基于GRU与迁移学习的滚动轴承故障诊断

Rolling bearing fault diagnosis based on GRU and transfer learning

现阶段基于深度学习的故障诊断需要大量的数据,而制作数据集是一项耗时耗力的工作。针对这一缺点,提出一种基于门控循环单元(Gate Recurrent Unit,GRU)与迁移学习的滚动轴承故障诊断方法。该方法利用与目标域特征相似且易获得源域数据的特点训练网络,确定网络结构和参数,冻结经过训练的卷积神经网络(Convolutional Neural Networks,CNN)和GRU,用小样本目标域数据训练该网络,微调全连接层和分类层,达到迁移的目的。实验对比分析表明,基于GRU与迁移学习的滚动轴承故障诊断方法明显优于基于BP神经网络和基于概率神经网络(Probabilistic Neural Network,PNN)方法的故障诊断,能够更加准确地进行故障分类,为小样本数据集下的故障诊断提出了新思路。

At present,fault diagnosis based on deep learning requires a lot of data,and making data sets is a time-consuming and labor-consuming work.Aiming at this shortcoming,a rolling bearing fault diagnosis method based on Gate Recurrent Unit(GRU)and migration learning was proposed.The method use the characteristics which similar to the target domain and easy to obtain the source domain data,the network trained,the network structure and parameters deter mined,the trained Convolutional Neural Networks(CNN)and GRU were froze,the network with small sample target domain data trained,the full connection layer and classification layer was fine tuned,and classification layer to achieve the purpose of migration was achieved.The experimental comparative analysis shows that the rolling bearing fault diagnosis method based on GRU and transfer learning is obviously superior to the fault diagnosis method based on BP neural network and Probabilistic Neural Network(PNN),which can classify the fault more accurately,and put forward a new idea for fault diagnosis under small sample data set.