Dynamic risk analysis with alarm data to improve process safety using Bayesian network

Chemical process industries often suffer from abnormal events of varying magnitudes, which may lead to different consequences, including incipient faults, near-miss, incidents, and accidents. Typically, when an abnormal event occurs, various safety systems, such as alarm systems and safety instrumented systems (SIS), come into play to prevent the event from propagating. To improve process safety, estimation of the failure probabilities of safety systems and probabilities of corresponding consequences are extremely important and should receive more attention. Risk analysis techniques, such as event tree analysis, have been used popularly for this purpose. However they suffer limitations of static structures and dependencies among events, which affect the performance of safety system analysis. In this paper, Bayesian network (BN) for dynamic risk analysis is proposed. BN is a probabilistic approach to modeling and representation of influences between safety systems. In a BN model, causal relationships reflect dependencies among safety systems represented as nodes. After building the BN causal structure, we estimate the parameters of the network, namely, the failure probabilities of safety systems and probabilities of corresponding consequences, using alarm data of alarm systems. In addition, advantages of BNs for dynamic risk analysis are discussed. Finally dynamic risk analysis using BNs is applied to a case study to show its performance.