S2S-FDD: Bridging Industrial Time Series and Natural Language for Explainable Zero-shot Fault Diagnosis
Authors: Baoxue Li, Chunhui Zhao
Organizations: State Key Laboratory of Industrial Control Technology, College of Control Science and Engineering, Zhejiang University, Hangzhou, China · School of Information and Electrical Engineering, Zhejiang University City College, Hangzhou, China
Fault diagnosis is critical for the safe operation of industrial systems. Conventional diagnosis models typically produce abstract outputs such as anomaly scores or fault categories, failing to answer critical operational questions like "Why" or "How to repair". While large language models (LLMs) offer strong generalization and reasoning abilities, their training on discrete textual corpora creates a semantic gap when processing high-dimensional, temporal industrial signals. To address this challenge, we propose a Signals-to-Semantics fault diagnosis (S2S-FDD) framework that bridges high-dimensional sensor signals with natural language semantics through two key innovations: We first design a Signal-to-Semantic operator to convert abstract time-series signals into natural language summaries, capturing trends, periodicity, and deviations. Based on the descriptions, we design a multi-turn tree-structured diagnosis method to perform fault diagnosis by referencing historical maintenance documents and dynamically querying additional signals. The framework further supports human-in-the-loop feedback for continuous refinement. Experiments on the multiphase flow process show the feasibility and effectiveness of the proposed method for explainable zero-shot fault diagnosis.
Figures & tables
Fig. 1: Overview of the proposed S2S framework. It contains two key components, including the S2S operator and multi-turn tree-structured diagnosis method. The former converts raw sensor data into concise and domain-aware natural language summaries. Based on the descriptions, the latter retrieves relevant historical maintenance documents and conducts zero-shot fault diagnosis.
Prompt Template for Temporal Description
Your task is to describe the deviation between the measured value and the ideal normal value of a measurement point, focusing on trends, periodic patterns, volatility, and key anomalies. • Industrial process: [PROCESS_INFO] • All measurement points: [ALL_SENSORS] • Target measurement point: [TARGET_SENSOR] • Data table: [TABLE] (measured value, ideal normal value, deviation, and time-varying deviation percentage, along with the average deviation and deviation percentage under normal conditions) Only focus on time intervals where the deviation or deviation percentage significantly exceeds that under normal conditions, as these intervals indicate possible faults. Ignore periods with deviation close to normal. For significantly deviated periods, describe: 1. The trend of the measured value (increasing, decreasing, stable, or cyclically fluctuating) 2. Whether the measured value is above or below the ideal value If no significant deviation is observed, state that the variable has no obvious abnormality. Provide both quantitative indicators (time intervals) and qualitative insights, focusing on observable patterns. Avoid speculating on root causes. Do not use subheadings or Markdown syntax. Keep the response concise—no more than 100 words.
Prompt Template for Fault Diagnosis
You are an expert in fault diagnosis for industrial processes. Your task is to analyze the following information to identify potential faults: • Industrial Process: [PROCESS_INFO] • Measurement Points (List of available sensors): [ALL_SENSORS] • Fault Knowledge (Known faults and their characteristics): [FAULT_KNOWLEDGE] • Time-Series Observations: [TIME_DESP] (Deviation = Measured - Predicted) Follow these steps: 1. Identify Key Sensors: Determine which sensors show significant deviations from expected values. 2. Cross-Reference with Fault Knowledge: Compare the observed patterns with known fault characteristics to hypothesize possible faults. Please note that sometimes different faults may exhibit similar patterns in certain variables. So, do not rely solely on the provided fault knowledge; you should combine fault causes and process workflow to reasonably infer the trends of different variables. 3. Check data sufficiency: • If the time series observations provide enough information to identify the fault, output the fault number in the following format: <reasoning>...</reasoning> and <answer>...</answer> • If the time series observations are insufficient or critical variables are missing, use the ‘get_target_table‘ tool to query detailed data for specific sensors. Ensure the sensor name exists in the Measurement Points before querying. Output the tool call in the following format: <tool>get_target_table("SENSOR")</tool> Important Notes: • Prioritize sensors listed in [ALL_SENSORS] • Only call get_target_table when necessary: – When time series observations are insufficient to determine the fault – When critical variables for fault diagnosis are missing from the observations
TABLE I: Prompt Templates
Fig. 2: Sketch of the multiphase flow process.
Methods
Case 1
Case 2
Case 3
Case 4
Case 5
Case 6
Case 7
Case 8
Case 9
Case 10
Case 11
Case 12
Case 13
ACC@5
Qwen2.5-7B-Instruct
1
1
1
0
0
0
0
0
0
0
0
0
0
23.08%
Qwen2.5-72B-Instruct
0
1
1
0
1
1
0
0
0
0
0
0
0
30.77%
DeepSeek-V3
0
1
1
0
1
0
0
0
0
0
0
0
0
23.08%
DeepSeek-R1-Distill-Qwen-7B
1
1
1
1
1
1
0
0
0
0
1
0
1
61.54%
DeepSeek-R1-Distill-Qwen-32B
1
0
1
0
1
1
1
0
0
1
0
0
1
53.85%
QwQ-32B
1
1
1
1
1
1
1
0
0
0
0
0
1
61.54%
TABLE II: Comparison of Different Methods
Results of Qwen2.5-7B-Instruct
<reasoning> Based on the significant deviations in the sensors and the known fault characteristics, the most likely fault is Fault Type 1 (Air Line Blockage), as the deviations in PT312 and PT401 align with the expected behavior of a blockage in the air line. </reasoning> <answer> 1 </answer>
Results of DeepSeek-R1-Distill-Qwen-7B
Step-by-Step Analysis: 1. Identify Key Sensors: The significant deviations in air and water flow rates, along with pressure changes, were observed in sensors like PT312, PT401, PT408, FT305, FT104, and FT406. 2. Cross-Reference with Fault Knowledge: The observed trends in decreasing air and water flow rates, along with pressure decreases, matched the expected behavior of a water line blockage (Fault Type 2). 3. Determine Data Sufficiency: The data from key sensors provided enough information to infer the fault, without needing additional data retrieval. Conclusion: The most probable fault is Fault Type 2: water line blockage. <answer> 2 </answer>
Results of Deepseek-R1
<reasoning> 1. Key Sensor Deviations:
- Air Delivery Pressure (PT312): Decreasing trend (158-200), lower than ideal.
TABLE III: Diagnosis results of three LLMs under Case 4