基于ARCN模型的轴承故障诊断

Bearing fault diagnosis based on ARCN model

提出注意力循环机制与胶囊网络融合的注意力循环胶囊网络(ARCN)的诊断模型。提取时序特征信息构建初级胶囊;自适应融合路由机制、注意力循环机制构建数字胶囊特征;基于西储大学轴承实验数据,验证了ARCN模型的准确率、鲁棒性、稳定性、收敛误差,其准确率相比Caps模型识别准确率提高1.2%、收敛误差达到0.2。基于实验仿真平台,采集正常、内环故障、外环故障和滚动体故障的振动信号,并通过小波基变换获取的时频图构建ARCN模型的数据集。仿真实验结果表明:ARCN模型下,每类故障被误诊的概率不超过总样本的1%。

The attention recurrent and capsule network(ARCN)diagnosis model was proposed by integrating the attention cycle mechanism and capsule network.Firstly,the bidirectional LSTM network was used to extract the time-series characteristic information to construct the primary capsule.Secondly,routing mechanism and attention cycle mechanism were used to construct adaptively digital capsule.The accuracy, robustness, stability and convergence error of ARCN model in bearing fault identification were verified by bearing experiment data of Western Reserve University.The accuracy of ARCN model was 1.2% higher than that of Caps model.The convergence error of the ARCN model reached 0.2.Based on the experimental simulation platform, the vibration signals of normal, inner ring fault, outer ring fault and rolling element fault were collected.The results showed that the misdiagnosis probability of each kind of fault was less than 1% of the total samples under ARCN model.