增强辛几何模态分解和自组织自编码卷积网络的电机轴承工况识别

Motor bearing condition identification of enhanced symplectic geometric mode decomposition and self-organizing auto-encoder convolution network

针对电机轴承振动信号特征提取与工况识别困难的问题,提出一种基于增强辛几何模态分解(ESGMD)和自组织自编码卷积网络(SOAECN)的电机轴承工况识别方法。在辛几何模态分解(SGMD)的基础上将电机轴承振动信号自适应分解为初始辛几何模态分量(ISGMCs),并利用改进凝聚聚类算法对ISGMCs重新组合得到聚类辛几何模态分量(CSGMCs);提出一种综合评价指标,利用此指标筛选能反映振动信号特征的CSGMCs分量并重构;结合卷积神经网络和小波自编码器,构造自编码卷积网络(AECN),并在AECN基础上改进其损失函数且引入自组织策略,进而构造SOAECN;将重构后的振动信号输入SOAECN进行自动特征提取与工况识别。实验结果表明:ESGMD‑SOAECN方法的工况识别率达到了98.76%,自动特征提取能力和工况识别能力优于深度稀疏自动编码器、深度降噪自动编码器和深度信念网络等深度学习方法,可为电机轴承自动工况识别提供参考。

Aiming at the difficulty of vibration signals feature extraction and condition identification of motor bearings,a method based on enhanced symplectic geometry mode decomposition(ESGMD)with self-organizing auto-encoder convolution network(SOAECN)is proposed.On the basis of symplectic geometry mode decomposition(SGMD),the vibration signals of motor bearing are adaptively decomposed into initial symplectic geometric mode component(ISGMCs).The ISGMCs are adaptively reorganized by improved condensed clustering method to obtain the clustering symplectic geometric modal components(CSGMCs).A new comprehensive evaluation index is proposed,which is used to screen and reconstruct the CSGMCs that can reflect the characteristics of vibration signals.Combined convolution neural network with wavelet auto-encoder,the auto-encoder convolution network(AECN)is constructed.Its loss function is improved and self-organizing strategy is introduced on the basis of AECN,then the SOAECN is constructed.The reconstructed vibration signals are fed into SOAECN for automatic feature learning and condition identification.Experimental results indicate that ESGMD-SOAECN reaches 98.98%of condition identification rate.The ability of condition automatic feature extraction and automatic condition identification is better than deep learning methods such as deep sparse auto-encoder,deep de-noising auto-encoder,deep belief network and so on.The results can provide a reference for the identification of motor bearing conditions.