摘要
在车辆悬架故障诊断过程中,深度学习故障诊断模型在面对少量样本数据时模型训练效果不佳,导致诊断模型的接收者操作特性曲线(receiver operating characteristic,ROC)的曲线下面积(area under curve,AUC)较小的问题,利用经验模态分解(empirical mode decomposition,EMD)方法,对采集的车辆悬架高频振动信号进行分解处理,根据每个经验模态分量(intrinsic mode functions,IMF)的能量,提取高频异常振动故障特征,构建了基于深度迁移学习的诊断模型;以深度卷积神经网络算法为基础,对小样本特征矢量信息进行故障知识迁移处理,通过参数微调更新权值,优化故障诊断模型。实验结果表明:优化后模型的AUC值为0.89,模型故障诊断结果具有较高准确性。
In the fault diagnosis process of vehicle suspension,the deep learning fault diagnosis model has poor training performance when facing a small amount of sample data,resulting in a smaller area under curve(AUC)of the receiver operating characteristic(ROC)curve of the diagnostic model.Empirical mode decomposition(EMD)method was employed to decompose the collected high-frequency vibration signals of vehicle suspensions.According to the energy of each intrinsic mode function(IMF),high-frequency abnormal vibration features were extracted,and the diagnosis model based on deep transfer learning was established.Based on deep convolutional neural network algorithm,the small-sample feature vector information was fault knowledge transferred.The fault diagnosis model was optimized by fine-tuning parameters and updating weights.Experimental results demonstrate that the AUC value of the optimized model is 0.89,and the fault diagnosis results of the proposed model have higher accuracy.
作者
牛礼民
胡超
万凌初
张代庆
NIU Limin;HU Chao;WAN Lingchu;ZHANG Daiqing(School of Mechanical Engineering,Anhui University of Technology,Ma’anshan 243032,Anhui,China)
出处
《重庆交通大学学报（自然科学版）》
CAS
CSCD
北大核心
2024年第3期121-127,共7页
Journal of Chongqing Jiaotong University(Natural Science)
基金
教育部产学合作协同育人项目(202102095077)
安徽省虚拟仿真实验教学项目(2021xnfzxm010)。
关键词
车辆工程
悬架
故障诊断
深度迁移学习
卷积神经网络
经验模态分量
vehicle engineering
suspension
fault diagnosis
deep transfer learning
convolution neural network
empirical mode component
