Distributed Operating Performance Assessment for Hot Strip Mill Process Based on Probabilistic Support Tensor Data Description with Feature Tensor

Abstract

In industrial processes, operating performance assessment is of great practical significance for guiding the production adjustment for operators. From the perspective of classification, operating performance assessment is considered as a multi-class classification problem. As a well-known one-class classifier, support vector data description (SVDD) are oriented to vector data and cannot deal with tensor data directly. Moreover, SVDD gives the target data set a spherically shaped description, which is a binary output. However, practical industrial data of different operating performance grade may have overlapping region, which is a knotty problem for classification. To handle above issues, a distributed operating performance assessment method based on probabilistic support tensor data description (PSTDD) is proposed in this work. First, the plant-wide process variables are selected and divided into several blocks. Then, a PSTDD model is developed in each block. Based on the assessment results of different blocks, a global assessment index is designed. If the process is running at non-optimal condition, the root cause are traced by variable contributions. Experimental results on a real hot strip mill process (HSMP) illustrate the effectiveness of the proposed method comparing to the traditional distributed SVDD.