Fuzzy Adaptation and Matrix Implementation of Smith Predictor for Dynamic Matrix Control in Continuous Processes With Large Dead Time

Abstract

Industrial processes with large and varying dead time are difficult to control and sustain a good controller performance over a wide range of process conditions. Since Dynamic Matrix Control is based on sampled-data model identification, and online inverse calculations are avoided, adaptation of the original model is not performed in commercial solutions. This paper presents an on-line adaptive form of the DMC control based on the Smith Predictor compensation strategy for processes with large dead time that does not require recalculating the inverse of the process matrix. The matrix form of the Smith Predictor was modeled in software and implemented in a lab scale continuously stirred tank reactor, where actual process noise was present. The adaptive algorithm included a Mamdani-type fuzzy inference system to modify the size of the dead time matrix. Implementation results demonstrated the algorithm ability to compensate for large dead time emulating the response of the non-delayed process. At the same time, the controller is able to adjust the size of the delay based on process response.