基于DCNN网络及Self-Attention-BiGRU机制的轴承剩余寿命预测

Residual life prediction of bearings based on DCNN network and Self-Attention-BiGRU mechanism

深度神经网络在剩余寿命预测(RUL)领域得到了广泛的应用。传统的滚动轴承寿命预测模型存在预测精确度较低、鲁棒性较弱的问题。为了进一步提升预测模型的精确度以及鲁棒性,提出了一种融合深度卷积神经网络(DCNN)、双向门控循环单元(BiGRU)以及自注意力机制(Self-Attention)三种模块的滚动轴承剩余使用寿命预测模型。首先,利用DCNN网络对原始振动信号的时域特征、频域特征进行了提取;然后,使用不确定量化的方法对提取到的特征进行了评价和筛选,利用筛选过后的特征构建了新的替代特征集;最后,利用Self-Attention-BiGRU网络对轴承的剩余使用寿命进行了预测,并在IEEE PHM2012数据集上进行了验证。实验结果表明:相较于BiGRU、GRU和BiLSTM三种模型的预测结果,基于DCNN及Self-Attention-BiGRU方法的预测结果最优,两项误差值:平均绝对误差(MAE)、均方根误差(RMSE)最低,其中工况一的一号轴承RUL预测的MAE值相较于BiGRU、GRU以及BiLSTM网络分别下降了7.0%、7.4%和6.5%,RMSE值相较于其他三种模型分别下降了7.6%、8.4%和6.9%,预测的Score值最高,分值为0.985。通过不同数据集的划分,证明了该方法在轴承RUL预测时的强鲁棒性。实验结果验证了基于DCNN网络及Self-Attention-BiGRU模型在轴承剩余使用寿命预测中的有效性。

Deep neural networks have been widely used in the field of remaining useful life prediction(RUL).The traditional rolling bearing life prediction model has low prediction accuracy and poor robustness.In order to further improve the accuracy and robustness of the prediction model,a rolling bearing residual life prediction model was proposed by integrating three modules,namely,deep convolutional neural network(DCNN),bidirectional gated recurrent unit(BiGRU),and self-attention mechanism(Self-Attention).Firstly,the time-domain features and frequency-domain features of the original vibration signals were extracted using the DCNN network.Then,the extracted features were evaluated and screened using uncertainty quantification,and the screened features were used to construct a new alternative feature set.Finally,the remaining service life of the bearing was predicted using Self-Attention-BiGRU network.The proposed method was validated on the IEEE PHM2012 dataset.The experimental results show that the DCNN network and Self Attention-BiGRU method provides the optimal prediction results,with the lowest two error values:mean absolute error(MAE),root mean squared error(RMSE)compared to the prediction results of the three models BiGRU,GRU and BiLSTM.Among them,the MAE value predicted by the RUL of the No.1 bearing in Case I is respectively decreased by 7.0%,7.4%and 6.5%compared to the BiGRU,GRU,and BiLSTM networks.The RMSE values is respectively decreased by 7.6%,8.4%and 6.9%compared to the three other models,and the highest Score value is predicted with a score of 0.985.The robustness of the proposed method for bearing RUL prediction is demonstrated by the partitioning of different datasets.The experimental results validate the effectiveness of the DCNN network and Self-Attention-BiGRU based model in bearing remaining useful life prediction.