基于CEEMDAN样本熵与卷积神经网络的轴承故障诊断

Bearing Fault Diagnosis Based on CEEMDAN Sample Entropy and Convolutional Neural Network

针对强噪声背景下滚动轴承振动信号故障特征难以提取和识别困难的问题,提出将自适应白噪声平均总体经验模态分解(CEEMDAN)样本熵与卷积神经网络(CNN)联合的故障诊断方法(CEEMDAN样本熵-CNN方法)。基于分形理论,采用CEEMDAN算法分解振动信号并提取其非线性特征,通过分形盒维数筛选最优IMF分量,以其样本熵构成的特征向量输入CNN模型,实现轴承故障的分类和诊断,并进行t-SNE聚类可视化分析。结果表明:在不同工况下,与经验模态分解(EMD)样本熵和集成经验模态分解(EEMD)样本熵方法相比,所提CEEMDAN样本熵-CNN方法具有良好的识别能力和泛化性能,其可视化分析结果更具直观性。

Aiming at the problem that it is difficult to extract and identify the fault features of rolling bearing vibration signals under strong noise background,a fault diagnosis method(CEEMDAN sample entropy-CNN)combining complete cnsemble empirical mode decomposition with adaptive noise(CEEMDAN)sample entropy and convolutional neural network(CNN)was proposed.Based on the fractal theory,CEEMDAN algorithm was used to decompose the vibration signal and extract its nonlinear features.The optimal IMF component was selected by fractal box dimension,and the feature vector composed of its sample entropy was input into CNN to realize the classification and diagnosis of bearing faults.Finally,the t-SNE clustering visual analysis was performed on the bearing faults.Results show that compared with empirical mode decomposition(EMD)sample entropy and ensemble empirical mode decomposition(EEMD)sample entropy,the proposed CEEMDAN sample entropy-CNN method has better recognition ability and generalization performance under different working conditions,and its visualization analysis results are more intuitive.