噪声与振动控制2025，Vol.45Issue(2)：90-96,104,8.DOI:10.3969/j.issn.1006-1355.2025.02.015
改进自注意力机制的滚动轴承寿命预测方法
Residual Life Prediction Method of Rolling Bearings Based on Improved Self-attention Mechanism
摘要
Abstract
In order to solve the problem of poor accuracy of the existing convolution and recurrent models in prediction of the residual life(RL)of rolling bearings,a new RL prediction model based on improved self-attention mechanism was proposed.Firstly,aiming at the issue of high memory occupation and noisy signal information of the self-attention mecha-nism in the transformer model,a probwindow based multi-head self-attention(PW-MSA)was proposed on the basis of the window based multi-head self-attention(W-MSA).Then,in order to address the issue of mismatch of multi-head information and lack of local information,the talking head method was employed to achieve the communication of multi-head informa-tion Finally,the depthwise separable convolution was added to the feedforward neural network layer to extract local informa-tion,thereby the model's prediction accuracy was increased.The self-attention models before and after the improvement were analyzed and compared each other using the PHM2012 bearing dataset,and compared with other advanced prediction models.The results demonstrate that the improved self-attention model can raise the prediction accuracy by 13.04%.关键词
故障诊断/滚动轴承/剩余使用寿命预测/概率窗口自注意力机制/Transformer模型Key words
fault diagnosis/rolling bearing/residual life prediction/probwindow based self-attention/Transformer model分类
信息技术与安全科学引用本文复制引用
史竞成,吴占涛,程军圣,杨宇..改进自注意力机制的滚动轴承寿命预测方法[J].噪声与振动控制,2025,45(2):90-96,104,8.基金项目
国家重点研发计划资助项目(2020YFB2009602) （2020YFB2009602）
国家自然科学基金资助项目(52275103) （52275103）
深圳自然科学基金重点资助项目(JCYJ20210324115413036) （JCYJ20210324115413036）
湖南省教育厅科学研究资助项目(21A0017) （21A0017）
