exascale needs & challenges for aeronautics industry · 2019. 12. 19. · confidential and...

Exascale Needs & Challenges for Aeronautics Industry PRACEdays15 - Dublin, May 26th 2015

May 2015Exascale Needs & Challenges for Aeronautics Industry - PRACEdays15

Flight-Physics Capability StrategyDr Eric CHAPUT

© AIRBUS S.A.S. All rights reserved. Confidential and proprietary document.


Content

HPC in Aeronautics a Historical Perspective

Current HPC Resources and Applications at Airbus

Exascale Vision

Exascale Challenges

Conclusions



A historical perspective

In the last 40 years, commercial aviation industry has achieved:

1960 1970 1980 1990 2000 2010 202020%

40%

60%

80%

100%

1950

Relative Fuel Burn per seat

- 70% fuel burn

- 70% CO2 emissions

Noi

se L

evel

1960 1970 1980 1990 2000

2nd Generation Turbofans

Turbojets1st Generation Turbofans

20 dB

More than 20dB improvement

Within Airbus, we have played a key role in reducing the environmental impact from aviation



Simulation and HPC in the A380 Design

The first Airbus aircraft to have exploited aerodynamics simulation & HPC

Quieter.

Cleaner. Less than 3 litres of fuel per pax per 100 kmLess than 75g of CO2 per pax per km

80828486889092949698

A340-300 A340-600 A380-800

EPNdB 275t

+90t365t

+285t

560t

A380 Bigger & Quieter

Twice as many passengers flown for the same noise level



Pushing the Boundaries through the A350

An evolution of the aerodynamics design: Heavy use of simulation and HPC – around 5000 data setsCryogenic Wind Tunnel Testing replicating flight conditions

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.10.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.10.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.10.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1

Before

After

Wing Span

Wav

e D

rag Not A350 Distribution

Advanced CFD + Optimisation

Flight Re Testing

Optimised aircraft design, finest aerodynamics and lowest fuel burn

40 % less test than in A380


Airbus HPC patterns

•Simulation frameworks• Higher process level (e.g. D.O.E./sensitivity analysis, optimization, trade-off studies…)• Massive data production in single process (e.g. aero-data direct production geometry

using HPC)

• Complex processes (e.g. multi-level, multi-disciplinary optimization) Advanced execution & data management

•Grid services / LSF profiles• Virtualization of HPC resources (e.g. separate & global optimization of HPC activities )• Meta-scheduling global virtualization of HPC resources through grid, local

queues management through LSF•FlowSimulator / Open-PALM

• FlowSimulator modular approach (e.g. share mesh parallel services , coupling strategies parallel scripted at solver time loop level)

• Capability to use FlowSimulator brick as a server being one of the component of a distributed simulation



-

-

-

Spain

United

Kingdom


AIRBUS Cloud & Grid Architecture

AIRBUS GRID SERVICE

AIBUS IndiaToulouse

Filton

HamburgBremen

Bangalore

Getafe

HPC Service

HPC Service HPC Service

AIRBUS IW GRID SERVICE

Suresne

AIRBUS HelicopterGRID SERVICE

Marignane

Donauwörth

Ottobrunn

Private HPC CloudGreen Computing : PUE < 1.2World Grid-Computing : Europe, America and IndiaLSF Platform-Computing(IBM) and SynfinyWay (Fujitsu)HP POD - Cluster PlatformInterconnect - Infiniband


AIRBUS HPC4 - Key Characteristics


Location : Toulouse, HamburgContainers : 3 POD‘sServers : 2 320 Cores : 55 680RMAX : 1 to 2 Pflops (Ivy

Bridge, Broadwell)Memory : 270 TBLocal Storage : 1 500 TBShared Midterm Storage : 750 TBShared ComputingStorage : 1 800 TBOutsourced serviceEnergy Efficiency (PUE < 1.2)Access / SchedulerResource Management : LSF / SynfiniwayInterconnect : Infiniband



Flight Physics current major HPC customer

Computational Fluid Dynamics• Design & Optimize External Aircraft Shapes• Design Sensitivities & Trades• High Speed an Low Speed Performance• Unsteady Aerodynamic/Aeroelastic Effects• Structural Flexibility Effects • Aircraft Data Model for the whole Envelope• Loads and Aeroelastic Data• Multiphysics : Aeroacoustics, Aerothermics…

