Scholarlink

Abstract <p>Motor current signal analysis offers a cost-effective and non-invasive approach to bearing fault diagnosis. The low signal-to-noise ratio (SNR) and the extremely weak bearing fault signature make the direct use of raw current signals challenging. Confronted with the difficulties of highlighting the weak bearing fault signature in current signals, a multi-stage framework is adopted in this paper.</p> Methods: <p>Firstly, time-shifting operation is undertaken on the raw current signal to attenuate the fundamental supply frequency and its odd harmonics producing a residual current signal with higher SNR. Variational mode extraction (VME) is, then, applied to the signal to minimize high-frequency noise arising from other motor components as well as interferences to further enhance the SNR of the residual signal. Since the use of VME requires careful selection of the penalty factor ( and the center frequency ( , the black widow optimization algorithm (BWOA) is exploited to determine the optimal VME parameters for current signal. The extracted modes are subsequently converted into 2-D symmetrized dot pattern images (SDP), providing a visual representation of the fault characteristics. Features from the SDP images of each signal are extracted using a Convolutional Neural Network (CNN) and a support vector machine (SVM) for pattern detection and fault classification.</p> Results: <p>Using an experimental benchmark setup and corresponding current signals measured at Paderborn University, the framework is tested and through extensive evaluation, the proposed framework demonstrated a fault diagnosis accuracy of α) ω) 95.83%.</p> Conclusions: <p>Comparative analysis showcased its advantages over established methods, reinforcing the proposed approach's effectiveness in improving bearing fault detection outcomes.</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 Framework for Motor Bearings Fault Diagnosis using a Heuristic based Adaptive Variational Mode Extraction and a Hybrid CNN-SVM Classifier of Current Signals

  • 期刊论文

  • 发表日期：

    2025年7月9日

查看全文
  • Emmanuel R. Jonjo,
  • Islam Ali,
  • Tamer F. Megahed,
  • Mohamed G. A. Nassef

摘要

Abstract

Motor current signal analysis offers a cost-effective and non-invasive approach to bearing fault diagnosis. The low signal-to-noise ratio (SNR) and the extremely weak bearing fault signature make the direct use of raw current signals challenging. Confronted with the difficulties of highlighting the weak bearing fault signature in current signals, a multi-stage framework is adopted in this paper.

Methods:

Firstly, time-shifting operation is undertaken on the raw current signal to attenuate the fundamental supply frequency and its odd harmonics producing a residual current signal with higher SNR. Variational mode extraction (VME) is, then, applied to the signal to minimize high-frequency noise arising from other motor components as well as interferences to further enhance the SNR of the residual signal. Since the use of VME requires careful selection of the penalty factor ( and the center frequency ( , the black widow optimization algorithm (BWOA) is exploited to determine the optimal VME parameters for current signal. The extracted modes are subsequently converted into 2-D symmetrized dot pattern images (SDP), providing a visual representation of the fault characteristics. Features from the SDP images of each signal are extracted using a Convolutional Neural Network (CNN) and a support vector machine (SVM) for pattern detection and fault classification.

Results:

Using an experimental benchmark setup and corresponding current signals measured at Paderborn University, the framework is tested and through extensive evaluation, the proposed framework demonstrated a fault diagnosis accuracy of α) ω) 95.83%.

Conclusions:

Comparative analysis showcased its advantages over established methods, reinforcing the proposed approach's effectiveness in improving bearing fault detection outcomes.