Multivariate statistical process control of chemical processes

Abstract

The thesis describes the application of Multivariate Statistical Process Control (MSPC) to chemical processes for the task of process performance monitoring and fault detection and diagnosis. The applications considered are based upon polymerisation systems. The first part of the work establishes the appropriateness of MSPC methodologies for application to modern industrial chemical processes. The statistical projection techniques of Principal Component Analysis and Projection to Latent Structures are considered to be suitable for analysing the multivariate data sets obtained from chemical processes and are coupled with methods and techniques for implementing MSPC. A comprehensive derivation of these techniques are presented. The second part introduces the procedures that require to be followed for the appropriate implementation of MSPC-based schemes for process monitoring, fault detection and diagnosis. Extensions of the available projection techniques that can handle specific types of chemical processes, such as those that exhibit non-linear characteristics or comprise many distinct units are also presented. Moreover, the novel technique of Inverse Projection to Latent Structures that extends the application of MSPC-based schemes to processes where minimal process data is available is introduced. Finally, the proposed techniques and methodologies are illustrated by applications to a batch and a continuous polymerisation process.