Scholarlink

<p>Actual industrial environments are complex and subject to noise interference, resulting in rolling bearing faults that exhibit complexity and coupling characteristics. Traditional fault diagnosis methods demonstrate limited feature extraction capabilities and low accuracy when handling complex and coupled faults. Therefore, this paper proposes a rolling bearing fault diagnosis method combining the Gramian Angular Field (GAF) with a dual-channel Convolutional Neural Network (CNN) and Least Squares Support Vector Machine (LSSVM). This method first converts raw vibration signals into two-dimensional images—Gramian Angular Difference Field (GADF) and Gramian Angular Sum Field (GASF) with GAF. These images are then inputted into two separate channels of the CNN for feature extraction and fusion. Finally, the extracted features are fed into Crested Porcupine Optimizer (CPO) optimized LSSVM for classification, enabling fault diagnosis. The method was validated through a rolling bearing hybrid fault experiment designed for this study and verified using the CWRU public dataset, achieving high accuracy in both cases. Furthermore, it demonstrated superior performance compared to several other fault diagnosis methods when dealing with limited training samples and noisy interference, confirming the effectiveness and superiority of the proposed approach.</p>

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 hybrid fault diagnosis method for rolling bearings combining GAF and dual-channel CNN

  • 期刊论文

  • 发表日期：

    2026年7月20日

查看全文
  • Kai Zhang,
  • Chen Yang,
  • Shuzhi Gao

摘要

Actual industrial environments are complex and subject to noise interference, resulting in rolling bearing faults that exhibit complexity and coupling characteristics. Traditional fault diagnosis methods demonstrate limited feature extraction capabilities and low accuracy when handling complex and coupled faults. Therefore, this paper proposes a rolling bearing fault diagnosis method combining the Gramian Angular Field (GAF) with a dual-channel Convolutional Neural Network (CNN) and Least Squares Support Vector Machine (LSSVM). This method first converts raw vibration signals into two-dimensional images—Gramian Angular Difference Field (GADF) and Gramian Angular Sum Field (GASF) with GAF. These images are then inputted into two separate channels of the CNN for feature extraction and fusion. Finally, the extracted features are fed into Crested Porcupine Optimizer (CPO) optimized LSSVM for classification, enabling fault diagnosis. The method was validated through a rolling bearing hybrid fault experiment designed for this study and verified using the CWRU public dataset, achieving high accuracy in both cases. Furthermore, it demonstrated superior performance compared to several other fault diagnosis methods when dealing with limited training samples and noisy interference, confirming the effectiveness and superiority of the proposed approach.