High Performance Computing for Modeling Fluid...
Transcript of High Performance Computing for Modeling Fluid...
19/07/2016
High Performance Computing for Modeling Fluid-Transport
and Electromagnetic Phenomena
Assoc. Prof. Antonis Papadakis
Purpose of Presentation
Educational Background
Employment History
Team Members
Hardware Resources
Numerical Models
Results - High Voltages
Future Work
Vision Forward (ΚΥΑ)
Present latest results and future plans on HPC for fluids and electromagnetic
phenomena
Exploit possibility of collaboration in HPC for Space and other applications
such as Computational Fluid Dynamics, Electromagnetics, Multiphase flows,
Plasmas, Renewables and Heat and Mass Transfer
Identify which areas are niche through collaborative effort
Explore possibility on participating in European and Research Promotion
Foundation (RPF) funded projects
Listen to suggestions regarding current and future work
University of Warwick, Department of Electrical Engineering (1996 - 1999)
BENG in Electrical Engineering
Cambridge University, Trinity College, Department of Electrical Engineering- Full Time
Scholarship (1999 - 2004)
PhD on Plasma MultiPhysics modeling
Skills developed
Maths (Linear algebra)
Modeling of partial differential equations using Finite Difference, Finite Element
and Finite Volume methods
Plasma physics
Software development (Programming language C++, Matlab )
University of Cyprus, Department of Physics, High Energy Physics group (2004 - 2007)
Cyprus representative for CERN – CMS experiment
Part of CMS group that discovered Higgs particle (Nobel Prize)
Electromagnetics (EM) of the ECAL detector of the CMS
Very Front Electronics (VFE’s) of the ECAL detector of the CMS
Fusion simulations for ITER/Hellenic Association of EURATOM
Frederick University, Department of Electrical Engineering (2007 - Today)
High Voltages
Electromagnetic Radiation (Overhead and underground transmission lines,
Substations of open and closed type, Mobile telephony antennas)
Plasmas (Fusion, Microplasmas, Atmospheric plasmas)
Renewables (Photovoltaic, Wind energy)
Computational Analysis – High Performance Computing, Adaptive mesh
techniques, Linear algebra and Computational fluid dynamics
EM
FE-FV
Fluid
PhD programme at Frederick University has recently begun – End of 2012
Started gathering a significant number of PhD students and Post-doctoral staff
PhD students – Erasmus INTACT and LEADERS funding
Mr. Muhammad Naveed Shaikh (Pakistan) – Computational fluid dynamics and
solar cells
Mr. Parash Acharya (Nepal) – Computational fluid dynamics and wind turbines
Mr. Almasy Wasif (Palestine) – Computational fluid dynamics
Mr. Gobinta Panta (Nepal) – Dielectric Barrier Discharges plasmas
Mr. Mujahid Hussain (Pakistan) – Computational fluid dynamics and solar cells
Mr. Kastrounis Nikolaos (Greece) – Computational fluid dynamics in Oil & Gas
Postdoctoral positions
Zafar Qayyum – Transport properties of solar cells
MSc candidates
Total of 10 MSc students
Birendra Suwal (PhD potential candidate) – Computational Fluid Dynamics
PERSONAL High Performance Computing Cluster (HPC)
Windows Server 2008 R2 Enterprise
Setup of a cluster
5 servers x 24 cores = 120 cores
RAM capabilities (228 Gbytes)
Visual Studio Environment: Automation, IntelliSense, Parallel
debugging
MPI already installed/enabled through HPC Pack
Private network (1 GBit/s)
Personal High Performance Computing Cluster (HPC) - Continued
Matlab incorporated in Visual Studio environment for
Viewing result purposes
Simulation purposes
Code written in object orientated software programming language C++
Existing EM/Fluid code is approximately 800 C++ files
Consists code for plasma, electromagnetic, photovoltaic and wind energy through
fluid models and EM solvers
Commercial software will also be used for validation purposes (COMSOL
Multiphysics, Ansys Fluent, PSCAD, Matlab)
Recently gained access to CYTERA supercomputer at CYI for analyzing large scalephenomena:
CYTERA (116 servers x 12 cores = 1392 cores) with large RAM capabilities(116 x 48 Gbytes = 5,568 GBytes)
Characterisation of plasma dynamics by solving:
Poisson equation for the electric field
Continuity equations of charged particles (electrons, positive and negative ions)
Navier-Stokes equations for the neutral gas (mass, momentum and energyconservations)
First to:
Couple charge and neutral gas dynamics in plasma applications back in 2000
Introduce neutral gas heating effects in plasmas from a single electron as aninitial condition
Numerically characterize secondary streamer propagation in Dielectric BarrierDischarges
