Loading...
Hyperparameter optimization for causal marching physics informed neural network for hyperelasticity
Files
Citations
Altmetric:
Publication Date
2024-06-03
Type
conference paper
Downloads
Citation
Pratap, Vikrant, Gilchrist, Michael D., & Tripathi, Bharat B. (2024). Hyperparameter optimization for causal marching physics informed neural network for hyperelasticity. Paper presented at the 9th European Congress on Computational Methods in Applied Sciences and Engineering ECCOMAS Congress 2024, 03-07 June
Abstract
This study presents an approach for hyperparameter optimization in the Causal Marching Physics-Informed Neural Networks (CMPINNs) framework, specifically designed to model hyperelasticity. Physics-Informed Neural Networks (PINNs) are powerful tools for solv ing governing partial differential equations (PDEs) in physical systems. The CMPINNs model proposed in this work enhances the PINN framework by minimizing the residuals of the gov erning PDEs while enforcing the boundary conditions for the nonlinear mechanical responses of hyperelasticity. We study the accuracy of using CMPINNs to solve the Neo-Hookean hy perelastic model using soft and hard constrained boundary conditions. Additionally, the study presented a hyperparameter optimization for CMPINNs to identify the best suitable set of hy perparameters for deformation like biaxial compression. This optimization process ensures that the CMPINN effectively captures the complex stress-strain relationships in hyperelastic mate rials under deformation. This research advances the development of robust, physics-informed computational models for hyperelastic materials, reducing reliance on labelled or synthetic data.
Funder
Publisher
University of Galway
Publisher DOI
Rights
CC BY-NC-ND