Extension to the multi-input and multi-output (MIMO) case is also included