2013

80%of HPC Capacity

Aerodynamics Loads & Aeroelastics

Flight-Physics


Systems, Structure, PowerPlant future big customers


20%of HPC Capacity

2013

SystemsSafety & Certification contribution• Electromagnetic compatibility in nominal

and hazardous conditions (EMH)

Fuel Tanks: thermal fluid modelling• Thermal fluid modelling for Fuel tanks is

used throughout the whole A/C development lifecycle

Major Component Optimisation• Box sizing (covers)• Composite fuselage preliminary sizing

(covers thickness)• Hundreds of variables and thousands of

constraintsVirtual testing - Non Linear FE• Generalization of NLFE models with

contact and complex materialStrength analysis

StructureNoise prediction• Integration of noise constraints in the A/C

design process• Robustness of noise assessment along

development cyclePropellers / contra propellers noise• Assessment of noise radiated by SRP and

CRP.• Optimisation of engines installations (MDO)• Extensive use of coupled CFD/CAA

numerical tools

Powerplant



Progress in Flight Physics Simulation Capability

Physical Complexity

Cap

abilit

y –

Flig

ht E

nvel

ope

Cov

erag

e

- Complete physical modelling - Laminar-Turbulent Transition- Multidisciplinary Optimisation

3D simulation of installed engine -CFD-based Aero Data -

Optimisation on Engine integration -Low speed High-Lift -

- Support to Flight-Tests- Unsteady Aerodynamics- Thermal Aircraft

CFD – Computational Fluid Dynamics

- High speed CFD-based scaling- Simplified geometry

- High fidelity Aerodynamics Simulation - Reducing standard wind tunnel testing- Flexible Aircraft Representation



Simulation Envelope


3g

0g

- 1g

2g

1g

LOA

D F

AC

TOR

3g

0g

- 1g

2g

1g

VC

VD

EAS(at a given altitude)

LOA

D F

AC

TOR

VF

Aerodynamic Flow Predictions vs. Flight Envelope

Buffet boundary

• The colour gradient is intended to show level of confidence in CFD flow solutions

• Beyond buffet boundary flow tends to become more and more separated from the aircraft surface

• Unsteady effects start to become dominant• High lift configuration

flow is physically very complex due to high loading

• Low g flow tends to separate on the lower side of the aircraft

• Local low Mach number / low compressibility flow only weakly coupled to main flow

• Transonic flow effects (shocks) get stronger with increasing Mach number and load

• Interaction of physical effects (shock flow, boundary layer flow) start to dominate

• Maximum lift borders are determined by onset of massive separation causing total lift loss

• Region of highest accuracy and robustness

• Region of most experience



3g

0g

- 1g

2g

1g

LOA

D F

AC

TOR

3g

0g

- 1g

2g

1g

VC

VD

EAS(at a given altitude)

LOA

D F

AC

TOR

VF

Aerodynamic Flow Predictions vs. Flight Envelope

Buffet boundary

• The colour gradient is intended to show level of confidence in CFD flow solutions

• Beyond buffet boundary flow tends to become more and more separated from the aircraft surface

• Unsteady effects start to become dominant• High lift configuration

flow is physically very complex due to high loading

• Low g flow tends to separate on the lower side of the aircraft

• Local low Mach number / low compressibility flow only weakly coupled to main flow

• Transonic flow effects (shocks) get stronger with increasing Mach number and load

• Interaction of physical effects (shock flow, boundary layer flow) start to dominate

• Maximum lift borders are determined by onset of massive separation causing total lift loss

• Region of highest accuracy and robustness

• Region of most experience

Uncertainties

Unsteadiness of flow

MD interaction

Prediction accuracy

Comprehensiveness of A/C data

Optimisation




Exascale Vision

Exascale computing key enabling technology for future aircraft design

Fully multi-disciplinary development and optimization process

Wide use of integrated MDA/MDO capabilities

Real-time/interactive way of working fostering innovation


Multi-disciplinary analysis and design, supported by affordable CFD-based aerodynamic and aero elastic data prediction will be a significant change of paradigm in aeronautics industry.


