An integrated identification and control design methodology for multivariable process system applications

Abstract

We present a way to take advantage of the favorable asymptotic properties of ARX estimators to develop an integrated methodology for identification and controller design for multivariable process plants. This method relies on well-established numerical tools and builds on an engineer's existing process and statistical intuition. Specifically, the ARX estimate serves as a suitable intermediate model for the design and analysis of MIMO process control systems. Guidelines for the design of pseudo-random binary sequence signals that take advantage of the engineer's prior knowledge of the process time constants are presented. Control-relevant model reduction is performed on elements of the ARX model to obtain low-order models conforming to the IMC-PID tuning rules. A simple analysis technique is used to assess stability of the decentralized and decoupled strategies. These techniques and full multivariable control are applied to the Shell heavy oil fractionator problem and the Weischedel-McAvoy distillation column model, respectively.