基于ResNet-ABiLSTM的滚动轴承剩余寿命预测

Residual life prediction of rolling bearing based on ResNet-ABiLSTM

传统数据驱动的方法过度依赖先验知识且特征提取能力不足,从而导致预测精度不高等后果。针对这一问题,提出了一种带有自注意力机制(SAM)的残差网络(ResNet)与双向长短时记忆网络(BiLSTM)结合的剩余使用寿命(RUL)预测方法(ResNet-ABiLSTM)。首先,对采集的原始监测信号进行了标准化处理,并采用滑窗法对处理后的数据进行了重采样,以实现数据的扩充目标;然后,通过采用残差网络和双向长短时记忆网络,分别提取了数据空间维度和时间维度上的深层特征,同时引入了自注意力机制,关注了时空维度上反映设备退化趋势的更重要的特征;最后,采用PHM2012轴承数据集对预测效果进行了验证,并将其结果与CNN-LSTM、ResNet-BiLSTM、HI-GRNN、CNN-HI、ResNet-CBAM、DRN-BiGRU等方法的预测结果进行了对比分析。研究结果表明:采用ResNet-ABiLSTM方法的两项误差值(RMSE、MAE)分别取得了0.037、0.029的最低值,其效果显著优于其他对比方法;该结果验证了ResNet-ABiLSTM方法对轴承RUL预测的准确性和有效性。

Aiming at the problem that traditional data-driven methods over-rely on prior knowledge and lack feature extraction ability,which leads to low prediction accuracy,a residual service life prediction method(ResNet-ABiLSTM)combining residual network(ResNet)with self-attention mechanism(SAM)and bidirectional long and short term memory network(BiLSTM)was proposed.Firstly,the original monitoring signals were standardized and resampled by sliding window method to realize data expansion.Then,the residual network and the bidirectional long and short term memory network were used to extract the deep features of the data in the spatial dimension and the temporal dimension respectively,and the self-attention mechanism was introduced to focus on the more important features reflecting the equipment degradation trend in the spatial and temporal dimension.Finally,the PHM2012 bearing data set was used for verification,and the results were compared with the predicted results of CNN-LSTM,ResNet-BiLSTM,HI-GRNN,CNN-HI,ResNet-CBAM,DRN-BiGRU and other methods.The results show that the two error values(RMSE and MAE)of ResNet-ABiLSTM method are 0.037 and 0.029,respectively,which are significantly superior to other comparison methods.The results verify the accuracy and effectiveness of the proposed method for predicting bearing RUL.