Chatter Detection in Hot Strip Mill Process based on Modified Independent Component Analysis

Abstract

Multivariate statistical process monitoring (MSPM) have been applied to process monitoring for industrial processes. The conventional method for a statistical monitoring model is principal component analysis (PCA). However, this is not sufficient to extract meaningful information in non-Gaussian data, which is the property of the process data in many industrial processes. Alternatively, the modified independent component analysis (MICA) method can be used to give meaningful information up to higher order statistics, which improves some drawbacks of independent component analysis (ICA) method. In this paper, we propose a protocol to monitor a chatter phenomenon in a hot strip mill process (HSMP) based on modified independent component analysis (MICA). First, we develop the chatter index (CI) that represent the degree of a chatter numerically. The statistical monitoring model for a chatter detection is constructed by using the chatter-free data, which is classified by CI. From the chatter monitoring model, a chatter detection rate of 86.7% is achieved.