Scholarlink

<p>Under constrained sensor installation scenarios requiring non-intrusive monitoring, this study proposes an acoustic-driven feature extraction method for rolling bearing fault diagnosis. Addressing the critical challenges of weak transient signatures and non-stationary noise in acoustic signals, this paper developed the Loudness-based Filter Parameter Determination (L-FPD) technique through three key innovations: (i) Time-adaptive loudness quantification via sliding-window Zwicker model analysis, (ii) Impact-sensitive pulse detection through loudness-time spectra thresholding, and (iii) Critical-band frequency localization using specific loudness distribution statistics. The method overcomes limitations of conventional vibration-optimized techniques by exploiting the relationship between mechanical impacts and transient loudness characteristics in time-domain signals. Experimental validation on defective bearings demonstrates superior performance over spectral kurtosis methods, with the L-FPD-processed envelope spectrum achieving greater fault frequency prominence. This acoustic-specific processing framework enables reliable fault detection at low signal-to-noise ratios, significantly advancing non-contact bearing condition monitoring capabilities for industrial applications.</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

Loudness-based impact feature extraction and adaptive filtering for acoustic signal of bearing fault diagnosis

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Feiyun Cong,
  • Xinyu Huang,
  • Qiming Bo,
  • Chen Liang,
  • Qihao Zhou,
  • Weizheng Zhao,
  • Gang Gao,
  • Yi Zhou

摘要

Under constrained sensor installation scenarios requiring non-intrusive monitoring, this study proposes an acoustic-driven feature extraction method for rolling bearing fault diagnosis. Addressing the critical challenges of weak transient signatures and non-stationary noise in acoustic signals, this paper developed the Loudness-based Filter Parameter Determination (L-FPD) technique through three key innovations: (i) Time-adaptive loudness quantification via sliding-window Zwicker model analysis, (ii) Impact-sensitive pulse detection through loudness-time spectra thresholding, and (iii) Critical-band frequency localization using specific loudness distribution statistics. The method overcomes limitations of conventional vibration-optimized techniques by exploiting the relationship between mechanical impacts and transient loudness characteristics in time-domain signals. Experimental validation on defective bearings demonstrates superior performance over spectral kurtosis methods, with the L-FPD-processed envelope spectrum achieving greater fault frequency prominence. This acoustic-specific processing framework enables reliable fault detection at low signal-to-noise ratios, significantly advancing non-contact bearing condition monitoring capabilities for industrial applications.