Development of a Generative AI-Based Model for the Design of Multi-Material Permanent Magnets in Electrical Machines

IFP Energies nouvelles - Mobilité et Systèmes; Rueil-Malmaison; On-site; EUR 10,000 - 13,000; Full time

Job summary

IFP Energies nouvelles - Mobilité et Systèmes recherche un(e) stagiaire ingénieur(e) de fin d’études ou Master en traitement du signal et/ou image, intelligence artificielle et apprentissage profond pour développer un modèle génératif destiné à la conception d’aimants permanents multi-matériaux dans les machines électriques.

Le stage, d’une durée de 6 mois entre février et août 2027, se déroulera à Rueil-Malmaison et comprendra une revue d’état de l’art, la construction d’une base de données

Qualifications

• Étudiant(e) en dernière année d’ingénierie ou Master en traitement du signal et/ou image, IA ou domaine connexe.
• Maîtrise du Python souhaitable.

Responsibilities

• Réaliser une revue d’état de l’art sur l’électromagnétisme et l’IA appliquée à la morphologie.
• Construire une base de données d’entraînement à partir de données existantes.
• Entraîner divers algorithmes d’apprentissage profond sous contraintes physiques.
• Valider les configurations générées par des simulations de champs magnétiques.
• Évaluer et comparer les méthodes selon la fidélité, la faisabilité physique et les performances magnétiques.

Skills

Python

Deep learning

Machine learning

Signal processing

Image processing

Education

Final-year engineering student

Master’s student

Job description

Development of a Generative AI-Based Model for the Design of Multi-Material Permanent Magnets in Electrical Machines

The electrification of mobility and the automotive sector has gradually shifted from a topological mix of electric propulsion systems toward a predominance of permanent-magnet machines, which account for 95% of the market. These magnets, which are exclusively imported, contain heavy rare earth elements to reduce their susceptibility to demagnetization at high temperatures. Elements such as dysprosium and terbium are distributed homogeneously throughout the magnet, regardless of the physical constraints. Although these magnets offer excellent magnetic performance and high resistance to demagnetization, they nevertheless pose major challenges in terms of supply and cost. As part of efforts to reduce the use of rare earth elements, several approaches are already being implemented. These include segmentation, grain boundary diffusion (GBD), and the development of magnets containing few or no rare earth elements.

The most severe magnetic stresses in a magnet are localized in specific regions. This observation paves the way for a promising alternative: the use of multi-material magnets. This approach consists in concentrating the rare earth elements only in the regions subjected to the highest demagnetizing-field-induced stresses. This makes it possible to reduce the overall amount of rare earth elements while maintaining or improving magnetic performance. This internship is situated within this context. Its objective is to optimize the distribution of materials within permanent magnets. The proposed approach aims to address two major challenges: local magnetic stresses, such as demagnetization, and the reduction of rare earth element usage.

The internship will be conducted in several stages:

• Conducting a state-of-the-art review covering the electromagnetism of magnets, as well as deep-learning methods applied to morphological reconstruction and morphology generation conditioned on physical data;
• Building a training database from existing data related to conventional homogeneous magnets. For multi-material magnets, electromagnetic simulations will be carried out to characterize the interactions between the different materials and assess their influence on magnetic properties;
• Training various deep-learning algorithms subject to physical and magnetic constraints;
• Validating the generated configurations through simulations of local magnetic fields and using appropriate metrics;
• Evaluating and comparing the methods according to several criteria, including the fidelity of the generated configurations, their physical feasibility, and their magnetic performance.

Profile :

Final-year engineering student or Master’s student in signal and/or image processing, artificial intelligence (machine learning/deep learning), or a related field. Proficiency in Python is desirable.

Key words : Signal and image processing; artificial intelligence; generative AI; deep learning; digital twin; electrical machines; permanent magnets.

Durée et période du stage : 6 mois entre février et août 2027
Location : Rueil-Malmaison