JAXA Supercomputer System Annual Report April 2017-March 2018 · 1) Sumi, Kurotaki,' Ubsteady...

Report Number：R17ECMP01 Subject Category：Competitive Funding

URL：https://www.jss.jaxa.jp/ar/e2017/4426/

Responsible Representative

Takashi Aoyama, Aeronautical Technology Directorate, Numerical Simulation Research Unit

Contact Information

Atsushi Hashimoto [email protected]

Members

Atsushi Hashimoto, Tsukasa Nagao

Abstract

We evaluate the hyperbolic equations for Navier-Stokes equations.

https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-15K18286/

Reasons for using of JSS2

We plan a large scale computation in the future.

Achievements of the Year

We studied a laminar flow over a flat plate. (Fig. 1) shows the drag coefficient convergence. We confirmed that the drag computed with the hyperbolic system converges more rapidly than that with the traditional system. (Fig. 2) shows the residual of the streamwise velocity, u, versus the number of iterations and CPU time. A comparison of the CPU time shows that the time for hyperbolic equations is approximately 75% less than that for traditional equations.

Fig.1 Drag coefficient convergence

Convergence Acceleration with Hyperbolic Navier-Stokes Equations

JAXA Supercomputer System Annual Report April 2017-March 2018

Fig.2 Residual versus number of iterations and CPU time

Publications

Non peer-reviewed papers 1) Nagao, et. al., "A Study on Time Evolution Method for Hyperbolic Navier-Stokes System," AIAA

SciTech 2018, AIAA 2018-0370.


Usage of JSS2

Computational Information

Resources Used

Fraction of Usage in Total Resources*1 (%): 0.00

Details Computing Resources

System Name Amount of Core Time (core x hours) Fraction of Usage*2 (%) SORA-MA 0.00 0.00 SORA-PP 0.00 0.00 SORA-LM 0.00 0.00 SORA-TPP 0.00 0.00

File System Resources

File System Name Storage assigned(GiB) Fraction of Usage*2 (%) /home 000.52 0.00 /data 488.28 0.01 /ltmp 097.66 0.01

Archiver Resources

Archiver System Name Storage used(TiB) Fraction of Usage*2 (%) J-SPACE 0.00 0.00

*1 Fraction of Usage in Total Resources: Weighted average of three resource types (computing, file system,

and archiver) *2 Fraction of Usage: Percentage of usage relative to each resource used in one year

Parallelization Methods N/A Thread Parallelization Methods N/A Number of Processes 1 Elapsed Time per Case 1.00 hours





Takashi Aoyama, Aeronautical Technology Directorate, Numerical Simulation Research Unit

Contact Information

Shingo Matsuyama [email protected]

Members

Shingo Matsuyama

Abstract

In this research, large scale direct numerical simulation of turbulent flow is performed on supercomputer to clarify the role of very small-scale turbulence in the mixing process of fuel and air. We try to clarify the role of fine-scale turbulence by analyzing the data obtained by simulations with varying the Reynolds number which is a parameter that governs the turbulence intensity.

https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-15K05817/


In order to clarify the role of fine-scale turbulence in scalar mixing process, statistical data by DNS is required for high Reynolds number condition. For performing DNS under high Reynolds number condition of Re > 104, a numerical mesh of the order of one billion points is required. Such large-scale simulation can be executed only on a supercomputer, and therefore, supercomputer system is indispensable for carrying out this research.


DNSs were performed for turbulent planar jets at Re = 3×103, 104 and 3×104 on 0.1billion to 1.3 billion grid points with 9-th order spatial accuracy. It was confirmed by comparison with past experimental data and evaluation of Kolmogorov length scale that Re-dependence of planar jet is correctly reproduced by the present DNS (Fig.1). By filtering DNS data, it is found that the contribution to turbulent scalar mixing due to turbulence with scales less than 60 and 100 times of kolmogrov length scale are negligible at Re = 3 ×103 and 104, respectively.

Investigation on Fine-scale Scalar Mixing in High Reynolds Number Turbulent Jets


Fig.1 Instantaneous contours of scalar dissipation rate on the x-y plane (z = 0) for

the DNS at Re = 3×103,104 and 3×104 (from top to bottom). Cited from non peer-reviewed paper [2].


