Stage Physically-Informed Neural Networks applied to high-temperature industrial processes H/F

TYPE DE CONTRAT : Stage NIVEAU DE FORMATION : BAC +4 / BAC +5 SPÉCIALITÉ(S) : Recherche & Développement PAYS / RÉGION : France / Ile-de-France DÉPARTEMENT : Yvelines (78) VILLE : Chatou EDF est labellisé Happy Trainees DESCRIPTION DE L'OFFRE INTERNSHIP: _Physically-Informed Neural Networks for residual stress distribution assessment applied to high-temperature industrial manufacturing processes._ at EDF R&D (Paris Chatou) and Mines ParisTech (Centre des Matériaux, Evry) Keyword: Computational Engineering, Applied Mathematics, Residual Stress, Welding, Finite Element Analysis, Data Assimilation, Machine Learning, Deep Learning, Neural Network, Industry 4.0 EDF is currently investigating the capabilities of emerging data-driven solutions to assess residual stress distribution induced by high-temperature manufacturing processes. Industrial context: Residual stresses are generated during manufacturing due to creation of permanent strain during material processing. For instance, during a welding operation, high thermal gradients spread through the welded component. The difference in cooling rates experienced in different parts of the component results in localized variations in thermal expansion and contraction. As a result, these phenomena develop non-compatible strains leading to-residual stresses. Residual stresses can negatively affect structural integrity. For example, thick-walled structures in the as-welded condition are more prone to brittle fracture than a structure that has been stress-relieved. The undesired stresses may also influence the fatigue performance. Hence, assessing the level of residual stress at the end of manufacturing can reduce design engineering justifications as their level and state may affect the fit for service of metal parts in pipes and pressure vessel. PROFIL SOUHAITÉ CANDIDATE’S PROFILE: Core skills: The candidate should have strong skills in applied Math and more specifically in Statistics, Machine Learning and Deep Learning. Proven experience in supervised and unsupervised learning methods will be asked. Besides, the candidate should be familiar with classical optimization algorithms. Fluency in Python programming is also a plus. Additional skills: The candidate should have knowledge in numerical simulation for Mechanics. Learning methods developed during this internship will be applied to computational welding mechanics (solid mechanics for welding applications). Hence, the candidate should show a strong interest in this field and more generally in the domain of manufacturing. Transversal skills: Internship success depends on the candidate’s scientific curiosity, his/her strong interest in digital industry as well as his/her ability to easily work in an interdisciplinary team. Langage: French or English.