Modélisation et techniques d'inversion avancées appliquées à l'imagerie hyperspectrale de flammes turbulentes // Forward Modeling and Advanced Inversion Techniques Applied to Hyperspectral Imaging of Turbulent Flames

Cette thèse s'inscrit dans le cadre du projet ANR « Radiative Analysis of Glass furnaces: Novel AppROaches for Computational Hyperspectral imaging » (Ragnaroch) et vise à développer des outils théoriques et numériques pour exploiter l'imagerie hyperspectrale dans l'analyse de flammes à haute température, y compris dans des configurations industrielles complexes. L'objectif est de permettre l'estimation non-intrusive de champs scalaires (température, concentrations de particules chimiques et suie) à partir de mesures spectrales. Contrairement aux diagnostics laser classiques souvent limités aux laboratoires, l'imagerie hyperspectrale offre une solution prometteuse, mais son application à des flammes turbulentes industrielles reste encore très peu explorée.

Le travail consistera à adapter des techniques issues de la télédétection atmosphérique (PCA, réseaux de neurones, ALD) aux besoins spécifiques de la combustion, en validant ces méthodes sur des flammes numériques simulées avec le logiciel de CFD ProLB, avant de les appliquer à des mesures expérimentales. Le doctorant participera également à des campagnes expérimentales et au développement d'algorithmes d'inversion pour reconstruire les champs de température et de composition à partir des données hyperspectrales.

************************************************************************************************************

This PhD is part of the ANR-funded “Radiative Analysis of Glass furnaces: Novel AppROaches for Computational Hyperspectral imaging” (Ragnaroch) project and aims to develop theoretical and numerical tools to enable the use of hyperspectral imaging for analyzing high-temperature flames, including in complex industrial environments. The goal is to retrieve non-intrusive, space- and time-resolved measurements of scalar fields (temperature, species concentrations, and soot) from spectral data. While traditional laser diagnostics are common in labs, their application in industrial settings is limited. Hyperspectral imaging offers a promising alternative, though its use in turbulent industrial flames remains underexplored.

The research will involve adapting methods from atmospheric sensing (e.g., Principal Component Analysis, Neural Networks, Augmented l-distribution) to combustion applications. These techniques will be validated on numerically simulated flames (using ProLB) before being applied to real experimental data. The PhD candidate will also take part in experimental campaigns (lab- and industry-scale) and contribute to the development of inversion algorithms to extract scalar field information from hyperspectral measurements using previously built forward radiation models.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Reliable space- and time-resolved measurements of temperature and species concentrations are essential in the development of future generations of clean and efficient combustion systems. Information about these scalar profiles is required: 1/ to understand the complex flame structures due to fluid flow / chemistry / radiative transfer / soot production / turbulence interactions, and, 2/ to improve / assess the physical models involved in numerical simulations, including chemical kinetics.

Emission spectroscopy techniques, including non-imaging and imaging (hyperspectral) measurements, are particularly attractive. Hyperspectral imaging devices record one flame emission spectrum for each pixel of a two-dimensional sensor producing a spectral image called hypercube. Their application in the field of combustion has benefited from the development of accurate high-resolution high-temperature spectroscopic databases to interpret the data. In practice, line-of-sight flame emission spectra are recorded. Scalar (temperature and species / soot concentration) profiles can then be inferred from these spectra by application of inversion techniques combined with accurate forward models. Application of the method to each pixel of the hyperspectral camera allows, in principle, reconstructing the entire combustion field. Hyperspectral devices have been increasingly used as optical diagnostics to probe laminar flames. It was shown that they can be used for turbulent flames too. However, turbulence / radiation interactions complicate significantly the inversion process. Gore and co-workers have shown that the first- and second-statistical moments of temperature, species mole fractions and soot volume fraction fluctuations can be estimated by combining hyperspectral measurements and tomography techniques in statistically-steady axi-symmetric turbulent flames. Nevertheless, the amount of scientific papers dedicated to the application of hyperspectral imaging to turbulent fields remains marginal at the present time. Moreover, industrial combustion chambers show complex configurations and geometries that usually depart significantly from the idealized axi-symmetric flame.
This PhD work is part of the project “Radiative Analysis of Glass furnaces: Novel AppROaches for Computational Hyperspectral imaging” (Ragnaroch) funded by the French National Research Agency (ANR) https://anr.fr/Projet-ANR-24-CE51-2798.

The aim of the PhD work is to contribute to the field of radiative analysis in high temperature configurations. Its objective is to develop and fully validate theoretical and numerical tools to allow the use of hyperspectral imaging in real flame scenarios, including actual industrial configurations.
For this purpose, the PhD work will first consist of adapting theoretical and numerical tools used in atmospheric sensing studies (Principal Component Analysis (PCA), Neural Networks, Augmented l-distribution (ALD)) to combustion applications. This will require understanding the methods before to suggest the modifications needed to apply the techniques to flame configurations. Model parameters will be constructed and fully validated on numerical flames calculated using ProLB as part of the project, before to be applied in real applications. The PhD candidate will then participate to the experiments on real flames (laboratory scale and industrial configuration) and also to the development of inversion methods to retrieve scalar fields in flames (temperature, species concentrations) from experimental radiative images. The inversion process will use the previously developed forward models of flame radiation.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Début de la thèse : 01/10/2025

Autre financement public

ANR Financement d'Agences de financement de la recherche

Université de Lorraine

608 SIMPPÉ - SCIENCES ET INGENIERIES DES MOLECULES, DES PRODUITS, DES PROCEDES ET DE L'ÉNERGIE

Ecole d'ingénieur ou master de physique. De solides compétences en méthodes numériques, de sérieuses connaissances en mathématiques et en statistiques et/ou une expérience préliminaire significative en transfert radiatif dans les gaz à haute température seront appréciées. Pour toute thèse proposée au sein de l'Ecole Doctorale, le futur doctorant devra bien être titulaire d'un master (diplôme de master/d'ingénieur français ou étranger, …) justifiant d'un parcours remarquable. Dans tous les cas (diplôme de master ou d'ingénieur français ou étranger, …) le dossier doit comporter : • le CV du candidat et lettre de motivation • les notes obtenues au diplôme conférant le grade de master, mention 'Assez Bien' requise au minimum et copie du diplôme s'il est disponible • des lettres de recommandations émanant du Responsable de la filière de formation et du tuteur de stage de fin d'études • des éléments tangibles sur l'initiation à la recherche (mémoire de recherche, publication, ...). Le dossier complet de candidature doit être envoyé à la direction de thèse par les adresses messageries des directeurs de thèses : Mr Parent : gilles.parent@univ-lorraine.fr Mr André : frederic.andre@univ-lille.fr
Engineering school or equivalent. Solid skills in numerical methods, serious background in mathematics and statistics and/or a significant preliminary experience in radiative transfer in high temperature gases will be appreciated All applicants to the Doctoral School SIMPPÉ must have successfully completed a Master degree or its equivalent with a grade comparable to or better than the French grade AB (corresponding roughly to the upper half of a graduating class). In all cases (French or foreign Master degree, engineering degree, etc.) the counsel of the doctoral school will examine the candidate's dossier, which must include: • CV and letter of motivation • the grades obtained for the Master (or equivalent) degree and a copy of the diploma if it is available • 2 letters of recommendation, preferably from the director of the Master program and the supervisor of the candidate's research project • written material (publications, Master thesis or report, etc.) related to the candidate's research project. The complete application file must be sent to the thesis supervisors by email : Mr Parent : gilles.parent@univ-lorraine.fr Mr André : frederic.andre@univ-lille.fr