基于门控循环单元胶囊网络的滚动轴承故障诊断  

Fault Diagnosis for Rolling Bearings Based on Capsule Network of Gated Recurrent Unit

作　　者：王超群 李彬彬[1] 焦斌[1] WANG Chaoqun;LI Binbin;JIAO Bin(School of Electrical Engineering,Shanghai Dianji University, Shanghai 201306, China)

机构地区：[1]上海电机学院电气学院,上海201306

出　　处：《轴承》2021年第5期56-62,共7页Bearing

摘　　要：针对目前噪声、变负载工况下基于深度学习的轴承故障诊断方法可能存在准确率下降的问题,提出一种基于门控循环单元(GRU)的胶囊网络模型,利用门控循环单元充分提取故障特征,再通过胶囊网络神经元中的向量提取更多细节特征并减少信息的丢失,最终完成故障的分类。试验及对比分析表明,该模型在0 dB信噪比的加噪状态下仍能达到94.375%的准确率,变负载工况下的平均准确率可达90%,优于CAPS,GRU,CNN,DNN等常用深度学习模型。Aimed at declining accuracy of fault diagnosis method for rolling bearings based on deep learning at present under the condition of noise and variable load,a capsule network model based on gated recurrent unit is proposed.The fault features are fully extracted by gated recurrent unit,and then more detailed features are extracted by vectors in neuron of capsule network to reduce the loss of information,and finally the classification of fault is completed.The experimental and comparative analysis show that the accuracy of the model can still reach 94.375%under the condition of 0 dB signal-to-noise ratio,and the average accuracy can reach 90%under the condition of variable load,which is superior to CAPS,GRU,CNN,DNN and other commonly used deep learning models.

关 键 词：滚动轴承 故障诊断 门控循环单元 胶囊网络 深度学习 神经网络 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP183[自动化与计算机技术—控制理论与控制工程]