Quality-related fault detection based on mutual information principal component analysis

Quality-related fault detection has received extensive attention in recent years. It requires an appropriate supervisory relationship between process variables and quality variables. While the traditional principal component analysis (PCA) doesn't consider the relationships between them. Thus we proposed the mutual information principal component analysis (MIPCA) to detect the quality-related faults. MIPCA fully integrates the advantages of mutual information (MI) and PCA. With MIPCA, process variables can be utilized to monitor the process under the supervision of quality variables and judge a fault is whether related to the quality or not. Finally, the feasibility and effectiveness of the MIPCA are verified in Tennessee Eastman Process (TEP).