Innovating the Future of Flight - HPC Advisory Council · 2020-01-16 · • A business driven innovation. Accelerate company’s transformation –take full advantage of new technologies.

HPCXXL Conference

April 2019

Innovating the Future of Flight

Content

2nd April 2019 HPCXXL Conference, Gerd Büttner2

Introduction to Aircraft Business

Driver for HPC in Aircraft Industry

Current HPC /Grid Infrastructure

Current Limitations

Upcoming Requirements

Future Challenges

3

Air traffic

doubles every

15 years

3.6 billion

Passengers

62.7million

Jobs supported

51.2 million

Tonnes of freight

$2.7trillion

Global GDB annually

Aviation drives our

global economy.

2nd April 2019 HPCXXL Conference, Gerd Büttner

4

Commercial Aircraft’s

global workforce is

united by a passion for

aviation and restless

desire to create better

ways to fly.

7,390 Backlog

400+ Operators

56,000 Employees

€53.8 billionAnnual revenue, restated

IFRS15

Passion for aviation.


5

19,241Aircraft sold

60Produced monthly

25,000+Daily flights

11,851Delivered

An Airbus takes off

or lands every 1.4 seconds.

End February 2019


New Commercial Aircraft Delivery Record

6

*including A220: 20 deliveries, 135 orders, Backlog 480 aircraft

Deliveries

800

2016 20162017 2017 20182018 2018

Orders

747net

Backlog

7,577 aircraft

688

6,874

718

7,2657,577*

800*

Orders

747*

Deliveries

+11%


Global Company

7

China1,150

Spain13,177 (10%)

USA2,707 (2%)

UK11,304 (9%)

Germany44,214 (34%) France

47,865 (37%)

133,700 employees

from 135 nationalities

Located across 35 countries

on more than 180 sites

International Sales

31% Europe

69% Non-EuropeOther countries

10,178 (8%)


Content

8




Current Limitations


Future Challenges


We have achieved a tremendous increase in Flight Physics simulation capability

over the last 20 years

9

• High fidelity Aerodynamics Simulation

• Reducing standard wind tunnel testing

• Flexible Aircraft Representation

• Complete physical modelling

• Laminar-Turbulent Transition

• Multidisciplinary Optimisation

• High speed CFD*-based scaling

• Simplified geometry

• 3D simulation of installed engine

• CFD*-based Aero Data

• Optimisation on Engine integration

• Support to Flight-Tests

• Unsteady Aerodynamics

Physical Complexity

Ca

pa

bili

ty –

Flig

ht

En

ve

lop

e C

ove

rag

e

* CFD = Computational Fluid Dynamics


Pushing the boundaries with the A350

Heavy use of simulation and HPC for aerodynamics design– around 5000 data sets

Cryogenic Wind Tunnel Testing replicating flight conditions

10

Advanced CFD +

Optimisation

Flight Re Testing

Optimised aircraft design, finest aerodynamics and lowest fuel burn

40 % less test than in A380


HPC enables the virtualization of aircraft development

11

• Engineering domains using HPC:

✓ Flight Physics - Aerodynamics design and data

✓ Structure - Optimization, composite, virtual testing, global

Finite Element Method

✓ Power plant - Noise of turbo machinery, CROR, airframe,

cabin & cockpit noise

✓ Systems - Lightning, fuel

✓Multi Disciplinary Optimization

• HPC at Airbus is a key enabler for Engineering to achieve

the main objectives of new Aircraft programmes:

✓ Optimize the level of performance

✓ Increase maturity at entry into service

✓ Deal with the complexity

✓ Answer to certification requirements

✓ Decrease lead time and costs


Aircraft Industry

12 2nd April 2019 HPCXXL Conference, Gerd Büttner

Content

13




Current Limitations


Future Challenges


Airbus Scientific Computing Architecture

14

AIRBUS GRID SERVICE

AEC

I

Toulouse

Filton

HamburgBremen

Bangalore

Getafe

HPC Service

HPC Service

HPC Service

Airbus Helicopters GRID

SERVICE

Marignane

Donauwörth

Ottobrunn

Airbus Defence &Space

29 location in Europe

Germany

France

Spain

UK

Italy


HPC4/5 Functional Architecture View

15

MIDTERM StorageSCRATCH

Login NodesTransfer NodesGrid NodesAdmin Nodes

Type1 Nodes Type2 Nodes

Infini-

bandMIDTERM StorageSCRATCH

Login NodesTransfer NodesGrid NodesAdmin Nodes

Type1 Nodes Type2 Nodes

Infini-

band

Toulouse Hamburg 10 Gb Ethernet

56 Gb Infiniband

8 x 10 Gb 4 x 10 Gb


Hardware-Details

• Compute nodes

(Skylake 16 core /socket, 2 sockets by node)

