# Fast structural analysis of concrete thin-shells using deep learning

## Overview

This repository contains scripts and data supporting the following research article:
Pollet, M., Shepherd, P., Hawkins, W., & Costa, E. (2026). Fast structural analysis of concrete thin-shells using deep learning. Computers & Structures, 320, 108042. https://doi.org/10.1016/J.COMPSTRUC.2025.108042

It contains the trained Multilayer Perceptrons, Convolutional Neural Networks, and Graph Neural Networks used to obtain the results presented in the article, as well as Python scripts that were used to train these models and assess their performance.
Running these scripts necessitates the ConcreteShellFEA dataset (available at: https://doi.org/10.15125/BATH-01519) to be downloaded.
Additionally the path to this dataset needs to be specified in each script, under the DATASET_ROOT variable.
Further details about this data can be found in the related research article.

## Repository structure

```
FastStructuralAnalysisOfConcreteThinShellsUsingDeepLearning/
│── models/                              # Trained models
│   ├── CNN_buckling_best_model.pth      # Buckling prediction CNN used in paper
│   ├── CNN_stress_best_model.pth        # Stress prediction CNN used in paper
│   ├── GNN_buckling_best_model.pth      # Buckling prediction GNN used in paper
│   ├── GNN_stress_best_model.pth        # Stress prediction GNN used in paper
│   ├── MLP_buckling_best_model.pth      # Buckling prediction MLP used in paper
│   ├── MLP_stress_best_model.pth        # Stress prediction CNN used in paper
│── results/raw_predictions/             # Predictions on testing set
│   ├── CNN_buckling_best_preds.pth      # Predictions made by buckling CNN
│   ├── CNN_stress_best_preds.pth        # Predictions made by stress CNN
│   ├── GNN_buckling_best_preds.pth      # Predictions made by buckling GNN
│   ├── GNN_stress_best_preds.pth        # Predictions made by stress GNN
│   ├── MLP_buckling_best_preds.pth      # Predictions made by buckling MLP
│   ├── MLP_stress_best_preds.pth        # Predictions made by stress MLP
│── scripts/                             # Scripts for model training and evaluation
│   ├── datasets.py                      # Definition of dataset classes
│   ├── metrics.py                       # Definition of metric classes
│   ├── models.py                        # Definition of model classes
│   ├── tools.py                         # Definition of various tool classes and functions
│   ├── train_CNN_buckling.py            # Buckling prediction using CNN
│   ├── train_CNN_stress.py              # Stress prediction using CNN
│   ├── train_GNN_buckling.py            # Buckling prediction using GNN
│   ├── train_GNN_stress.py              # Stress prediction using GNN
│   ├── train_MLP_buckling.py            # Buckling prediction using MLP
│   ├── train_MLP_stress.py              # Stress prediction using MLP
│── requirements.txt                     # List of dependencies
│── README.md                            # Repository documentation
```

## License

The contents of this dataset are licensed under Creative Commons: Attribution 4.0 (CC BY), which allows for the derivation of new resources from the data, and for the redistribution of the data and their derivations, both openly and commercially, as long as the creators of this repository are credited for the original creation.

## Crediting/citing the creators of this dataset

The creators of this repository can be acknowledged by citing the related research article:

Pollet, M., Shepherd, P., Hawkins, W., & Costa, E. (2026). Fast structural analysis of concrete thin-shells using deep learning. Computers & Structures, 320, 108042. https://doi.org/10.1016/J.COMPSTRUC.2025.108042

or using BibTeX:

```
@article{Pollet2026,
   author = {Maxime Pollet and Paul Shepherd and Will Hawkins and Eduardo Costa},
   doi = {10.1016/J.COMPSTRUC.2025.108042},
   issn = {0045-7949},
   journal = {Computers \& Structures},
   month = {1},
   pages = {108042},
   publisher = {Pergamon},
   title = {Fast structural analysis of concrete thin-shells using deep learning},
   volume = {320},
   url = {https://linkinghub.elsevier.com/retrieve/pii/S0045794925004006},
   year = {2026}
}
```

## Article abstract

The present paper investigates the use of deep learning models as fast structural analysis tools for the design of concrete thin-shells. A dataset of 20,000 thin-shells with various geometric and material properties is generated. The buckling factor and the stress fields of each thin-shell under design loads are determined using Finite Element analysis. Three different types of deep learning models – Multilayer Perceptron (MLP), Convolutional Neural Network (CNN) and Graph Neural Network (GNN) – are then trained for buckling and stress prediction. For both prediction tasks, the MLP and the CNN are found to be the best performing models, reaching errors below 0.31 % for buckling prediction, and below 0.51 % for peak stress prediction. These results demonstrate the ability of such models to act as fast structural analysis tools for concrete thin-shells. Deep learning models could therefore enable faster and wider design space exploration during the shape optimisation of concrete thin-shells.