Large language models struggle to improve industrial fault diagnosis reliability
When LLMs Sit Above Diagnostic Tools: Unrealized Complementarity in Industrial Fault Diagnosis
Computation and Language
Summary
When big AI language models are combined with specialized diagnostic tools to find industrial equipment problems, the results aren’t always better. The study shows that sometimes the AI gets confused by conflicting information from the tools, leading to worse decisions than the tools alone. Although the AI helps somewhat, it misses many corrections the specialist tools would catch by themselves. This means just stacking AI on top of expert systems might not unlock their full potential in real-world fault detection.
What this means in practice
- •For industrial maintenance teams: Evaluate diagnostic systems by comparing combined AI-tool outputs against specialist tools alone to improve fault detection reliability.
- •For manufacturing equipment operators: Use specialist diagnostic tools independently rather than relying on large language model integrations to avoid reduced fault diagnosis accuracy.
Authors
Donghwan Kim
Abstract
Large language models are increasingly used as integration layers above specialized tools, but a stronger component does not necessarily produce a stronger combined system. Across five diagnostic datasets (bearing vibration, process monitoring, semiconductor equipment), we study whether an LLM can reliably use external diagnostic information; paired repeat calls separate advice effects from output instability. In all five, conflicting external information overturned initially correct LLM judgments. Among the four datasets with direct integration comparisons, none showed a consistent advantage for implicit LLM integration over the stronger standalone source. On a Tennessee Eastman confirmation set whose protocol was fixed before evaluation, unaided accuracy was 64.67%, implicit LLM-specialist integration 77.43%, and the specialist alone 83.33%. Specialist information improved the LLM by 12.8 points (95% interval 9.7 to 15.9), yet the integrated output stayed 5.9 points below the specialist (95% interval -12.0 to -0.7). A two-source selector oracle reached 92.76%, indicating complementarity that the integrated output did not fully realize. The integrated output missed 140 of 295 specialist corrections (47.5%) but lost 15 of 99 initially correct LLM judgments (15.2%). The deficit remained under prompt and specialist sensitivity analyses. Among CWRU cases solved under both evidence presentations, task-aligned physical evidence yielded lower estimates of susceptibility to incorrect advice in six of seven models (five intervals excluding zero); higher reasoning effort gave no reliable reduction in five models, and a separate four-model TEP analysis gave no clear evidence that it resolves the integration problem. Source quality and integration quality should be evaluated separately: an integration layer should be compared with its stronger standalone component, not only with the unaided LLM.