par Savarese, Matteo
Président du jury Coussement, Axel
Promoteur Parente, Alessandro
Co-Promoteur De Paepe, Ward
Publication Non publié, 2024-11-27
Thèse de doctorat
Résumé : Energy transition is unquestionably a priority in modern society. Political institutions are aligning with this goal, implementing policies to decrease reliance onfossil fuels and encourage the use of renewable energy sources. Consequently, theshare of renewables in the energy mix is poised to increase significantly within thenext few years. However, the intermittent nature of renewable sources necessitatesthe development of efficient storage techniques for the medium and long term.Smart Energy Carriers and carbon-free fuels offer a promising path to reducingCO2 emissions while ensuring reliable energy storage and supply. Among theseoptions, hydrogen and its carbon-free derivatives have garnered growing optimism.These molecules can store excess renewable energy in gas form, which can then betransported and converted into thermal energy via combustion, primarily producingharmless byproducts such as water and nitrogen.Nevertheless, combustion remains the dominant method of energy conversion, andthe different combustion properties of e-fuels present several technical challenges.A key concern is the formation of pollutants like NOx during high-temperature oxidation, which contributes to environmental issues such as acid rain. It is, therefore,imperative to efficiently adapt current combustion technologies and develop moreefficient ones to produce energy through combustion processes that are sustainable,efficient, and extendable to a wide range of fuels.Numerical modeling and simulations have become strategic tools for industry, helping to optimize the design of new technologies while reducing the need for costlyexperimental campaigns. Computational Fluid Dynamics (CFD) is a widely usedapproach, but its high computational demands make large-scale deployment challenging, especially for high-fidelity simulationsThe primary barrier to widespread CFD use is the high computational cost of high-fidelity modeling. Accurately representing chemical and physical processes requires solving complex partial differential equations on grids consisting of millionsof elements. Even for simple fuels, combustion involves intricate chemical reactions, necessitating detailed kinetic mechanisms with numerous species and reactions.Within the last few years, the development of more cost-effective strategies for numerical modeling has been an important topic for research. In particular, alternativetools, such as Chemical Reactor Networks, represent a strategic way to complementand limit high-fidelity simulations, providing kinetics simulations of realistic combustion systems in a convenient amount of time. Based on the assumption that acomplex flow field can be represented through a highly simplified network of connected chemical reactors, they allow for the solution of conservation equations onlyin a few elements, paving the way for the use of extremely detailed kinetics mechanisms in a highly reduced time frame.Reactor networks, however, are characterized by a lower degree of fidelity, and theirgrey-box nature necessitates expertise and careful attention in their design, whichmust be meticulously tailored for different test cases. Consequently, the correct design and the potential applications of these tools remain active areas of research.In this work, new methods and procedures for the efficient design of such physics-based reduced-order models are studied and analyzed. The development of moreefficient design techniques, including those based on machine learning algorithms,offers reliable models that can be used for parametric design, optimization, andcontrol in a straightforward and cost-effective manner. This approach complementsthe use of complex simulations and reduces their number, limiting the computationalburden and allowing for smarter combustor design processes.

Documents en relation

DI-fusion