摘要
提出一种神经网络初步诊断和D-S证据理论融合决策诊断相结合的分层信息融合故障诊断策略,并建立相应的功能模型。在神经网络初步诊断层中,首先,把待诊断设备的故障特征参数空间划分为若干个子参数空间,同时根据各子参数空间构造相应的故障子空间;其次,根据各子参数空间的定义和相应的子故障空间构造独立的诊断子网络以及相应的学习样本;然后,构造多个独立的诊断子网络,利用差分进化算法的全局寻优能力,得到神经网络的权值和阈值,对网络进行训练;最后,对训练好的网络进行测试,确定出各个子网络的性能,为以后的融合决策诊断做准备。在融合决策诊断层中,用初步诊断层中各子网络的输出结果构造证据体,通过证据融合推理分析得出最终诊断结论。通过差分进化神经网络与D-S证据理论的结合,解决了随着诊断参数的增多,神经网络结构逐渐庞大而造成网络学习困难、网络识别精度下降等问题。借助于D-S证据理论可以有效地把各诊断子网络的输出结果进行融合决策,最后以变速箱轴承故障诊断为例,验证该方法的可行性和有效性。
This paper presents a fault diagnosis strategy of stratified information fusion that combines initial diagnosis of neural network with fusion decision diagnosis of D-S evidence theory,and sets up a corre-sponding function model.In the initial diagnosis of neural network,firstly,fault feature parameter space of the equipment to be diagnosed is divided into several sub-parameter space,and corresponding fault subspace is constructed according to each sub-parameter space;secondly,the corresponding sub-network learning samples are independently constructed according to the definition of each sub-parameter space and the corresponding sub-fault space;thirdly,by constructing multiple independently diagnosis of sub-networks and using the global optimization ability of differential evolution algorithm can fast get neural network weights and thresholds in order to train the network;finally,test the trained network to determine the performance of each sub-network,preparing for the later fusion of decision-making diagnosis.The body of evidence is constructed by using the output of each sub-network in the initial diagnosis of the integration of decision-making diagnosis.Then the final diagnosis conclusion is made by reasoning and analyzing the fusion of evidence.This paper,combining differential evolutionary neural network with D-S evidence theory,solves a series of problems produced by the increase of diagnostic parameters and network learning difficulties,identification accuracy decline caused by the gradually huge of neural network structure.At the same time,D-S evidence theory can effectively make fusion diagnosis of the output results of each sub-network.Finally,this paper takes the gearbox bearing fault diagnosis for example to verify the feasibility and effectiveness of the method.
出处
《振动.测试与诊断》
EI
CSCD
北大核心
2013年第S2期137-143,224-225,共9页
Journal of Vibration,Measurement & Diagnosis
基金
国家自然科学基金资助项目(51175179)
关键词
差分进化
神经网络
D-S证据
信息融合
故障诊断
differential evolution,neural network,D-S evidence,information fusion,fault diagnosis