基于多尺度卷积神经网络的滚动轴承故障诊断方法研究

Research on Fault Diagnosis Method of Rolling Bearings Based on Multi-Scale Convolution Neural Network

针对滚动轴承提取故障特征时容易被噪声干扰、依赖人工经验等问题,本文提出一种基于多尺度卷积神经网络(MCNN)的滚动轴承故障诊断方法。该方法首先利用多尺度数据预处理方法滤除振动信号中的高频噪声并进行数据增强,然后把多个时间尺度的振动信号输入到不同尺度下的一维卷积神经网络中,利用不同尺度的CNN提取滚动轴承振动信号中的不同特征,并将特征进行融合,从而得到滚动轴承更为全面的特征。该方法通过转子实验台验证,实验结果表明基于MCNN的滚动轴承故障诊断方法可以准确诊断出滚动轴承的故障类型,准确率高达100%。

Aiming at the problems that rolling bearings are easily disturbed by noise and rely on manual experience when extracting fault features,a rolling bearing fault classification method based on multi-scale convolution neural network was proposed in this chapter.In this method,the vibration signal is filtered out by multi-scale data preprocessing method and the data is enhanced,then the rolling bearing vibration signals of multiple time scales are input into one-dimensional convolution neural network with different scales.Different features of rolling bearing vibration signals are extracted by CNN of different scales,and the features are fused to obtain more comprehensive features of rolling bearings.The method was verified by the rotor test bench,and the experimental results showed that the rolling bearing fault classification method based on MCNN could accurately diagnose the fault types of rolling bearings,and the accuracy was 100%.