Application of machine learning in analysing flexible plate governed by the Mooney Rivlin model

Shravan Kumar Bhadoria, Gyan Vikash, Ramesh Gupta Burela,
Application of machine learning in analysing flexible plate governed by the Mooney Rivlin model,
Next Research,
Volume 2, Issue 4,
2025,
100901,
ISSN 3050-4759,
https://doi.org/10.1016/j.nexres.2025.100901.
(https://www.sciencedirect.com/science/article/pii/S3050475925007687)
Abstract: In the present study, a supervised physics-based machine learning model is employed to analyze the response of a geometrically and materially nonlinear flexible plate governed by the compressible Mooney–Rivlin model. The required physics-based dataset is generated using Isight Simulia. The Design of Experiments method, specifically Latin Hypercube Sampling, is employed to create the dataset for Machine Learning implementation. The problem involves seven input parameters: three geometric parameters (plate length, width, and thickness), three material parameters, and one load parameter. The three output parameters are the maximum displacement, the maximum principal nominal strain, and the maximum von-Mises stress. A dataset of 4000 design points is generated using the Latin Hypercube method in Isight Simulia. An artificial neural network utilizing the Swish activation function and 3 hidden layers with the steepest descent gradient method as the optimization method is used to best fit the physics-based data. For the test cases, artificial neural network predictions are compared with three-dimensional finite element analysis and the variational asymptotic method; displacement and von-Mises stress show close agreement with three-dimensional finite element analysis (98.3% and 97.7% averaged accuracy, respectively), while noticeable differences are observed in the maximum principal nominal strain.
Keywords: Finite element method; Applied machine learning; Neural network; Design of experiment; Hyperelasticity; Variational asymptotic method