Scholarlink

<p>The remaining useful life (RUL), defined as the estimated time remaining in a machine's life cycle or usage until it requires repair or replacement, is a critical parameter in predictive maintenance. Accurate prognosis of RUL is a primary goal of predictive-maintenance algorithms. This research presents a novel data-driven approach for estimating RUL, utilising a hybrid architecture combining deep convolutional neural networks (DCNN) with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) layers (CNN-LSTM-GRU). A key advantage of this deep learning-based method is that it eliminates the need for manual feature extraction or selection and does not require prior knowledge of machine health prognostics or signal processing. The proposed 1DCNN-LSTM-GRU model achieved superior Remaining Useful Life (RUL) prediction accuracy on the C-MAPSS datasets, outperforming methods like SVM, Random Forest, and standard LSTMs. It yielded the lowest Root Mean Square Error (RMSE), notably 12.85 for FD001 and 12.93 for FD003. The findings demonstrate that this CNN-LSTM-GRU strategy offers a distinctive and promising approach to RUL estimation.</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

Estimation of the remaining useful life of aircraft engines using a CNN-LSTM-GRU hybrid model

  • 期刊论文

  • 发表日期：

    2025年8月8日

查看全文
  • Krishna Chandra Patra,
  • Rabinarayan Sethi,
  • Dhiren Kumar Behera

摘要

The remaining useful life (RUL), defined as the estimated time remaining in a machine's life cycle or usage until it requires repair or replacement, is a critical parameter in predictive maintenance. Accurate prognosis of RUL is a primary goal of predictive-maintenance algorithms. This research presents a novel data-driven approach for estimating RUL, utilising a hybrid architecture combining deep convolutional neural networks (DCNN) with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) layers (CNN-LSTM-GRU). A key advantage of this deep learning-based method is that it eliminates the need for manual feature extraction or selection and does not require prior knowledge of machine health prognostics or signal processing. The proposed 1DCNN-LSTM-GRU model achieved superior Remaining Useful Life (RUL) prediction accuracy on the C-MAPSS datasets, outperforming methods like SVM, Random Forest, and standard LSTMs. It yielded the lowest Root Mean Square Error (RMSE), notably 12.85 for FD001 and 12.93 for FD003. The findings demonstrate that this CNN-LSTM-GRU strategy offers a distinctive and promising approach to RUL estimation.