Scholarlink

In the field of fault diagnosis, the significant imbalance between normal and fault data poses a considerable challenge. To overcome this obstacle, an Improved Unet Diffusion Model (IUnet-DM) has been introduced for data generation, which can combine with a diagnostic network to enhance fault diagnosis performance in scenarios with imbalanced data. To enhance the feature extraction prowess for time-series data, causal convolutional networks and Long Short-Term Memory (LSTM) networks are integrated into the model as a TLBlock. Furthermore, the Wide Depth Convolutional Neural Network (WDCNN) and one-Dimensional Residual Network (1DResNet) are harnessed to provide precise diagnostic capabilities. To benchmark the performance of the proposed approach, the Deep Convolutional Generative Adversarial Network (DCGAN) and the Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) are explored as comparison methods. Experiments conducted on bearing and gearbox datasets demonstrate the high accuracy and exceptional performance of the proposed method under conditions of data imbalance.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Fault Diagnosis Method for Data Imbalance Conditions Based on the Improved Unet Diffusion Model

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Tianliang Zhao,
  • Haitao Ni,
  • Wenbo Zhang,
  • Liuyang Song,
  • Huaqing Wang

摘要

In the field of fault diagnosis, the significant imbalance between normal and fault data poses a considerable challenge. To overcome this obstacle, an Improved Unet Diffusion Model (IUnet-DM) has been introduced for data generation, which can combine with a diagnostic network to enhance fault diagnosis performance in scenarios with imbalanced data. To enhance the feature extraction prowess for time-series data, causal convolutional networks and Long Short-Term Memory (LSTM) networks are integrated into the model as a TLBlock. Furthermore, the Wide Depth Convolutional Neural Network (WDCNN) and one-Dimensional Residual Network (1DResNet) are harnessed to provide precise diagnostic capabilities. To benchmark the performance of the proposed approach, the Deep Convolutional Generative Adversarial Network (DCGAN) and the Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) are explored as comparison methods. Experiments conducted on bearing and gearbox datasets demonstrate the high accuracy and exceptional performance of the proposed method under conditions of data imbalance.