Fluid simulation
Simulating a variety of fluid flows accurately and efficiently can find many applications in the prototyping process of industrial product design, special effects production in movies and advertisements, virtual reality, blood flows for medical diagnosis, as well as aerial and underwater robot training in virtual environments. Our lab aims at developing stable and accurate high-performance fluid simulation algorithms and the related multi-node multi-GPU
parallel simulation system to produce different kinds of complex
fluid flow phenomena, with coupling to dynamic solids, in order to support these applications. Unlike the traditional approaches solving Navier-Stokes equations, our whole framework is built upon the advanced
statistical kinetic formulation by solving Boltzmann equation, which is local and conservative without solving any global equations at each time step. This is quite beneficial for handling complex geometric shapes. In addition, the whole algorithms can be easily parallelizable on the GPU and scalable across different GPUs, with a significant performance boost of 2 to 3 orders of magnitude faster than the GPU-based Navier-Stokes solvers under similar accuracy. By coupling with different solid dynamics simulators, efficient multi-physics simulations can be readily achieved.
Until now, the lab has studied and developed different types of kinetic fluid simulation algorithms, including the low-dissipation and low-dispersion single-phase subsonic flow solver for a very large range of Reynolds numbers (up to 50,000,000), multiphase flows with turbulent interactions among air, liquid and solid, where the density ratio between liquid and air can be as large as 800~1000 with a rich set of multiphase flow phenomena including turbulent splashing, surface tension effects, bubbling and wetting on solid surfaces, as well as the one-way and two-way coupling of single-phase fluid flow solver with dynamic solids. The lab also studied the simulation of quantum fluids (both vortex dynamics and turbulence) for their scientific study. Now, the lab is developing algorithms for more versatile coupling with arbitrary geometric shapes, for both single-phase and multiphase fluids with turbulence, the simulation of aeroacoustics under complex environments, as well as the simulation of transonic and supersonic flows with shock waves and turbulence, also in the presence of complex geometric shapes.
Visualization and rendering
Fluid simulations can produce a large amount of data sets, which can
be very useful for both industrial and scientific studies.
Visualizing and rendering these datasets can reveal important
structures, such as vortices and turbulence. Our lab develops
efficient visualization and rendering techniques for fluids. For
visualization, we focus on visualizing vortex and turbulence
structures present in classical and quantum fluids. Both volumetric
and particle-based approaches have been studied, as well as their
hybrid combination. For rendering, we are developing hybrid
techniques, which combine particle-based and ray-based approaches
for creating realistic volumetric 3D fluids. In addition, due to the
huge data size (fluid dynamics data are 4D data), we are also
developing machine-learning-based approach to effectively compress the fluid data, for the purpose of both efficient storage and transmission, with which visualization can also be done more efficiently.
Unmanned aerial vehicles
Modern unmanned aerial vehicles (UAV) are very useful flight platforms. They can be used as transport vehicles, flying cameras or flying robots. However, they are far from perfect now, and we want them to be safer and smarter, fly higher and farther, and stay in the air longer. This requires intelligent design, fabrication, control and navigation. However, realizing this goal is not easy in practice. Our lab is currently working on developing novel geometric design and autonomous navigation algorithms of UAVs based on the
"virtual-to-real" concept. For design, we rely on our high-performance fluid (aerodynamic) simulation system, by developing geometric optimization algorithms. For autonomous navigation, we have built a UAV
flight simulation system, integrating the hardware into the simulation
system, to facilitate developing intelligent obstacle avoidance and navigation algorithms, with machine learning techniques to enable UAV fly freely and safely in a complex environment in the future. Both hardware and software systems are being built in our lab.