基于1 DCNN-GWO-SVM的柴油机喷油系统故障诊断方法研究  

Fault Diagnosis Method of Diesel Engine Fuel System Based on 1 DCNN-GWO-SVM

作　　者：冯海波 毛玉欣 孔祥鑫 张探军 刘峰春 叶俊杰 FENG Haibo;MAO Yuxin;KONG Xiangxin;ZHANG Tanjun;LIU Fengchun;YE Junjie(China North Engine Research Institute(Tianjin),Tianjin 300406,China)

机构地区：[1]中国北方发动机研究所(天津),天津300406

出　　处：《车用发动机》2024年第4期85-92,共8页Vehicle Engine

摘　　要：准确、有效的故障诊断是柴油机安全可靠运行的重要保障。基于热工参数诊断的方法存在测点多、专业性强等问题,传统机器学习结合振动信号诊断方法存在人为影响因素过高、不确定性大等问题,因此提出了一种基于1DCNN-GWO-SVM的柴油机喷油系统故障诊断方法。首先利用一维卷积神经网络(one-dimensional convolutional neural network,1DCNN)对时域下的柴油机振动加速度信号进行自学习特征提取,然后利用提取到的特征向量训练支持向量机(support vector machine,SVM)分类模型,并利用灰狼优化算法(grey wolf optimization,GWO)对SVM的C,g等超参数进行寻优,以此来实现对柴油机的“端对端”故障诊断。在实例验证中,1DCNN-GWO-SVM在测试集上能达到99.10%的诊断准确率,优于传统的机器学习故障诊断方法,并且在信噪比为分别10 dB,20 dB,30 dB的干扰环境下,依然能保持90%以上的诊断准确率。结果表明:1DCNN-GWO-SVM是一种预测精度高、泛化能力强、抗干扰能力强的柴油机“端对端”喷油系统故障诊断方法,具有实际工程应用价值。Accurate and effective fault diagnosis is an important guarantee for the safe and reliable operation of diesel engine.A diesel engine fault diagnosis method based on 1DCNN-GWO-SVM was proposed to address the problems of multiple measurement points and strong profession in thermal parameter diagnosis methods,as well as the high human influencing factors and high uncertainty in traditional machine learning combined with vibration signal diagnosis methods.A one-dimensional convolutional neural network(1DCNN)was used to extract self-learning features of diesel engine vibration acceleration signals in the time domain.Then the extracted feature vectors were used to train the support vector machine(SVM)classification model.The grey wolf optimization algorithm(GWO)was used to optimize the hyperparameters of SVM such as C and g in order to achieve end-to-end fault diagnosis of diesel engine.For the sample verification,1DCNN-GWO-SVM could achieve a diagnostic accuracy of 99.10%on the training set,which was superior to various traditional machine learning fault diagnosis methods.Moreover,it could still maintain a diagnostic accuracy of over 90%in interference environments with signal-to-noise ratios of 10 dB,20 dB,and 30 dB,respectively.The results indicate that 1DCNN-GWO-SVM is an end-to-end fault diagnosis method for diesel engine fuel injection systems with high prediction accuracy,strong generalization ability,and strong anti-interference ability,which has practical engineering application value.

关 键 词：卷积神经网络 支持向量机 灰狼优化算法 柴油机 故障诊断 

分 类 号：TK423.8[动力工程及工程热物理—动力机械及工程]