基于自注意力机制与LSTM的轴承质量状态监测方法

Bearing quality state monitoring method based on self-attention mechanism and LSTM

针对传统轴承质量状态监测方法复杂,手工特征设计明显,网络模型参数量大的问题,提出一种基于自注意力机制与LSTM的轴承质量状态监测方法。首先对原始轴承信号进行预处理,依次将其输入到一维卷积和LSTM网络中进行时序信号自适应特征学习,接着利用自注意力模块得到轴承信号的全局语义特征,最后输入到全连接层完成从特征域到类别域的映射。实验结果表明,在公开的CWRU数据集上,该模型与其他五种轴承质量状态识别模型相比,具有更少的参数量、更短的训练时间和较高的准确率,为轴承质量状态实时监测提供了便利。

Aiming at the problems of complex traditional bearing quality condition monitoring methods,obvious manual feature design and large number of network model parameters,a bearing quality condition monitoring method based on self-attention mechanism and LSTM was proposed.Firstly,the original bearing signals are preprocessed and input into one-dimensional convolutional and LSTM networks for adaptive feature learning of time sequence signals.Then,the global semantic features of bearing signals are obtained by using self-attention module.Finally,the bearing signals are input into the fully connected layer to complete the mapping from feature domain to class domain.The experimental results show that compared with the other five bearing quality status recognition models on the open CWRU data set,the proposed model has fewer parameters,shorter training time and higher accuracy,which provides convenience for real-time monitoring of bearing quality status.