Novel methodologies for integrated identification and robust process control

Modern approaches for designing robust control systems require: selection of a nominal dynamic model of the process, characterization of the uncertainty between the plant, and the model, (in certain cases) a model reduction procedure, and finally an algorithm that uses plant uncertainty and user-desired performance specifications for design and analysis of robust controllers. In this dissertation, three problems which integrate these four steps for designing robust decentralized Proportional-Integral-Derivative (PID) and multivariable Model Predictive Controllers (MPC) are developed. The problems examined are: Integrated identification and PID control design methodology for Single-Input Single-Output (SISO) systems, Control-relevant identification of multivariable systems, and finally integrated multivariable identification and robust control using loop shaping ideas. In all these methodologies one begins with dynamic modeling from plant data and concludes with parameter settings for high performance PID controllers. The logical sequence of the steps involved in these methodologies are: plant-friendly input signal design and execution, identification of the nominal model plus the uncertainty bounds (plant-set), nominal performance bounds estimation, control-relevant parameter estimation or control-relevant model reduction, and PID or Model Predictive Control (MPC) controller design. In: the first step, plant-friendly input signals such as Pseudo Random Binary Sequence (PRBS) and Schroeder-phased are designed using a priori time constant information. In this research, systematic guidelines to design these signals are presented. Secondly, a nominal model along with uncertainty bounds on each element of the transfer function matrix are identified using three different identification procedures: Zhu's Asymptotic method, Bayard's Frequency Domain method and the Coprime Uncertainty Estimation method. The nominal model along with element-by-element bounds are then used in a Structured Singular Value framework to derive robust performance and stability bounds. These bounds are then used to choose loop shapes required for tuning PID controllers. Finally, the designed controllers are validated by closed-loop simulations as well as by applying the Structured Singular Value (μ) theorem. Practical applicability of these integrated methodologies are shown by applying them to meaningful industrial case studies such as the Shell Heavy Oil Fractionator problem, the Weishedel-McAvoy high-purity distillation column and a paper machine simulator developed by Honeywell-Measurex.