Scholarlink

Purpose <p>Rolling bearings are vital machine components that support rotating shafts and reduce friction between moving parts, ensuring smooth and efficient operation. Accurate fault diagnosis of rolling bearings is essential for predictive maintenance, as it enhances machine reliability, prevents unexpected breakdowns, and extends service life. However, training deep learning models for fault diagnosis remains challenging when only limited fault data is available.</p> Methods <p>To tackle the issue of limited labeled data, this study integrates Convolutional Neural Networks (CNN) and Support Vector Machines (SVM). CNNs are employed for automatic feature learning from raw vibration signals, while SVMs provide strong generalization ability for classification under small-sample conditions. An adaptive cut-off strategy was implemented, enabling the system to automatically decide when to transition from CNN-based feature extraction to SVM-based classification, without manual intervention. This hybrid CNN-SVM framework was trained and validated using bearing fault datasets under realistic constraints of limited fault samples.</p> Results <p>The proposed CNN-SVM approach demonstrated superior performance compared to conventional CNN. While the CNN model alone achieved a high validation accuracy of 99.09% and classification accuracy of 99.10%, the CNN-SVM model further improved results, yielding 99.32% validation accuracy and 99.31% classification accuracy. These improvements highlight the robustness and efficiency of the hybrid framework, especially under data-scarce conditions.</p> Conclusion <p>The integration of CNN and SVM offers a promising solution for rolling bearing fault diagnosis when limited fault data is available. By combining the deep feature learning ability of CNN with the generalization strength of SVM, and introducing adaptive cut-off conditions, the proposed framework achieves enhanced diagnostic accuracy and stability. This study demonstrates that hybrid deep learning models can effectively address the challenge of small-sample fault detection, contributing to more reliable predictive maintenance and improved operational safety of machinery.</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

Efficient Rolling Bearing Fault Diagnosis with a CNN-SVM System Enabled by Automatic Cut-off Conditions for Small Sample Data

  • 期刊论文

  • 发表日期：

    2025年12月12日

查看全文
  • Varun Sharma,
  • Rajvardhan Jigyasu,
  • Sachin Singh

摘要

Purpose

Rolling bearings are vital machine components that support rotating shafts and reduce friction between moving parts, ensuring smooth and efficient operation. Accurate fault diagnosis of rolling bearings is essential for predictive maintenance, as it enhances machine reliability, prevents unexpected breakdowns, and extends service life. However, training deep learning models for fault diagnosis remains challenging when only limited fault data is available.

Methods

To tackle the issue of limited labeled data, this study integrates Convolutional Neural Networks (CNN) and Support Vector Machines (SVM). CNNs are employed for automatic feature learning from raw vibration signals, while SVMs provide strong generalization ability for classification under small-sample conditions. An adaptive cut-off strategy was implemented, enabling the system to automatically decide when to transition from CNN-based feature extraction to SVM-based classification, without manual intervention. This hybrid CNN-SVM framework was trained and validated using bearing fault datasets under realistic constraints of limited fault samples.

Results

The proposed CNN-SVM approach demonstrated superior performance compared to conventional CNN. While the CNN model alone achieved a high validation accuracy of 99.09% and classification accuracy of 99.10%, the CNN-SVM model further improved results, yielding 99.32% validation accuracy and 99.31% classification accuracy. These improvements highlight the robustness and efficiency of the hybrid framework, especially under data-scarce conditions.

Conclusion

The integration of CNN and SVM offers a promising solution for rolling bearing fault diagnosis when limited fault data is available. By combining the deep feature learning ability of CNN with the generalization strength of SVM, and introducing adaptive cut-off conditions, the proposed framework achieves enhanced diagnostic accuracy and stability. This study demonstrates that hybrid deep learning models can effectively address the challenge of small-sample fault detection, contributing to more reliable predictive maintenance and improved operational safety of machinery.