Analyze streamer branching in long point-plane gaps from a single electron as aninitial condition
Numerically characterize normal and abnormal glow discharges from a singleelectron as initial condition
Analyze positive and negative streamer propagation in long RF gaps
Develop a highly efficient 2D-Adaptive mesh generator producing nearly idealtriangular elements
Poisson’s Equation
Continuity Equations
( ) ( )
r
o
p n ee
N N N 0
e2
eeeeppeeeeee
ND)WN.(NNWNWNSt
N
)WN.(NNNNWNSt
Npppnpneppeee
p
)WN.(NNWNSt
Nnnpnpnee
n
Poisson and Continuity model is coupled with the Navier-Stokesmodel via E/N on which , , , We, Wp, Wn depend on
nppneppeeeee NN2NN|W|N|W|N)v.(t
ssCP)vv.(
t
)v(
Navier-Stokes equations :
Coupled Problem with Continuity equations via
E.JfvC))T(k.()vP.()v.(t
ths
s
Solution Procedure
PO
Next Step
Nn
Start
PO
TPCON
NS
NS
TP CON
n
En
TRn
n+1/2
TRn+1/2
En+1/2
n+1
Nn+1
Nn+1/2
Finite Element-Flux Corrected Transport method (FE-FCT)
Discretization occurs using the Galerkin Finite Element method
Two-step Lax–Wendroff technique is used
Utilizes an accurate High-Order scheme
Utilizes a diffusive Low-Order scheme
Combines the High and Low-Order schemes in some clever way
Produce positive, monotonic, highly conservative results withnon-physical oscillations
Create New Mesh
Initial Coarse Mesh
Error Calculation
Edge swapping and Node Addition/Removals
Mesh Jiggling
Apply Initial Conditions to the Adapted New Mesh
Run the Simulation Decision Time to Re-mesh
Interpolation from the Adapted New Mesh back to the Initial Coarse Mesh
Error Calculation
Create New Mesh1
Edge Swapping and Node Addition/Removals
Mesh Jiggling
Interpolation from Adapted New Mesh to the Adapted New Mesh1
Create Adapted New Mesh
Create Adapted New Mesh1
Bar chart displaying the number ofelements that have similar elementquality values of the Mesh2 createdin a commercially mesh generationsoftware.
Bar chart displaying the number ofelements that have similar elementquality values of the Mesh2 created ina commercially mesh generationsoftware after being treated by theelement quality improvementalgorithm.
Test
Number
Before treatment After treatment Computer
Time (s)
RAM Memory
(kBytes)
Minimum
Value
Average
Value
Minimum
Value
Average
Value
1 0.785 0.960 0.859 0.971 2.568 3504
2 0.614 0.951 0.859 0.971 4.145 3788
3 0.605 0.949 0.875 0.982 5.961 5568
4 0.605 0.948 0.854 0.984 23.781 15736
5 0.602 0.936 0.851 0.981 49.109 22916
6 0.602 0.933 0.836 0.981 63.970 27016
Schematic diagram of the Delaunaytriangulation of a point-planeconfiguration mesh created in acommercially mesh generationsoftware.
Schematic diagram of theDelaunay triangulation of a point-plane configuration mesh createdin a commercially meshgeneration software.
1x10-30.50-1x10-3 -0.5
1x10-3
0.5
0
2
4
6
8
10
12x1021
0
Radial (R) – Coordinates (m)
m-3
2x10-40.2 0.4 0.80300
380
360
340
320
400
Axial (Z) – Coordinates (m)
0.6 1.81.61.0 1.2 1.4
420
440
460
480
Tools to be implemented
MODULAR 3D MIXED FE-FV TVD CODE
3D-ADAPTIVE MESH GENERATOR
PARALLEL PROCESSING - MPI
Utilized in a number of research areas under investigation such as simulation
of:
Electromagnetics (Maxwell’s equation solution)
Plasmas (Fusion, Atmospheric microplasmas, Atmospheric plasmas)
Renewables (Thin film and organic photovoltaics, Wind energy)
MODULAR 3D FE-FV-TVD CODE
C++ templated classes
MPI Implementation
Time-dependent, convection-diffusion schemes for laminar and
turbulent flows
Developing a combination of FV and FE method in 3D
3D-ADAPTIVE MESH GENERATOR
Tetrahedral elements
h and r-refinement
Laplacian smoothing
PARALLEL PROCESSING - MPI
Modular 3D FE-FV-TVD
3D-Adaptive Mesh Generator
Work already done:
Parallel 3D-FE-FCT code using OPENMP software
Intermediate stage into MPI processing
Mesh decomposition software METIS for MPI purposes
Installed and working to provide load balancing between the server cores
Poisson equation using MPI
Currently validating Helmholtz equation using MPI
Currently developing 3rd order FV–TVD and 3rd and 5th order WENO schemes
ELECTROMAGNETICS
Maxwell equation solution using Finite Elements, Finite Difference and Finite
Volume methods:
Human electromagnetic radiation absorption from mobile phones at high
frequencies
Simulation of the EM radiation generated from mobile telephony antennas
using either:
Commercial software (HFFS, COMSOL Multiphysics, OPERA-ELEKTRA)
Developed software
Active and passive shielding of EM sources both at the source and close to the
object under protection
Electromagnetic radiation from overhead transmission and distribution lines,
underground power lines and within substations of closed and open type
