基于MHMM模拟电路早期故障诊断

MHMM模型是一种基于高斯混合密度的连续隐马尔科夫模型,具有很好的模式识别能力,对于高混叠样本优势明显。模拟电路结构复杂,早期软故障呈现多样化,故障样本混叠严重,难以辨识。针对这个特点,提出了将MHMM模型应用于模拟电路早期故障诊断的新思路。首先,通过线性判别分析(LDA)技术将由仿真电路采集的数据样本进行降维处理,产生低维观测序列,并对样本初步划分;然后,使用高斯混合模型(GMM)对观测序列逼近,并完成MHMM模型的参数训练;最后,通过实例验证,并与BP网络进行比较。结果表明,MHMM对于早期故障的检测更具有优越性。

Incipient Fault Diagnosis of Analog Circuits Based on MHMM

MHMM is a Hidden Markov Model (HMM) model with mixture of Gaussians output,which has fine pattern recognition capability and is suitable for dealing with mixed samples.Analog circuit is complicated,the incipient faults are various,and the samples of incipient faults are mixed severely.Therefore,we put forward a MHMM based method for incipient fault diagnosis of analog circuits.First,the dimensions of the experimental samples were decreased with the Linear Discriminant Analysis (LDA) technology,and the observation sequences with the lower dimensions were obtained,and the samples were partitioned primarily.Then,the observation sequences were approached by Gaussian Mixture Model (GMM),and the MHMM model was built up by using them.The experimental results showed that:compared with BP neural network,MHMM method is more advantageous for incipient fault diagnosis.