A novel common and specific features extraction-based process monitoring approach with application to a hot rolling mill process

Abstract

In this paper, a new common and specific features extraction-based process monitoring method is proposed for multimode processes with common features. Based on the common basis vectors, the common features that reflect the common information among multimode data can be obtained. The specific features corresponding to the individual properties of each mode are likewise obtained using the specific basis vectors. Moreover, the two basis vectors can be updated using a migration method when the new mode data are available in the database. A Kullback–Leibler distance-based metric is developed to measure the changes occurred in both two features. A derivative contribution plot-based method is finally proposed to isolate the root-cause variables leading to abnormal changes. The whole proposed methods are applied to an actual hot rolling mill (HRM) process, where common settings for different steel products and specific characteristics for each steel product exist. It is shown that the proposed method can successfully extract common features in an HRM process, and can present better monitoring performance compared with existing methods.