A Simple Machine in a Complex Material: Scaling CalculiX for Screw Fixation in Trabecular Bone¶
Benjamin Schnabel (HLRS)
Abstract¶
Femoral neck fractures are treated with screw osteosynthesis, and healing depends on how stable the implant is. Predicting this stability requires finite element models that resolve the trabecular microstructure and capture its viscous, plastic and damage behaviour. We implemented a viscoelastic-plastic damage model as a CalculiX UMAT. We identified its parameters from published experiments on bovine trabecular bone in two phases: a parallel differential evolution search followed by a local least-squares refinement. Models built from micro-CT images quickly reach several million degrees of freedom, which is more than shared-memory direct solvers handle comfortably. We therefore added the MPI-parallel oneMKL PARDISO solver to CalculiX through an interface. On HLRS Vulcan, the solver runs 1.7× faster on 16 nodes for a bone model with 10 million degrees of freedom. Nonlinear analyses finish up to 2.6× faster when the symbolic analysis is reused. The main limits are serial preprocessing, the centralized analysis phase and memory on the master rank. We discuss these limits and the next steps towards screw-in simulations.