采用注意力DBN-GRU的滚动轴承剩余寿命预测方法

Method of Predicting Remaining Useful Life of Rolling Bearing Adopting Attention Mechanism DBN-GRU

针对滚动轴承剩余寿命预测中特征提取困难以及预测精度低等问题,提出一种引入注意力机制的深度置信网络(DBN)和门控循环单元(GRU)组合方法用于滚动轴承寿命预测。首先利用Nadam优化器改进DBN网络,直接从原始数据中挖掘出深度特征;其次将注意力机制引入改进DBN中对深度特征进行权重分配形成全局特征;最后通过GRU网络进行寿命预测试验。结果表明,所提方法的预测结果均方根误差和平均相对误差分别比ADBN-LSTM、DBN-GRU、DBN-LSTM和DBN-BP平均低36.81%和34.15%,预测寿命的准确率更高,为滚动轴承健康管理提供了实际价值。

Aiming at the difficulty of feature extraction and low prediction accuracy in predicting remaining life of rolling bearings,a combined method of attention mechanism deep belief network(DBN)and gated recurrent unit(GRU)was proposed to apply to predict remaining life of rolling bearing.Firstly,the Nadam optimizer is used to improve the DBN network,and the deep features are directly mined from the original data;Secondly,the attention mechanism is introduced into the improved DBN to distribute the weight of deep features to form global features;Finally,and the life prediction experiment is carried out through GRU network.The results show that the root mean square error and mean relative error of the proposed method are 36.81%and 34.15%lower than those of ADBN-LSTM、DBN-GRU、DBN-LSTM and DBN-BP respectively,and the accuracy of life prediction is higher,which provides practical value for health management of rolling bearing.