Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Syn-Diag: An LLM-based Synergistic Framework for Generalizable Few-shot Fault Diagnosis on the Edge

Zijun Jia, Shuang Liang, Jinsong Yu

TL;DR

Syn-Diag addresses data scarcity and edge deployment constraints in industrial fault diagnosis by integrating cross-modal semantic alignment, content-aware synergistic reasoning with dynamic prompts, and cloud-edge knowledge distillation with online updating. The framework aligns time-frequency spectrogram features with LLM semantic space, enables a multiple-choice style inference, and distills knowledge to a lightweight edge model that can update the cloud model iteratively via a shared decision space. Across six datasets under 1-, 3-, 5-, 7-shot and cross-condition settings, Syn-Diag achieves state-of-the-art performance, while the edge model reduces size by $83\%$ and latency by $50\%$ relative to the cloud model. This work presents a practical, robust paradigm for scalable, adaptive intelligent diagnostics on resource-constrained devices.

Abstract

Industrial fault diagnosis faces the dual challenges of data scarcity and the difficulty of deploying large AI models in resource-constrained environments. This paper introduces Syn-Diag, a novel cloud-edge synergistic framework that leverages Large Language Models to overcome these limitations in few-shot fault diagnosis. Syn-Diag is built on a three-tiered mechanism: 1) Visual-Semantic Synergy, which aligns signal features with the LLM's semantic space through cross-modal pre-training; 2) Content-Aware Reasoning, which dynamically constructs contextual prompts to enhance diagnostic accuracy with limited samples; and 3) Cloud-Edge Synergy, which uses knowledge distillation to create a lightweight, efficient edge model capable of online updates via a shared decision space. Extensive experiments on six datasets covering different CWRU and SEU working conditions show that Syn-Diag significantly outperforms existing methods, especially in 1-shot and cross-condition scenarios. The edge model achieves performance comparable to the cloud version while reducing model size by 83% and latency by 50%, offering a practical, robust, and deployable paradigm for modern intelligent diagnostics.

Syn-Diag: An LLM-based Synergistic Framework for Generalizable Few-shot Fault Diagnosis on the Edge

TL;DR

Syn-Diag addresses data scarcity and edge deployment constraints in industrial fault diagnosis by integrating cross-modal semantic alignment, content-aware synergistic reasoning with dynamic prompts, and cloud-edge knowledge distillation with online updating. The framework aligns time-frequency spectrogram features with LLM semantic space, enables a multiple-choice style inference, and distills knowledge to a lightweight edge model that can update the cloud model iteratively via a shared decision space. Across six datasets under 1-, 3-, 5-, 7-shot and cross-condition settings, Syn-Diag achieves state-of-the-art performance, while the edge model reduces size by and latency by relative to the cloud model. This work presents a practical, robust paradigm for scalable, adaptive intelligent diagnostics on resource-constrained devices.

Abstract

Industrial fault diagnosis faces the dual challenges of data scarcity and the difficulty of deploying large AI models in resource-constrained environments. This paper introduces Syn-Diag, a novel cloud-edge synergistic framework that leverages Large Language Models to overcome these limitations in few-shot fault diagnosis. Syn-Diag is built on a three-tiered mechanism: 1) Visual-Semantic Synergy, which aligns signal features with the LLM's semantic space through cross-modal pre-training; 2) Content-Aware Reasoning, which dynamically constructs contextual prompts to enhance diagnostic accuracy with limited samples; and 3) Cloud-Edge Synergy, which uses knowledge distillation to create a lightweight, efficient edge model capable of online updates via a shared decision space. Extensive experiments on six datasets covering different CWRU and SEU working conditions show that Syn-Diag significantly outperforms existing methods, especially in 1-shot and cross-condition scenarios. The edge model achieves performance comparable to the cloud version while reducing model size by 83% and latency by 50%, offering a practical, robust, and deployable paradigm for modern intelligent diagnostics.

Paper Structure

This paper contains 29 sections, 14 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Few-shot Fault Diagnosis
  4. Multimodal Large Language Models
  5. Method
  6. Cross-modal Semantic Alignment Pre-training
  7. Model Architecture
  8. Multi-level Alignment Loss Function
  9. Advantages and Feature Generalization
  10. Dynamic Synergistic Prompt Fusion and Deep Prompt Tuning
  11. Dynamic Synergistic Prompt Fusion
  12. Deep Prompt Tuning
  13. Few-shot Fault Diagnosis Fine-tuning based on Synergistic Reasoning
  14. Parameter-Efficient Fine-Tuning
  15. Synergistic Reasoning and Classification
  16. ...and 14 more sections

Figures (5)

  • Figure 1: Overall architecture diagram
  • Figure 2: Feature distribution using t-SNE on the CWRU dataset: before and after image-text alignment
  • Figure 3: Feature distribution using t-SNE on the SEU dataset: before and after image-text alignment
  • Figure 4: Effects of edge updates and cloud aggregation under different conditions on CWRU data (the teacher model here refers to Syn-Diag-Cloud, and the student model here refers to Syn-Diag-Edge)
  • Figure 5: Resource Occupation and Efficiency Test for Syn-Diag-Cloud and Syn-Diag-Edge