Enhancing FMEA with ChatGPT: Structured Outputs, Qualitative Evaluations, and AI-Human Hybrid FMEA

SUMMARY & CONCLUSIONS The integration of ChatGPT into Failure Mode and Effects Analysis (FMEA) marks a significant advancement in risk assessment methodologies within reliability engineering. By leveraging ChatGPT's extensive knowledge base and sophisticated natural language processing capabilities, the conventional FMEA process can be significantly automated and accelerated. This study demonstrates the feasibility and benefits of incorporating ChatGPT into FMEA through a structured approach that includes creating a system hierarchy and detailing the FMEA in JavaScript Object Notation (JSON) format. The proposed method streamlines the identification and evaluation of failure modes and enhances the thoroughness and accuracy of the analysis. Our case study on an electric scooter patent illustrates how ChatGPT can generate relevant failure modes with detailed causes and mitigation actions, showcasing its potential to uncover failure modes that human analysts might overlook. Another case study on car seats shows that ChatGPT can reveal 81% of failure modes mentioned by customers. Despite these advancements, several challenges remain. ChatGPT's performance in predicting Severity, Occurrence, and Detection (SOD) values was found to be inconsistent, highlighting the need for improved training data and methodologies. Future research should focus on developing public datasets for FMEA, enhancing domain-specific knowledge integration, and improving the accuracy of SOD value predictions. By addressing these challenges, the reliability engineering community can fully harness the potential of AI-human hybrid FMEA to enhance product safety and reliability in complex manufacturing environments.