This cutting-edge paper presents a domain-augmented LLM framework combining Chain-of-Thought prompting with retrieval-augmented generation (RAG) to classify 1,182 BSEE offshore incident investigation reports into ABS Root Cause Map categories. Using GPT-4o-mini, the Domain CoT + RAG model achieved F1-scores above 0.75 for major categories including Equipment Reliability, Procedure, and Human Factors. The framework reproduced expected clusters of human-related failures consistent with risk-based PSM theory. This work represents the state of the art in applying generative AI to PSM Incident Investigation, demonstrating that LLMs with domain-specific context can achieve scalable, low-cost automated root cause analysis that closely approximates human expert classification.
