李泽东, 李志农, 王成军. 深度卷积神经网络在轴承多故障复合诊断中应用研究[J]. 南昌航空大学学报(自然科学版), 2020, 34(1): 12-20. DOI: 10.3969/j.issn.1001-4926.2020.01.003
引用本文: 李泽东, 李志农, 王成军. 深度卷积神经网络在轴承多故障复合诊断中应用研究[J]. 南昌航空大学学报(自然科学版), 2020, 34(1): 12-20. DOI: 10.3969/j.issn.1001-4926.2020.01.003
Ze-dong LI, Zhi-nong LI, Chen-jun WANG. Application of Deep Convolution Neural Network in Bearing Multi Fault Diagnosis[J]. Journal of nanchang hangkong university(Natural science edition), 2020, 34(1): 12-20. DOI: 10.3969/j.issn.1001-4926.2020.01.003
Citation: Ze-dong LI, Zhi-nong LI, Chen-jun WANG. Application of Deep Convolution Neural Network in Bearing Multi Fault Diagnosis[J]. Journal of nanchang hangkong university(Natural science edition), 2020, 34(1): 12-20. DOI: 10.3969/j.issn.1001-4926.2020.01.003

深度卷积神经网络在轴承多故障复合诊断中应用研究

Application of Deep Convolution Neural Network in Bearing Multi Fault Diagnosis

  • 摘要: 基于深度学习具有强大的自特征提取能力和较优的分类能力,将深度卷积神经网络引用到轴承的故障诊断中,提出了基于一维深度卷积神经网络的轴承复杂工况故障诊断方法。在提出的方法中,将轴承的多故障振动信号作为模型的直接输入,通过训练深度卷积神经网络模型,利用模型中多个卷积层和池化层对输入的振动信号进行自特征提取,并进行故障分类。从而以基于数据驱动的方式形成端到端的故障诊断。研究表明,在一维深度卷积神经网络中直接输入轴承振动信号进行故障诊断,与提取时域和频域特征结合支持向量机进行故障诊断的方法相比,深度卷积神经网络可以更好地反映时域振动信号与特征间的关系,获得了比传统智能诊断方法更高的识别效率。

     

    Abstract: Based on deep learning with strong self-feature extraction capabilities and better classification capabilities, 1D-deep convolutional neural networks are used for bearing composite fault diagnosis. A method for bearing composite fault diagnosis based on deep convolutional neural networks is proposed. In the proposed method, the multi-fault vibration signal of the bearing is used as the direct input of the model. By training a deep convolutional neural network model, multiple convolution layers and pooling layers in the model are used to self-extract the input vibration signal. And classify the fault. Thus, an end-to-end composite fault diagnosis is formed in a data-driven manner. Experiments show that directly inputting bearing vibration signals in a deep convolutional neural network for fault diagnosis, compared with the method of extracting time and frequency domain features combined with support vector machines for fault diagnosis, the deep convolutional neural network can better reflect the time The relationship between domain vibration signals and features has achieved higher recognition efficiency than traditional intelligent diagnostic methods.

     

/

返回文章
返回