Fig.2 Movie of scalar dissipation rate for the DNS at Re = 3×103.

Fig.3 Movie of scalar dissipation rate for the DNS at Re = 104.

Fig.4 Movie of scalar dissipation rate for the DNS at Re = 3×104.


Publications

Non peer-reviewed papers 1) Shingo Matsuyama, "Direct Numerical Simulation of a Turbulent Plane Jet with Scalar Mixing",

proceedings of the 48th JSASS annual meeting, 2C15, 2017. 2) Shingo Matsuyama, "DNS of a Turbulent Plane Jet with Scalar Mixing", proceedings of the JSFM

Annual Meeting 2017, 2017. Presentations 1) Shingo Matsuyama, "Direct Numerical Simulation of a Turbulent Plane Jet with Scalar Mixing", the

48th JSASS annual meeting, 2017. 2) Shingo Matsuyama, "DNS of a Turbulent Plane Jet with Scalar Mixing", the JSFM Annual Meeting

2017, 2017.


Usage of JSS2


Resources Used



System Name Amount of Core Time (core x hours) Fraction of Usage*2 (%) SORA-MA 22,602,550.63 3.00 SORA-PP 0.00 0.00 SORA-LM 0.00 0.00 SORA-TPP 0.00 0.00


File System Name Storage assigned(GiB) Fraction of Usage*2 (%) /home 598.94 0.41 /data 2,849.03 0.05 /ltmp 488.28 0.04

Archiver Resources




Parallelization Methods MPI Thread Parallelization Methods OpenMP Number of Processes 286 - 924 Elapsed Time per Case 1,000.00 hours





Takuji Kurotaki, Aeronautical Technology Directorate, Numerical Simulation Research Unit

Contact Information

Takuji Kurotraki [email protected]

Members

Takuji Kurotaki

Abstract

Atomization of fuel of aeronautical engines is focused in because of fuel consumption and environmental issues. Information from CFD results is very helpful for solutions of them. The research of numerical simulation of compressible muti fluid flow is conducted to apply for these problems.


Super computer system can be used from simple problems for the validation of basic method to relatively large application for unsteady and multi dimensional problems. It is convenient for the speed up of development of numerical cord.

This kinds of simulation deal with unsteady problems and in order for the reduction of computational time, the ability of handling of parallel computing is indispensable.


Two main approach to the simulation of compressible multiphase flow are investigated; diffuse interface approach which treats the interface as mixture of both fluids and sharp interface approach with levelset method.

In the development of diffuse interface approach, basic equations are extended to six equation model and by considering the chemical equilibrium it became to possible to treat phase change such as cavitation. Atomization of fuel from 2D nozzle is simurated by this chords.

In the development of sharp interface approach, new HLLC scheme which is extended to treat phase change models at the interface and it was confirmed that the problem of the cavitation collapse are simulated accurately and the methods to inprove the mass prezervation was investigated.

Adbanced Numerical Simulation of Compressible Two-phase Flow of diesel atomization


Fig.1 Volume fraction of fuel and numerical Schlieren

Fig.2 Volume fraction of vapor of fuel and air

Publications

Presentations 1) Sumi, Kurotaki,' Ubsteady Numerical Simulation of Compressible Gas-Liquid Two-Phase Flow for

Ultra-High-Pressure Fuel Injection,'The 28th Internal Conbustion Symposium (2017), 13.


Usage of JSS2


Resources Used



System Name Amount of Core Time (core x hours) Fraction of Usage*2 (%) SORA-MA 0.00 0.00 SORA-PP 4,466.74 0.06 SORA-LM 0.00 0.00 SORA-TPP 0.00 0.00


File System Name Storage assigned(GiB) Fraction of Usage*2 (%) /home 005.09 0.00 /data 078.01 0.00 /ltmp 1,041.67 0.08

Archiver Resources




Parallelization Methods N/A Thread Parallelization Methods OpenMP Number of Processes 1 Elapsed Time per Case 10.00 hours





Yuko Inatomi, Institute of Space and Astronautical Science, Department of Interdisciplinary Space Science

