07-12, 12:00–12:30 (Europe/Amsterdam), Else (1.3)
This talk will present the latest developments in Peridynamics.jl, a Julia package designed for dynamic fracture simulations. We rewrote the package from scratch to enhance multithreading and enable MPI support, redesigned the API to catch errors earlier, and implemented new material models to match the state-of-the-art peridynamics research. The talk will focus on how and why we changed the package and provide simple tutorials on the new API using fascinating examples of dynamic fracture.
Peridynamics is a nonlocal continuum formulation that uses an integro-differential equation to describe the relative displacement and forces between material points. This equation is also fulfilled for material discontinuities, allowing for modeling of crack propagation, fragmentation, and many other effects. Peridynamics has been shown to correctly map complex fracture problems with different materials, making it suitable for various engineering applications.
Our ongoing development aims to address the package's critical limitations and needed functionality enhancements. The newest version of the package acknowledges these challenges, particularly the limited support for high-performance computing. To improve this, we have rewritten the package from scratch, incorporating a dual strategy: multithreading via Threads.jl for an interactive user experience and MPI.jl integration for scaling simulations on HPC clusters. This effort involved rethinking the underlying data structure and implementing a customized halo-exchange approach to communicate between threads and MPI ranks. Additionally, a new API focuses on early error detection, addressing past issues where wrong user input could only be caught at the start of a simulation. Inspired by user feedback, the new API offers a more streamlined experience for single and multi-body simulations.
The bond-based model is the first peridynamics formulation with significant limitations and a high sensitivity to discretization errors. New material models that overcome these limitations exist. In the new version of the package, we introduce some of them, like the correspondence formulation, allowing for advanced simulations with higher accuracy and fewer discretization errors. The talk will feature simple tutorials on example simulations, highlighting the functionality of the new API and the expanded capabilities due to new material models and MPI integration.
PhD Candidate, University of Siegen, Germany
Specializing in solid mechanics, with a focus on dynamic fracture, peridynamics, and continuum mechanics.