基于信息融合及双连接注意力残差网络的轴承故障诊断

Bearing fault diagnosis based on double-connected attention residual network and information fusion

针对单一传感器获取特征信息不足、深层卷积神经网络提取的故障特征缺乏重要性区分的问题,提出一种传感器信息融合与双连接注意力残差网络相结合的轴承故障诊断方法。首先,对从不同位置采集的振动信号,采用传感器信息融合策略转换成多通道输入,以获取更全面的故障特征信息;其次,针对融合后的多通道输入,设计了一种双连接残差网络来增强模型对特征信息的提取,同时引入通道注意力机制模块,对输出的特征赋予不同权重,使模型提取到的特征更具有鉴别性,改善了识别准确率。将所提方法应用于复杂工况下的轴承数据集,试验表明,该方法在变工况和噪声环境干扰下,具有良好的故障分类精度。

Aiming at problems of insufficient feature information obtained by a single sensor and lack of importance distinction of the fault features extracted by the deep convolutional neural network,a fault diagnosis method combining double-connected attention residual network with sensor information fusion was proposed.Firstly,aiming to obtain more comprehensive fault feature information,the vibration signals collected from different locations were converted into multi-channel inputs using a sensor information fusion strategy.Secondly,a double-connected residual network was designed for the fused input to enhance the capability of fault feature extraction,and the channel attention mechanism module was introduced to assign different weights to the output features,so that the features extracted by the model became more discriminative,thereby the fault diagnosis accuracy can be improved.The proposed method was applied to the bearing data set under complex operating conditions.Experimental results show that the proposed method has good classification accuracy under variable working conditions and noise interference.