A 2D discrete model with a phase-field regularisation applied to dynamic crack propagation

When a brittle material like glass or hardened steel breaks, it does not simply snap in half. Instead, a crack races through the solid, often branching into a complex web of fractures that shatters the object into many pieces. Predicting exactly where these cracks will travel and how fast they will move is one of the most difficult challenges in engineering. If a model used to simulate this destruction is flawed, it might suggest a bridge will fail in a safe direction when it actually collapses catastrophically. For decades, scientists have tried to build computer models that mimic the atomic structure of materials, treating them as a grid of tiny points connected by springs. While these models are powerful, they suffer from a peculiar flaw: the path the crack takes often depends on how the computer grid is rotated, rather than on the physics of the material itself. It is as if a crack in a real piece of glass would suddenly change direction simply because the engineer turned the computer screen.

To solve this problem, Matías Braun, a researcher at Universidad Nebrija in Spain, has developed a new way to simulate dynamic fracture. His approach combines a traditional grid of points with a "phase-field" method, a technique that treats the crack not as a sharp line that jumps from point to point, but as a smooth, fuzzy zone of damage that spreads through the material. By blending these two ideas, Braun created a model that allows cracks to travel in any direction, regardless of how the underlying computer grid is oriented. He tested this new method on three distinct scenarios: a steel plate struck by a high-speed impact, a notched plate pulled apart until it snaps, and a thick metal cylinder exploding from the inside out. In every case, the new model produced results that matched real-world experiments and eliminated the artificial bias caused by the computer grid, offering a clearer, more reliable view of how materials truly fail.

The core of the problem Braun addressed lies in how computers handle the invisible world of atoms. In his earlier work, he used a discrete model where the material was represented by a triangular lattice, a pattern of dots connected by lines. When the material was stressed, the connections between the dots would break, simulating a crack. However, because the dots were arranged in a fixed pattern, the crack was forced to follow the lines of the grid. If the grid was rotated, the crack would change its path, even though the material itself had not changed. This is a significant issue because, in reality, a crack should follow the path of least resistance determined by the material's properties, not the arbitrary angle of a computer mesh.

Braun's solution was to introduce a phase-field regularisation. Instead of forcing the crack to break specific connections between dots, he assigned a value to each dot that represents how damaged it is. This value ranges from zero, meaning the material is perfectly intact, to one, meaning it is completely broken. The crack becomes a region where this value gradually increases, creating a smooth transition rather than a jagged line. This allows the crack to curve and branch in any direction, effectively ignoring the rigid structure of the grid. Crucially, Braun calibrated this system using two fundamental properties of the material: its tensile strength, which is the force required to start a crack, and its fracture energy, which is the energy needed to keep the crack growing. By anchoring the simulation to these real physical constants, the model ensures that the results depend on the material, not the math.

To prove that this new method worked, Braun ran three rigorous tests. The first was the Kalthoff–Winkler test, a classic experiment where a steel plate with a pre-cut notch is struck by a projectile at high speed. In this scenario, a crack shoots out from the notch at a specific angle. Previous models often predicted the wrong angle depending on how the grid was rotated. In Braun's simulations, the crack consistently traveled at an angle of approximately 70 degrees, matching the experimental observations perfectly. This result held true whether the grid was aligned with the plate or rotated by 15 or 30 degrees. The crack speed also remained stable, traveling at about 70 percent of the speed of sound in the material, independent of the grid's orientation.

The second test involved a single-edge-notched plate under tension, a setup used to study how cracks grow slowly under a steady pull. Here, the goal was to see if the crack would travel in a straight line from the notch to the opposite edge. Traditional grid models often forced the crack to zigzag or deviate based on the grid's alignment. Braun's model, however, produced a straight crack for every orientation of the grid. The force required to break the plate was also consistent, varying by less than two percent regardless of how the grid was rotated. This demonstrated that the model could accurately predict both the path and the strength of the material without the artificial influence of the computer mesh.

The final and most complex test was the fragmentation of a thick cylinder subjected to an internal explosion. In this scenario, no single crack is guided by a notch; instead, many cracks must nucleate spontaneously and race outward, shattering the cylinder into dozens of pieces. This is a chaotic event where the symmetry of the grid could easily dictate the pattern of the fragments. When Braun ran the simulation with the grid centered on the cylinder's axis, the cracks formed a symmetric pattern. However, when he shifted the grid slightly so it was no longer perfectly centered, the cracks distributed themselves irregularly, and the number of fragments changed. This confirmed that the model was sensitive to the physical setup rather than just the grid's symmetry. The simulation predicted between 12 and 19 fragments, a number that fell within the range of other advanced methods but was lower than some, likely because the smooth damage zone in the model prevented cracks that were too close together from forming.

Throughout these tests, the researchers carefully tracked the energy of the system. They compared the energy lost as the material broke against the energy required to create the new crack surfaces. In some cases, there was a small mismatch between the energy calculated by the damage model and the energy actually released by the breaking connections. This is a known limitation of the hybrid approach, where the driving force for the crack is calculated differently from the energy of the broken bonds. However, the mechanical energy balance remained accurate to within half a percent, confirming that the simulation was physically sound. The results suggest that while the model is not perfect, it represents a significant step forward in understanding dynamic fracture.

The implications of this work extend beyond just better computer graphics. By removing the dependence on grid orientation, engineers can trust that their simulations of crash tests, structural failures, or material testing are reflecting the true physics of the material. The model successfully bridges the gap between the discrete nature of atoms and the continuous behavior of solid objects. It shows that with the right mathematical tools, it is possible to simulate the chaotic, violent process of breaking without letting the computer's grid dictate the outcome. The study concludes that this approach offers a robust way to predict how materials fail under extreme conditions, providing a clearer path for designing safer and more resilient structures.