基于改进Transformer的滚动轴承剩余寿命预测方法

Remaining useful life prediction method for rolling bearing based on enhanced Transformer

针对现有的滚动轴承剩余使用寿命预测方法存在预测准确度不足、训练效率不高等问题,提出一种时频分析结合改进Transformer的轴承剩余使用寿命预测方法。首先用短时傅里叶变换提取轴承的时频特征,为了改善Transformer的特征提取能力,研究了基于膨胀因果卷积的可变长度数据分析结构,并设计了自适应位置编码模块替代Transformer的传统编码方式,改进的模型增强了对时频数据的分析能力,实现了高效、准确的端到端的滚动轴承剩余寿命预测。在PHM2012轴承数据集上的实验结果表明提出的方法的效率比LSTM高20%,同时预测精度相比于多种现有传统方法均具有16%以上的提升。

To address the problems of insufficient prediction accuracy and low training efficiency of existing rolling bearing remaining life prediction methods,a time⁃frequency analysis combined with an improved Transformer is proposed for bearing remaining life prediction.First,the short⁃time Fourier transform is used to extract the time⁃frequency features of complex vibration signals.To improve the feature extraction capability of the Transformer,a variable length data analysis structure based on dilated causal convolution is investigated,and an adaptive position coding module is designed to replace the traditional coding method of the Transformer.The improved model enhances the analysis of time⁃frequency data and achieves an efficient and accurate end⁃to⁃end prediction of the remaining life of rolling bearings.Experimental results on the PHM2012 bearing dataset show that the proposed method has higher training efficiency and better prediction accuracy than traditional methods.