Application of Hierarchical Bayesian Belief Networks to Fault Detection and Diagnosis of Industrial Processes

فاقد چكيده فارسي

Implementation of effective process monitoring is of critical importance nowadays due to enforcement of strict environmental regulations, mandatory plant safety standards and highly competitive market. In this paper, application of Hierarchical Bayesian Belief Networks (HBBNs) for fault detection and diagnosis is studied in the benchmark chemical industrial process, known as, Tennessee Eastman Process (TEP). All twenty one faults in TEP are classified into six sub-spaces and specific Bayesian Belief Networks (BBNs) were developed, trained and tested for each one. Fault detection and diagnosis results using HBBN are presented in terms of missed detection rate and misclassification rate. The effectiveness of this method was compared to the other methods such as principal component analysis and hierarchical neural network HBBN is shown to work ideally for fault detection with a zero missed detection rate. Misclassification rate is presented for each TEP fault and HBBN is ranked compared to other FDD methods for which data on misclassification rate were available. HBBN outperforms other exclusively fault classification methods by the overall misclassification rate of 17.9%.