Abstract
This article discusses a method for controlling variation in industrial processes when the model parameters are estimated from data and subject to uncertainty. A static input/output relationship with multiple input variables and an integrated moving average disturbance model are assumed. Most robust control methods use deterministic measures of uncertainty and a control objective that focuses on worst-case performance. This work uses a probabilistic measure of uncertainty and a control objective that relates more closely to minimizing variation, where parameter estimation errors are treated simply as an additional source of variability. We show that this approach results in a higher probability of closed-loop stability than the standard minimum variance control and can substantially lessen the adverse impact of parameter uncertainty on closed-loop variance. Guidelines for designing and evaluating the experiment used to estimate the model parameters are also discussed.