基于LSSVM-FNFN的模拟电路故障诊断  

Fault Diagnosis of Analog Circuits Based on LSSVM-FNFN

作　　者：赵力 史贤俊 秦玉峰 ZHAO Li;SHI Xianjun;QIN Yufeng(Naval Aviation University,Yantai 264000)

机构地区：[1]海军航空大学,烟台264000

出　　处：《舰船电子工程》2023年第8期204-211,235,共9页Ship Electronic Engineering

摘　　要：为解决传统故障诊断方法在模拟电路的故障诊断中存在故障定位率低、软故障诊断性弱等问题,提出了一种LSSVM-FNFN的智能诊断方法。参考基于残差评估的诊断思想,设计了基于最小二乘支持向量机(Least Squares Support Vector Machine,LSSVM)的系统残差生成器,并利用高阶统计量作为特征向量,完成了对故障特征的有效提取;借鉴神经网络在模式识别方面的优势,设计了基于泛函模糊神经网络(Functional Neural Fuzzy Network,FNFN)的故障分类器,实现了对不同响应下的输出残差曲线进行故障识别;同时利用改进的粒子群算法优化选取LSSVM参数以及FNFN网络的权值,最后以某型装备的惯性测量组合为例进行实例验证,仿真结果表明,该方法生成的残差包含了丰富的故障信息且高阶统计量特征对故障信息进行了有效的提取,取得了很好的识别效果。An intelligent diagnosis method of LSSVM-FNFN is proposed to solve the problems of low fault localization rate and weak diagnosability of soft faults in the traditional fault diagnosis method of analog circuits.With reference to the diagnostic idea based on residual evaluation,a system residual generator based on least squares support vector machine(LSSVM)is designed,and high-order statistics are used as feature vectors to complete the effective extraction of fault features.With the advantage of neural net⁃works in pattern recognition,a generalized FNFN(Functional Neural Fuzzy Network)based fault classifier is designed to achieve fault identification of the output residual curves under different responses.An improved particle swarm algorithm is proposed to opti⁃mize the LSSVM parameters and FNFN network weights,which greatly improves the efficiency of the parameter selection.Finally,an example of a type of equipment the simulation results show that the residuals generated by the method contain rich fault informa⁃tion and the higher-order statistical features can effectively extract the fault information and achieve a good recognition effect.

关 键 词：最小二乘支持向量机 高阶统计量 泛函模糊神经网络 粒子群算法 故障诊断 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置]