Generative Artificial Intelligence (GenAI)-driven method for hazard scenario generation in chemical process systems

Accurate and comprehensive hazard scenario generation is the first step in quantitative risk analysis (QRA). Traditional approaches to scenario generation rely heavily on subject matter experts, making the process time-consuming and labor-intensive. Besides, scenarios are generally presented in a descriptive form, which is useful but offers limited insights into how they unfold within process systems. This work introduces a novel framework combining generative artificial intelligence (Gen-AI) and human knowledge to address these gaps. First, new scenarios are automatically generated using process flow diagrams (PFDs) and predefined scenarios through a combined graph isomorphism network (GIN) and a transformer-based sequence decoder. Human experts then verify the generated scenarios. Thus, the proposed framework can make the scenario-generation process more exhaustive, faster, and more reliable. Two case studies have been demonstrated using U.S. Chemical Safety Board (CSB) reports. Results suggest that the proposed method can improve the robustness of current hazard identification and scenario generation practices.