Scholarlink

<p>Remaining useful life (RUL) prediction of aircraft engines is of great significance for the safety and reliability of aircraft operations. However, the high feature dimension and noise of the raw data cause difficulties for existing methods in extracting long sequence time features and allocating weights. In this study, we propose a RUL prediction network named BLTTNet with enhanced feature extraction ability to address these difficulties. We utilize the efficient implicit feature extraction capability of BiLSTM to represent high-dimensional features. Then, DCEFormer and TCN are used to process the global and local information of the time series respectively. Specifically, DCEFormer with a Transformer structure enhances the allocation of feature weights by processing the contributions of different features in both the time step dimension and the sensor dimension, thereby improving the accuracy of RUL prediction for mechanical equipment. Meanwhile, an adaptive fusion method is employed to fuse the model. Finally, we conducted experiments on CMAPSS dataset. On the FD002 subset, it achieved excellent RMSE and score performance, with average improvements of 5.91 and 13.23% respectively.</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

BLTTNet: feature fusion based on BiLSTM-Transfomer-TCN for prediction of remaining useful life of aircraft engines

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月29日

查看全文
  • Yixu Yang,
  • Xiaoying Su,
  • Chaoyong Wang,
  • Hongxi Liu,
  • Kunhao Fu,
  • Tao Xie,
  • Zishuo Zhang

摘要

Remaining useful life (RUL) prediction of aircraft engines is of great significance for the safety and reliability of aircraft operations. However, the high feature dimension and noise of the raw data cause difficulties for existing methods in extracting long sequence time features and allocating weights. In this study, we propose a RUL prediction network named BLTTNet with enhanced feature extraction ability to address these difficulties. We utilize the efficient implicit feature extraction capability of BiLSTM to represent high-dimensional features. Then, DCEFormer and TCN are used to process the global and local information of the time series respectively. Specifically, DCEFormer with a Transformer structure enhances the allocation of feature weights by processing the contributions of different features in both the time step dimension and the sensor dimension, thereby improving the accuracy of RUL prediction for mechanical equipment. Meanwhile, an adaptive fusion method is employed to fuse the model. Finally, we conducted experiments on CMAPSS dataset. On the FD002 subset, it achieved excellent RMSE and score performance, with average improvements of 5.91 and 13.23% respectively.