基于SSA-VMD和SDP的双通道CNN轴承故障识别方法  

Fault identification method for bearings based on dual channel CNN using SSA-VMD and SDP

作　　者：蒋丽英 高铭悦 李贺 JIANG Liying;GAO Mingyue;LI He(College of Automation,Shenyang Aerospace University,Shenyang 110136,China)

机构地区：[1]沈阳航空航天大学自动化学院,辽宁沈阳110136

出　　处：《机电工程》2025年第2期257-266,共10页Journal of Mechanical & Electrical Engineering

基　　金：国家自然科学基金资助项目(62003223)。

摘　　要：针对滚动轴承故障振动信号具有非线性和非平稳性等特征,以及单通道卷积神经网络(CNN)提取故障特征不显著的问题,提出了一种基于麻雀算法-变分模态分解(SSA-VMD)和对称点模式(SDP)的双通道CNN滚动轴承故障诊断方法。首先,结合样本熵和皮尔逊相关系数,构建了新的综合适应度函数,利用麻雀算法(SSA)进行了自适应寻优,确定了最佳的变分模态分解(VMD)参数K和α。将原始振动信号经过VMD分解后,得到了本征模态函数(IMF)分量,通过计算各IMF分量的峭度值进行了筛选,将筛选出的信号进行重构后得到了一维特征信号;然后,根据互相关系数选择了合适的对称点模式(SDP)参数值,将原始振动信号转化为极坐标下的SDP图像,获得了具有良好可分性的二维特征图;最后,将一维和二维特征作为双通道CNN的输入进行了联合训练,将训练好的网络用于故障类型识别,在西储大学和江南大学的轴承数据集上对其有效性进行了验证。研究结果表明:通过网络训练,其故障诊断的准确率分别达到了98.5%和100%。该结果验证了该方法在准确识别故障特征方面具有优越性和普适性。Aiming at the nonlinear and non-stationary characteristics of rolling bearing fault vibration signal and the unobviousness of the fault features extracted by single-channel convolutional neural network(CNN),a dual-channel CNN-based fault diagnosis method was proposed by incorporating sparrow search algorithm-variational mode decomposition(SSA-VMD) and symmetrized dot pattern(SDP).Firstly,a new comprehensive fitness function was constructed using sample entropy and Pearson correlation coefficient.VMD's parameters,K and α,were adaptively optimized by SSA.The intrinsic mode functions(IMF) obtained by VMD were filtered and reconstructed based on kurtosis to generate a one-dimensional feature signal.Then,the original vibration signal was transformed into SDP image in polar coordinates by choosing appropriate parameters through cross-correlation coefficient to obtain two-dimensional feature map with good separability.Finally,as two different input channels,the one-dimensional and two-dimensional features were used to jointly train CNN in order to identify fault types.The research results show that the method respectively achieves diagnostic accuracies of 98.5% and 100% on the Case Western Reserve University and Jiangnan University bearing datasets,demonstrating its superior performance and broad applicability in fault feature recognition.

关 键 词：一维特征信号构建 二维特征转换 卷积神经网络 麻雀算法 变分模态分解 对称点模式 

分 类 号：TH133.3[机械工程—机械制造及自动化]