Abstract: In view of the problems of traditional bearing life prediction methods such as excessive dependence on prior knowledge,lack of adaptability and large prediction error caused by difficult extraction of degradation characteristics,a bearing residual life prediction method based on Deep Residual Shrinkage Network (DRSN) and Bidirectional Long-Short-Term Memory network (BiLSTM) with adaptive feature extraction was proposed.Without any prior knowledge,DRSN was used to automatically learn the characteristics of the original signal of the bearing,extract the degradation characteristics and construct the health index.Then,the number of hidden layer neurons and learning rate of BiLSTM were optimized by sparrow search algorithm,and the remaining life prediction model of bearing was established based on the optimized BiLSTM.The performance of health index extracted by DRSN,residual network and mean feature and different bearing residual life prediction models were compared.The experimental results showed that the health index extracted by DRSN network had the best performance,and the error of the optimized BiLSTM bearing residual life prediction model was the smallest.The root means square errors of the three bearing residual life prediction models based on the optimized BiLSTM,BiLSTM and Long-Short-Term Memory network (LSTM) were 1.41%,2.71% and 5.64% respectively,which verified the effectiveness of the proposed method.

Key words: deep residual shrinkage network, bidirectional long short term memory network, residual life prediction, sparrow search algorithm

摘要： 针对传统轴承寿命预测方法过度依赖先验知识、缺乏自适应性及退化特征难以提取导致的预测误差大的问题,提出一种自适应特征提取的基于深度残差收缩网络(DRSN)和双向长短时记忆网络(BiLSTM)的轴承剩余寿命预测方法。首先,无需任何先验知识利用DRSN对轴承原始信号进行自动特征学习,提取退化特征并构建健康指标;然后,采用麻雀搜索算法优化BiLSTM隐藏层神经元个数和学习率,基于优化的BiLSTM网络建立轴承剩余寿命预测模型;最后,进行对比实验验证:分别对比DRSN、残差网络、均值特征3种方法提取的健康指标的性能和不同的轴承剩余寿命寿命预测模型进行对比实验。实验结果表明DRSN网络提取的健康指标性能最优,同时基于优化后的BiLSTM轴承剩余寿命预测模型的误差最小,基于优化后BiLSTM、BiLSTM和长短时记忆网络(LSTM)的3种轴承剩余寿命预测模型的均方根误差分别为1.41%、2.71%、5.64%,验证了方法的有效性。

关键词: 深度残差收缩网络, 双向长短时记忆网络, 剩余寿命预测, 麻雀搜索算法