Research On Statistical Monitoring For Multimode Process With PCA Mixture Model

Process safety and product quality are two important issues in modern industry, which bring about process monitoring techniques. For the complex industrial processes, it is difficult to achieve the exact mathematical model and the expert knowledge of processes. Thus the application of process monitoring methods based on system theory or priori knowledge is limited. With the wide use of the distribution control system (DCS) and intelligent instruments in industrial processes, large amount of data are sampled and collected. The multivariate statistical process monitoring (MSPM) methods based on process data have developed significantly over the past decade, and become popular in process monitoring area. Based on the existing research works, this dissertation focuses on the multimode process monitoring, and the main contents are as follows:(1) Propose a novel multimode process monitoring approach based on PCA mixture model. This PCA technique is used to de-correlate and reduce the dimension of process variables, ensuring the covariance matrix nonsingular.(2) The expectation maximization (EM) algorithm, which is used to estimate the mixture model's parameters, needs the clustering number to be known as a priori. Hence, this thesis introduces and compares two improved EM methods including Figueiredo-Jain (F-J) algorithm and Bayesian Yin-Yang (BYY) algorithm.(3) The proposed method is applied to the monitoring of continuous stirred-tank heater (CSTH) process and the benchmark Tennessee Eastman (TE) process. By comparing with the traditional PCA-based method and GMM-based method, the validity and effectiveness of proposed PCA mixture approach is illustrated.