Comprehensive quantitative dynamic accident modelling framework for chemical plants

• Main Author: Al-shanini, Ali Hassan Ali
• Format: Thesis
• Language: English
• Published: 2015
• Subjects: TP Chemical technology
• Online Access: http://eprints.utm.my/id/eprint/77822/1/AliHasanAliPFChE2015.pdf; http://eprints.utm.my/id/eprint/77822/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:94625
• Institution: Universiti Teknologi Malaysia

This thesis introduces a comprehensive accident modelling approach that considers hazards associated with process plants including those that originate from the process itself; human factors including management and organizational errors; natural events related hazards; and intentional and security hazards in a risk assessment framework. The model is based on a series of plant protection systems, which are release, dispersion, ignition, toxicity, escalation, and damage control and emergency management prevention barriers. These six prevention barriers are arranged according to a typical sequence of accident propagation path. Based on successes and failures of these barriers, a spectrum of consequences is generated. Each consequence carries a unique probability of occurrence determined using event tree analysis. To facilitate this computation, the probability of failure for each prevention barrier is computed using fault tree analysis. In carrying out these computations, reliability data from established database are utilized. On occasion where reliability data is lacking, expert judgment is used, and evidence theory is applied to aggregate these experts’ opinion, which might be conflicting. This modelling framework also provides two important features; (i) the capability to dynamically update failure probabilities of prevention barriers based on precursor data, and (ii) providing prediction of future events. The first task is achieved effectively using Bayesian theory; while in the second task, Bayesian-grey model emerged as the most promising strategy with overall mean absolute percentage error of 18.07% based on three case studies, compared to 31.4% for the Poisson model, 22.37% for the first-order grey model, and 22.4% for the second-order grey model. The results obtained illustrated the potentials of the proposed modelling strategy in anticipating failures, identifying the location of failures and predicting future events. These insights are important in planning targeted plant maintenance and management of change, in addition to facilitating the implementation of standard operating procedures in a process plant.