A Comparison of Adaptive Controllers: Academic vs Industrial

Abstract

The performance of Foxboro's Expert Adaptive Controller Tuning (EXACT) controller and Turnbull Control System's (TCS) Auto-Tuning Controller were compared versus the Generalized Minimum Variance (GMV) controller of Clarke and Gawthrop [2]. The EXACT controller's pattern recognition approach to automatic tuning of PID parameters gave good control for a wide range of operating conditions on two real processes. However, it was sometimes slow to adapt because it adjusts PID parameters only once per set point or disturbance transient. The TCS controller did not automatically update the PID controller parameters and hence acted more as an advisor than a self-tuning controller. The GMV controller is adaptive-predictive, but not self-tuning in the sense that parameters in the 1/Q control law (P=R=1) are user specified and not adapted. It is shown that a self-tuning strategy analogous to that used by the EXACT's tuner can be combined within the GMV algorithm so that the P+I controller parameters in I/Q can be tuned based on measured closed loop performance.