Development of Large-Eddy Simulation frameworks on massively parallel environments represented by the supercomputer "K" and "Fugaku", and their applications to industrial problems are mentioned. One of the frameworks is based on the unstructured finite volume method to reproduce the complicated geometries typically required for industrial applications. The validity of the method on the applied aerodynamics such as golf balls and road vehicles on the supercomputer "K" and its limit to High Performance Computing are discussed from two points: tuning on the new hardware architectures on many-core processors with lower memory bandwidth, and total turn-around time including the mesh generation. To overcome the problem, new complex unified simulation framework based on the hierarchically structured finite volume method and its applications on the supercomputer "Fugaku" is introduced. In addition to the world-largest vehicle aerodynamics simulation utilizing tens of billions of numerical cells, these applications include direct feedback noise prediction emitted from a gap on the vehicle's surface, and coupling aerodynamics and vehicle's six-degrees-of-freedom motion simulation, as examples of "capability computing" on engineering problems. It should be noted that these simulations are realized on very complicated geometries equivalent to actual vehicles. Finally, airborne infection risk assessment based on droplet/aerosol dispersion simulation in indoor environment and proposal of the countermeasures is introduced as a response of the framework on "Fugaku" to the COVID-19 pandemic. (ACM Gordon Bell Special Prize for HPC-Based COVID-19 Research Awarded) In this simulation more than 50 infection scenes over 1,000 test cases were able to be evaluated in a short period of one year, by drastically decreasing the total turn-around time of the simulation. The scenes include public transportations like a bus, a taxi, an airplane, and a commuter train, and pubic locations like an office, a school, a hospital, a concert hall, a live music club and so on. In fact, "capacity computing" like this is also becoming important applications on the massively parallel environment from the viewpoint of "big data" analysis. Thus, perspective of coupling of data science and HPC turbulence simulation for computer aided engineering is discussed to end the talk.
|