MEchanics GAthering –MEGA- Seminar: Hydroelectric Dam Expansion and Damage: 3D Multiscale Simulation using FFT Solvers and Neural Networks
Event details
| Date | 12.03.2026 |
| Hour | 13:05 › 14:00 |
| Speaker | Lucas Fourel (LSMS, EPFL) |
| Location | Online |
| Category | Conferences - Seminars |
| Event Language | English |
Abstract: Hydroelectric dams are a crucial source of energy, particularly for Switzerland, where they constitute the majority of electricity production. Unfortunately, a significant portion of these structures undergoes expansion and damage due to the alkali-silica reaction, a phenomenon affecting concrete elements exposed to water.
We have developed a 3D multiscale numerical model to simulate fracture in concrete mesostructures and analyze its impact on the dam's macroscopic behavior. To reduce the computational cost of these simulations, we employ efficient approaches based on Fast Fourier Transform (FFT) solvers. Furthermore, we investigate a second approach in which mesoscale simulations are precomputed to train a neural network on simulated concrete fracture data. This network is then integrated into dam simulations to further accelerate computations. In this seminar, we will present an overview of these methods, demonstrating their efficient application to multiscale problems and discussing their computational performance.
Bio: Following my Master's and PhD in mechanical engineering at INSA Lyon, I joined the Computational Solid Mechanics Laboratory (LSMS) at EPFL in Lausanne as a postdoctoral researcher. My work focused on advancing numerical methods for solid materials, leveraging finite elements, FFT-based solvers, and neural networks. I applied these computational techniques to a diverse range of materials, from polycrystalline metals to composites like concrete and porous ceramics, addressing critical challenges in mechanical and civil engineering, including rolling element bearings, gears, and hydroelectric dams.
We have developed a 3D multiscale numerical model to simulate fracture in concrete mesostructures and analyze its impact on the dam's macroscopic behavior. To reduce the computational cost of these simulations, we employ efficient approaches based on Fast Fourier Transform (FFT) solvers. Furthermore, we investigate a second approach in which mesoscale simulations are precomputed to train a neural network on simulated concrete fracture data. This network is then integrated into dam simulations to further accelerate computations. In this seminar, we will present an overview of these methods, demonstrating their efficient application to multiscale problems and discussing their computational performance.
Bio: Following my Master's and PhD in mechanical engineering at INSA Lyon, I joined the Computational Solid Mechanics Laboratory (LSMS) at EPFL in Lausanne as a postdoctoral researcher. My work focused on advancing numerical methods for solid materials, leveraging finite elements, FFT-based solvers, and neural networks. I applied these computational techniques to a diverse range of materials, from polycrystalline metals to composites like concrete and porous ceramics, addressing critical challenges in mechanical and civil engineering, including rolling element bearings, gears, and hydroelectric dams.
Practical information
- General public
- Free
Organizer
- MEGA.Seminar Organizing Committee