模糊核聚类相关向量机模拟电路故障诊断

Fault Diagnosis in Analog Circuit Based on Relevance Vector Machine

针对模拟电路故障与特征间存在的模糊组及交叠,首先建立基于Fisher准则函数的最佳聚类数自适应估计方法,采用模糊核聚类选择最优可诊断故障集,然后提出一种基于稀疏贝叶斯相关向量机(RVM)理论的模拟电路故障诊断模型,提高了RVM模拟电路故障分类的效率和准确度;模型可以在贝叶斯框架下对分类函数的权重进行推断,而且得到各分类的验后概率,从而能判断分类结果的置信度,辅助进行诊断决策;仿真结果表明提出的模拟电路诊断模型在精度提高的情况下,比支持向量机需要的向量更少,更具稀疏性和泛化性,是一种有效的模拟电路故障诊断方法。

In order to reducing fuzzy group and overlap of analogous circuit between fault and feature, this article , first, establishes auto--adapted estimating approach of best cluster number based on Fisher criterion function, fuzzy nuclear cluster is used to select best diagnos-able fault component set, and then proposes an analog circuit fault diagnosis model based on relevant vector machine （RVM） from the sparse Bayesian theory, RVM can infer the discriminant function under the Bayesian framework. Moreover, it can obtaining posterior probability of each classification, thus can judge the degree of confidence of classification result, assist diagnosis decision--making. The result indicate that RVM need less RVs than SVs with comparative default accuracy, sparser and generalizing, it is an effective approach for analog circuits fault diagnosis.