Scholarlink

In practical industrial production, the scarcity of fault data is one of the primary challenges in fault diagnosis. Data generation based on deep generative models has been proven to be an effective approach to address the scarcity of fault data. However, the instability of most existing GAN-based models during training is a significant issue, and the quality of samples directly generated from raw industrial process data is often unsatisfactory. To address these issues, a novel enhanced fault diagnosis scheme called PCA-DDPM-CNN is developed. PCA is utilized to filter out noise interference from original industrial process data while adjusting the dimensionality of the original data to the specific input dimensionality of DDPM. Then the data processed by PCA is added noise and reconstructed step by step to continuously train the DDPM's network. After then, realistic fault samples can be generated using the trained DDPM, the augmented data can be used as a new training set for the diagnostic model to enhance its performance. Experiment based on TE process dataset verifies that the model can stably generate more realistic samples for enhancing fault diagnostics in industrial processes.

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 Data Generation Method for Enhanced Fault Diagnosis Based on PCA-DDPM-CNN Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Pengchao Wang,
  • Haoxin Gu,
  • Yujie Cheng,
  • Lixiang Jiang

摘要

In practical industrial production, the scarcity of fault data is one of the primary challenges in fault diagnosis. Data generation based on deep generative models has been proven to be an effective approach to address the scarcity of fault data. However, the instability of most existing GAN-based models during training is a significant issue, and the quality of samples directly generated from raw industrial process data is often unsatisfactory. To address these issues, a novel enhanced fault diagnosis scheme called PCA-DDPM-CNN is developed. PCA is utilized to filter out noise interference from original industrial process data while adjusting the dimensionality of the original data to the specific input dimensionality of DDPM. Then the data processed by PCA is added noise and reconstructed step by step to continuously train the DDPM's network. After then, realistic fault samples can be generated using the trained DDPM, the augmented data can be used as a new training set for the diagnostic model to enhance its performance. Experiment based on TE process dataset verifies that the model can stably generate more realistic samples for enhancing fault diagnostics in industrial processes.