Physics-Informed Neural Network Approach for Surface Wave Propagation in Functionally Graded Magnetoelastic Layered Media

TL;DR

This paper presents a physics-informed neural network (PINN) framework for analyzing surface wave dispersion in complex layered magnetoelastic media, validated against analytical solutions and systematically studied for various parameters.

Contribution

The study introduces a novel PINN-based method for dispersion analysis in layered magnetoelastic media, demonstrating high accuracy and exploring effects of different neural network configurations.

Findings

PINN results closely match analytical solutions.

The framework effectively captures phase velocity variations with material parameters.

Systematic investigation of neural network architectures improves model performance.

Abstract

This paper investigates propagation of SH-waves in a layered composite structure consisting of a pre-stressed functionally graded magnetoelastic orthotropic layer overlying a pre-stressed functionally graded orthotropic half-space under the influence of gravity. The study introduces a physics-informed neural network (PINN) framework for the dispersion analysis of SH-waves in the considered composite medium. As a benchmark, an analytical solution to the dispersion relation is derived and used to validate accuracy and reliability of the proposed PINN formulation. In the developed PINN model, the phase velocity corresponding to a prescribed wave number is treated as a trainable parameter, enabling the determination of the dispersion relation associated with the nonlinear eigenvalue problem. The Adam optimizer is employed to minimize the loss function during the training process. In addition, the effects of different activation functions and network architectures, including variations in number of hidden layers and neurons, are systematically investigated to study the performance of the proposed framework. Error analysis is carried out using several norms, namely $L_1$, $L_2$, RMSE, relative absolute error, and $L_\infty$, to assess the accuracy of the predictions. Furthermore, the variation of phase velocity with wave number under different material parameters is investigated. The comparison between the analytical and PINN-based results demonstrates excellent agreement, confirming the effectiveness of the proposed deep learning approach for analysing dispersion relations in complex layered composite structures.