基于并行双向时间卷积网络和双向长短期记忆网络的轴承剩余使用寿命预测方法

在基于深度学习的轴承剩余使用寿命(RUL)预测方法中,时间卷积网络(TCN)忽略了振动数据中未来时间信息的重要性,长短期记忆网络(LSTM)难以有效地学习振动数据的长时间序列特征.针对以上问题,提出一种基于并行双向时间卷积网络(Bi-TCN)和双向长短期记忆网络(Bi-LSTM)的轴承RUL预测方法.首先,对多传感器数据进行归一化处理,并将每个传感器数据进行通道合并,实现多传感器数据的高效融合;然后,采用Bi-TCN和Bi-LSTM构建并行的双分支特征学习网络,其中Bi-TCN提取数据的双向长时间序列特征, Bi-LSTM提取数据的时间相关特征;同时,设计一种特征融合注意力机制,该机制分别计算Bi-TCN和Bi-LSTM的输出权重,以实现两种网络输出特征的自适应加权融合;最后,融合特征通过全连接层并输出轴承RUL的预测结果.利用西安交通大学轴承数据集和PHM 2012轴承数据集进行RUL预测实验,实验结果表明,与其他先进的预测方法相比,所提出方法可以准确预测更多类型轴承的RUL,同时具有更低的预测误差.

Remaining useful life prediction method for bearing based on parallel bidirectional temporal convolutional network and bidirectional long and short-term memory network

In remaining useful life(RUL) prediction methods for bearings based on deep learning, temporal convolutional networks(TCNs) does not consider the future time information of vibration data, long and short-term memory(LSTM) networks are difficult to learn long time series data features effectively. To solve the above problems, a bearing RUL prediction method based on the parallel bidirectional temporal convolutional network and bidirectional long and short-term memory network is proposed. First, the multi-sensor data are normalized, and the data of each sensor are merged by channel to achieve efficient fusion of multi-sensor data. Then, a parallel dual network structure is constructed by using the Bi-TCN and Bi-LSTM, in which the Bi-TCN goes to learn the bi-directional long time series features and the Bi-LSTM goes to learn the time-dependent features, so the parallel dual network structure can learn richer vibration signal features. Meanwhile, a feature fusion attention mechanism is developed to fuse the output features of the dual network structure, which calculates the output weights of the Bi-TCN and Bi-LSTM to achieve adaptive weighted fusion of the output features. Finally, the fused features are passed through the fully connected layer to output the prediction results of the bearing RUL. RUL prediction experiments are conducted using Xián Jiaotong University bearing dataset and PHM 2012 bearing dataset respectively. The results show that, compared with the advanced prediction methods, the proposed method can accurately predict the RUL of more types of bearings and has lower prediction errors.