PeriFast/Dynamics: A MATLAB Code for Explicit Fast Convolution-based Peridynamic Analysis of Deformation and Fracture

Abstract

We present PeriFast/Dynamics, a compact and user-friendly MATLAB code for fast peridynamic (PD) simulations for deformation and fracture. PeriFast/Dynamics uses the fast convolution-based method (FCBM) for spatial discretization and an explicit time marching scheme to solve large-scale dynamic fracture problems. Different from existing PD solvers, PeriFast/Dynamics does not require neighbor search and storage, due to the use of the Fast-Fourier Transform and its inverse. Run-times and memory allocation are independent of the number of neighbors inside the PD horizon, leading to faster computations and lower storage requirements. The governing equations and discretization method are briefly reviewed, the code structure explained, and individual modules described in detail. A 3D demonstrative example on dynamic brittle fracture is solved using three different constitutive models (a bond-based, an ordinary state-based, and a correspondence model). Differences between the results are explained. Users are provided with a step-by-step description of the problem setup and execution of the code. PeriFast/Dynamics is a branch of the PeriFast suite of codes, and is available for download at the GitHub link provided in reference [1].

Wave interactions and fracture evolution in a thin glass plate under impact: a combined experimental and peridynamic analysis

“…These values are roughly within 5% from the corresponding values in soda-lime glass. While a state-based PD model (see [30]) would be able to exactly match all of the elastic constants in the materials used here, the computational cost compared with the bond-based formulation would be at least twice as large [31], [32]. Given that the variability in material properties of glass, for example, is often higher than 5%, and the fact that we are concerned here with understanding the evolution of the damage and failure phenomenon in a brittle target, in this study we select the bond-based model.…”

“…Larger 𝑚 values are always better for reducing the grid-dependence but they come with an increase in computational cost, and a balance has to be achieved between accuracy and cost. Larger values of 𝑚 do not penalize efficiency in some recent fast algorithms for PD models (see [32], [41]).…”