Modeling of 2D Functionally Graded Circular Plates with Artificial Neural Network

Abstract

The thermo-mechanical properties of the functionally graded material (FGM) depend on the volumetric distribution that determines the material character, which is very important in order to overcome different operating conditions and stress levels. Three different training algorithms are used in an Artificial Neural Network (ANN) to determine the equivalent stress levels of a hollow disc that is functionally graded in two directions. The data set was created by choosing the most important four different equivalent stress values (σ_(eqv max max) ,σ_(eqv max min) ,σ_(eqv min max) ,σ_(eqv min min)) that determine the material structure in thermo-mechanical analysis. Performance estimation was performed in three different training algorithms (Gradient Descent Backpropagation, Gradient Descent with Momentum Backpropagation, BFGS Quasi-Newton Backpropagation Algorithm). In this study, termomechanical behaviour was numerically determined by using finite difference method at different compositional gradient upper values to train ANN.