Quality prediction and control in rolling processes using logistic regression

Abstract

With the advancement of distributed sensing technologies, abundant data are generated in rolling processes. While these data contain rich information about the process and product, it is a challenging task to develop a systematic method to model the relationship between process and product quality variables for quality improvements. This paper addresses this challenge by using logistic regression in which the quality measure is binary. Efforts are made to select minimum number of process variables in the model, based on which product qualities can be adequately predicted. If the predicted quality is worse than a target value, active control is initiated by adjusting key process variables. Considering the constraints of quality target, control costs and control feasibility, selecting appropriate control actions is formulated as mathematical optimization problems. Solutions and sensitivity studies are provided. Case studies using the data from real rolling lines are reported to demonstrate the effectiveness of this method.