# Learning to fail: Predicting fracture evolution in brittle material models using recurrent graph convolutional neural networks

arXiv: https://arxiv.org/abs/1810.06118 · HTML: https://mlsanity.com/paper/1810.06118/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 14 Oct 2018.

## Built on by (most cited first, 6 of 6)

- [Materials Property Prediction with Uncertainty Quantification: A Benchmark Study](https://mlsanity.com/paper/2211.02235/) (2022)
- [A Data-Driven Approach to Full-Field Damage and Failure Pattern Prediction in Microstructure-Dependent Composites using Deep Learning](https://mlsanity.com/paper/2104.04485/) (2021)
- [Applications of physics-informed scientific machine learning in subsurface science: A survey](https://mlsanity.com/paper/2104.04764/) (2021)
- [Spatiotemporal Predictions of Toxic Urban Plumes Using Deep Learning](https://mlsanity.com/paper/2406.02582/) (2024)
- [Neuro-DynaStress: Predicting Dynamic Stress Distributions in Structural Components](https://mlsanity.com/paper/2301.02580/) (2022)
- [HOSSnet: an Efficient Physics-Guided Neural Network for Simulating Crack Propagation](https://mlsanity.com/paper/2306.08783/) (2023)

## Abstract (arXiv)

We propose a machine learning approach to address a key challenge in materials science: predicting how fractures propagate in brittle materials under stress, and how these materials ultimately fail. Our methods use deep learning and train on simulation data from high-fidelity models, emulating the results of these models while avoiding the overwhelming computational demands associated with running a statistically significant sample of simulations. We employ a graph convolutional network that recognizes features of the fracturing material and a recurrent neural network that models the evolution of these features, along with a novel form of data augmentation that compensates for the modest size of our training data. We simultaneously generate predictions for qualitatively distinct material properties. Results on fracture damage and length are within 3% of their simulated values, and results on time to material failure, which is notoriously difficult to predict even with high-fidelity models, are within approximately 15% of simulated values. Once trained, our neural networks generate predictions within seconds, rather than the hours needed to run a single simulation.