Scholarlink

<p>To address the limitations of blind detection in stochastic resonance (SR) systems based on signal-to-noise ratio (SNR), this paper proposes an adaptive bearing fault detection algorithm that integrates weighted iterative autocorrelation coefficients and a stochastic Hopfield network (SHN). The proposed algorithm comprises three main modules: a signal preprocessing module, a period estimation module, and a feature extraction module. First, a stochastic Hopfield network is constructed by combining array stochastic resonance (ASR) with a Hopfield neural network to achieve fault feature extraction. The performance and practicality of the SHN are validated through numerical simulations, and the effects of system order and parameters on performance are analyzed. Next, to achieve accurate signal period estimation, weighted iterative autocorrelation coefficients are constructed by combining iterative autocorrelation functions with Enhanced Periodogram. To further improve estimation accuracy, the signal preprocessing module is implemented using dual-tree complex wavelet packet transform (DTCWPT) and sub-band average kurtosis (SAK). Finally, the proposed method’s performance and practicality are validated using bearing data from the PADERBORN dataset under various operating conditions. Comparative analysis with the ASR system demonstrates that the proposed method consistently exhibits superior performance and robustness across different experimental 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

Application of weighted iterative autocorrelation coefficient and stochastic hopfield network in bearing fault detection under variable speed conditions

  • 期刊论文

  • 发表日期：

    2025年6月5日

查看全文
  • Gang Zhang,
  • Wenhao Liu,
  • Qing Xiong,
  • Luming Lei

摘要

To address the limitations of blind detection in stochastic resonance (SR) systems based on signal-to-noise ratio (SNR), this paper proposes an adaptive bearing fault detection algorithm that integrates weighted iterative autocorrelation coefficients and a stochastic Hopfield network (SHN). The proposed algorithm comprises three main modules: a signal preprocessing module, a period estimation module, and a feature extraction module. First, a stochastic Hopfield network is constructed by combining array stochastic resonance (ASR) with a Hopfield neural network to achieve fault feature extraction. The performance and practicality of the SHN are validated through numerical simulations, and the effects of system order and parameters on performance are analyzed. Next, to achieve accurate signal period estimation, weighted iterative autocorrelation coefficients are constructed by combining iterative autocorrelation functions with Enhanced Periodogram. To further improve estimation accuracy, the signal preprocessing module is implemented using dual-tree complex wavelet packet transform (DTCWPT) and sub-band average kurtosis (SAK). Finally, the proposed method’s performance and practicality are validated using bearing data from the PADERBORN dataset under various operating conditions. Comparative analysis with the ASR system demonstrates that the proposed method consistently exhibits superior performance and robustness across different experimental conditions.