Contact Information

Ryoji Takaki [email protected]

Members

Ryoji Takaki, Taku Nonomura, Seiji Tsutsumi, Yuma Fukushima, Soshi Kawai, Yuki Kawaguchi, Ikuo Miyoshi, Satoshi Sekimoto, Shibata Hisaichi, Hiroshi Koizumi, Yuichi Kuya, Tomohide Inari, Ryota Hirashima

Abstract

We develop a high-speed/high-precision computational program using a quasi-first principle method, which can faithfully reproduce the actual flight environment to understand the true nature of fluid phenomena. Specifically, we develop a high-precision compressible flow solver with geometric wall models and LES (Large Eddy Simulation) wall models based on hierarchical, orthogonal and equally spaced structured grids.


We need large computer like JSS2 because our calculations must be large scale computations. Moreover, JSS2 has a similar architecture to the our target computer called Post-K.


We advance the development of FFV-HC-ACE: a compressible Navier-Stokes equation solver using hierarchical, orthogonal and equally spaced structured grid. Following new functions are added to the program; calculation function of convective and viscous fluxes using image point (IP) and data communication function between oblique blocks. A flow around a NACA0012 2D airfoil is calculated as a preliminary validation of the program. Fig. 1 shows the Mach number distributions around the airfoil and block boundaries. Mach number is 0.3, Reynolds number is 10,000 and attack angle is 3 degree.

Post-K Priority Issue 8D: Research and development of core technology to innovate

aircraft design and operation


Fig.1 Mach number distributions around NACA0012 (Mach number is 0.3, Reynolds

number is 10,000 and attack angle is 3 degree)

Publications

Presentations 1) H. Shibata, M. Sato, Y. Fukushima, S. Tsutsumi, T. Nonomura, S. Kawai and R. Takaki,

Development of a compressible flow solver using hierarchical Cartesian grids - toward the age of Exa-scale supercomputers-, 29th international conference on parallel computational fluid dynamics.

2) H. Shibata, M. Sato, S. Tsutsumi, Y. Fukushima, T. Nonomura, S. Kawai and R. Takaki, Investigation of flux based wall boundary condition for hierarchical cartesian grid method, 49th Fluid Dynamics Conference/35th Aerospace Numerical Simulation Symposium.

3) H. Shibata, Y. Fukushima, S. Tsutsumi, Y. Kuya, S. Kawai and R. Takaki, Assessment of wall boundary conditions for a cartesian grid method, 31th Computational Fluid Dynamics Symposium.

4) R. Takaki, Characterization of PRIMEHPC FX100 by CFD solver using structured grid method, 1st Workshop of HPC MONOZUKURI on Post-K Priority Issue 6 6 and 8.

5) R. Takaki, A challenge to analysis of real complicated configurations of aircraft -Sub-issue D-, 1st Workshop of HPC MONOZUKURI on Post-K Priority Issue 6 6 and 8.

6) R.Takaki, Characterization of many core CPU by CFD programs, 49th Fluid Dynamics Conference/35th Aerospace Numerical Simulation Symposium.

7) R. Takaki, H. Shibata, S. Kawai, Y. Fukushima, S. Tsutsumi and Y. Kuya, A challenge to analysis of real flight Reynolds number flows by Post-K computer, 55th Aircraft Symposium.


8) R. Takaki, High performance computing by many-core based CPU, 3rd Symposium on Space

Science Informatics. 9) R. Takaki, Toward the achievement of the aerodynamic characteristic evaluation for real

configurations and real flight environments of aircraft, 3rd Symposium on Post-K computer Priority Issue 8.

