← Latest papers
🤖 machine learning

A mesh-free deep energy method for phase-field modeling of thermally driven brittle fracture

This paper introduces a mesh-free deep energy method utilizing multiresolution neural networks to simulate spontaneous nucleation, propagation, and arrest of cracks in thermally driven brittle fracture without requiring pre-defined defects or meshes, demonstrating accuracy against finite element references and experimental benchmarks.

Original author: Han Zhang, Mehrisadat Makki Alamdari, Babak Shahbodagh, Mohammad Vahab, Cosmin Anitescu, Timon Rabczuk, Elena Atroshchenko

Published 2026-11-16
📖 5 min read🧠 Deep dive

Original author: Han Zhang, Mehrisadat Makki Alamdari, Babak Shahbodagh, Mohammad Vahab, Cosmin Anitescu, Timon Rabczuk, Elena Atroshchenko

Original paper dedicated to the public domain under CC0 1.0 (http://creativecommons.org/publicdomain/zero/1.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the author. For technical accuracy, refer to the original paper. Read full disclaimer

When a hot piece of glass or ceramic is suddenly plunged into cold water, it does not simply cool down; it often shatters in a startlingly organized way. The surface contracts rapidly while the interior remains hot and expanded, creating immense internal tension. This stress forces the material to crack, but the cracks do not appear randomly. Instead, they often form neat, parallel lines that grow deeper into the material, or they might wiggle back and forth in a wave-like pattern before splitting into branches. For engineers designing everything from jet engine components to smartphone screens, predicting exactly where these cracks will start, how far they will travel, and how they will arrange themselves is a critical challenge. Traditional computer simulations struggle with this task because they rely on a fixed grid, like graph paper, to calculate the physics. When a crack forms, it cuts through this grid, forcing the computer to constantly redraw the map to follow the jagged path. This process is slow, and if the computer guesses the wrong place to start the crack, the entire simulation can fail to capture the complex, spontaneous patterns that nature creates.

A team of researchers has developed a new way to simulate this violent process without using a grid at all. Instead of relying on a fixed mesh of points, their method uses artificial intelligence to treat the entire object as a continuous, smooth surface that can be analyzed from any angle. They built a system where two neural networks, which are computer programs designed to recognize patterns, work together to solve the physics of the problem. One network tracks how the material moves and where it breaks, while the other calculates how heat flows through it. Crucially, this system does not need to be told where a crack will begin. It allows the crack to nucleate spontaneously, just as it does in reality, emerging from the stress of the cooling process itself. By letting the computer learn the shape of the crack as it forms, rather than forcing the crack to fit a pre-drawn path, the researchers can observe how complex fracture patterns evolve under thermal shock.

The researchers tested this approach on five different scenarios to see if it could match the behavior of real materials and existing computer models. In one set of tests, they simulated a metal plate with a small notch cut into it, subjecting it to different temperatures while pulling it apart. The new method predicted the exact force required to start the crack with an accuracy within about one percent of a standard, trusted computer model. This confirmed that the AI system could handle the basic physics of heat and stress working together. In another test, they simulated a strip of glass being pulled from a hot region into a cold bath. As the temperature difference increased, the simulated crack path changed from a straight line to a wavy, oscillating line, and finally to a branched pattern. This sequence of changes matched what scientists have observed in real experiments and what other theories predict, showing that the method captures the subtle transitions in how materials fail.

Perhaps the most striking results came from simulating a ceramic slab being quenched, or cooled rapidly, from all sides. In these simulations, the material spontaneously developed a dense array of parallel cracks, with some growing deep into the slab while others stopped short. The spacing and depth of these cracks closely matched a famous benchmark calculation and real-world photographs of shattered ceramic. However, the new method revealed something interesting: it predicted a slightly denser pattern of cracks than the traditional grid-based models did. This suggests that the way the new method handles the "irreversibility" of cracking—meaning that once a material is damaged, it cannot heal itself—allows it to capture more of the fine details of the fracture process. The researchers also applied the method to a curved ceramic ring, where the cracks formed in a similar hierarchical pattern, confirming that the approach works on complex shapes without needing to reshape the computer's internal map.

The success of this work lies in how it handles the two competing demands of the problem: the thin layer of heat that forms on the surface and the narrow band where the material breaks. In the past, computer models had to choose between resolving the heat or the crack, often missing one or the other. The new method uses a flexible, multi-layered representation that can zoom in on the heat boundary layer when it is thin and follow the crack as it widens, all without ever needing to redraw a grid. While the researchers note that their simulations are still being refined and that the method currently takes longer to run than some traditional approaches for simple problems, the ability to predict spontaneous crack formation without human intervention is a significant step forward. It offers a powerful new tool for understanding how brittle materials fail under thermal stress, potentially helping engineers design components that are more resistant to the sudden temperature changes they face in the real world.

Drowning in papers in your field?

Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.

Try Digest →