Vibration Characteristics Prediction of Functionally Graded Material Plates using FEM and Artificial Neural Networks

Abstract

The present work proposes a precision model to predict the frequencies of FGM plates using Artificial Neural Network (ANN). The material properties of the ceramic-metal composite change with the thickness using a power-law distribution. The Finite Element Method (FEM) simulation of bending and shear deformation with Third-Order Shear Deformation Theory (TOSDT) was used to generate a comprehensive dataset and depicts the bending and shear deformation without requiring any shear correction factors. The Levenberg-Marquardt algorithm is used to train a feed-forward back propagation network. The 7-26-2 architecture is determined as the most robust in the systematic optimization of the hidden layer architecture, producing regression coefficients that were above 0.9998 with low Mean Squared Errors. Comparison of results with numerical values of various boundary conditions, including all sides clamped, one side clamped, and the rest of the sides free supported showed very good agreement with minimal deviations. The parametric analysis revealed that both plate thickness and index value of power law are the most significant variables, which have a direct proportional relationship with natural frequencies. Conversely, material gradation index and mass density have a negative correlation with the dynamic response. These findings validate the fact that trained ANN model is a very effective surrogate tool which can be used to make near-instantaneous predictions. The strategy is a valuable and fast evaluation measure of engineering design and optimization of advanced FGM structural components.