Analog fault diagnosis: a fault clustering approach. European Test Conference

A novel analog circuii f irul t dicignosis ineihod is proposed. This method uses a neural network paradigm to cluster different faulis. I t is capable of dealing with ihe common fault models in ~ n a l o g circuiis, nomely ihe catastrophic and paranieiric fauiis. The proposed technique is independent of ihe lineariiy or nonlineariiy of the circuii. The process pciramcier drip fis rind coinponeni iolerance effects of ihe circuii ore well iciken carc of: Several fault diagnosis siraicgies for differeni problem complexities are described. The proposed meihodology is illustrated by mean.7 of (in Opcruiinnnl 7r~tiscondrtci~t~1:c Amplifier (O lA) example. npproacli capahlc to ovcrcomc thc abovc limitations. Thc basic idea of this incthodology consisLs of training a ncural nctwork paradigm. Thc training input vcclor has a set of data points consisting of familics o f simulatcd faulty as well a s friult frcc bchaviors of thc circuit. Thcse inpct vcctors arc callcd signuiures. Thc neural nctwork palati igm cl assi fics these signa turcs into di ffercn t clusters. The actual expcrimcntlil vector callcd the lest vector, oblaincd from thc circuit undcr tcst, is thcn prescnted to the ncural network. The network outputs thc clustcr of known signaturc(s) that bcst matchcs this tcst vector. This approach allows thc tcsting as wcll as thc diagnosis of the circuit. Scvcral dcgrccs of complcxity of the proposed technique arc introduccd in this papcr. The practical and thcorctical aspccts of this techniquc arc verified and illustrated by inciins ol 'an IC cxample.