采用“GA+LM”优化BP神经网络的电液伺服阀故障诊断  

Fault Diagnosis of Electro Hydraulic Servo Valves Based on GA+LM Algorithm Optimized BP Neural Networks

作　　者：权凌霄[1,2] 郭海鑫 盛世伟 李雷 

机构地区：[1]燕山大学机械工程学院,秦皇岛066004 [2]河北省重型机械流体动力传输与控制实验室,秦皇岛066004 [3]中国航发北京航科发动机控制系统科技有限公司,北京102200

出　　处：《中国机械工程》2018年第5期505-510,共6页China Mechanical Engineering

基　　金：国家重点基础研究发展计划(973计划)资助项目(2014CB046405);国家自然科学基金资助项目(51375423;51505410)

摘　　要：针对标准BP神经网络用于故障诊断时学习效率低、收敛速度慢、易陷入局部极小点及对初始参数较为敏感等不足,提出了一种组合优化的方法,即采用遗传算法(GA)确定BP神经网络的最佳初始权值矩阵,以规避BP神经网络对初始参数较为敏感的不足;应用LM(Levenberg-Marquardt)算法在局部解空间里对BP神经网络进行精确训练,搜索全局最优解。该方法在保留BP神经网络的广泛映射能力的前提下,提升了网络的学习速度和精确搜索能力,进而大幅提高了基于BP神经网络的电液伺服阀故障诊断的效率和精度。通过对MOOG D761-2716A机械反馈伺服阀进行故障诊断,进一步说明了该方法的实用性和高效性。The BP neural network had some shortcomings,such as the learning speed was very slow,easy to fall into local minima,and was sensitive to initial parameters.To improve the precision and speed of fault diagnosis of electro hydraulic servo valves,a new combination optimization method was proposed.GA was used to optimize the initial weights and thresholds of the neural network to improve the training speed and to reduce the BP neural network's sensitivity to initial parameters.And LM algorithm was used to train accurately and search for the global optimal solution in the local solution space.In the premise of preserving the mapping ability of the BP neural network,the method may improve the learning speed and accuracy of the networks,and thus the efficiency and accuracy of fault diagnosis of electro hydraulic servo valves may be greatly improved.At last,the MOOG D761-2716 Aservo valve's diagnosis results were given,which explained the practicality and efficiency of this method.

关 键 词：机械装备 电液伺服阀 故障诊断 BP神经网络 GA+LM算法 

分 类 号：TH137[机械工程—机械制造及自动化]