Multidisciplinary Design Capability

Twist / Camber /Thickness

-Initial Geom.

-Step 1-Step 2

Parametric Geometry / Shape Generation

Mesh Generation / Deformation

Finite Element Model Generation

Aerodynamic Coefficients

Structural Forces

CFDFlow Solver CSMLoads

Deformation

LOAD

FAC

TOR

3g

2g

1g

0g

-1g

A C D

E

FH

VF VD

EAS

Vs1g

Design SpaceMapping

Full Flight Envelope



Exascale Challenges: Fly the Virtual Aircraft

Real-time Simulation of Manoeuvre Flight of a Complete Aircraft Full Unsteady aerodynamics simulation and Multi-disciplinary interactions

Full Finite Element Modelling of the Airplane

Full Simulation of Manoeuvre FlightLoads and Stress for the

Airplane in the Whole Envelope

Digital Prediction of “Flight Performance” and “Handling” Prior to First Flight



RANS to LES

Move from RANS to unsteady Navier-Stokes simulation, (ranging from current RANS-LES to full LES) and/or Lattice Boltzmann methodSignificantly improve predictions of complex flow phenomena around full aircraft configurations with advanced physical modelling.

Exascale

Petascale


Complex high lift case: RANS-LES vs LBM


Nose Landing Gear wake interacting with Main Landing Gear & cavities

Mesh data• 46.8 M nodes• 131 M cells• 72.5 M prisms• 58.5 M tetras• 37 prism layers



Petaflops to Exaflops Challenges

1980 1990 2000 2010 2020 2030

Capacity: # of Overnight

Loads cases run

Capacity: # of Overnight

Loads cases run Capacity [Flop/s]

CFD-basedLOADS

& HQ

CFD-basedLOADS

& HQ

Aero Optimisation& CFD-CSM

Aero Optimisation& CFD-CSM Full MDOFull MDO

Real timeCFD based

in flightsimulation

Real timeCFD based

in flightsimulation

x106

1 Zeta (1021)

1 Peta (1015)

1 Tera (1012)

1 Giga (109)

1 Exa (1018)

102

103

104

105

106

LES

CFD-basednoise

simulation

CFD-basednoise

simulation

RANS LowSpeed

RANS HighSpeed

HS Design

Data Set

Capability achieved during one night batchCapability achieved during one night batch

UnsteadyRANS

“Smart” use of HPC power:• Algorithms• Data mining• knowledge

1980 1990 2000 2010 2020 2030

CFD-basedLOADS

& HQ

CFD-basedLOADS

& HQ

Aero Optimisation& CFD-CSM

Aero Optimisation& CFD-CSM Full MDOFull MDO

Real timeCFD based

in flightsimulation

Real timeCFD based

in flightsimulation

x106

1 Zeta (1021)

1 Peta (1015)

1 Tera (1012)

1 Giga (109)

1 Exa (1018)

106

105

104

103

102

LES

CFD-basednoise

simulation

CFD-basednoise

simulation

RANS LowSpeed

RANS HighSpeed

HS Design

Data Set

UnsteadyRANS

“Smart” use of HPC power:• Algorithms• Big Data• Knowledge

# of Overnight Loads cases run

Capacity

Achieved during one night batchCapability

Available Computational

Capacity (Flops/s)



Exaflops Challenges

Hardware challenges:• Billion of components• Energy consumption • Size of system memory• System resilience• Energy Efficiency• System resilience

Software Co-design challenges:• Extreme Parallelism• Compute, I/O & Storage Performance Balance• Simple rule of thumb design• Middleware aware of failure resilience

Optimize trade-off among• Supporting new workloads, e.g. Big Data• New HPC Delivery models, e.g. Cloud• applied mathematics, algorithms, computing challenges

Educational challenge

2018-2020 - CFD New Generation



CFD Technology Roadmap : NASA Vision 2030 CFD Code


High Fidelity CFD Strategy

Converge the current solutions

Exascale Needs & Challenges for Aeronautics Industry - PRACEdays15 May 2015


Evolution of CFD towards Exascale


Knowledge & Visualization: POD RACER GappyPOD Unsteady Data extract.

