Simulation & Identification of Sheet Defects Using FEM & ANN

Abstract

In this work simulation and identification of defect in metallic wall is done using the Finite element method (FEM) and Artificial Neural Network (ANN). The problem of defect identification is an inverse problem which has been solved in this work. Recently, ANNs are introduced to solve the inverse problems in most of the research applications in industrial nondestructive testing, mathematical modeling, medical diagnostics, geophysical prospecting for petroleum and minerals, and detection of earthquakes. Here FEM is used for the analysis of change in magnetic induction properties on the account of defect present in the metallic sheet. Now for the simulation of these defects on metallic wall both variations in width and height of the defects are considered. The simulation is done using the Pdetool of Matlab. The obtained results are used to generate a set of vectors for the training of neural network (NN). The basic problem of training of NN is that it requires some input data for training by simulation using FEM this problem has been solved as this provide an input data for training of ANN. Therefore NN has been trained for the given dataset, finally, the obtained neural network is used to identify a group of new defects, simulated by the finite element method, but not belonging to the original dataset. The network can be embedded in electronic devices in order to identify defects in real metallic walls. The association of FEM and ANN techniques seems to be a useful alternative for identification of defects trough inverse analysis. Future works are intended to be done in this field, such as the use of more realistic FEM, computer parallel programming, in order to get quickly solutions.