基于加窗和卷积神经网络的柴油机拉缸故障诊断

张永祥; 王宇; 姚晓山

车用发动机 ›› 2019, Vol. 0 ›› Issue (6) : 84-89.

栏目

基于加窗和卷积神经网络的柴油机拉缸故障诊断

张永祥¹，王宇¹，姚晓山²

作者信息 +

Fault Diagnosis of Cylinder Scuffing for Diesel Engine Based on Windowing and Convolutional Neural Network

ZHANG Yongxiang¹,WANG Yu¹,YAO Xiaoshan²

Author information +

文章历史 +

摘要

柴油机发生拉缸故障时，其振动信号较为复杂，而且伴有较强的背景噪声。为实现柴油机拉缸故障诊断，提出了基于加窗和卷积神经网络（Convolution Neural Network，CNN）的方法，实现了端到端的故障模式识别。首先将已产生拉缸故障的活塞缸套装入6-135柴油机，进行拉缸故障试验，测取典型工况下机体表面的加速度信号；然后根据柴油机的结构和配气相位分布，分析确定加速度信号的加窗位置并进行加窗处理，得到加窗后信号样本，分为训练集和测试集；再将训练集输入到CNN中不断学习，更新模型参数；最后将训练好的CNN模型应用于测试集，输出故障识别结果。结果表明：基于加窗和CNN的方法可以有效地实现柴油机拉缸故障的诊断。

Abstract

The vibration signal of cylinder scuffing for diesel engine was complicated and accompanied by strong background noise. A method of windowing and CNN was proposed to realize the endtoend fault pattern recognition, which made it easy to diagnose the fault of cylinder scuffing. The piston and cylinder pair with scuffing failure was first put into 6-135 diesel engine and the cylinder scuffing failure test was carried out to measure the acceleration signal of body surface under typical working conditions. Then the windowing position of acceleration signal was determined and the window processing was conducted to obtain the windowing signal samples according to the structure of diesel engine and the distribution of valve timing. The sample was divided into a training set and a test set. The training set was input into the CNN to perform continuous learning in order to update the model parameters and the test set was input into the trained CNN model in order to output the fault recognition result. The results show that the method of windowing and CNN can effectively diagnose the scuffing fault of diesel engine.