Empirical Modeling of Control Valve Layer with Application to Model Predictive Control-Based Stiction Compensation

Abstract

Abstract: In this work, we develop an empirical modeling procedure for feedback loops comprised of sticky valves and linear controllers for the valves and incorporate the models in a model predictive control (MPC)-based stiction compensation strategy. The empirical models are developed from data on the measured values of the valve outlet flowrates and the set-points for these flowrates. They utilize standard empirical model structures but are defined in a piecewise fashion, with different branches identified for set-point changes that correspond to sticking of the valve and to sliding of the valve, and modifications to account for the effect of the linear controller on the response of the valve outlet flowrate to a set-point change. Through a chemical process example, it is shown that the use of the empirical models in the MPC-based stiction compensation strategy can decrease the computation time of the method without significantly jeopardizing the constraint satisfaction of the closed-loop process, but preventing the need for a valve layer model with parameters and/or details about the valve that are difficult to obtain.