Neural network-based expert system to predict the results of finite element analysis

Abstract

Realizing the fact that the performance of a finite element (FE) analysis depends on the type of elements, mesh topology, mesh density, node numbering and others, an attempt is made in the present work, to develop a neural network-based expert system to predict stress analysis results of an FEM package, within a reasonable accuracy. A rubber cylinder is compressed diametrically between two plates, whose induced stresses and deformed shape are to be determined using an FE analysis. By varying two parameters, namely element size and shape ratio, results (obtained through FE analysis) in terms of induced stress and deformed shape of the cylinder are recorded, which are utilized to train a neural network (NN)-based expert system, by using a back-propagation algorithm and a genetic algorithm, separately. Results of two NN-based expert systems are compared, in terms of accuracy in prediction of the results. It is interesting to note that the expert system can predict the results within a fraction of a second, whereas an actual FE analysis may take several seconds depending on the complexity of the problem.