MEchanics GAthering –MEGA- Seminar: How segmented 3D cracks store energy + Converging Periodic Orbits in Fluid Flows Using Data-Driven Methods
Event details
| Date | 04.12.2025 |
| Hour | 13:05 › 14:00 |
| Speaker |
Xinyue Wei (EMSI, EPFL) Pierre Beck (ECPS, EPFL) |
| Location | Online |
| Category | Conferences - Seminars |
| Event Language | English |
Abstract 1: Crack propagation is one of the main causes of material failure. While the classical theory of linear elastic fracture mechanics is based on planar assumptions, cracks in reality are intrinsically three-dimensional. A typical complex crack often consists of segmented crack fronts bridged by a thin material ligament. These ligaments are universal features at the complex crack fronts, yet their mechanical role has remained unclear. Using light sheet microscopy and particle tracking, we directly resolve the in-situ kinematics around 3D crack fronts in brittle hydrogels. Our results show that the apparent fracture toughness of the complex cracks is proportional to the elastic energy stored within the material ligament, linking local crack-front structure to global fracture toughness and underscoring the need for fully three-dimensional energetic descriptions of brittle failure.
Abstract 2: Unstable periodic orbits (UPOs) are the non-chaotic, dynamical building blocks of spatio-temporal chaos. Their computation, however, is challenging due to two main issues: the system's chaotic nature and the large number of spatial discretization variables. We tackle both of these issues at once by using data-driven dimensionality reduction in order to implement a convergence algorithm for UPOs directly within a low-dimensional latent space. The convergence algorithm avoids time-integration, thus taming the chaos, and is based on a latent dynamics obtained by pulling the physical equations into the latent space using the chain rule. Crucially, this preserves the structure of the attractor, which we show by demonstrating an equivalence between the latent UPOs and their physical counterparts for a model PDE and the 2D Navier-Stokes equations.
Bio 1: Xinyue is a PhD student at EMSI lab at EPFL, where she investigates the fracture of brittle hydrogels. Xinyue received her bachelor’s degree at Shanghai Jiao Tong University and her master’s degree at the University of Pennsylvania
Bio 2: Pierre Beck received his Bachelor's and Master's degrees in Mathematics from the University of Cambridge. After spending two years at the Central Bank of Luxembourg, he began his PhD in the Emergent Complexity in Physical Systems (ECPS) lab at EPFL in September 2022. In his research, he is interested in the dynamical systems approach to fluid dynamics. In particular, he cares about the computation of simple invariant solutions of chaotic fluid flows with recent tools from machine learning.
Abstract 2: Unstable periodic orbits (UPOs) are the non-chaotic, dynamical building blocks of spatio-temporal chaos. Their computation, however, is challenging due to two main issues: the system's chaotic nature and the large number of spatial discretization variables. We tackle both of these issues at once by using data-driven dimensionality reduction in order to implement a convergence algorithm for UPOs directly within a low-dimensional latent space. The convergence algorithm avoids time-integration, thus taming the chaos, and is based on a latent dynamics obtained by pulling the physical equations into the latent space using the chain rule. Crucially, this preserves the structure of the attractor, which we show by demonstrating an equivalence between the latent UPOs and their physical counterparts for a model PDE and the 2D Navier-Stokes equations.
Bio 1: Xinyue is a PhD student at EMSI lab at EPFL, where she investigates the fracture of brittle hydrogels. Xinyue received her bachelor’s degree at Shanghai Jiao Tong University and her master’s degree at the University of Pennsylvania
Bio 2: Pierre Beck received his Bachelor's and Master's degrees in Mathematics from the University of Cambridge. After spending two years at the Central Bank of Luxembourg, he began his PhD in the Emergent Complexity in Physical Systems (ECPS) lab at EPFL in September 2022. In his research, he is interested in the dynamical systems approach to fluid dynamics. In particular, he cares about the computation of simple invariant solutions of chaotic fluid flows with recent tools from machine learning.
Practical information
- General public
- Free
Organizer
- MEGA.Seminar Organizing Committee