(189k) Heterogeneous Supercomputing with Multi-Scale Modeling - Towards Virtual Reality in Process Engineering

Authors: 
Ge, W. - Presenter, Institute of Process Engineering, Chinese Academy of Sciences
Dynamic simulation in process engineering can be very time-consuming but still of low accuracy due to the strong coupling between different scale (e.g., molecular, elemental and systematic) and deferent processes (e.g., reaction, diffusion and flow). A natural and effective approach to address this challenge is multi-scale modeling which combines or cascades models at different scales in a single simulation so that the effects of smaller-scale behaviors on larger scales can be accounted without describing their full details. However, multi-scale modeling may also complicate the computation by mixing different numerical methods and algorithms, e.g., continuum-based and discrete methods. Heterogeneous computing is, therefore, especially suitable for simulation using multi-scale models, with processors of different architectures for different algorithms accordingly.

In this talk, we will demonstrate this possibility by introducing a series of multi-scale simulations on particle-fluid systems such as fluidized bed, pneumatic conveyor, mixing tanks etc. Highly parallel codes for continuum-based finite volume method, discrete element method and lattice Boltzmann method are implemented on the central processing units (CPUs), many-integrated cores (MICs) and graphics processing units (GPUs), respectively, using the petaflops heterogeneous supercomputer Mole-8.5E [1,3]. The accuracy, speed and efficiency of the simulation can be improved dramatically to quasi-realtime level for lab-scale systems with particle-scale resolution. In contrast, traditional simulation may cost weeks for seconds of physical process. Based on this capability, interactive simulation and visualization with online control of the operating conditions has been realized also, suggesting a more exciting prospect of virtual-reality (VR) style simulation in process engineering. Some preliminary results can be found in our recent publications [2,3] and the most recent developments will be presented in the talk.

[1] Wang, X. & Ge, W. 2013. Chapter 5: The Mole-8.5 Supercomputing System. In: Vetter, J. S. eds. Contemporary High Performance Computing from Petascale toward Exascale. pp. 75-98. Boca Raton: Chapman & Hall / CRC.

[2] Ge, W. et al. 2011. Meso-scale oriented simulation towards virtual process engineering (VPE)-The EMMS Paradigm. Chemical Engineering Science 66, 4426-4458.

[3] Ge, W. et al. 2015. Multiscale discrete supercomputing - a game changer for process simulation? Chemical Engineering & Technology 38, 575-584.