Digitizing Tacit Manufacturing Knowledge into Graph-Grounded, Human-Aligned Coaching Agents

Current manufacturing AI systems fail to capture structured, explainable reasoning chains during dynamic troubleshooting, leaving critical tacit knowledge invisible to digital agents and leading to extended changeovers and material waste. To close this human-to-digital loop, this project introduces an automated pipeline that extracts expert operator decision patterns into structured Observe-Orient-Decide-Act (OODA) Operational Knowledge Graphs. This machine-queryable foundation is then paired with a Retrieval-Augmented Generation (RAG) assistant fine-tuned through Reinforcement Learning with Human Feedback (RLHF). This approach ensures that the resulting coaching advice is explicitly aligned with actual shop floor preferences, safety norms, and sequence dependencies, successfully transforming implicit human expertise into actionable and safe guidance.