// – parallel computingAMR – Automatic Mesh RefinementCFD – Computational Fluid DynamicsDG – Discontinuous Galerkin methodIGA – Iso-Geometric Analysis

LBM – Lattice Boltzmann MethodLES – Large-Eddy SimulationMB – Multi-BlockMDA/MDO – Multi-Disciplinary Analysis / Optimisation

POD – Proper Orthogonal DecompositionRANS –Reynolds Averaged Navier-StokesRSM – Reynolds Stress turbulence ModelSA – Spalart-Almarass turbulence modelSDM – Spectral Difference Method

SST – Shear Stress Transport turbulence model

Process Automation: CFD Manager, FlowSimulator Rapid CFD for MDO

Mesh Generation & Adaptation: MB, Hybrid Chimera, AMR Automatic Hyperflex, Uncertainty

Solver Algorithms: RANS 2nd Order // effectiveness Uncertainty High-Order (DG, SDM, IGA), LBM ; massively //

Hardware: TeraFlop/s 2004 PetaFlop/s 2014 ExaFlop/s 2024

Physical Modelling: RANS (SA, SST RSM) Unsteady RANS-LES, LBM LES

Multi-Disciplinary / Multiphysics: Parametric Shape, Adjoint-based Optimisation Coupling Validation & Verification MDA/MDO


Challenge #1 – Visualization/Identification of aircraft shape in flight


•Design of Experiment of CFD-CSM computations•POD of surface pressure•Visualization of sensitivity to flight conditions

LES

Design of Experiments

Proper Orthogonal Decomposition

Non-Linear Interpolation

CFD – Computational Fluid DynamicsCSM – Computational Structural MechanicsDOE – Design of ExperimentsPOD – Proper Orthogonal Decomposition


•Derivative A/C design• Minimum change in geometry vs legacy• Capturing the flow physics (jet, ice shape…)

•New A/C design • Immersed boundary in LES

Challenge #2 – From Automated Chimera to Immersed CAD LES


LES

RANS


Challenge #3 – Multidisciplinary Design Optimization


MDOAdjoint-based Aero Optimization

Awareness of design change effect on other disciplines - Adjoint-based flexible shape Optimization- Sensitivity of Aero design on Structure Reserve Factor



Direct Noise Computation

LES - 2x109 cells - HPC 8000 CPUcores – 20days

Challenge #4 – Flow control to reduce noise and vibration

RANS/LES - 7x107 cells - HPC 100 CPUcores – 15days

LES – Large-Eddy SimulationRANS –Reynolds Averaged Navier-Stokes

Constrained design



Conclusion

Exascale computing is seen as a key enabling technology for future aircraft design to be developed and optimised in a fully multi-disciplinary way, making a wide use of design systems that provide integrated analysis and optimisation capabilities which allow for a real-time/interactive way of working.The move from RANS to unsteady Navier-Stokes simulation, (ranging from current RANS-LES to full LES) and/or Lattice Boltzmann method will significantly improve predictions of complex flow phenomena around full aircraft configurations with advanced physical modelling. For instance moving LES capability from Petascale to Exascale computing will accelerate the understanding of noise generation mechanisms and will enable the elaboration of flow control strategy for noise reduction. Multi-disciplinary analysis and design, and real time simulation of aircraft manoeuver, supported by affordable CFD-based aerodynamic and aero elastic data prediction will be a significant change of paradigm in aeronautics industry.



© Airbus S.A.S. All rights reserved. Confidential and proprietary document. This document and all information contained herein is the sole property of AIRBUS. No intellectual property rights are granted by the delivery of this document or the disclosure ofits content. This document shall not be reproduced or disclosed to a third party without the express written consent of AIRBUS S.A.S. This document and its content shall not be used for any purpose other than that for which it is supplied. The statementsmade herein do not constitute an offer. They are based on the mentioned assumptions and are expressed in good faith. Where the supporting grounds for these statements are not shown, AIRBUS S.A.S. will be pleased to explain the basis thereof.AIRBUS, its logo, A300, A310, A318, A319, A320, A321, A330, A340, A350, A380, A400M are registered trademarks.

exascale needs & challenges for aeronautics industry · 2019. 12. 19. · confidential and...

Documents