Usage of JSS2


Resources Used



System Name Amount of Core Time (core x hours) Fraction of Usage*2 (%) SORA-MA 6,186,602.05 0.81 SORA-PP 759.84 0.01 SORA-LM 0.00 0.00 SORA-TPP 0.00 0.00


File System Name Storage assigned(GiB) Fraction of Usage*2 (%) /home 3,310.17 2.29 /data 21,116.11 0.39 /ltmp 10,782.88 0.81

Archiver Resources




Parallelization Methods MPI Thread Parallelization Methods OpenMP Number of Processes 2 - 20 Elapsed Time per Case 20.00 hours





Takahiro Yamada, Institute of Space and Astronautical Science, Division of Spacecraft Engineering

Contact Information

Tomoya Fukami [email protected]

Members

Tomoya Fukami, Hirobumi Saito, Kengo Ando, Pyne Budhaditya, Prilando Rizki Akbar

Abstract

We are goint to develop synthetic aperture radar system that can be onboard on a 100kg class satellite and to verify it on ground. Conventionally radar observations have required large or medium satellite with 500-1000kg mass. Our research and development may change earth observation drastically.

http://www.jst.go.jp/impact/en/program/13.html


We utilize super-computer for design of slot array antennas dedicated for small synthetic aperture radar. Design of slot array antenna requires a super-computer that is provided with high speed and large memory.


We performed simulation for 4 panels antenna and antenna network feeder system for antenna panels array. Here the, we simplified the simulation of network feeder system for 7 antenna panels array with 3.5 antenna panels array simulation. The simulation results are shown in Fig.1. In this simulation, in-phase and uniform excitation to the parallel-plate antenna are expected. Furthermore, the simulation results which are showing performance of each antenna panels is shown in Fig.2.

Small synthetic aperture radar satellite


Fig.1 Tournament feeder network design for 7 antenna panels array.

Fig.2 Antenna Panels simulation results.


Publications

Peer-reviewed papers 1) B. Pyne, P.R. Akbar, V. Ravindra, H. Saito, J. Hirokawa, T. Fukami, "Slot-Array Antenna Feeder

Network for Space-borne X-band Synthetic Aperture Radar", submitted to the Proceedings of IEEE Transactions on Antennas and Propagation, under review as of March 12, 2018.

Presentations 1) B. Pyne, V. Ravindra, P.R. Akbar, H. Saito, J. Hirokawa, "Performance Analysis of Edge-feed

Rectangular Parallel Plate Slot-array Antenna Panel for Compact Space-borne X-band SAR System", Proceedings of the International Conference on Electromagnetics in Advanced Applications (ICEAA) - IEEE Topical Conference on Antennas and Propagation in Wireless Communications (APWC), Verona, September 2017, pp. 407-410.

URLs for the Research Results on the Web 1) http://www.isas.jaxa.jp/topics/001134.html


Usage of JSS2


Resources Used



System Name Amount of Core Time (core x hours) Fraction of Usage*2 (%) SORA-MA 153.80 0.00 SORA-PP 443,619.62 5.55 SORA-LM 92,167.94 47.46 SORA-TPP 175.00 0.02


File System Name Storage assigned(GiB) Fraction of Usage*2 (%) /home 991.82 0.69 /data 22,842.42 0.42 /ltmp 11,718.76 0.88

Archiver Resources




Parallelization Methods MPI Thread Parallelization Methods Intel MKL Number of Processes 2 - 80 Elapsed Time per Case 12.00 hours





Yasutada Tanabe, Aeronautical Technology Directorate, Next Generation Aeronautical Innovation Hub Center

Contact Information

Yasutada Tanabe [email protected]

Members

Yasutada Tanabe, Hideaki Sugawara, Masahiko Sugiura, Ryousuke Satou

Abstract

Aerodynamic interaction between a rotor and a fixed wing on a compound helicopter is numerically simulated to clarify the increase of the arodynamic drag caused by the interaction during high speed flight.


There is a large number of test cases and an aboundance of computing sources are required.


As the first step, the performance of the isolated rotor and the aerodynamic characteristics of the wing models have been clarified.

Publications

N/A

Rotor/Wing Interaction Simulations


Usage of JSS2


Resources Used



System Name Amount of Core Time (core x hours) Fraction of Usage*2 (%) SORA-MA 150,265.58 0.02 SORA-PP 41,331.21 0.52 SORA-LM 0.00 0.00 SORA-TPP 0.00 0.00



Archiver Resources










Contact Information

Masahiko Sugiura [email protected]

Members

Yasutada Tanabe, Masahiko Sugiura, Hideaki Sugawara

Abstract

Research on advanced technologies of aircraft (helicopter) is conducted to construct and validate an optimization design tool.

http://www.aero.jaxa.jp/eng/research/frontier/rotary/


It is absolutely necessary to utilize a super computer, regarding understanding aerodynamic performances which require large scale computations.


