Handling cyclic loops for accurate root cause diagnosis of rare events in chemical processes using modified Bayesian network

In chemical processes, an accurate root cause diagnosis of rare events is essential to take an appropriate corrective action and mitigate hazardous consequences. Bayesian-based probabilistic approaches have been widely used for root cause diagnosis of rare events since they can effectively handle data scarcity for low-frequency rare events. However, these probabilistic approaches do not account for cyclic loops that are commonly present in chemical processes. This results in an inaccurate root cause diagnosis. To deal with this limitation, we propose a modified Bayesian network (mBN) that transforms a cyclic loop into an acyclic network. First, the mBN determines the weakest causal relation in the cyclic loop, which is then converted into a temporal relation. As a result of this modification, the cyclic loop is transformed into an acyclic network over time horizon, and is utilized for the root cause diagnosis of rare events. Since considering cyclic loops is important to find true causality, an improved root cause diagnosis is observed using the proposed mBN. For demonstration purposes, we apply the proposed mBN to diagnose a rare event in an industrial benchmark chemical process.