电机噪声信号特征提取与故障预测模型建立  

Motor Noise Signal Feature Extraction and Fault Prediction Model Establishment

作　　者：谢尚海 刘吉乔 李敬兆[1] Xie Shanghai;Liu Jiqiao;Li Jingzhao(School of Electrical and Information Engineering,Anhui University of Science and Technology,Huainan 232000,China)

机构地区：[1]安徽理工大学电气与信息工程学院,安徽淮南232000

出　　处：《煤矿机械》2023年第11期172-176,共5页Coal Mine Machinery

基　　金：国家自然科学基金资助项目(61170060)。

摘　　要：针对电机传统故障检测方法中存在特征信息提取效果差、故障诊断精度低的问题,提出了一种基于自适应噪声完备集合经验模态分解(CEEMDAN)与多尺度排列熵(MPE)结合的卷积神经网络-门控循环单元(CNN-GRU)故障诊断方法。研究结果表明,CEEMDAN-MPE方法能有效地克服经验模态分解(EMD)和集合经验模态分解(EEMD)方法产生模态混叠等问题。与卷积神经网络(CNN)、长短时记忆(LSTM)、CNN-LSTM模型相比较,准确率分别高出了23.92%、16.61%、10.12%。这说明CNN-GRU模型具有更高的预测精度。Aiming at the problems of poor feature information extraction and low fault diagnosis accuracy in the traditional fault detection methods of motor,a convolutional neural networks-gate recurrent unit(CNN-GRU)fault diagnosis method based on the combination of complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)and multiscale permutation entropy(MPE)was proposed.The research results show that the CEEMDAN-MPE method can effectively overcome the problems such as empirical mode decomposition(EMD)and ensemble empirical mode decomposition(EEMD)method generating modal blending.Compared with convolutional neural networks(CNN),long shortterm memory(LSTM)and CNN-LSTM models,the accuracy is 23.92%,16.61%and 10.12%higher respectively.This indicates that the CNN-GRU model has higher prediction accuracy.

关 键 词：CEEMDAN 故障诊断 音频信号 MPE 健康预测 CNN-GRU 

分 类 号：TM307.1[电气工程—电机]