Aerodynamically optimal design of helicopter rotor blade.

Research of Advanced Rotorcraft


Fig.1 Computational grids of a hovering helicopter rotor

Publications


Peer-reviewed papers 1) Atthaphon Ariyarit, Masahiko Sugiura, Yasutada Tanabe, and Masahiro Kanazaki, "Hybrid

surrogate-model-based multi-fidelity efficient global optimization applied to helicopter blade design", Engineering Optimization, September, 2017.

Usage of JSS2


Resources Used



System Name Amount of Core Time (core x hours) Fraction of Usage*2 (%) SORA-MA 176,463.66 0.02 SORA-PP 40,651.28 0.51 SORA-LM 3,184.92 1.64 SORA-TPP 4,780.06 0.53



Archiver Resources










Contact Information

Yasutada Tanabe [email protected]

Members

Yasutada Tanabe, Hideaki Sugawara

Abstract

Represented by the multiple rotor type UAV (Unmanned Aerial Vehicle), or so-called drone, the flying robot has been applied to disaster relief and other operations. However, the wind and gust resistence of this type of drone need to be further improved. Especially in the observation and survey of existing infrastructures, buidings and bridges, the drones are often required to fly near walls. The flowfields around the drones are significantly changed with walls nearby and highly depend on the distance between the drone and the walls. It is considered very effective to incorporate the wall influence on the flight performance into the flying control rules.

This research is carried out as a part of the Tough Robotics Challenge funded by the ImPACT Program of Council for Science, Technology and Innovation (Cabinet Office, Government of Japan). The influence of an upper wall and a side wall on the aerodynamics of the multiple rotors is investigated through CFD simulations. Also the flight performance of a drone inside narrow tunnels is under studies.

https://www.jst.go.jp/impact/en/program/07.html


Numerical simulations of multiple rotors are carried out using a CFD code, rFlow3D, developed at JAXA. The rotor blades are modeled with high fidelity. A large amount of flight conditions must be considered and each takes a long time of CPU hours. JSS2 supercomputer is a prerequisite of this kind of research.

CFD Analysis of a Multiple Rotor UAV near Walls



A hexa-rotor drone hovering near an upper wall (ceiling) and a side wall is numerically simulated. When the distance between the drone and the wall because shorter than 1.5 times of the rotor diameter, the aerodynamic forces generated by the rotors changes remarkably. The flight performance of a drone flying inside narrow tunnels is also investigated.

Fig.1 Changes of the flowfields around a hexa-rotor drone due to distance changes between the drone and a side wall

Publications

Non peer-reviewed papers 1) Koichi Yonezawa, Hironori Matsumoto, Kazuyasu Sugiyama, Hiroshi Tokutake, Yasutada Tanabe,

Shigeru Sunada, "Development of High-Performance Multi-Rotor Drone", Fourteenth International Conference on Flow Dynamics, November 1-3, 2017, Sendai International Center.

2) Yasutada Tanabe, Masahiko Sugiura, Takashi Aoyama, Hideaki Sugawara, Shigeru Sunada, Koichi Yonezawa, Hiroshi Tokutake, "Influences of Upper and Side Walls on the Performance of a Multiple Rotor Drone", 6th Asian-Australian Rotorcraft Forum & Heli Japan 2017, November 7-9, 2017, Kanazawa, Japan.

3) Koichi Yonezawa, Hironori Matsumoto, Kazuyasu Sugiyama, Hiroshi Tokutake, Yasutada Tanabe, Shigeru Sunada, "Development of a Ducted Rotor for Multicopters", 6th Asian-Australian Rotorcraft Forum & Heli Japan 2017, November 7-9, 2017, Kanazawa, Japan.

4) Raijo Torita, Takuji Kishi, Hiroshi Tokutake, Shigeru Sunada, Yasutada Tanabe, Koichi Yonezawa, "Modeling of Aerodynamic Characteristics of Drone and Improvement of Gust Response", 6th Asian-Australian Rotorcraft Forum & Heli Japan 2017, November 7-9, 2017, Kanazawa, Japan.


