基于多尺度多路径集成网络的轴承故障诊断方法

BEARING FAULT DIAGNOSIS METHOD BASED ON MULTI⁃SCALE AND MULTI⁃PATH ENSEMBLE NETWORK

针对卷积神经网络参数量大、层数深、泛化性能弱的问题,提出了一种基于多尺度多路径集成网络(Multi⁃scale Multi⁃path Convolutional Neural Network,MCS⁃CNN)的轴承故障诊断方法。首先,提出多尺度卷积块,利用深度卷积和逐点卷积来降低网络参数量、增加网络宽度,有效提取多尺度特征;其次,提出集成块,利用多路径的连接方式将低级和高级特征连接起来以增加网络深度,进而提高模型诊断精度;最后,在凯斯西储大学轴承数据集上进行验证。结果表明,在高噪声和跨负载场景下的故障诊断精度分别可达97.5%和98.25%,在混合场景下相较于现有的诊断方法精度提高了15百分点以上,有效表明了此方法的鲁棒性和泛化性。

To address the problems such as complex convolutional neural network parameters,deep layers and weak generalization performance,a multi⁃scale multi⁃path convolutional neural network(MCS⁃CNN)based bearing fault diagnosis method cas proposed.Firstly,a multi⁃scale convolution block was proposed to reduce the number of network parameters and increase the network width by using depthwise convolution and pointwise convolution,so as to extract multi⁃scale features effectively.Secondly,an ensemble block was proposed to increase the network depth by connecting low⁃level and high⁃level features through multiple paths,thereby improving the diagnostic accuracy of the model.Finally,the effectiveness of the method was verified on the Case Western Reserve University bearing dataset.The results show that the fault diagnosis accuracy of the proposed method can reach 975%and 9825%in high⁃noise and cross⁃load scenarios,and the accuracy in mixed scenarios is improved by more than 15%compared to existing diagnosis methods,which prove the robustness and generalisability of the proposed method.