Fast GPU simulation of reinforced concrete at the scale of reinforcement ribs by the discrete element method

Abstract

The paper presents the development of the GPU-based discrete element method (DEM) code for simulating damage and fracture of cohesive solids with application to reinforced concrete at the scale of reinforcement ribs. The solid volume of concrete and steel is modelled by bonded spherical particles. Very fine discretization, containing more than million particles, is applied to describe the 3D reinforcement bar geometry at the scale of ribs and to investigate cracking behaviour of concrete near the reinforcement bar. The numerical model is validated by using experimental results of the double pull-out test. Influence of the discretization scale to the numerical solution is evaluated by using the reinforcement strain profiles and the cracking patterns. The developed GPU-based DEM algorithm efficiently handles interaction of particles, does not require any atomic operation and allows performing fast damage and fracture simulations with large number of particles. The performance measured on GPU is compared with that attained on different CPUs for varying number of particles. The high value of the Cundall number (particle number multiplied by time steps computed per second) equal to 4.3E+07 is measured on NVIDIA® Tesla™ P100 GPU in the case of 1858560 particles.