Usage of JSS2


Resources Used



System Name Amount of Core Time (core x hours) Fraction of Usage*2 (%) SORA-MA 570,860.89 0.08 SORA-PP 232,987.32 2.92 SORA-LM 0.00 0.00 SORA-TPP 0.00 0.00



Archiver Resources









Takashi Takahashi, Aeronautical Technology Directorate, Numerical Simulation Research Unit

Contact Information

Yuya Ohmichi [email protected]

Members

Yuya Ohmichi, Kenji Kobayashi

Abstract

Knowledge extraction techniques for large data sets are important because, recently, the computers and numerical simulation techniques have been highly developed and they produced massive datasets . In this study, we are developing knowledge extraction tools which extract patterns from a large data obtained by unsteady fluid simulations.


Large amount of memory is necessary for performing the developed knowledge extraction tools.


A numerical tool to conducte dynamic mode decomposition and compressed sensing techniques was developed. The tool was applied to a flow field around a re-entry capsule and we found a novel fluid phenomenon that had very long time scale.

Publications

Peer-reviewed papers 1) Y. Ohmichi, Preconditioned dynamic mode decomposition and mode selection algorithms for large

datasets using incremental proper orthogonal decomposition, AIP Advances, 7(7), 075318, 2017.

Modal decomposition analysis of a flow field around a re-entry capsule


Usage of JSS2


Resources Used



System Name Amount of Core Time (core x hours) Fraction of Usage*2 (%) SORA-MA 24,711.62 0.00 SORA-PP 129.30 0.00 SORA-LM 1,051.62 0.54 SORA-TPP 0.00 0.00



Archiver Resources









Yutaka Iwahori, Aeronautical Directorate, Structure and Advanced Composite Technology Reasearch Unit

Contact Information

Akinori Yoshimura [email protected]

Members

Akinori Yoshimura

Abstract

JAXA contributes to the SIP (Cross-ministerial Strategic Innovation Promotion) program, which is governed by Cabinet Office, Government of Japan. JAXA takes charge high fidelity modeling of the polymer matrix composites for aircraft structure. The calculation for the modeling is performed in this topic.

http://www.jst.go.jp/sip/dl/k03/jst_pamphlet_english.pdf


Damage simulation of the open hole tensile (OHT) test against CFRP was performed using explicit finite element damage simulation. Huge number of DOFs was necessary for detailed damage simulation in CFRP. Application of the supercomputer system makes such high-computational-cost calculation possible.


Damage simulation of CFRP under open-hole tensile (OHT) test was performed using ECDM (Enhanced Continuum Damage Mechanics) model, which was damage model developed in JAXA.

Simulation was performed using ABAQUS/Explicit 2016 on JSS2 system. Each ply of the CFRP laminate was modeled by one continuum shell element in the thickness direction. ECDM model was implemented using user-subroutine of ABAQUS. In FY29, ECDM model was modified and final failure criteria was modified. These modification improved the precision of the simulation results.

Figure 1 shows model overview, and Figure 2 shows the predicted damage.

SIP CFRP high fidelity modeling technology


Fig.1 Model overview of the Open-hole-tension simulation

Fig.2 Predicted damage

Publications

N/A


Usage of JSS2


Resources Used



System Name Amount of Core Time (core x hours) Fraction of Usage*2 (%) SORA-MA 0.00 0.00 SORA-PP 3,968.24 0.05 SORA-LM 0.00 0.00 SORA-TPP 0.00 0.00


File System Name Storage assigned(GiB) Fraction of Usage*2 (%) /home 003.18 0.00 /data 031.79 0.00 /ltmp 651.04 0.05

Archiver Resources




Parallelization Methods MPI Thread Parallelization Methods N/A Number of Processes 12 - 120 Elapsed Time per Case 60.00 hours


JAXA Supercomputer System Annual Report April 2017-March 2018 · 1) Sumi, Kurotaki,' Ubsteady...

Documents

Transcript of JAXA Supercomputer System Annual Report April 2017-March 2018 · 1) Sumi, Kurotaki,' Ubsteady...