华中科技大学学报（自然科学版）2026，Vol.54Issue(5)：1-8,8.DOI:10.13245/j.hust.240918
基于CNN-Transformer encoder-BiLSTM模型的轴承剩余寿命预测
Remaining life prediction for bearing based on CNN-Transformer encoder-BiLSTM model
摘要
Abstract
To address the problem that the bearing degradation process exhibited strong nonlinearity and long-term dependencies were difficult to be effectively modeled under complex operating conditions,a remaining useful life prediction method based on an improved convolutional neural network(CNN)-Transformer encoder-bidirectional long short-term memory(BiLSTM)model was proposed.In this method,the CNN block was used to focus on local information for better feature extraction.The improved Transformer encoder was used to introduce three different attention masks so that the attention calculation process only focused on the important part of the long-term information.The BiLSTM was used to focus on long-term dependencies of all information.The accuracy of the model was validated on the C-MAPSS and XJTU-SY datasets.Experimental results show that the model achieves better prediction performance than other existing methods,and demonstrates better stability after Gaussian noise is added.关键词
剩余寿命预测/注意力掩码机制/卷积神经网络/Transformer encoder/双向长短期记忆网络Key words
remaining life prediction/attention mask/convolutional neural network/Transformer encoder/BiLSTM分类
机械制造引用本文复制引用
张代林,孔康,朱晨曦,杨奕婷..基于CNN-Transformer encoder-BiLSTM模型的轴承剩余寿命预测[J].华中科技大学学报（自然科学版）,2026,54(5):1-8,8.基金项目
山东省重点研发计划资助项目(2024TSGC0186) （2024TSGC0186）
国家重点研发计划资助项目(2023YFB3406604). （2023YFB3406604）
