Scholarlink

<p>Accurate Remaining Useful Life (RUL) prediction is crucial for maintaining the safety of critical equipment, such as turbofan engines. Existing approaches often fail to capture the complex degradation patterns present in multi-sensor data under diverse operating conditions. To overcome these limitations, this study proposes a deep learning model that integrates a Dual Attention Residual Dilated (DARD) module with a Bi-directional LSTM network. The DARD module extracts multi-scale temporal and spatial features, while the BiLSTM captures bidirectional temporal dependencies. An adaptive gated fusion mechanism further enhances prediction accuracy by dynamically weighting spatial and temporal features. Experimental results on the C-MAPSS dataset demonstrate that the proposed DARD-BiLSTM model achieves superior prediction accuracy and robustness, with average reductions of 10.79% in RMSE and 10.19% in Score compared to state-of-the-art methods. These results highlight the model’s potential for reliable predictive maintenance and health monitoring in real-world industrial scenarios.</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

DARD-BiLSTM with gated fusion for predicting the remaining useful life of turbofan engines

  • 期刊论文

  • 发表日期：

    2026年3月24日

查看全文
  • Liang Jiang,
  • Xinyang Zhang,
  • Keqing Wang,
  • Yonghong Zhang

摘要

Accurate Remaining Useful Life (RUL) prediction is crucial for maintaining the safety of critical equipment, such as turbofan engines. Existing approaches often fail to capture the complex degradation patterns present in multi-sensor data under diverse operating conditions. To overcome these limitations, this study proposes a deep learning model that integrates a Dual Attention Residual Dilated (DARD) module with a Bi-directional LSTM network. The DARD module extracts multi-scale temporal and spatial features, while the BiLSTM captures bidirectional temporal dependencies. An adaptive gated fusion mechanism further enhances prediction accuracy by dynamically weighting spatial and temporal features. Experimental results on the C-MAPSS dataset demonstrate that the proposed DARD-BiLSTM model achieves superior prediction accuracy and robustness, with average reductions of 10.79% in RMSE and 10.19% in Score compared to state-of-the-art methods. These results highlight the model’s potential for reliable predictive maintenance and health monitoring in real-world industrial scenarios.