| Wen Juan (文 娟)* ** ***,Wu You* ***,Song Yang* ***,Pan Baisong* ***.[J].高技术通讯(英文),2026,32(2):109~120 |
|
| Remaining useful life prediction for bearings based on Transformer-BiLSTM network optimized by the Newton-Raphson-based optimizer |
| |
| DOI：10. 3772 / j. issn. 1006-6748. 2026. 02. 001 |
| 中文关键词: |
| 英文关键词: Transformer, bidirectional long short-term memory, Newton-Raphson-based optimizer, remaining useful life prediction, bearing |
| 基金项目: |
| Author Name | Affiliation | | Wen Juan (文 娟)* ** *** | (* College of Mechanical Engineering, Zhejiang University of Technology, Hangzhou 310014, P. R. China)
(** Technology and Equipment of Rail Transit Operation and Maintenance Key Laboratory of Sichuan Province, Chengdu 610031, P. R. China)
(*** Key Laboratory of Special Purpose Equipment and Advanced Processing Technology,Ministry of Education, Zhejiang University of Technology, Hangzhou 310014, P. R. China) | | Wu You* *** | | | Song Yang* *** | | | Pan Baisong* *** | |
|
| Hits: 133 |
| Download times: 131 |
| 中文摘要: |
| |
| 英文摘要: |
| The rolling bearing is one of the critical components in mechanical equipment, and predicting its remaining useful life (RUL) is of great significance in enterprise production processes. While deep learning-based approaches have achieved great success for bearing prognosis, most of them are not capable of mining both global and local information from the run-to-failure data. In addition,hyperparameters such as the number of hidden layer neurons, learning rate, and regularization parameters in neural networks still rely heavily on manual experience for setting. To address these issues, a novel framework for predicting the RUL of bearings based on the Transformer and the bidirectional long short-term memory ( Transformer-BiLSTM) is proposed, and the Newton-Raphson-based optimizer (NRBO) is introduced to determine the crucial parameters of the network. Firstly,degradation sensitive features are extracted and selected from the raw vibration signals, forming the input for the prediction model. Secondly, the mean absolute error (MAE) between the predicted and actual values is utilized as the fitness function of the NRBO algorithm to optimize the Transformer-BiLSTM model, searching for the optimal values of the key hyperparameters. Finally, the optimized model is used for RUL prediction, and its performance is validated on publicly available datasets. The results demonstrate that the proposed method can achieve the optimal hyperparameter combination without relying on empirical guidance. Compared with the unoptimized model, the optimized prediction model reduces the MAE and root mean squared error (RMSE) by 6. 50% and 9. 91% , respectively. |
|
View Full Text
View/Add Comment Download reader |
| Close |
|
|
|
