基于格拉姆角场和卷积神经网络的滚动轴承微弱故障位置辨识研究

Weak fault location detection of rolling bearing based on Gramian angular field with CNN

为实现卷积神经网络(CNN)对滚动轴承微弱故障位置的辨识,首先在保留整体信息的情况下,使用分段聚合近似(PAA)对轴承信号降维压缩;其次引入了格拉姆角场(GAF)将降维压缩后的轴承一维时间序列转换成了二维图像;然后引入批量归一化层、小批量法(minibatch)等方法设计卷积神经网络;最后将训练样本图像输入卷积神经网络进行训练和验证。结果表明,格拉姆和/差角场图均可有效识别滚动轴承不同零件的故障,格拉姆差角场在准确度上较格拉姆和角场高,更适合用于微弱故障的识别。

In order to implement the convolutional neural network(CNN)identification of the weak fault position of the rolling bearing,the piecewise aggregate approximation(PAA)which could compress the amount of signal data as much as possible while preserving the whole signal information is introduced.Using Gramian angular field(GAF)to transform the 1d bearing vibration signals time series into 2d images,the CNN is properly designed by using batch normalization layer and minibatch method.Finally,the training sample images are input into the CNN for training and verification.The results show that both GASF and GADF could effectively identify the faults of different parts,while GADF has some advantages over GASF in verifying accuracy and improving accuracy,which makes GADF more suitable for identifying the faults of different parts.