基于改进谱峭度与一维卷积神经网络的轴承故障诊断

Bearing Fault Diagnosis Based on an Improved Kurtogram Kurtosis and One-Dimensional Convolution Neural Network

由于轴承所处工况的复杂性,其振动信号中包含了各种噪声和干扰,导致传统信号处理方法效果不理想。因此,很多研究将信号处理方法与神经网络相结合对故障进行诊断,卷积神经网络(CNN)因其对图像具有独特的特征提取能力而被引入故障诊断领域。而通过信号处理构造的图像可能存在信息冗余问题,将信息冗余图像直接作为网络输入会增加其复杂度。针对上述问题提出了一种基于改进谱峭度与一维CNN的故障分类方法。改进谱峭度方法克服了非高斯噪声和偶然性冲击的影响,能很快地选择正确的滤波频带。考虑到构造谱峭度图的原理,将谱峭度图转换成一维序列信号,作为一维CNN输入进行故障分类,相比于直接将谱峭度图输入二维CNN中,该方法去除了图像的冗余信息,减少了网络结构参数,降低了网络复杂度。通过二组数据分析验证了该文方法的有效性和泛化性。

Because of the complexity of the working condition of the bearing,the vibration signal contains various noise and interference,which leads to poor performance of traditional signal processing methods.Therefore,many researchers combine the signal processing method and the neural network to diagnose the fault,and the CNN(convolution neural network)is introduced into the field of fault diagnosis because of its unique image feature extraction ability.However,the image constructed by signal processing may have the problem of information redundancy,which increases the complexity of the network.To solve the above problems,this paper presents a fault classification method based on improved kurtosis and one-dimensional CNN.The improved kurtosis method overcomes the influence of non-Gaussian noise and accidental impact,and can quickly select the correct filtering frequency band.At the same time,considering the principle of constructing kurtogram,the kurtogram is transformed into onedimensional sequence signal,which is used as the input of one-dimensional CNN for fault classification.Compared with inputting the kurtogram into the two-dimensional CNN directly,this method can remove the redundant information in the image,reduce the number of structural parameters and complexity of the network.The validity and generalization of the proposed method are verified by two sets of data analysis.