Analysing the use of Real-time Physics Engines for Scientific Simulation: Exploring the Theoretical and Practical Benefits for Discrete Element Modelling

Within computer science, reusability of specific modular software components is generally accepted as best practice. Simulation techniques such as Discrete Element Modelling (DEM) rely on the well defined problems of Newtonian physics, and while differences exist in the methods defined to compute solutions to these problems, each method follows the same basic set of premises. Recently, libraries termed Physics Engines (PE) have been released that are designed to solve physics based problems. This paper considers the features of a range PEs and explores whether the techniques and design methodologies can be applicable to the design and implementation of a working simulation. The NVIDIA PhysX engine has been utilised in a practical DEM implementation to simulate the evolution of extensional fault systems in rock. Through understanding the general processing pipeline implemented by a PE, obvious similarities with a range of DEM implementations has became apparent. Discussed are areas that are compatible and also areas within the PE that have been proved unsuited to large scale DEM simulation. It is shown that current versions of PEs may not provide access to techniques giving high enough numerical accuracy for certain applications, but the basic premise of an easy to use and highly optimised library, designed to allow researchers to construct complex simulation scenarios is compelling.