Ph.D. Offer on the Development of New Strategies for Optimizing Thermodynamic Predictive Models for Materials Design (M/F)

Descrizione

Ph.D. Offer on the Development of New Strategies for Optimizing Thermodynamic Predictive Models for Materials Design (M/F) Sintesi in italiano (traduzione automatica): L'organizzazione offre un'opportunità di dottorato per lo sviluppo di nuove strategie per ottimizzare modelli predittivi termodinamici per la progettazione di materiali. Il candidato sarà supervisionato da esperti di IMN e LS2N, situati nel campus scientifico dell'Università di Nantes. Le mansioni principali includono la creazione di strumenti digitali per migliorare l'affidabilità dei modelli termodinamici, utilizzando dati sperimentali forniti da utenti industriali e accademici. È richiesta una laurea in ingegneria dei materiali o un campo correlato. Il lavoro si svolgerà principalmente presso LS2N, con viaggi regolari al CEA per collaborazioni industriali. Il candidato dovrà affrontare sfide legate alla complessità dei modelli attuali e sviluppare approcci innovativi per l'ottimizzazione dei processi. The Ph.D. student will be supervised by Isabelle Braems-Abbaspour (ID2M, IMN), Alexandre Goldsztejn, and Thomas Gouhier (LS2N), as well as Christine Guéneau (CEA). The first three have a long-standing partnership that has previously led to the development of software for plotting guaranteed phase diagrams. The two host laboratories (IMN and LS2N) are located on the University of Nantes science campus (Lombarderie), a 5-minute walk apart. Day-to-day work will primarily take place at LS2N. Regular trips to the CEA are planned. The IMN is the Jean Rouxel Institute of Materials in Nantes (IMN, UMR 6502, http://www.cnrs-imn.fr ). The IMN is a joint research center of the CNRS and the University of Nantes, with more than 200 staff members, including over 120 permanent staff (professors, CNRS researchers, engineers) and approximately 80 doctoral and postdoctoral students. LS2N is the Nantes Laboratory for Science and Digital Technology (LS2N, UMR 6004, http://www.ls2n.fr ). LS2N is also a joint research center of the CNRS and the University of Nantes, with a staff of approximately 500 people, including nearly 200 permanent staff (professors, CNRS researchers, engineers) and 200 doctoral and postdoctoral students. A trip outside the region, specifically to the industrial partner (CEA), will take place during the course of the thesis. Predictive thermodynamic models developed using the CALPHAD method [LFS2007] play a strategic role in the development of new materials by making it possible to anticipate the behavior of systems before conducting costly and time-consuming experiments in order to: a) be extrapolated to describe the thermodynamic landscapes of more complex materials, b) anticipate extreme behaviors or hostile environments (high temperatures, reactive atmospheres, etc.) and prevent the formation of harmful secondary phases, c) serve to link energy properties to in-service properties using an integrative approach. The quality and reliability of these models determine whether it will be possible to design the materials of the future in a more rational, rapid, and innovative (and therefore competitive) manner. However, the community is questioning all current models due to their complexity and lack of precision, despite the high cost of the software used to create them. As a result, materials specialists still rely on their own expertise to conduct a preliminary critical review of the various data points, uncertainties, model selection, number of parameters to adjust, and weightings to assign, all while having to initialize the model parameters to begin the adjustment process. In this era of automation and large-scale screening, it is crucial to replace these delicate and time-consuming steps with a reliable, autonomous, and fast digital tool that can then be integrated into exploration workflows. We propose to develop a reliable tool for optimizing thermodynamic models (free enthalpies) based solely on experimental data provided by industrial and academic users. The goal is to provide them with reliable, carefully validated, proprietary, and modular models that are as simple as possible, so that they can develop their strategies with complete peace of mind—such as evaluating new process conditions or optimizing manufacturing ranges—without relying on overly uncertain extrapolations. This will be achieved by defining new models (variables, constraints, objective functions) for parameter estimation tailored to this thermodynamic framework, bridging the gap between the specific experimental data available in this context and modern nonlinear programming frameworks (such as linear matrix inequalities [NP2023]). The models will also be adapted for optimization using modern optimization approaches, such as cubic-regularized Newton [M2023], which require, for example, the calculation or approximation of Hessian matrices. LFS2007] Hans Lukas, Suzana G. Fries et Bo Sundman, Computational thermodynamics: the CALPHAD method, Cambridge Un Annuncio in ing