Scholarlink

Purpose <p>This study presents a novel methodology for diagnosing and classifying compound gearbox faults under variable loading conditions. The research focuses on the simultaneous detection of pitting and full tooth breakage using multivariate vibration signals and evaluates the effectiveness of a two-stage NA-MEMD framework for improving diagnostic performance.</p> Methods <p>Multidimensional vibration signals were collected from a two-stage gearbox operating under torque loads of 0 Nm, 10 Nm, 20 Nm, and 30 Nm. A two-stage NA-MEMD decomposition approach was applied, and the most informative Intrinsic Mode Functions (IMFs) were selected using Improved Multiscale Permutation Entropy (IMPE) and correlation coefficient criteria. Features extracted from the reconstructed effective IMFs were classified using the W-kNN algorithm. FFT analysis was performed to identify gear mesh frequencies (GMFs), harmonics, and sidebands associated with compound faults. The proposed method was further compared with CEEMD.</p> Results <p>The two-stage decomposition strategy improved classification accuracy compared with the first-stage decomposition. Using RMS as the feature, accuracy increased by approximately 10.0%, 8.6%, and 6.2% under 10 Nm, 20 Nm, and 30 Nm loading conditions, respectively. FFT analysis successfully identified two fundamental GMFs and their harmonics corresponding to faults on different gearbox shafts. Comparative results demonstrated superior fault diagnostic capability of NA-MEMD over CEEMD.</p> Conclusion <p>The proposed framework provides a robust and effective approach for compound fault diagnosis in gearboxes operating under varying load conditions. The integration of multivariate signal decomposition, optimized IMF selection, frequency-domain analysis, and W-kNN classification significantly enhances fault detection and classification performance in noisy environments.</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 Novel Approach for Gearbox Compound Fault Diagnosis and Classification Using Multidimensional Vibration Signal: Integrating Two Stage NA-MEMD, IMPE-Correlation Based IMFs Selection, and W-kNN

  • 期刊论文

  • 发表日期：

    2026年6月23日

查看全文
  • Manish Pandit,
  • Sanjay Kumar Chaturvedi,
  • Rajiv Nandan Rai

摘要

Purpose

This study presents a novel methodology for diagnosing and classifying compound gearbox faults under variable loading conditions. The research focuses on the simultaneous detection of pitting and full tooth breakage using multivariate vibration signals and evaluates the effectiveness of a two-stage NA-MEMD framework for improving diagnostic performance.

Methods

Multidimensional vibration signals were collected from a two-stage gearbox operating under torque loads of 0 Nm, 10 Nm, 20 Nm, and 30 Nm. A two-stage NA-MEMD decomposition approach was applied, and the most informative Intrinsic Mode Functions (IMFs) were selected using Improved Multiscale Permutation Entropy (IMPE) and correlation coefficient criteria. Features extracted from the reconstructed effective IMFs were classified using the W-kNN algorithm. FFT analysis was performed to identify gear mesh frequencies (GMFs), harmonics, and sidebands associated with compound faults. The proposed method was further compared with CEEMD.

Results

The two-stage decomposition strategy improved classification accuracy compared with the first-stage decomposition. Using RMS as the feature, accuracy increased by approximately 10.0%, 8.6%, and 6.2% under 10 Nm, 20 Nm, and 30 Nm loading conditions, respectively. FFT analysis successfully identified two fundamental GMFs and their harmonics corresponding to faults on different gearbox shafts. Comparative results demonstrated superior fault diagnostic capability of NA-MEMD over CEEMD.

Conclusion

The proposed framework provides a robust and effective approach for compound fault diagnosis in gearboxes operating under varying load conditions. The integration of multivariate signal decomposition, optimized IMF selection, frequency-domain analysis, and W-kNN classification significantly enhances fault detection and classification performance in noisy environments.