Researches On Fault Diagnosis And Prognosis Of Process Industry Based On MCUSUM-ICA-MSPCA

The fault diagnosis and prediction of production process has been one major concern of process industry. The purpose of fault diagnosis is to find abnormal situation and causes, solve the problen, make the production back to normal operation. This is done by real-time monitoring the operational status of the production process.There are many fault diagnosis ways, which one to choose need take into account the different objects and the complexity of monitoring and control efficiency. With the increasing complexity of industrial production systems, Monitoring system algorithm using has emerged a lot of problems. Since the multivariate statistical methods were introduced to fault diagnosis, it shows strong vitality and adaptability. The multivariate statistical methods have been widely development in recent years.Independent component analysis is a new introduced multivariate statistical method in recent years. Based on independent component analysis and principal component analysis, by analyzing the status of existing research methods, this paper made the multi-scale improvement of traditional PCA method, uses differences in principal component space D as a statistical control index, improve the analysis and diagnostic capabilities of algorithm highly. Based on the detailed research of independent component analysis in the theory and application, this paper made Tennessee - Eastman process as simulation platform, in-depth studied the ICA method and its application. For the small faults in process industry, this paper introduced the multi-variable accumulation and control chart methods, this method played a good monitoring result in the case of early prediction of small faults. Based on ICA-MSPCA information extraction step in the proposed strategy, a new synthesis process monitoring algorithm based on MCUSUM-ICA-MSPCA approach of the process of industrial production was put forward. This algorithm made up for deficiencies of other methods of monitoring, and was simulated by the standard Tennessee Eastman process simulation model, the result indicated that this algorithm achieved good control effect.