基于注意力时间卷积网络和双向门控循环单元的轴承故障诊断

Bearing Fault Diagnosis Based on Attention Temporal Convolutional Network and Bidirectional Gated Recurrent Unit

针对传统特征提取方法依赖人工经验以及传统神经网络未充分利用时间序列信息的问题,首先通过时间卷积网络的空洞因果卷积、随机丢弃层和残差结构跨时间步提取不同振动信号的特征;然后引入注意力机制获取关键信息,实现特征优化选择;再利用双向门控循环单元捕捉长期依赖关系;最后通过归一化指数函数进行故障分类。实验结果表明,在不同训练样本比例下,该方法的识别精度高于一维卷积神经网络、双向长短期记忆网络、双向循环神经网络;用该方法能够有效识别轴承故障类型,且模型的泛化能力较强。

In order to solve the problem that traditional feature extraction methods rely on artificial experience and traditional neural networks do not make full use of time series information,firstly,the features of different vibration signals are extracted across time steps through the dilated causal convolution,random discard layer and residual structure of the temporal convolutional network;then,the attention mechanism is introduced to obtain key information and optimize feature selection;bidirectional gated recurrent units are used to capture long-term dependencies;finally,the normalized exponential function is used to classify faults.The experimental results show that the recognition accuracy of this method is higher than that of one-dimensional convolutional neural network,bidirectional long short-term memory network and bidirectional recurrent neural network under different training sample proportions;this method can identify bearing fault types effectively,and the model has strong generalization ability.