Modeling and analysis of process failures using probabilistic functional model

Abstract

Failure analysis is an important tool for effective safety management in the chemical process industry. This thesis applies a probabilistic approach to study two failure analysis techniques. The first technique focuses on fault detection and diagnosis (FDD), while the second is on vulnerability analysis of plant components. In formulating the FDD strategy, a class of functional model called multilevel flow modeling (MFM) was used. Since this model is not commonly used for chemical processes, it was tested on a crude distillation unit and validated using a simulation flowsheet implemented in Aspen HYSYS (Version 8.4) to demonstrate its suitability. Within the proposed FDD framework, probabilistic information was added by transforming the MFM model into its equivalent fault tree model to provide the ability to predict the likelihood of component’s failure. This model was then converted into its equivalent Bayesian network model using HUGIN 8.1 software to facilitate computations. Evaluations of the system on a heat exchanger pilot plant highlight the capability of the model in detecting process faults and identifying the associated root causes. The proposed technique also incorporated options for multi – state functional outcomes, in addition to the typical binary states offered by typical MFM model. The second tool proposed was a new methodology called basic event ranking approach (BERA), which measures the relative vulnerabilities of plant components and can be used to assist plant maintenance and upgrade planning. The framework was applied to a case study involving toxic prevention barriers in a typical process plant. The method was compared to some common importance index methodologies, and the results obtained ascertained the suitability of BERA to be used as a tool to facilitate risk based decisions in planning maintenance schedules in a process plant.