基于多通道数据融合及CNN的电机故障诊断方法

On the Motor Fault Diagnosis Method Based on Multi-channel Data Fusion and CNN

为解决因电机结构复杂、信号非平稳等因素导致电机故障诊断困难问题及传统故障诊断算法对专家经验的依赖,提出一种基于多通道数据融合与卷积神经网络(CNN)相结合的电机故障诊断方法。该方法首先采集电机驱动端的振动信号和定子电流信号并对其进行时频域转换,再将两者频域信号进行归一化处理并转变为二维图谱数据,最后构建CNN网络模型,确定网络层数、学习率等超参数,并将样本输入模型进行故障特征提取和分类诊断。结果表明,在合适的参数下采用该方法的电机故障诊断准确率为100%,对比单独采用振动信号或电流信号的传统故障诊断方法和1D-CNN模型,该方法能够更有效地对电机各类故障进行诊断。

In order to solve the problem of difficult motor fault diagnosis caused by complex motor structures and non-stationary signals,as well as the dependence of traditional fault diagnosis algorithms on expert experience,a fault diagnosis method for motors based on multi-channel data fusion and convolutional neural networks(CNN)was proposed.The method first collects vibration signals and stator current signals at the motor drive end and converts them into frequency domain signals,then normalizes the frequency domain signals of the two signals and converts them into two-dimensional spectrum data.Finally,a CNN network model is constructed,the hyperparameter such as network layers and learning rate are determined,and samples are input into the model for fault feature extraction and classification diagnosis.The results showed that the accuracy of motor fault diagnosis using this method under appropriate parameters is 100%.Compared with traditional fault diagnosis methods and the 1D-CNN model using vibration or current signals alone,this method can more effectively diagnose various motor faults.