基于迁移学习与深度残差网络的滚动轴承快速故障诊断算法EI北大核心CSCD

针对现有基于深度学习的滚动轴承故障诊断算法训练参数量大,训练时间长且需要大量训练样本的缺点,提出了一种基于迁移学习(TL)与深度残差网络(ResNet)的快速故障诊断算法(TL-ResNet)。首先开发了一种将短时傅里叶变换(STFT)与伪彩色处理相结合的振动信号转三通道图像数据的方法;然后将在ImageNet数据集上训练的ResNet18模型作为预训练模型,通过迁移学习的方法,应用到滚动轴承故障诊断领域当中;最后对滚动轴承在不同工况下的故障诊断问题,提出了采用小样本迁移的方法进行诊断。在凯斯西储大学(CWRU)与帕德博恩大学(PU)数据集上进行了试验,TL-ResNet的诊断准确率分别为99.8%与95.2%,且在CWRU数据集上TL-ResNet的训练时间仅要1.5 s,这表明本算法优于其他的基于深度学习的故障诊断算法与经典算法,可用于实际工业环境中的快速故障诊断。

Fast fault diagnosis algorithm for rolling bearing based on transfer learning and deep residual network

Here,aiming at shortcomings of the existing rolling bearing fault diagnosis algorithm based on deep learning,such as,large amount of training parameters,long training time and a large number of training samples,a fast fault diagnosis algorithm(TL-ResNet)based on the transfer learning(TL)and the deep residual network(ResNet)was proposed.Firstly,a method of converting a vibration signal into 3-channel image data by combining short-time Fourier transform(STFT)and pseudo-color processing was developed.Then,ResNet 18 model trained on ImageNet data set was taken as the pre-training model,and it was applied in the field of rolling bearing fault diagnosis with TL method.Finally,a small sample transfer method was proposed to do fault diagnosis of rolling bearing under different working conditions.Tests were conducted on Case Western Reserve University(CWRU)and Padborn University(PU)data sets.Test results showed that diagnostic accuracies using TL-ResNet are 99.8%and 95.2%,respectively;the training time of TL-ResNet on CWRU data set is only 1.5 s;the proposed algorithm is superior to other fault diagnosis algorithms based on deep learning and classical algorithms,and it can be used for rapid fault diagnosis in practical industrial environment.