基于K-means聚类与概率神经网络的模拟电路故障诊断方法

Analog circuit fault diagnosis method based on K-means and probabilistic neural network

模拟电路已广泛应用于航空电子系统,模拟电路的失效会影响系统的功能,引起系统故障,甚至引发灾难性的安全事故。为快速准确地实现模拟电路的故障诊断,该文引入概率神经网络方法,并针对传统概率神经网络方法中的诊断准确性、诊断效率问题,提出基于K-means与概率神经网络的模拟电路故障诊断方法,定义聚类有效性指标,采用K-means聚类分析与有效性指标分析相结合的方式,选取聚类中心作为模式层神经元训练概率神经网络模型,从而降低模型的复杂程度,大大减少故障诊断时间。最后,以有源滤波电路为对象,通过与传统概率神经网络方法以及随机概率神经网络方法的对比分析,验证该文方法在故障诊断准确性以及故障诊断效率上的优越性能。

Analog circuit has been widely used in avionics system.The failure of analog circuit will affect the function of the system,cause system failure and even catastrophic accident.This paper proposes a combined method based on K-means and probabilistic neural network for analog circuit fault diagnosis.In view of the low diagnostic accuracy and efficiency of traditional probabilistic neural network,K-means clustering analysis is adopted to select clustering centers as pattern neurons.Thus,the complexity of the diagnostic model and the fault diagnosis time are reduced.Finally,the proposed method is compared with traditional probabilistic neural network and random probabilistic neural network,the experimental results of an active filter circuit verify the superior performance of the method in fault diagnosis accuracy and fault diagnosis efficiency.