BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:First steps of numerical simulation using Artificial Intelligence
DTSTART;VALUE=DATE-TIME:20191009T153000Z
DTEND;VALUE=DATE-TIME:20191009T160000Z
DTSTAMP;VALUE=DATE-TIME:20230208T065704Z
UID:indico-contribution-4947@indico.esa.int
DESCRIPTION:Speakers: Vincent Vadez (Dorea)\, Pierre Alliez (INRIA Sophia
Antipolis)\nAs for the thermal simulation of a satellite\, its design is c
rucial. It involves the integration in the thermal and mechanical model of
reduced parts of equipment or supports (additive manufacturing) from 3D m
odels finite elements. The goal of the thesis is the drastic reduction of
3D geometric models that minimizes the alteration of the physical properti
es (here the calculation of the view surfaces) induced by the model reduct
ion during the numerical simulation.\n\nWe are currently working on **an a
pproximation guided by view factors**. The aim is to design a geometric ap
proximation method where each atomic simplification or optimization operat
ion is guided by preserving the view factors of the reference model\, rath
er than preserving the geometry. The ultimate aim being a radiative therma
l computation on a "small" model made of a few hundreds of facets instead
of millions of facets. \nThe next step consists in **a supervised learning
of the geometric error metric** (using **deep learning**). Deep learning
is a subset of machine learning (which is also a subset of Artificial Inte
lligence). In both cases\, algorithms appear to learn by analyzing huge am
ounts of data. Deep learning processes data using computing units\, called
neurons\, arranged into ordered sections\, called layers. A stack of thos
e layers form what we call a neural network\, which is the kind of model w
e will use.\nWhen working with supervised algorithms\, the input data is l
abeled and has a specific expected result. We train a model to predict the
labels of the given training examples. As training progresses\, the predi
ctions or classifications become more accurate. Our idea is to use deep le
arning as follows : from a database of expertly generated 3D models\, we w
ill design a supervised machine learning method using multiple layers to l
earn the geometric error metric able to govern an automatic approximation
algorithm so that the resulting thermal simulation is as accurate as possi
ble to a reference calculation. \nWe already developed and are still impro
ving a software relying on a new hierarchical geometric data structure for
the efficient computation of view factors\, which allows us to have a pre
cise reference case.\n\nhttps://indico.esa.int/event/308/contributions/494
7/
LOCATION: Newton 2
URL:https://indico.esa.int/event/308/contributions/4947/
END:VEVENT
END:VCALENDAR