← Computational Mechanics

Discretization technologies

Solving partial differential equations more efficiently.

Key works

GPU-accelerated simulation

Graphics cards accelerate highly parallel, repetitive computations, which numerical simulations can exploit for much shorter runtimes. cuwave is a differentiable finite difference code for wave propagation on a single GPU, and mlhp can run parts of its finite element pipeline (matrix-free assembly and iterative solvers) on the GPU.

wave propagation

Wave propagation in the Discobolus: 28 million degrees of freedom over 1500 time steps, simulated in 0.21 s on an NVIDIA RTX 5090. powered by cuwave

topology optimization

Compliance topology optimization: 373 million voxels (quarter MBB beam) at 28.6 s per iteration on an NVIDIA RTX 5090, so 300 iterations took 2 h 23 min. powered by mlhp

Phase-field fracture

Phase-field models describe discrete cracks as a narrow band of damage. mlhp supports the most common phase-field fracture models with adaptive refinement, as well as GPU acceleration to limit computational effort.

refinement

Adaptive refinement: the mesh follows the crack tip and coarsens behind it, so the computation stays small. powered by mlhp

as-built

From CT scan to crack path: a 3D-printed concrete prism simulated on its scanned, as-built geometry with 914 million degrees of freedom, 136 load steps in 21 min on an NVIDIA RTX 5090. powered by mlhp

Publications