CUDA-based Parallel Computation Of Discrete Elements For Road Engineering

The discrete element method is widely used in road engineering as a typical discontinuous computational method,but the current commercial software for discrete element simulation is very inefficient in solving the simulation of larger size pavement structures,which is fundamentally due to the low computational parallelism;in addition,its high learning cost and the lack of corresponding intrinsic models such as asphalt concrete have affected the use of discrete element methods in road engineering.In recent years,GPUs,as the mainstream high-performance computing acceleration devices,are increasingly used in parallel computing in many fields,especially for processing big data analysis.This paper focuses on the development of parallel computing,the application of unified computing architecture(CUDA)technology in discrete element numerical simulation,and the principles and methods of GPU multithreading to provide ideas for using parallel computing technology to improve the efficiency of discrete element large-scale numerical computation in road engineering.In this study,a road material discrete element simulation system GAPSYS with a visual interface is developed.The innovative features of this program are the implementation of GPU parallel accelerated computing,the ability to adapt to the material calculation needs of asphalt mixtures and various pavements,and the accomplishment of a fast modeling approach.This system is divided into three main parts,pre-processing,computational core and post-processing.The pre-processing is to select and set parameters and models through the visualization interface to generate the corresponding particle,wall and particle cluster models;the computational core mainly refers to the data processing and discrete element iterative computation implemented using CUDA algorithm;the post-processing is mainly the result output in the form of files,curves and pictures.In this paper,the following work is mainly accomplished:(1)design and development of the visualization interface.With the help of Visio 2016,the visualization interface was analyzed,designed and programmed using the object-oriented approach.The concept of short code was introduced in the interface development,and the pre-processing function of the program was realized through Im GUI graphics library and Open GL graphics display tool.(2)Kernel development.Based on Visual Studio 2013 Integrated Development Environment(IDE),CUDA encapsulation of data structures and basic algorithms was implemented for the basic modules of discrete elements using C++ language.(3)Parallelization of discrete element algorithms.We analyzed the parallelizability of the discrete element,realized the porting of the discrete element algorithm from serialization to parallelization,and completed the parallel accelerated computation of the discrete element iteration loop.(4)Algorithm testing.After the software development work is completed,the reliability of key programs is tested by arithmetic examples.