基于改进CEEMDAN-CNN的轴承故障诊断研究

Research on Bearing Fault Diagnosis based on Improved CEEMDAN-CNN

为保证旋转机械安全稳定运行和实现轴承早期疲劳损伤阶段故障诊断,提出了改进自适应白噪声平均总体经验模态分解(CEEMDAN)与卷积神经网络融合的故障诊断方法。通过CEEMDAN方法分解原始故障信号,分形盒维数作为指标筛选核主成分分析降维的最佳重构分量,输入卷积神经网络实现非线性故障特征提取。结果表明:在西安交通大学轴承数据集不同信噪比下,该诊断方法与单一的传统经验模态分解-卷积神经网络(EMD-CNN)、集合经验模态分解-卷积神经网络(EEMD-CNN)方法及其对应的改进重构最佳模态分量方法进行对比,其低信噪比下准确率达87.1%,且在各种信噪比下均保持最高准确率;该方法应用于西储大学轴承数据集的准确率达到98.7%,证明其具有较强的鲁棒性与泛化性;该方法可有效解决传统轴承故障诊断方法信号非线性特征提取不充分的局限性。

In order to ensure the safe and stable operation of rotating machinery and realize the fault diagnosis in the early fatigue damage stage of bearings,a fault diagnosis method was proposed to improve the fusion of complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)and convolutional neural network(CNN).The original fault signal was decomposed by CEEMDAN method,the fractal box dimension was used as an index to screen the best reconstruction component of the kernel principal component analysis(KPCA)dimensionality reduction,and the nonlinear fault feature extraction was realized by inputting the CNN.The results show that the accuracy rate of this diagnostic method is 87.1%at low signal-to-noise ratio and maintains the highest at various signal-to-noise ratios when compared with a single traditional empirical mode decomposition-convolutional neural network(EMD-CNN)method,an ensemble empirical mode decomposition-convolutional neural network(EEMD-CNN)method and its corresponding improved reconstruction of the optimal mode component method in the bearing dataset of Xi′an Jiaotong University at various signal-to-noise ratios;the accuracy rate of this method in the bearing dataset of Case Western Reserve University(CWRU)reaches 98.7%,which proves that it has strong robustness and generalisation and can effectively solve the limitations of the traditional bearing fault diagnosis methods with insufficient signal nonlinear feature extraction.