CEEMD与卷积神经网络特征提取的故障诊断方法研究
Fault Diagnosis of a Bearing using Feature Extraction Method based on CEEMD Algorithm and CNN

轴承动力学行为具有非线性的特点，导致其振动信号特征与运行状态之间存在较强的非线性关系；且振动信号的特征提取与选择往往需要大量的先验知识，导致特征的设计难以准确反映不同的运行状态。针对以上问题，提出一种基于互补集合经验模态分解（Complementary ensemble empirical mode decomposition，CEEMD）与卷积神经网络（Convolutional neural networks，CNN）的特征提取方法，从振动信号时频图中自适应提取其敏感特征，反映设备运行状态。首先采用CEEMD算法分解得到振动信号的固有模态函数（Intrinsic mode function，IMF）分量，构造各个IMF时频图，并采用CNN提取时频图的特征；然后，将提取到的特征与小波包分频带能量值相结合，组建特征指标向量，用于构建轴承故障诊断模型。将该方法应用于不同负载、不同故障深度的轴承试验中，结果表明该方法能够在多种工况下有效地提高故障识别率。
The dynamic behavior of a bearing is nonlinear, resulting in the strong nonlinear relationship between its vibration signal features and operation state. The feature extraction and selection of vibration signals often require a large amount of prior knowledge, which makes it difficult to accurately reflect different operation states for feature design. In order to solve the above problems, a feature extraction method based on Complementary Ensemble Empirical Mode Decomposition (CEEMD) algorithm and Convolutional Neural Networks (CNN) is proposed to extract sensitive features of self-adaptive vibration signals from their time-frequency diagram. These features can effectively reflect the operating status of the bearing. Firstly, the feature extraction method uses the CEEMD algorithm to obtain the Intrinsic Mode Function (IMF) components of vibration signals and then constructs each IMF time-frequency diagram; it also uses CNN to extract the features of time-frequency diagram. Secondly, it combines the extracted features with the wavelet packet frequency band energy values to form feature index vectors, which are used to build the bearing's fault diagnosis model. The experimental results show that this method can effectively improve the fault diagnosis rate