基于改进DBD算法的神经网络在柴油机故障诊断中的应用  

Application of Neural Network in Diesel Engine Fault Diagnosis Based on Modified DBD Algorithm

作　　者：路锋[1,2] 司爱威[1] 程慧[2] 

机构地区：[1]装甲兵工程学院机械工程系,北京100072 [2]北京特种车辆研究所,北京100072

出　　处：《计算机测量与控制》2013年第2期321-323,342,共4页Computer Measurement &Control

摘　　要：针对BP神经网络和传统DBD算法的不足,引入了一种结合遗传算法思想的新算法;通过对学习率的修正量有指导随机互换、修改学习率的增量因子和修改作用函数等方法,较好地解决了BP神经网络收敛速度慢,容易陷入局部最小,以及对参数选择过于敏感的问题;该算法在柴油机燃油供给系统的故障诊断中取得了很好的效果;同时,利用燃油压力波中提取的主要特征点的特征值进行燃油供给系统的故障诊断,具有较高的准确度。To the drawbacks of the neural network based on back propagation （BP） algorithm, this paper proposed a new modified algorithm , which combined the genetic algorithm （GA） and delta--bar--delta （DBD） algorithm. By the methods of study rate stochastic changes, the new algorithm resolved the problem of the BP neural network, especially like slow convergence and being trapped at local minimum, as well as the sensitivity to the choice of parameters. It has been applied in the fault diagnosis of the fuel supply system of the diesel engine successfully. At the same time, the use of the feature points extracted from the fuel pressure wave characteristic value of the fuel supply sys tem of fault diagnosis, with high accuracy.

关 键 词：DBD算法 遗传算法 神经网络 故障诊断 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]