# LLMFuxian: Reproduction of Multimodal Large Model for Mechanical Fault Diagnosis in Intelligent Operation and Maintenance

> Code repository for reproducing the paper 'Multimodal data-enabled large model for machine fault diagnosis towards intelligent operation and maintenance', focusing on single-dataset experiments using the HUSTbearing dataset.

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-05-23T08:47:55.000Z
- 最近活动: 2026-05-23T08:55:54.734Z
- 热度: 159.9
- 关键词: 故障诊断, 多模态大模型, 智能运维, 轴承数据集, 知识图谱, 零样本学习, PyTorch, 工业AI
- 页面链接: https://www.zingnex.cn/en/forum/thread/llmfuxian
- Canonical: https://www.zingnex.cn/forum/thread/llmfuxian
- Markdown 来源: floors_fallback

---

## Introduction / Main Floor: LLMFuxian: Reproduction of Multimodal Large Model for Mechanical Fault Diagnosis in Intelligent Operation and Maintenance

Code repository for reproducing the paper 'Multimodal data-enabled large model for machine fault diagnosis towards intelligent operation and maintenance', focusing on single-dataset experiments using the HUSTbearing dataset.

## Original Author and Source

- **Original Author/Maintainer**: wangwen2077-boop
- **Source Platform**: GitHub
- **Original Title**: llmfuxian
- **Original Link**: https://github.com/wangwen2077-boop/llmfuxian
- **Release Date**: May 23, 2026
- **Tech Stack**: Python, PyTorch
- **Original Paper Title**: Multimodal data-enabled large model for machine fault diagnosis towards intelligent operation and maintenance

## Project Overview

LLMFuxian is an academic paper reproduction project aimed at reproducing the methods in the paper 'Multimodal data-enabled large model for machine fault diagnosis towards intelligent operation and maintenance'. This project focuses on the field of intelligent operation and maintenance of industrial equipment, using multimodal large model technology to achieve automatic diagnosis of mechanical faults.

Currently, it has mainly completed the reproduction of single-dataset experiments using the **HUSTbearing dataset** from the paper, and does not yet include the gear dataset, FRC/CAME external baselines, or complete ablation experiments.

## Challenges in Industrial Equipment Fault Diagnosis

In modern industrial production, the health status of rotating machinery (such as bearings and gearboxes) directly affects production efficiency and safety. Traditional fault diagnosis methods face the following challenges:

1. **Data Heterogeneity**: Various sensor data (vibration signals, temperature, current, etc.) have different formats
2. **Complex Fault Types**: From single faults to compound faults, with diverse patterns
3. **Scarce Labeled Data**: Fault samples are difficult to obtain, especially in compound fault scenarios
4. **Cross-Working Condition Generalization**: Difficulties in model transfer under different speed and load conditions

## Solution Approach of Multimodal Large Model

The LMM-FD (Large Multimodal Model for Fault Diagnosis) method proposed in the paper combines the knowledge understanding ability of large language models with time series analysis, and enhances entity representation through knowledge graphs to achieve zero-shot compound fault diagnosis capability.

## Current Reproduction Status

The project has completed the following core functions:

## Data Preprocessing

- HUSTbearing data reading and parsing
- Tab-delimited text parsing of original `.xls` files
- Filtering of three working conditions: `20Hz / 40Hz / 60Hz`
- Window segmentation processing of vibration signals for `X/Y/Z` three axes

## Dataset Division Strategy

A carefully designed training/test division strategy is adopted:

- **Training/Test Data**: Healthy state (healthy) + single fault (single-fault)
- **Zero-shot Validation**: Compound fault (compound fault) time series are isolated separately

This design simulates real industrial scenarios—compound fault samples are scarce, and the model needs to perform zero-shot inference on unseen compound fault types.