An artificial neural network approach to enrich HAZOP analysis of complex processes

This paper proposes an innovative approach to enrich Hazard and operability (HAZOP) analysis for complex processes using process simulators and artificial neural networks (ANNs). HAZOP study is a systematic qualitative procedure aimed at identifying potential hazards and operability issues. It heavily relies on the collective knowledge and experience of the team during brainstorming sessions. Traditionally, HAZOP considers only "one failure at a time," overlooking the effects of deviation causes amplitudes, their propagation, and subsequent domino effects. This simplification is necessary to manage time and costs during sessions. However, in complex systems, neglecting certain scenarios may result in overlooking critical situations. In our proposed method, we leverage process simulators to simulate upset scenarios comprehensively. By systematically varying all possible deviation causes and their combinations, we generate a substantial amount of simulation data. To facilitate evaluation, we introduce novel evaluation indexes. Additionally, we define a sensitivity index for ranking HAZOP scenarios based on severity of consequences. Furthermore, we classify scenarios into three severity levels according to their consequences. To enhance HAZOP analysis, we employ ANNs. These networks learn process behaviors and predict the evaluation indexes. They also classify scenarios based on pre-simulated data. With this approach, the HAZOP team can efficiently analyze the consequences of nearly any combination of deviation causes and failures, even with varying amplitudes. We validate our method by applying it to a real-world complex polymerization plant, demonstrating its value in practical scenarios.