Generative artificial intelligence (Gen-AI)-driven method for hazard scenario generation in chemical process systems

This research proposes a human-in-the-loop GenAI framework for generating hazard scenarios in chemical process systems. The method represents process information as graphs and combines a graph isomorphism network with a transformer-based sequence decoder to generate additional scenarios from process flow information and predefined scenarios. Human experts then verify the generated results. The framework is demonstrated with two case studies based on U.S. Chemical Safety Board accident information. Its relevance to Process Safety Management is direct: comprehensive scenario identification is a foundational step in HAZOP, What-If analysis, and quantitative risk assessment. By systematically proposing additional credible scenarios, the method can help PHA teams challenge the completeness of their existing scenario set and reduce dependence on unaided brainstorming. It also illustrates a useful PSM pattern in which AI expands the search space, while experienced engineers remain responsible for technical validation and risk decisions. The strength of this paper is the emphasis on scenario generation, which is the critical first step in HAZOP. As a university research team the authors have done a careful analysis by creating their own models; however, the approach might be more difficult to apply when using commecially available frontier models?