基于GGD-EfficientNet和声纹识别的风力发电机齿轮箱故障诊断  

FAULT DIAGNOSIS OF WIND TURBINE GEARBOXES BASED ON GGD-EFFICIENTNET AND VOICEPRINT RECOGNITION

作　　者：廖力达[1] 陈伟克 罗晓 舒王咏 张芝铭 代军 Liao Lida;Chen Weike;Luo Xiao;Shu Wangyong;Zhang Zhiming;Dai Jun(School of Energy and Power Engineering,Changsha University of Science and Technology,Changsha 410114,China)

机构地区：[1]长沙理工大学能源与动力工程学院,长沙410114

出　　处：《太阳能学报》2025年第4期570-578,共9页Acta Energiae Solaris Sinica

基　　金：湖南省自然科学基金(2024JJ9181)。

摘　　要：针对风力发电机齿轮箱齿轮故障时的噪声提出一种基于分组全局上下文网络(GE-GCNet)与深度可分离卷积(DSCConv)结合的效率神经网络(GGD-EfficientNet)和声纹识别的齿轮箱故障诊断方法。首先通过实验获取齿轮箱故障齿轮的噪声信号,并根据齿轮状态分为6类。然后,使用Log-Mel谱的方法提取噪声信号语谱图。考虑到效率卷积神经网络(EfficientNet)对齿轮故障语谱图特征提取能力不足等缺点,在EfficientNet的基础上,结合分组卷积改进的GE-GCNet和DSCConv,提出一种高性能的齿轮故障诊断模型GGD-EfficientNet。实验表明:所提方法能在齿轮箱故障齿轮语谱图数据集下准确率达到99.7%。所提模型能从数据集中对故障类型进行有效分类,可有效帮助诊断齿轮箱中齿轮故障。This study presents a novel approach for diagnosing gearbox faults based on noise generated by faulty gears,utilizing a highperformance neural network called Grouped GCNet and Depthwise Separable Convolution with EfficientNet(GGD-EfficientNet).Noise signals from faulty gears were experimentally acquired and categorized into six distinct types according to their operational states.the Log-Mel spectral method was employed to extract the spectrograms of these noise signals.Recognizing the limitations of EfficientNet in feature extraction from gearbox fault spectrograms,we enhanced its performance by integrating the Group-based Enhanced Global Context Network(GE-GCNet)and Depthwise Separable Convolution(DSCConv).Experimental results demonstrate that the proposed GGD-EfficientNet achieves an impressive accuracy of 99.7%on the spectrogram dataset of faulty gears,effectively classifying fault types and providing valuable insights for gearbox fault diagnosis.

关 键 词：风力发电机 齿轮 故障检测 GGD-EfficientNet 声纹识别 

分 类 号：U226.81[交通运输工程—道路与铁道工程]