Scholarlink

<p>Bearing fault diagnosis in high-noise environments remains a critical challenge in condition monitoring, where conventional methods struggle with signal denoising, feature extraction, and feature selection. Traditional IMF selection techniques fail to capture signals’ nonlinear and time-varying nature while existing feature extraction methods overlook localised energy distribution and frequency variations. Additionally, feature selection methods often struggle with high-dimensional noisy datasets, reducing classification performance. To address these issues, this study proposes an integrated approach combining Dynamic Time Warping (DTW)-based IMF selection, Wavelet-Based Fused Time-Frequency Features (F-TFF) extraction, and Bat Algorithm (BA)-based feature selection. DTW enhances IMF selection by accurately identifying noise-free IMFs, F-TFF extracts centroid-based, energy-based, and statistical features for a more representative signal characterisation, and BA optimises feature selection for improved robustness in noisy datasets. The proposed method is validated on a bearing fault dataset under five different noise levels, achieving a classification accuracy of 100% in noise-free conditions and 70.04% at -20 dB SNR, outperforming benchmark techniques. These results confirm the approach’s effectiveness in enhancing fault diagnosis under high-noise conditions.</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

Noise-resilient bearing fault diagnosis using DTW-based IMF selection and wavelet-fused features with the Bat Algorithm

  • 期刊论文

  • 发表日期：

    2025年5月27日

查看全文
  • Andrews Athisayam,
  • Manisekar Kondal

摘要

Bearing fault diagnosis in high-noise environments remains a critical challenge in condition monitoring, where conventional methods struggle with signal denoising, feature extraction, and feature selection. Traditional IMF selection techniques fail to capture signals’ nonlinear and time-varying nature while existing feature extraction methods overlook localised energy distribution and frequency variations. Additionally, feature selection methods often struggle with high-dimensional noisy datasets, reducing classification performance. To address these issues, this study proposes an integrated approach combining Dynamic Time Warping (DTW)-based IMF selection, Wavelet-Based Fused Time-Frequency Features (F-TFF) extraction, and Bat Algorithm (BA)-based feature selection. DTW enhances IMF selection by accurately identifying noise-free IMFs, F-TFF extracts centroid-based, energy-based, and statistical features for a more representative signal characterisation, and BA optimises feature selection for improved robustness in noisy datasets. The proposed method is validated on a bearing fault dataset under five different noise levels, achieving a classification accuracy of 100% in noise-free conditions and 70.04% at -20 dB SNR, outperforming benchmark techniques. These results confirm the approach’s effectiveness in enhancing fault diagnosis under high-noise conditions.