Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Integrating LLMs for Explainable Fault Diagnosis in Complex Systems

Akshay J. Dave, Tat Nghia Nguyen, Richard B. Vilim

TL;DR

The paper tackles explainability in fault diagnosis for complex, safety-critical systems by integrating a physics-based diagnostic tool (PRO-AID) with a Large Language Model as an explanation agent. It implements PRO-AID's analytical redundancy relations and virtual sensors within a Diagnostics System that uses a Symbolic Engine to manage context, enabling grounded, queryable explanations via GPT-4. Validation is conducted on the METL molten-sodium purification facility, where the agent correctly identifies faults, maps residuals to sensor faults, and analyzes historical sensor anomalies. The work demonstrates that combining physics-based diagnostics with AI-driven explanations can enhance operator understanding, trust, and decision-making in autonomous diagnostic workflows.

Abstract

This paper introduces an integrated system designed to enhance the explainability of fault diagnostics in complex systems, such as nuclear power plants, where operator understanding is critical for informed decision-making. By combining a physics-based diagnostic tool with a Large Language Model, we offer a novel solution that not only identifies faults but also provides clear, understandable explanations of their causes and implications. The system's efficacy is demonstrated through application to a molten salt facility, showcasing its ability to elucidate the connections between diagnosed faults and sensor data, answer operator queries, and evaluate historical sensor anomalies. Our approach underscores the importance of merging model-based diagnostics with advanced AI to improve the reliability and transparency of autonomous systems.

Integrating LLMs for Explainable Fault Diagnosis in Complex Systems

TL;DR

The paper tackles explainability in fault diagnosis for complex, safety-critical systems by integrating a physics-based diagnostic tool (PRO-AID) with a Large Language Model as an explanation agent. It implements PRO-AID's analytical redundancy relations and virtual sensors within a Diagnostics System that uses a Symbolic Engine to manage context, enabling grounded, queryable explanations via GPT-4. Validation is conducted on the METL molten-sodium purification facility, where the agent correctly identifies faults, maps residuals to sensor faults, and analyzes historical sensor anomalies. The work demonstrates that combining physics-based diagnostics with AI-driven explanations can enhance operator understanding, trust, and decision-making in autonomous diagnostic workflows.

Abstract

This paper introduces an integrated system designed to enhance the explainability of fault diagnostics in complex systems, such as nuclear power plants, where operator understanding is critical for informed decision-making. By combining a physics-based diagnostic tool with a Large Language Model, we offer a novel solution that not only identifies faults but also provides clear, understandable explanations of their causes and implications. The system's efficacy is demonstrated through application to a molten salt facility, showcasing its ability to elucidate the connections between diagnosed faults and sensor data, answer operator queries, and evaluate historical sensor anomalies. Our approach underscores the importance of merging model-based diagnostics with advanced AI to improve the reliability and transparency of autonomous systems.
Paper Structure (10 sections, 4 figures)

This paper contains 10 sections, 4 figures.

Table of Contents

  1. Introduction
  2. Explainability
  3. Objectives
  4. Methods
  5. Diagnostics Tool
  6. LLM
  7. Diagnostics System
  8. METL Experimental Facility
  9. Results
  10. Concluding Remarks

Figures (4)

  • Figure 1: Dependency structure of a generic model residual in PRO-AID.
  • Figure 2: A layout of the system and its interconnections.
  • Figure 3: The purification system in METL.
  • Figure 4: Sensor data and residual values during operation of the purification loop. The location of each sensor is presented in \ref{['fig:METL']}. An annotation in the TC 117 sensor subplot indicates the approximate time the facility operator injects a bias.