Où docteurs et entreprises se rencontrent
Menu
Connexion

Vous avez déjà un compte ?

Nouvel utilisateur ?

Development of an approach combining physical model and artificial intelligence for modeling the magnetic properties of electrical steels

ABG-96410 Sujet de Thèse
03/03/2021 Contrat doctoral
L2EP, Université de Lille
Villeneuve-d'Ascq - Les Hauts de France - France
Development of an approach combining physical model and artificial intelligence for modeling the magnetic properties of electrical steels
  • Sciences de l’ingénieur
  • Matériaux
  • Mathématiques
magnetic materials, artificial intelligence, numerical modelling, deep learning

Description du sujet

Controlling the energy performance of modern electrical motors, in particular for industry and electric mobility, requires reliable and accurate design tools. One of the important parameters to achieve this goal is related to the electrical steels that are used to manufacture the magnetic circuits of these motors. The properties of these materials define performance and energy efficiency during the energy conversion process. Accurate modeling of these electric steels must therefore be available for an optimal design of electric drives. However, the manufacturing processes and the severe operating constraints of these electrical motors require now to take into account the multi-physical behavior of electrical steels. Indeed, in modern applications, the mechanical and thermal stresses to which these materials are subjected impact the properties of interest for energy conversion (magnetic constitutive law and iron losses). These strong multi-physical couplings which appear in electrical steels (magneto-mechanical and magneto-thermal) often lead to the degradation of the properties of electrical steels and therefore of the performance of the motors. In order to model these materials taking into account the impact of manufacturing processes and operating conditions, conventional approaches result in heavy models in terms of complexity of implementation in design tools but also in terms of computing time. However, designers need multi-physical models of materials that are both accurate and fast to evaluate, especially for electrical motors design procedures involving optimization. These elements therefore require rethinking the way of representing physical behavior, especially that of magnetic materials, within a design and optimization procedure.

In this thesis, we propose to combine multi-physical models of electrical steels with Artificial Intelligence (AI) to reinforce the models which will necessarily be less accurate because these are suitable for design and optimization procedures, i.e. models that are fast to evaluate. To achieve this goal, we will rely on Deep Learning (DL) which allows to consider a larger number of layers for learning the neural network. This will be based on a Convolutional Neural Network (CNN) architecture that can be fed using carefully selected experimental data, especially those that complement the shortcomings of the multi-physics model, during the learning phase.

The ability of the multi-physics model, associated with AI, to represent the properties of interest for the design will be assessed. Two main axes will then be considered in this work. The first relates to the development and validation of multi-physical models of electrical steels associated with AI. The second axis will deal with the implementation of the models in a numerical finite element calculation code to validate the approach on academic test cases. Throughout the work, care will be taken to rely on experimental validations.

Prise de fonction :

01/10/2021

Nature du financement

Contrat doctoral

Précisions sur le financement

Financement ANR / ISITE ULNE

Présentation établissement et labo d'accueil

L2EP, Université de Lille

UniverSIty of LIlle, L2EP

The Laboratory of Electrical Engineering and Power Electronics (L2EP) is structured into 4 research teams, covering all aspects related to the field of electrical energy (design, modeling and management of electrical energy). The team “Numerical Tools and Methods”, involved in the proposed PhD project, has been developing for 25 years models of electrical devices mainly based on the finite element method. The development of these models requires a very good knowledge of ferromagnetic materials which play a key role within electrical devices. In that context, the team conducts research activities for the characterization and modeling of these materials. To that end, an experimental platform has been developed for the implementation of different characterization methods and techniques, including multi-physical characterizations. Also, the team developps advanced numerical tools to model low and intermediate frequency electromagnetic phenomena. The team activity is recognized in these research fields, which is reflected in particular by numerous industrial collaborations.

 

Intitulé du doctorat

PhD in Applied Physics

Pays d'obtention du doctorat

France

Etablissement délivrant le doctorat

Université de Lille

Ecole doctorale

Sciences pour l'Ingénieur (ED 72)

Profil du candidat

The candidate will:

  • Have a diploma (Master’s degree, Engineer degree or equivalent) relevant to the Ph.D thesis topic, with a good academic record
  • Demonstrate his/her motivation to make a Ph.D
  • Have a good level of written and oral English

Be either fluent in French, or willing to learn

23/03/2021
Partager via
Postuler
Fermer

Vous avez déjà un compte ?

Nouvel utilisateur ?

Besoin d'informations ?

Vous souhaitez recevoir nos infolettres ?

item1
Ils nous font confiance