基于多通道卷积神经网络的柴油机复合故障诊断

Compound fault diagnosis for diesel engine based on multi-channel convolutional neural network

针对复合故障诊断精度较低的问题,开展了柴油机多故障模拟实验,构建了基于AlexNet改进的多通道二维卷积神经网络模型,采用短时傅里叶变换将一维振动信号转换为二维时频图,导入构建的模型进行训练,实现特征自适应提取的故障诊断。将诊断结果与单通道卷积神经网络诊断结果比较发现:单通道卷积神经网络诊断只有在测点设置靠近故障源的情况下才能够获得较高的故障诊断准确率,否则诊断准确率明显降低,且复合故障诊断精度较低;多通道卷积神经网络的单故障和复合故障诊断精度均得到了提升,其中复合故障诊断精度提升了11.4%。

Aiming at the problem of low diagnosis accuracy of compound faults,a diesel engine multi-fault simulation experiment was carried out in which a multi-channel two-dimensional convolutional neural network model improved by AlexNet was constructed and one-dimensional vibration signals were converted into two-dimensional time-frequency graphs by short-time Fourier transform,which was imported into the constructed model for training,so that fault diagnosis with adaptive feature extraction was realized.Comparison of the diagnosis results indicates that the accuracy of the single channel convolutional neural network can be higher only when the measurement point is set close to the fault source and otherwise,the accuracy is significantly reduced and thus the diagnosis accuracy of the compound fault is low;that the diagnosis accuracy of the multi-channel convolutional neural network is improved,i.e.,the accuracy of compound fault diagnosis can be improved by 11.4%.