THIN FILM AND ORGANIC SOLAR CELLS
Capture the physics of the sunlight propagation and charged particle’s generation
(electrons and holes) and of the background material heating behaviour in
semiconductors
Optical modeling using Helmholtz equation at different frequencies of the
sunlight radiation (Already done using commercial software)
Maxwell equations for light propagation in solar cells
Simultaneous solution of the conservation of mass, momentum and energy
equations for electron, holes (excitons)
Lattice heat equations for the background material by solving a fully coupled
non-isothermal energy balance model
Materials of interest (Perovskites) – Efficiency surpassed silicon since discovered in
2009 - Very promising
ATMOSPHERIC MICROPLASMAS
Analyze atmospheric microplasmas by moving at smaller sizes by operating at
higher microwave frequencies: - Microstrip technology and Ring resonators
Proposed model:
Simultaneous solution of:
Maxwell equations for the calculation of the EM field
Charged and neutral gas particle conservation equations
ATMOSPHERIC PLASMAS
Arc characterization: Avalanche, Primary streamer, Normal glow (Redistribution of
the field on electrode), Abnormal glow, Secondary streamer, Spark and Arc
Incorporate into the existing code: (a) Momentum and energy balance equations
for the charged particles (b) Monte-Carlo simulations for the sheath-region near
the electrodes
Head of Center of Computational Analysis (KYA) – (http://www.kya.frederick.ac.cy)
Vision: Group all numerical modeling researchers of the University under one umbrella
Form a research funded center specialized in certain areas of Engineering such as:
Computational Fluid Dynamics, High Performance Computing, Numerical
modeling, Plasmas, Power Systems, Energy, Artificial Intelligence, Software, Oil &
Gas, Control, etc…
Provide scholarships to attract PhD students
Generate fast training programmes and seminars on High Performance Computing,
Linear algebra, Computational Fluid Dynamics, Adaptive mesh techniques
Setup of a new cluster for running MPI and GPU processing
Permanent staff running daily administrative jobs
Funding through collaboration with industry and through European and local funded
projects
CFD can be used into the design and characterization of the unsteady aerodynamic
environments in existing rocket engine turbines and rocket engine pumps
CFD Applications include modelling of (a) various types of plasma waves and instabilities
in planetary magnetospheres and radiation belts, (b) the confinement and stability of
plasmas in fusion devices, and (c) the propagation of discontinuities and shock waves in
the solar wind
Modelling of spacecraft behaviour due to experiencing of (a) energetic tenuous plasma
‘storms’ under geostationary orbit (b) energetic, dense, directed Auroral plasmas under
polar orbits and (c) cold, dense plasmas under low Earth orbits
Plasma wind tunnel testing for modeling spacecraft during uncontrolled atmospheric re-
entry at the end of their working lifetimes – Comparison with plasma wind tunnel tests
being performed in the L2K and L3K facilities of the German Aerospace Center (DLR) at
its Cologne facility
Simulation of temperatures, pressures, and flow patterns (velocities, accelerations, and
directions) in space vehicles and their propulsion systems
Complex internal flows
Turbopump flows with high speed rotation effects with tip vortex, leakage and
cavitation
Combustion instability for chemical propulsion systems
Simulation of launch vehicles at ascent:
Separated flows, typically along the long missile-type body
Shear layer interaction during stage separation
Plume induced flow separation
Plume with multiple nozzles (engines)
Jet impingement into cavity (flame trench ) during launch
Transient flows at launch
Source: Dochan Kwak and Cetin Kiris, “Current CFD Practices in Launch Vehicle
Applications”, International Workshop on “Future of CFD and Aerospace Sciences”, 23-
25th of April, 2012
High Performance Computing – Advection schemes
Mesh Decomposition software (METIS) for load balancing between the processors
Finite Volume - Total Variation Diminishing (FV-TVD) schemes using MessagePassing Interface (MPI)
Highly accurate advection schemes both in 2D and 3D
Produce positive, monotonic, highly conservative results with non-physicaloscillations
Can capture shock waves extremely well
Standard benchmark tests using MPI (square wave, shock tube, shock on a wedge)
Parallel linear algebra and optimization such as GMRES for asymmetric matricesand Conjugate Gradient methods for symmetric matrices
THANK YOU VERY MUCH
FOR LISTENING TO MY PRESENTATION