Recent experiences in the industrial exploitation of principal component based fault detection methods

Summary form only given. There has been a plethora of research and many industrial studies involving the application of multivariate statistical techniques for the detection and isolation of faults on process plants. Despite all this work and reported success of the studies, there remain very few online applications of the technology. To address this issue and to determine the capabilities and limitations of the multivariate statistical techniques, the Control Technology Centre Ltd. has undertaken a comprehensive study aimed specifically at the development of online multivariate solutions for process plants. This paper provides details of several of these investigations. These investigations highlight a number of factors which should be considered when conducting multivariate analysis. For example, there appears to be a need for consideration of cause and effect relationships, which is often ignored in other application studies. In addition the integration of the multivariate statistical routines into a hierarchical system is demonstrated to provide the potential for robust and self-diagnosing control systems.