Generating Troubleshooting Trees for Industrial Equipment using Large Language Models (LLM)

Troubleshooting trees play a pivotal role in industrial diagnostics and fault analysis of complex systems and equipment by serving as a systematic guide in identifying and resolving issues. Constructing a troubleshooting tree involves a rigorous process that integrates multidisciplinary expertise and begins with an in-depth analysis of system architecture, operations, and known failure modes. Information from diverse sources are typically used to draft the initial tree structure, which is then iteratively refined through real-world data and field feedback. To expedite this process, this paper explores the use of generative large language models (LLMs) to automatically extract and structure information from unstructured text sources like service manuals, maintenance records, and design documents. In this work, we investigate three different product manuals and propose a method for generating an initial troubleshooting tree. Our results shows that the proposed method has data extraction coverage ranging from 36% to 64% and an extraction precision from 88% to 100%. We also performed a detailed analysis on the potential hallucination of the method and discuss the bottlenecks of the current process. We envision this work to establish a robust and reliable generative LLM-based pipeline for automated generation of an initial troubleshooting tree for diverse industrial processes and operations.