Dictionary method for multiple soft and catastrophic fault diagnosis based on evolutionary computation

In this paper a new method for detection, location and identification of multiple soft and catastrophic faults in linear and nonlinear analog circuits is described. It is based on the node approach and two evolutionary methods: gene expression programming (GEP) and genetic algorithm (GA). On the before test stage the algorithm needs repeated analyses of circuit-under-test (CUT) with different values of parameters which can be faulty. The results form the training set for constructing a dictionary. It consists of all sets created with groups of possible faulty parameters, associated with them formulas for computing these parameters, the lower and the upper limits of node voltages of the circuit with faulty parameters. These formulas are determined using GEP, one of the new evolutionary methods. On the after test stage the method obtains the set of possible faulty elements and calculates the values of possible faulty elements. The accuracy of the fault identification of this results is usually unsatisfactory. The requested precision of fault detection can be with the already known method for soft fault diagnosis accomplished. The time consumed for obtaining formulas enabling us to calculate faulty parameters increases with the number of possible faulty elements and the number of simultaneously faulty elements. The results of the method depend on parameters of GEP process.