Coupled Multiphysics FEM Datasets for Deep Learning Supporting I-FENN with DeepONets

This repository contains the datasets supporting the paper “I-FENN with DeepONets: Accelerating simulations in coupled multiphysics problems”. DOI: https://doi.org/10.1016/j.cma.2025.118645 The source code used for dataset generation, DeepONet training, and implementation of the Integrated Finite Element Neural Network (I-FENN) framework is publicly available at:https://github.com/FouadAmin/I-FENN-DeepONet-Multiphysics The datasets consist of finite element method (FEM) simulation data generated for coupled multiphysics problems, specifically thermoelasticity and poroelasticity, and are used for training, validation, and testing of Deep Operator Networks (DeepONets) within the Integrated Finite Element Neural Network (I-FENN) framework. The data were generated using fully coupled, monolithic FEM solvers implemented using the deal.II library and represent time-dependent multiphysics responses under diverse loading conditions. The datasets include simulations over two-dimensional and three-dimensional domains. The provided data covers three examples: 1-Thermoelasticity (3D cube, thermal body load): Time-dependent FEM datasets of a 3D cube under spatially and temporally varying thermal body loads, providing strain-trace, temperature, and displacement fields. 2-Thermoelasticity (3D thick-walled tube, thermal surface load): Time-dependent FEM datasets of a 3D thick-walled tube under spatially and temporally varying thermal surface flux, providing strain-trace, temperature, and displacement fields. 3-Poroelasticity (2D excavation, fluid flux): Coupled poroelastic FEM datasets of a 2D excavation problem with transient fluid flux boundary conditions, containing displacement, strain-trace, and pressure fields. The datasets are stored in binary HDF5 (.h5) format to enable efficient handling of large-scale spatiotemporal data and high-dimensional fields. Due to their size, the datasets are distributed across multiple files. These data are intended to support: Reproducibility of the numerical results reported in the associated paper Benchmarking of hybrid FEM–machine learning frameworks Further research on operator learning, multiphysics simulation acceleration, and generalization in data-driven and hybrid computational mechanics