Francisco)(Paco))CHINESTA)

ESI)Chair)Professor)at)ECN)(Ecole)Centrale)de)Nantes)))High)Performance)Compu?ng)Ins?tute)–)Computa?onal)Physics)Group)

President)of)ESI)Group)Scien?fic)CommiDee)Fellow)of)the)«Ins?tut)Universitaire)de)France»)

Fellow)of)the)Spanish)Royal)Academy)of)Engineering))

Simulation, données, information, connaissance et prise de décision :

configurant la 4ème révolution industrielle

I’m an engineer luck or disgrace?

Scientists discover the world that exists, engineers

create the world that never was

Theodore'von'Karman'

Predic'ng*should*precede*crea'ng!*

But,*how*predic'ng*the*behavior*of*something*that*never*was?*

Many*op'ons*…*

The best way to predict the future is invent it

Alan%Cur)s%Kay%

It rarely works …

Tarot&reading&

10/02/2015&

a particular state of mind is required!

The$scien)fic$approach$

Example:)Astronomy)

First)was)the)data)6)Babylonian)astronomy)6)

then)were)the)modelS))

Almageste)from)Claudius)Ptolemy)))

F = G M ⋅md 2

Newton) Einstein)

All models are wrong but some are useful

George&Box&

Numerical*simula,on*based*on*models*

! Present

! Future

Model

• !Material!parameters:!• !Geometrical!parameters:!• !Applied!loads:!

U(x,t;M1,!,Mm ,G1,!,Gg ,L1,!,Ll )

ISSUE%Combinatorial%explosion%

Solu5on%for%each%choice%of%parameters%

10%parameters%10%values%/%parameter%%=%%%%%%%%%%%%%%%%%%%%%1010%%possibili5es%!!%

When%all%“possible”%solu0ons%have%been%%pre3calculated%!!%

Towards(modern(vademecums(…(

PGD$constructor$circumvents$curse$of$dimensionality$

U(x,t, p1,!, pm ) U(x,t; p1,!, pm ) (3+1+ m)D

U(x,t, p1,!, pm ) ≈ Xi (x) ⋅Ti (t) ⋅Pi

1(p1)!Pim (pm )

i=1

N

∑

Automated)Tape)Placement))

T (x, y, z, t, V,Q, o1, · · · , on, h1, h2, h3)

First was the data, then were the models,

today data-driven applications systems appears as the new

paradigm in SBES

From data to information: removing correlations

p1

p2S(p1, p2 )

S(p1) or S(p2 )

Patient-specific surgical planning Liver data

Only two uncorrelated parameters

New patient liver

Data interpolation

Real-time simulator patient specific

alleviates(modeling((

Experiments

Nowadays

Model Simulation

Includes modeling errors

Future

Experiments Simulation

Collaborative enrichment

Manifold learning based model

KIAS?

From%a%methodological%viewpoint:%If%Hooke%had%never%existed,%linear%elas;city%finite%element%simula;ons%(based%on%data)%would%have%existed?%

YES,%and%not%only%elas/city!%

Cheap&manifold&constructor&

Trial&behavior&manifold&

Model7free&simula8on&

Manifold&upda8ng&

Test,&IC&

Comparison&

Phase&A&:&behavior&model&A&

Phase&B&:&behavior&model&B&

What&behavior&?&What&model&?&

Model6Free&Upscaling&

uv

⎛⎝⎜

⎞⎠⎟= α

x0

⎛⎝⎜

⎞⎠⎟+ β

0y

⎛⎝⎜

⎞⎠⎟+ γ

yx

⎛⎝⎜

⎞⎠⎟

DNS$

σ (x)ε(x)⎧⎨⎩

⎫⎬⎭⇒

ΣΕ⎧⎨⎩

⎫⎬⎭∈!12

Manifold$learning$–>$Dim$=$2$

Model$free$macro$simula8on$

Methodological$example$

Macrostructure*manifold*

Behavior*manifold*

Data$Driven,+Data$Mining+and+Machine+Learning,+Deep$learning,+…+open+plenty+of+

possibili:es.+

BUT$be$careful,$surprises$are$not$excluded$

Faithless)electors:)wrong)data)Sampling)deficiency)Data8mining))

In#the#4th#industrial#revolu1on#there#is#s1ll#room#for#the#human#

being#