Fault Diagnosis with Statistical Properties and Implementation Using MKSVM for Flash ADC

Abstract

This paper focuses on the Fault Diagnosing methodologies crucial for attaining reliability and maintainability of all electronic circuits. Fault Diagnosis (FD) is implemented for analog to digital converter (ADC) with a wide range of faults. FD is considered as the pattern recognition problem and solved by machine learning theory. Functional test is is needed instead of structural test for testing complex circuits. Fault diagnosis using Fault Dictionary, Artificial Neural Networks (ANNs) and Fuzzy logic are enigmatic or inconclusive diagnosis results which have more debug duration and even inaccurate repair actions that exponentially rises service overhead. The effectiveness of these methods are considered, which cover ability in detecting, identifying and localization of faults, the ability of analysing linear and nonlinear circuits, etc. Recent machine learning techniques like support vector machines (SVM) with kernel functions improve the preciseness of functional FD which reduces the product cost through correct repair process. The proposed Multikernel SVM (MKSVM) methodology give better results than earlier methods as it works with the fundamentals of machine learning and generalization for FD.