Thursday, December 10th , 2020
3:45 – 4:45 p.m.
WEBEX
 
  
Speaker: Prof. Huidan (Whitney) Yu
Department of Mechanical and Energy Engineering, Indiana University-Purdue University, Indianapolis (IUPUI), USA 
 
“Unified Modeling and GPU-accelerated Lattice Boltzmann Method for Imaged-based Computational Fluid Dynamics”
 
 
 
Abstract: 
 
Solving complex flow systems based on imaging data has attracted increasing attention in the community of computational fluid dynamics (CFD) due to a variety of application needs. However, the existing numerical approaches involving image process, CFD, and the connection of both are laborious, complex, and error-prone. An ad-hoc coalition of software packages may have to be used. The multidisciplinary tasks of image processing, computational fluid modeling, and high performance computing together with their integration are overwhelming. We present a unique and powerful computational platform for image-based CFD with physical and computational advantages. It is featured with unified mesoscopic modeling for both image segmentation and CFD via GPU (Graphic Processing Units) parallel computing. The multidisciplinary tasks are innovatively integrated into one computational setup thus no software coalition is needed. We use lattice Boltzmann method (LBM) to solve the level-set equation and NS equations successively. The image segmentation and CFD are seamlessly connected avoiding extra grid and mesh generation or data transfer. The unified LBM modeling enables high efficient GPU parallelism for scalable acceleration. Application studies of imaged-based pores-scale porous media flow and image-based computational hemodynamics demonstrate the feasibility and applicability of this novel for complex flow systems. In this talk, the general methods of unified LBM modeling and GPU parallelization will be described first. The modeling reliability and computation efficiency, together with numerical results will be presented. The further development of the computation platform for more complicated flow system involving multiphase, fluid-structure interaction, non-Newtonian effect, and turbulence will be discussed. Parallelism via multiple GPU cards becomes necessary implemented to meet the computation demands for such systems.
 
About the Speaker 
Dr. Huidan (Whitney) Yu is an Associate Professor in Mechanical and Energy Engineering Department of Indiana University-Purdue University Indianapolis (IUPUI). Prior to the affiliation with IUPUI, she successively completed two PhD degrees in Physics at Peking University in China and Aerospace Engineering at Texas A&M University in the US, followed by two postdoctoral research positions at Los Alamos National laboratory and the Johns Hopkins University. Prof. Yu’s research field is, in general, on CFD for modeling and simulating thermal fluids. Her specific expertise is to use kinetic-based lattice Boltzmann method (LBM) to solve complex flow systems with multiscale and multi-physics. It has been well know the lattice Boltzmann method has the advantages of suitability for multiphase flow, ease for dealing with complicated geometry, and idealization for GPU (Graphic Processing Units) parallel computing. She has 20+ years of experience creating models and performing simulations for various fluid/flow systems including multiphase flow and turbulence. Her research over the years has produced 80+ prestigious journal papers and similar number of invited/conference talks. After joining IUPUI, Prof. Yu has dedicated herself to develop reliable and applicable computational tools for image-based computation analysis of biomedical flows through close collaboration with clinicians. She is the inventor of a clinically practical software named InVascular for noninvasive assessment of hemodynamic abnormalities within clinical favorite time, aiming to aid clinical diagnosis and treatment making.