Analog fault diagnosis in nonlinear DC circuits with an evolutionary algorithm

In this paper we present a technique for analog fault diagnosis (AFD)based on an evolutionary algorithm. The nonlinear DC circuit equations are written with modified nodal analysis (MNA), Parametric fault models and/or fault compensation models are implemented for linear circuit elements and for common nonlinear devices (diodes, BJTs, MOSFETs, OPAMPs), The diagnosis is performed by solving a nonlinear program (NLP) built with the circuit equations, with the measurements taken in the circuit and with the tolerances of the circuit elements and devices. This NLP is solved by minimizing a function of the fault variables and of the unknown circuit variables, with an evolutionary algorithm. The population in the evolutionary algorithm consists of subpopulations (with one or more individuals) each of them related to a specific fault. The technique was implemented with a suite of Perl programs and was applied to several examples. Some conclusions on the effectiveness and robustness of the evolutionary algorithm to perform analog fault diagnosis are presented.