− Type1 – 96 GB Mem/node (3 GB/core) – 1TB HD local storage

− Type2 – 384GB Mem/node (12 GB/core) – 7,8 TB SSD local storage

• Login nodes enable user interactive logon on Cluster

• Transfer nodes to Transfer Data to/from the Cluster

• Scratch (DDN Lustre) to store temporary data during execution

• Midterm Storage (PANASAS) to store data for multidisciplinary optimization

• Grid Nodes to run LSF resource management

• Administration nodes to support DCX cluster Administration


POD – Performance Optimized Datacenter

17

PODs

Key characteristics:

- Super computer in a container

- Outsourced service

Key advantages:

- High flexibility for growth and technology refresh

- High density

- Energy Efficiency (PUE 1.25)

Key Performance Factors:

- Expected ~ 3,3 PFLOP by 60000 cores with 235 TB Memory

- 2,8 PB Scratch and 750 TB Mid-Term Storage

Chillers

Secured

Power

Supply

Power Generator

TLS POD area


Content

18




Current Limitations


Future Challenges


Current Technology

• Number of core will

increase

• Energy by socket will be

stable

• CPU clock frequency will

decrease

• Single core performance

will decrease

Need to Scale


Top 6 – November 2018

6 different processor types /acceleratorIBM Power/NVIDIA, Sunway FPGA, Intel Xeon / THE-2, Intel Xeon/NVIDIA, Intel Xeon / Intel Phi


Software Scalability

21

Software is the #1 roadblock


Outlook

• New high scalable CFD Code will be available

• AMD „Rome“ will be used.

− The AMD „Rome“ chip has ¼ more memory capacity /bandwidth compared with Intel Skylake

• General the CPU become data mover, Compute Power are from Accelerators (GPUs or Vector Units)

• ARM become a serious Player on CPU market (not at Skylake level, but not far away)


Content

23




Current Limitations


Future Challenges


Emerging Technologies in Scope



HPC User Forum Forecast AI, ML, DL


Additional Methods for Engineering Design and Science

27

Model

Physics + Maths

+HPC

Prediction

Prediction

Model

Big Data & Machine Learning

Top Down

deductive

Bottom Up

inductive

Deep

Learning

Machine

Learning

Artificial

Intelligence

Learn from

past models

& Simulations

Future Simulations will take benefit from HPC and BigData/ML2nd April 2019 HPCXXL Conference, Gerd Büttner

Manufacturing Capabilities explained

… geo-located

product-DMU,

enriched with

process &

product info

Augmented Reality

through …

Bracket

check

overlay

mode▪ Sys. Installation

▪ Inspect aircraft

brackets, ESN

fasteners

▪ Manage Non-

Conformities

Example use cases

▪ Verify build status

▪ Capture & report

▪ Assess & map

damages

▪ Trouble shoot

▪ Identify parts &

error root causes

▪ Initiate corrective

actions

To do what?

geo-located &

mapped items

Fastener

Conc.

NC

NC

NC

NC

HPC@ Airbus; Gerd Büttner


Content

29




Current Limitations


Future Challenges


30

A commercial aircraft manufacturer with two other Divisions - Defence and

Space and Helicopters.

We are a leader in designing, manufacturing, and delivering aerospace products,

services and solutions to customers on a global scale.

Airbus is an international pioneer in the aerospace industry.



Hybrid

HPC

AI

HPC

Cloud

HPDA

Future Architecture view

Data Centric Architecture

32

Data

Data

Lakes

PLM

MES

0


Factory of the future - Implementation strategy Planning


Page 34

• Integration between HPC, BigData and AI

• IoT as one driver for Factory of the Future

➢ BigData Analytics on lifestream data from

the factory

➢HPC requirements to

o Analyse data

o Run simulation optimization models

(Digital Twin)

➢ AI decision processes based on Data

Analytics input

Factory of the future - Implementation requirements to Information

Management


Quantum Computing


Five Problem Statements

Aircraft Climb Optimisation Computational Fluid Dynamics

Wingbox Design Optimisation Aircraft Loading Optimisation

• 5 distinct flight physics problems

with varying degrees of

complexity

Quantum Neural Networks for Solving

Partial Differential Equations


37

• 1st twin-engine widebody aircraft, A300

• 1st full fly-by-wire commercial airliner

• 1st double decker airliner, A380

• 1st manufacturer to make extensive use of

advanced materials

• A business driven innovation. Accelerate company’s

transformation – take full advantage of new

technologies.

• Speed up innovation – Co-innovate with external

partners.

• Airbus Ventures funds start-ups around the globe.

Industry firsts Innovation creates value.


Thank you

Innovating the Future of Flight - HPC Advisory Council · 2020-01-16 · • A business driven innovation. Accelerate company’s transformation –take full advantage of new technologies.

Documents