基于格拉姆角场和多尺度CNN的轴承故障诊断

Fault Diagnosis for Bearings Based on Gram Angle Field and MSCNN

为解决噪声干扰导致轴承故障分类准确率降低的问题,提出了一种基于格拉姆角场(GAF)和多尺度卷积神经网络(MSCNN)的端到端故障诊断方法。通过对故障信号进行GAF图像变换保留时间序列的相关性与依赖性,并将图像输入MSCNN中进行分类,利用改进的激活函数克服传统CNN的梯度下降问题以获得更好的分类结果。试验结果表明,所提方法在轴承故障诊断中的分类准确率能够达到99.67%,而且具有较高的鲁棒性,适用于不同工况下轴承振动信号的故障诊断。

In order to solve the problem of low accuracy of bearing fault classification caused by noise interference,an end-to-end fault diagnosis method is proposed based on gram angle field(GAF)and improved multi-scale convolutional neural network(MSCNN).Through GAF image transformation of fault signals,the correlation and dependence of time series are retained,and the images are input into MSCNN for classification.The improved activation function is used to overcome the gradient descent problem of traditional CNN to obtain better classification results.The experimental results show that the classification accuracy of proposed method in bearing fault diagnosis can reach 99.67%,and it has high robustness.It is suitable for fault diagnosis of bearing vibration signals under different working conditions.