Analysis, Monitoring and Fault Diagnosis of Industrial Processes Using Multivariate Statistical Projection Methods

Abstract

Multivariate statistical procedures based on various versions of principle component analysis (PCA), and partial least squares (PLS), have recently been proposed for the analysis monitoring and diagnosis of industrial processes. These methods are capable of treating processes with large numbers of highly correlated process and quality variables, and can easily handle missing data. The only information needed to exploit them is a good database on historical process operations.In this paper, industrial experiences with these methods based on recent applications in many different industries are presented. Both continuous and batch processes are treated. Multi-block methods are shown to be very useful for treating large continuous processes or multistage batch operations, and multi-way methods are used to treat batch processes where one has time-varying trajectory data on many variables.