基于小波变换和CNN的船用机械故障诊断

Fault diagnosis of marine machinery based on wavelet transform and CNN

针对船用机械故障特征自适应提取与智能化诊断问题,采用连续小波变换与卷积神经网络的船舶机械故障诊断方法。以船用风机为例,首先模拟船用机械不同故障并采集振动信号,通过连续小波变换将一维振动信号转化为特征图谱,其包含大量的时频信息。然后通过多次训练后,确定网络结构参数,建立卷积神经网络结构,将时频图作为卷积神经网络输入,挖掘更深层次的高度抽象的故障特征信息。最后在卷积神经网络的输出层接入softmax分类器,实现船用机械的故障诊断。实验结果表明:所提方法能准确识别故障类型,且具有较强的鲁棒性和泛化能力,诊断准确率可达99.3%。与集成经验模态分解、极限学习机故障诊断方法相比,该方法有更高的诊断精度。

Aiming at the problem of self-adaptive extraction and intelligent diagnosis of marine machinery fault features,a marine machinery fault diagnosis method using continuous wavelet transform and convolutional neural network is proposed.Taking marine wind turbines as an example,firstly,different faults of marine machinery are simulated and vibration signals are collected,and the one-dimensional vibration signals are converted into feature maps by continuous wavelet transform,which contain a large amount of timefrequency information.Then after several trainings,the network structure parameters are determined,the convolutional neural network structure is established,and the time-frequency graph is used as the input of the convolutional neural network to mine deeper and highly abstract fault feature information.Finally,the softmax classifier is connected to the output layer of the convolutional neural network to realize the fault diagnosis of marine machinery.The experimental results show that the proposed method can accurately identify fault types and has strong robustness and generalization ability,and the diagnostic accuracy can reach 99.3%.Compared with integrated empirical mode decomposition and extreme learning machine fault diagnosis methods,this method has higher diagnostic accuracy.