基于CNN-SVM的深度卷积神经网络轴承故障识别研究

Bearing fault identification by using deep convolution neural networks based on CNN-SVM

针对传统智能诊断方法过分依赖于信号处理和专家经验提取故障特征以及模型泛化能力差的问题,基于深度学习理论,提出将卷积神经网络算法结合SVM分类器搭建适于滚动轴承故障诊断的改进型深度卷积神经网络模型。从原始实测轴承振动信号出发,模型逐层学习实现特征提取与故障识别,引入批量归一化、Dropout处理并改进模型分类器来提升轴承故障识别准确率、模型收敛速度和泛化能力。实验结果表明,优化后的深度学习模型可快速准确地提取轴承故障特征,针对不同类型、不同损伤程度的轴承可实现99%的识别准确率,并且模型有较强的泛化能力和强化学习能力。

Considering that the traditional intelligent diagnosis methods rely too much on the signal processing and expert experience to extract fault features and are of poor model generalization ability,based on the deep learning theory,a deep convolution neural network algorithm combined with SVM classifier was proposed to build an improved fault diagnosis model for rolling bearings.Starting from the original measured bearing vibration signals,the model learns from each layer to achieve feature extraction and fault recognition,and introduces the batch normalization,Dropout processing and improved model classifier to improve the bearing fault recognition accuracy,model convergence speed and generalization ability.The experimental results show that the optimized deep learning model can quickly and accurately extract the characteristics of bearing faults.99%recognition accuracy can be achieved for bearings of different types and degrees of damages,and the model has strong generalization ability and enhanced learning ability.