基于加权域适应卷积神经网络的滚动轴承故障诊断

Rolling Bearing Fault Diagnosis Based on Weighted Domain Adaptive Convolutional Neural Network

针对工作负载变化引起滚动轴承振动信号的数据分布变化,导致故障诊断算法的泛化性变差的问题,提出一种基于加权域适应卷积神经网络(weighted domain adaptive convolutional neural network,WDACNN)的滚动轴承故障诊断方法。在卷积神经网络中嵌入域适应算法,使基于源域的分类器在目标域上取得良好的泛化性,并引入权重系数对源域样本进行加权,降低类权重偏差的影响。采用6组迁移任务验证所提方法的有效性,结果表明:平均故障诊断精度达到96.6%,证明了本文方法在不同工作负载条件下的有效性。

A rolling bearing fault diagnosis method based on a weighted domain adaptive convolutional neural network(WDACNN)is proposed to solve the problem that the data distribution of vibration signals of rolling bearings changes due to workload changes,which leads to poor generalization of fault diagnosis algorithm.In this method,the domain adaptation algorithm is embedded in the convolutional neural network to make the classifier based on the source domain achieve excellent generalization in the target domain,and the weight coefficient is introduced to weight the samples in the source domain to reduce the influence of the class weight deviation.In the simulation experiment,six migration tasks are used to verify the effectiveness of the proposed method.The average fault diagnosis accuracy of the proposed method reaches 96.6%,which proves the effectiveness of the proposed method under different workload conditions.