Hybrid method for process fault detection and diagnosis

Mallick, Md. Raihan (2013) Hybrid method for process fault detection and diagnosis. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (3MB)

Abstract

This thesis demonstrates a real time automated hybrid method for process monitoring. Motivation of this research comes from the fact that there is hardly any single techniques available which is decent enough for process fault detection and diagnosis simultaneously. Process history based methods are well known as early fault detectors but operators require complex analysis to find out the root cause of the fault. Knowledge based qualitative models are worthy for root cause analysis but mostly done in off-line fashion. Moreover, modern processes are equipped with thousands of variables and structurally they are very complex in nature. All these influences make manual diagnostic task more complicated for the operators. Therefore, there is a need for automated process monitoring tool that has good detection and diagnosis performance. -- In this work, a hybrid method based on principal component analysis (PCA) and Bayesian belief network (BBN) is described for process monitoring. PCA is very proficient as early fault detector but not for fault diagnosis. On the other hand, BBN is good for diagnosis. This hybrid method combines t he strong features of both PCA and BBN to an automated monitoring system that can detect fault early as well as diagnose the root cause precisely. Upon successful detection of fault from PCA, diagnostic information from the PCA is passed to the BBN for root cause analysis. Pearl's message passing algorithm is used for belief updating. This monitoring tool integrates prior process knowledge along with the present observed evidence processed by the multivariate statistical method to come up with the most probable explanation of process fault. Efficacy of the proposed method is verified by simulating different scenarios on a simulated dissolution tank model. The monitoring tool is also validated using industrial data from a pure terephthalic acid (PTA) plant.

Actions (login required)

View Item