Hybrid Computational Model (FE and ANN) for Prediction of Modal Frequency Response of Damaged Hyperelastic (Hydrogel) Panel Structure

Authors:

Abstract
This study presents an integrated finite element (FE) and artificial neural network (ANN) framework for predicting the modal frequency response of damaged agar-based hydrogel plates under varying geometry, boundary conditions, and damage severity. The aim of the work is to develop a computationally efficient hybrid methodology in which validated FE simulations are used to generate a reliable dataset, and ANN-based surrogate models are subsequently employed for rapid frequency prediction. The hydrogel material parameters were obtained from experimental compression data through curve fitting using an Ogden-type hyperelastic model. Thereafter, the frequency characteristics were computed using ABAQUS-based FE analysis, and the numerical consistency was ensured through mesh convergence and validation studies. A systematic parametric investigation was conducted by varying stiffness-sensitive parameters, namely aspect ratio, thickness ratio, boundary conditions, and damage length, to establish physically consistent frequency trends. Unlike earlier studies that mainly focus on either numerical simulation or data-driven prediction independently, the present work combines FE-based modal analysis with ANN surrogate modelling for frequency-based assessment of damage-induced stiffness degradation in hydrogel plates. Based on the generated dataset, five ANN surrogate configurations were trained and compared to identify the most efficient prediction model. The proposed FE-ANN workflow enables rapid frequency estimation and reduces the need for repeated FE simulations. The frequency increased from 6.4977 Hz to 30.093 Hz when the aspect ratio changed from 1 to 3.