Donnons un sens à l'innovation


Offre de thèse : Machine learning of stable dynamic models of capsule suspension flows in microvessels

Sujet : Machine learning of stable dynamic models of capsule suspension flows in microvessels

Laboratoire/équipe : UMR CNRS 7338 Bio­mé­canique et Bio­ingénierie

Mots clés : Micro­cap­sules, flu­id-struc­ture inter­ac­tion, arti­fi­cial intel­li­gence, neur­al net­work, reduced order mod­els, fia­bil­i­ty

Micro-cap­sules, which are flu­id droplets enclosed in a thin elas­tic mem­brane, are cur­rent in nature (red blood cells, phos­pho­li­pidic vesi­cles) and in var­i­ous indus­tri­al appli­ca­tions (biotech­nol­o­gy, phar­ma­col­o­gy, cos­met­ics, food indus­try). They are used to pro­tect and trans­port active prin­ci­ples, by iso­lat­ing them from the exter­nal sus­pend­ing flu­id. One appli­ca­tion with high poten­tial is the use of micro­cap­sules for active sub­stance tar­get­ing, but sci­en­tif­ic chal­lenges remain to be met, such as find­ing the opti­mal com­pro­mise between pay­load and mem­brane thick­ness, char­ac­ter­iz­ing the mem­brane resis­tance and con­trol­ling the moment of rup­ture. Once inject­ed in an exter­nal flow, the par­ti­cles are indeed sub­ject­ed to dynam­i­cal load­ing con­di­tions, which result from the com­plex 3D cap­sule-flow inter­ac­tions. To mod­el them numer­i­cal­ly, one needs to account for the non-lin­ear large defor­ma­tions, which results in large sys­tems of equa­tions and thus in long com­pu­ta­tion­al times.

The objec­tive of the PhD project is to explore the use of advanced numer­i­cal meth­ods to speed up the sim­u­la­tions when solv­ing the inter­ac­tions between the internal/external flows and the defor­ma­tion of the hyper­e­las­tic mem­brane.
We will study both « neur­al net­work » approach­es and mod­el order reduc­tion, with the aim of com­par­ing their reli­a­bil­i­ty, sta­bil­i­ty and long-term pre­dictabil­i­ty. These algo­rithms will be trained and val­i­dat­ed from a data­base of numer­i­cal sim­u­la­tion results,
obtained with the open-source HEMOCELL cou­pled code, ded­i­cat­ed to the high per­for­mance sim­u­la­tion of dense sus­pen­sions of cells and capsules.