卷积门控循环残差网络在轴承故障诊断中的研究

Research on convolution gating cyclic residual network for bearing fault diagnosis

针对滚动轴承故障特征较难提取及许多深度学习方法因模型简单而导致准确率偏低的问题,提出一种基于残差网络的门控循环网络(GRU),该算法可以减少时序信息的丢失及解决由于网络较深而出现性能下降的问题。该模型包含2个卷积层、2个GRU层、1个残差块以及1个输出层,其先利用具有强大特征提取能力的卷积神经网络(CNN)提取轴承振动信号中的信息,然后将提取到的信息输入GRU中以保证时序信息不丢失,再通过残差模块解决神经网络深度较深问题,最后通过输出层输出结果。结果表明:该方法可以一次性诊断多种轴承的不同位置及不同尺寸的故障,且对比其他深度学习网络,该算法具有更高的准确性。

Aiming at the problem that it is difficult to extract the fault features of rolling bearing and the low accuracy of many deep learning methods due to the simple model,a gated cyclic network based on residual network is proposed.This algorithm can reduce the loss of timing information and solve the problem of performance degradation due to the deep network.The model includes two convolution layers,two gated cyclic network layers,one residual block and one output layer.Firstly,the convolution neural network with strong feature extraction ability is used to extract the information in the bearing vibration signal,and then the extracted information is input into the gated cyclic network to ensure that the timing information is not lost,and then the deep problem of the neural network is solved through the residual module,Finally,the results are output through the output layer.The experimental results show that this method can diagnose the faults of different positions and sizes of various bearings at one time,and the comparative experiments show that the algorithm has higher accuracy and stability than other deep learning networks.