Rising Star...

Rising Star...

In this edition, we profile Dr. Jin Seok Park, Postdoctoral Research Associate, Imperial College, London, focusing in the area of Flux Reconstruction and PyFR. 

Rising Star Main 2What is your background & area of study?

I have a PhD in Aerospace Engineering, where I studied at Seoul National University, graduating in 2014. Throughout my undergraduate study, I had an interest in fluid dynamics, mathematics and computational science. I decided to study computational fluid dynamics in my previous study. In the middle of my graduate study, I started to study high-order CFD methods, and after graduation I learnt that Dr Peter Vincent, Imperial College was seeking a Postdoctoral Researcher for High-Order CFD Methods and developing the high-order CFD code PyFR. So I  applied for the position, was successful and started in September 2014. During my role as a postdoctoral researcher, I have been involved in the Hyperflux project, developing accurate and robust numerical algorithms for higher-order CFD methods, collaborating with a number of industry organisations.

What are your career aspirations?

Primarily, I would like to pursue a career in academia in a university or as a research scientist at a research institute, focusing in the area of high-order CFD methods and GPU related simulation.

Why are you interested in pursuing a career in this area?

Aerospace engineering has come a long way since the Wright brothers invented the world’sNaca0021 Crop first successful aeroplane. There is a big paradigm shift in this field and there are still challenges to address. As air travel increases, the impact of aviation becomes a key issue in the form of emissions of carbon dioxide and other greenhouse gases from aeroplanes. Another area is aircraft noise, where a reduction in noise during takeoff and landing is required. The key to developing solutions and addressing these issues is in aircraft aerodynamic design, where current CFD technology is not able to correctly capture unsteady flow. High-order CFD can successfully resolve unsteady flow features, though a major consideration is the computing cost.  Researchers are making significant progress in the acceleration of high-order CFD simulations on modern manycore hardware - achieving 10 times speedup over conventional CFD technology. I think there needs to be more effort to study high-order CFD, and heterogeneous computing - and to promote this technology to industry.

Is this is a key area of interest for you? (Aerospace and the application of HO CFD)

My interest lies in the area of high-order CFD technologies and supercomputing, and its application across aerospace, where this sector is experiencing significant growth.

What are you most notable projects/awards/areas of study?

During my PhD research I discovered supersonic flow using high-order CFD methods, which is a published paper [see references below]. I also researched and developed a novel shock capturing method for high-order CFD.  For me personally, I am very excited and confident about the potential of high-order CFD methods. The Hyperflux project is the first research project to target and use simulations on complex industrial engineering geometry.  The project has also enabled opportunities to collaborate with industry, which is a brilliant experience. In 2014, I was awarded an Outstanding Research Award at Seoul National University, and in 2011, a Korean Society for Industrial and Applied Mathematics Young Researcher Award.

Tell us about the nominations for the 2016 ACM Gordon Bell Prize and SC16 Best Paper

My colleagues and I at Imperial College, London have been shortlisted for the 2016 ACM Gordon Bell Prize for High Performance Computing Applications and also for SC16’s Best Paper Award. The title is ‘Towards Green Aviation with Python at Petascale’ (Peter Vincent, Freddie Witherden, Brian Vermeire, Jin Seok Park, and Arvind Iyer, all of Imperial College London) which covers the topic of achieving accurate simulation of unsteady turbulent flow, which is critical for improved design of ‘greener’ aircraft that are quieter and more fuel-efficient. I am really excited about the event, which will provide the opportunity to meet with the supercomputing specialists, hence my interest in simulation, and understanding further advances in petascale technology.

As a fan of the game Go, Lee Sedol, one of the world's best players lost to AlphaGo, the artificially intelligent player developed by Google DeepMind. Tell us more about your interest and what this means for aerospace engineering.

I learned the game ‘Go’ when I was 9 years old and I still enjoying playing it!. I was very surprised at the result of the AlphaGo match, and prior to this, I didn't expect Lee Sedol to lose. When Deep Blue beat the world chess champion in 1997, it was expected that artificial intelligence (AI) would beat a human at Go by the end of 21st century. This estimate was based on the complexity of Go, and the development pace of supercomputing technology at that time. Go uses a 19x19 grid board and there are almost 130,000 possible next moves after the first two moves, while chess has only 400 possible moves. 

However, this expectation has been destroyed after just 20 years. I think there are two paradigm shifts that are leading this incredible result - deep learning technology and heterogeneous supercomputing. Although the concept of AI and deep learning technology was proposed in 1990s, available computing power was not enough to realise it at that time. With the enhancement of supercomputing technology and efficient implementation, AI can win against a human in Go and it can be utilized in many fields. 

I think that a similar advancement can happen in aerospace engineering. It is hard to simulate highly unsteady turbulent flow because there’s a wide range of length and time scales in turbulent flow. It is expected that we will conduct the large eddy simulation, which resolves 90% of turbulence of flow over whole aircraft in around the 2070s. However, I think that recent high-order accurate numerical methods and heterogeneous supercomputing will realise such a realistic simulation in the near future.

Who do you most aspire to and why?

There are two people that I aspire to. The first is my former supervisor in Seoul University, Professor Chongam Kim, he has almost 20 years of experience, teaching in university and managing a number of research projects and groups. My ambition is for my career to progress in a similar way. The other person I aspire to is Dr Peter Vincent, who is my current supervisor at Imperial College, London.

Jin-Seok Park - Recent Papers, Conferences & ReferencesNaca0021 Snapshot Copy

Simulating Unsteady Flow over a NACA 0021 Airfoil in Deep Stall with PyFR. J. S. Park†, F. D. Witherden, P. E. Vincent. Oral Presentation, 13th US National Congress on Computational Mechanics, 27-30 July 2015. San Diego, California, USA. 

Hierarchical multi-dimensional limiting strategy for correction procedure via reconstruction. J. S. Park, C. Kim. Journal of Computational Physics, Volume 308, 1 March 2016, Pages 57–80.http://dx.doi.org/10.1016/j.jcp.2015.12.020

Comparative study of shock-capturing methods for high-order CPR: MLP and artificial viscosity. JS Park, M Yu, C Kim, ZJ Wang - the 8th ICCFD Conference, ICCFD8-2014, 2014

Further scholar citations:  https://scholar.google.com/citationshl=fr&user=LtkrFVUAAAAJ&view_op=list_works&sortby=pubdate

ACM Gordon Bell Award Nomination

Toward Green Aviation with Python at Petascale. Peter Vincent, Freddie Witherden, Brian Vermeire, Jin Seok Park, and Arvind Iyer, all of Imperial College London.

Accurate simulation of unsteady turbulent flow is critical for improved design of ‘greener’ aircraft that are quieter and more fuel-efficient. We demonstrate application of PyFR, a Python based computational fluid dynamics solver, to petascale simulation of such flow problems. Rationale behind algorithmic choices, which offer increased levels of accuracy and enable sustained computation at up to 58 percent of peak DP-FLOP/s on unstructured grids, will be discussed in the context of modern hardware.

A range of software innovations also will be detailed, including use of runtime code generation, which enables PyFR to efficiently target multiple platforms, including heterogeneous systems, via a single implementation. Finally, results will be presented from a full-scale simulation of flow over a low-pressure turbine blade cascade, along with weak/strong scaling statistics from the Piz Daint and Titan supercomputers, and performance data demonstrating sustained computation at up to 13.7 DP-PFLOP/s.


Sign Up To Our Newsletter and Events