Scholarlink

<p>This paper presents a new deep learning method for predicting bearing remaining useful life (RUL). Current prognostics and health management (PHM) techniques struggle with complex degradation patterns in bearings. The proposed model features a hybrid architecture that includes Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), Bidirectional Gated Recurrent Units (BiGRU), and a Temporal Convolutional Network (TCN). This design uses CNNs to extract spatial features from vibration signals. Meanwhile, BiLSTM and BiGRU capture dependencies in both past and future data, while TCN models long-term temporal dependencies. The study explains the feature extraction process through time-domain and frequency-domain analysis, showing how these features feed into hybrid architecture. Experimental results reveal a significant improvement in RUL prediction accuracy compared to traditional methods, such as recurrent neural networks (RNNs), support vector machines (SVMs), and long short-term memory (LSTM) models. The model performs well in predicting bearing degradation under varying operational conditions, especially when coupled degradation modes are present. Its strength lies in capturing complex relationships within data from industrial machinery, improving the accuracy and reliability of RUL predictions. The key novelty of this work is the synergistic integration of CNN for spatial feature extraction, BiLSTM for bidirectional temporal modeling, and TCN for long-range dependency capture within a unified health indicator construction and RUL prediction pipeline—achieving state-of-the-art performance across varying operating conditions without manual feature engineering. This innovative approach offers a robust prognostic tool for industrial applications, effectively overcoming current limitations.</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

Robust bearing remaining useful life prediction using a hybrid deep learning framework across operating conditions

  • 期刊论文

  • 发表日期：

    2026年2月20日

查看全文
  • Anping Wan,
  • Fei Zhang,
  • Khalil AL-Bukhaiti,
  • Xiaomin Cheng,
  • Xiaosheng Ji

摘要

This paper presents a new deep learning method for predicting bearing remaining useful life (RUL). Current prognostics and health management (PHM) techniques struggle with complex degradation patterns in bearings. The proposed model features a hybrid architecture that includes Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), Bidirectional Gated Recurrent Units (BiGRU), and a Temporal Convolutional Network (TCN). This design uses CNNs to extract spatial features from vibration signals. Meanwhile, BiLSTM and BiGRU capture dependencies in both past and future data, while TCN models long-term temporal dependencies. The study explains the feature extraction process through time-domain and frequency-domain analysis, showing how these features feed into hybrid architecture. Experimental results reveal a significant improvement in RUL prediction accuracy compared to traditional methods, such as recurrent neural networks (RNNs), support vector machines (SVMs), and long short-term memory (LSTM) models. The model performs well in predicting bearing degradation under varying operational conditions, especially when coupled degradation modes are present. Its strength lies in capturing complex relationships within data from industrial machinery, improving the accuracy and reliability of RUL predictions. The key novelty of this work is the synergistic integration of CNN for spatial feature extraction, BiLSTM for bidirectional temporal modeling, and TCN for long-range dependency capture within a unified health indicator construction and RUL prediction pipeline—achieving state-of-the-art performance across varying operating conditions without manual feature engineering. This innovative approach offers a robust prognostic tool for industrial applications, effectively overcoming current limitations.