Quality-Related Fault Diagnosis Based on Improved PLS for Industrial Process

Abstract

As a common method, partial least squares (PLS) is so important in the quality-related process monitoring. Nevertheless, PLS separates variables insufficiently, thus it cannot provide accurate results in quality-related process monitoring. To make up for this deficiency, in this paper, an improved PLS is described in detail which promotes the performance of PLS based on the coefficient matrix between input and output measurements. According to the detection results of IPLS, the contributions plots of the square of T statistics is calculated to judge the faulty variables of the fault. Taking the Tennessee Eastman (TE) process as a description, it verifies the effectiveness of IPLS in quality-related fault detection and fault diagnosis.