Stochastic estimation and adaptive feed forward in a nonlinear process control application

The focus of this work is to investigate improvements in process control capability with the application of modern control techniques to a nonlinear process plant. The plant selected is the finishing mill section of the 84 inch Hot Strip Mill at the integrated steel works of LTV Steel, Cleveland. The integrated plant processes iron ore to steel coils primarily for use in the automobile industry, and as part of the overall process the finishing mill rolls hot steel bars into steel strip. The finishing mill hot rolling process has precise tolerance requirements on the steel strip thickness (or gauge), width, profile, flatness and temperature. There is a continuously increasing demand for tighter product tolerances and improved gauge control in the finishing mill, which is a motivation for this work.; The dynamic control of steel gauge in the finishing mill represents a nonlinear and time-varying process with stochastically varying inputs and process noise. The existing mill automation and control includes nonlinear process models to calculate initial set-points for the mill actuators as well as in-bar dynamic control based on linear control theory to achieve target gauge. In this work, to improve gauge control capability, stochastic estimates of process uncertainty were introduced in the control using an Extended Kalman Filter framework, which also allowed for the use of nonlinear process models in an adaptive feed forward control architecture, to augment the existing gauge control system. The stochastic models were obtained by fitting probability distributions to measured process data and incoming product data. In the adaptive feed forward control technique, Kalman filter based estimation of the state variables of a bar as it is rolled through a stand is used to improve the nonlinear model set-points of the following rolling stands of the bar. The technique was successfully implemented on the finishing mill, resulting in a significant improvement of gauge performance. The process improvements are statistically analyzed and validated at the 84 inch Hot Strip Mill for head-end and in-bar gauge control performance, using bars rolled before and after the use of adaptive feed forward to illustrate the improvements achieved with the technique.