RecurDyn/Particles is used to a model which includes the particle systems. To solve many particles, the DEM method is applied. The discrete element method (DEM) is a method of the numerical methods for computing the motion of a large number of particles which are the micrometer-scale size and above.
As the GPU can boost the calculation speed of parallel programming, the DEM (Discrete Element Method) is becoming widely used in granular and discontinuous materials, especially in the granular flows, powder mechanics, and rock mechanics. Some examples are the granular matters (rock, sand, and soil), powders, liquid, and solutions.
And, the typical industries using the DEM are the chemical industry, the pharmaceutical industry, the mining industry, the agriculture industry and the food handling industry, the powder metallurgy industry and the digital printing industry. With advances in computing the power and numerical algorithms, it has become possible to simulate millions of particles numerically.
The main disadvantage of particle-based simulation with a very large number of particles is to require very heavy computing resources for obtaining the satisfactory results. But, it is relatively not so difficult to parallelize the particle systems. Perhaps the GPU computing is the best way to efficiently handle a large number of particle system. Therefore, RecurDyn adopted the DEM solver, the SAMADIITM (Metariver Technology co., ltd.) which was developed to simulate the particle-based system using the GPU.
The GPU is massively parallel multithreaded device for executing a large amount of active threads. The GPU has multiple streaming multiprocessors, each of which contains multiple scalar processor cores.
A function that executes on the GPU consists of multiple threads executing codes in a single instruction, multiple data (SIMD) fashion. It is that each thread in a kernel executes the same code, but on different data. The libraries of CUDA (Compute Unified Device Architecture) from NVIDIA allow one to use the streaming microprocessors mounted in the high-end graphic cards as the general-purpose computing hardware. Now, the raw computational power of these multiprocessors can reach one teraflop, which is several hundred times the throughput of a modern scalar CPU.
SAMADIITM was developed with the hardware acceleration by the GPU as well as the software acceleration using the cell-linked list algorithm to accelerate the DEM simulation. The simulation space is partitioned into cells and the particles are assigned to the cells so that it is easier to find the neighbors of a given particle.
At the beginning of a simulation, an array that contains a list of cell neighbors for each cell is created. This method dramatically reduces the number of unnecessary inter-particle distance calculations.
The simulation of many particles in the multi body systems needs a strategy of integration. Many particles are solved with the explicit integration and the multi body systems are solved with the implicit integration.
Figure 1 shows a sample idea to the integration of MBD and particles. The whole modeling is consisted of a spring, a box and many spheres. The spring and the box are solved with RecurDyn while many particles are solved with SAMADIITM.