高压共轨系统小样本故障诊断方法  

Small-Sample Fault Diagnosis Method for High Pressure Common Rail System

作　　者：李良钰 苏铁熊[1] 马富康 Li Liangyu;Su Tiexiong;Ma Fukang(College of Mechatronic Engineering,North University of China,Taiyuan 030051,China;School of Energy and Power Engineering,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学机电工程学院,山西太原030051 [2]中北大学能源动力工程学院,山西太原030051

出　　处：《内燃机学报》2023年第3期255-262,共8页Transactions of Csice

基　　金：山西省基础研究计划资助项目(202103021224216).

摘　　要：为解决高压共轨柴油机在故障诊断时因故障样本获取成本较高或故障样本获取难度较大造成的故障诊断训练样本不足问题,提出了一种使用基于数据增强的小样本学习方法与基于GA_BP神经网络的故障诊断方法的高压共轨系统小样本故障诊断方法.针对高压共轨小样本故障诊断的特性,使用最小二乘—生成式对抗神经网络(LSGANs)进行训练集的数据合成,提高了合成数据的质量与多样性,3种故障状态的合成数据与真实数据集的平均地球搬运距离为2.25,在小样本集上的平均诊断正确率可达95.53%.试验结果表明:本故障诊断方法在解决小样本故障诊断问题时,生成合成数据质量更高、多样性更强及普适性强,可以合成多种状态的轨压信号.In order to solve the problem of insufficient diagnostic training samples caused by the high cost of acquiring fault samples or the difficulty of acquiring fault samples in fault diagnosis of high pressure common rail diesel engines,a small-sample learning method based on data enhancement and a small-sample fault diagnosis method based on GA_BP neural network fault diagnosis method were proposed for a high pressure common rail system.In view of the characteristics of fault diagnosis,the least square-generative adversarial neural networks(LSGANs)were used to synthesize the data of the training set,which improves the quality and diversity of the synthetic data.The average earth move distance between the synthetic data of the three fault states and the real data set is 2.25,and the average diagnosis accuracy rate on the small-sample set can reach 95.53%.The experimental results show that this fault diagnosis method can generate higher quality,diversity and universality of synthetic data when solving the problem of small-sample fault diagnosis,and can synthesize rail pressure signals in various states.

关 键 词：柴油机 高压共轨 小样本学习 故障诊断 

分 类 号：TK428[动力工程及工程热物理—动力机械及工程]