ED Mathématiques et Informatique
Optimisation of GANs and Application to Affective Computing
by Zakariae BELMEKKI (ESTIA-Recherche)
The defense will take place at 10h00 - Amphi C Bâtiment A29, Pessac Talence Gradignan, 33400, Talence
in front of the jury composed of
- Stephane AYACHE - Professeur - Université Aix Marseille, amU - Rapporteur
- Elisabeth ANDREE - Professeure - Université Augsburg - Rapporteur
- Stuart BARNES - Maître de conférences - Université de Cranfield - Examinateur
- Thierry COLIN - Professeur - Sophia Genetics - Examinateur
Generative Adversarial Networks (GANs) are one of the most powerful classes of generative models in deep learning, capable of synthesizing highly realistic data across diverse domains. However, their training remains computationally demanding and unstable. In this work, we focus on CNN-based GANs, which are prone to information redundancy within their convolutional layers. This thesis addresses these challenges by introducing a novel measure of feature redundancy, Convolutional Similarity, and by demonstrating its theoretical foundations, optimized implementation, and empirical impact on generative and discriminative models. The first contribution of this work is a rigorous theoretical derivation and analysis of redundancy in CNNs establishing that traditional orthogonality constraints and kernel decorrelation losses are insufficient for reducing redundancy in CNNs. Contrary to the prevailing assumption in the literature, minimizing linear similarity between kernels does not necessarily lead to lower similarity between feature maps. Building on this insight, Convolutional Similarity is proposed as a loss function that measures the overlap of convolutional responses directly in the spatial domain, independently of the data, thereby enabling redundancy reduction in a principled, model-agnostic way. We also present a highly optimized CUDA implementation that leverages shared memory, coalesced access, and warp-level reductions to achieve substantial speedups compared to its PyTorch prototype, with greatly reduced memory requirements. Extensive experiments were conducted using the ResNet18 architecture on benchmark datasets including CIFAR10, CIFAR100, and SVHN. The results demonstrate that using Convolutional Similarity as a weight initialization method accelerates convergence and yields higher test accuracies. When used as a regularization term during training, Convolutional Similarity improves generalization by promoting filter diversity even when the training loss remains higher. The final part of the thesis applies these findings to Affective Computing, a field concerned with enabling machines to interpret and simulate human emotions. A custom GAN model is designed to generate sequences of complex facial expressions represented through the Facial Action Coding System (FACS). The model is evaluated both quantitatively, using the proposed Action Unit Fréchet Inception Distance (AUFID), and qualitatively through human assessments. The results show that the GAN successfully synthesizes temporally coherent and anatomically consistent expressions even with scarce, imbalanced data, marking a novel contribution to affective facial behavior generation. Overall, this work establishes Convolutional Similarity as a theoretically grounded, computationally efficient, and empirically validated method for reducing redundancy in CNNs, thereby optimizing GAN performance and stability. Beyond GANs, the method provides a generalizable framework for enhancing feature diversity in convolutional architectures, paving the way toward more efficient, interpretable, and resource-conscious deep learning models.
Consequential and territorial modelling of the carbon footprint of digital devices and their use of the mobile network for agriculture
by Pierre LA ROCCA (LaBRI - Laboratoire Bordelais de Recherche en Informatique)
The defense will take place at 9h30 - Amphi LaBRI LaBRI 351, cours de la Libération F-33405 Talence cedex
in front of the jury composed of
- Aurélie BUGEAU - Professeure des universités - Université de Bordeaux - Directeur de these
- David BOL - Professeur - Université Catholique de Louvain - Rapporteur
- Peter STURM - Directeur de recherche - Centre Inria de l'Université Grenoble Alpes - Rapporteur
- Vincent PADOIS - Directeur de recherche - Centre Inria de l'Université de Bordeaux - Examinateur
- Éléonore LOISEAU - Ingénieure de recherche - INRAE - Examinateur
- Kevin MARQUET - Maître de conférences - Insa Lyon - Examinateur
The agricultural sector is undergoing a triple crisis, not only social but also economical and ecological, which requires a change in the production model. For political institutions, the solution lies in a twin transition, using information and communication technologies (ICT) as a lever to make the sector more sustainable. However, there are many uncertainties regarding the consequences of such a strategy. In particular, from an environmental point of view, few studies assess the effects of large-scale deployment of digital agriculture technologies. Yet the harmful effects of ICT on the environment are well known, and an increase in digital equipment and infrastructure for agriculture could be a threat to the sector's sustainability. This thesis introduces methods for assessing the environmental impacts of large-scale deployment of digital agricultural equipment and its dependence on infrastructure. The general framework adopts a prospective scenario-based approach and parametric and consequential modeling. A first contribution concerns the estimation of the impacts of digital technologies deployed on farms in a given agricultural context. This context is defined as a type of production and a distribution of farms of varying sizes. This method is applied to two case studies in mainland France: the identification and monitoring of oestrus in dairy cattle, and robots for mechanical weeding and automated sowing in large-scale cereal crops. The results reveal heterogeneous impacts, depending on the number of pieces of equipment deployed, their mass and their technological complexity. Scenarios involving the most advanced technologies and using a larger mass of equipment tend to have the highest impacts. A second contribution extends the case study dedicated to weeding robots and focuses on the impacts associated with their use of the mobile network. This method simulates the deployment of robots in a given area, based on a set of agricultural parcels and existing mobile network infrastructure, with each site supporting a maximum data volume. The scenarios considered vary according to the design of the robots, and the volumes of data sent by the robots differ greatly from one design to another. Based on a traffic simulation of data and its impact on the mobile network, we estimate the additional impact on existing infrastructure in terms of embodied carbon footprint of required network equipment, and that associated with their energy consumption. We also estimate the proportion of agricultural land that can be managed by the deployed robots. The results show that for intensive use of the network, the capacity of the current network greatly limits the number of robots that can be deployed, even when the capacity of existing sites is maximised. Furthermore, for the most intensive scenarios, the additional power consumption of the network is of the same order of magnitude than that of the robots themselves. Thus, for these scenarios, only a small fraction of the parcels can be managed without adding sites, and increasing this fraction would imply a significant increase in the overall footprint of the mobile network in question. This manuscript provides an initial overview of the consequential carbon footprint associated with the deployment of digital equipment on farms and their use of the mobile network.
ED Sciences Chimiques
Cellulose nanocrystals-stabilized polyhydroxyurethane and polyacrylate latexes: design, functionalization, and performance
by Hsin-Chen CHEN (Laboratoire de Chimie des Polymères Organiques)
The defense will take place at 10h00 - FD5 Roslagstullsbacken 21, AlbaNova, 11421, Stockholm
in front of the jury composed of
- Sylvain CAILLOL - Directeur de recherche - University of Montpellier - Rapporteur
- Valérie RAVAINE - Full professor - Université de Bordeaux - Examinateur
- Monica EK - Full professor - KTH Royal Institute of Technology - Examinateur
- Liyang LIU - Assistant professor - Chalmers University of Technology - Examinateur
- Christine GERARDIN - Full professor - Université de Lorraine - Rapporteur
Cellulose nanocrystals (CNCs) have emerged as highly effective solid stabilizers for Pickering emulsions and latexes, offering a sustainable alternative to conventional surfactants. Their amphiphilic nature allows partial wettability at the oil-water interface, creating robust colloidal stabilization in fully waterborne systems. Because CNCs originate from renewable lignocellulosic biomass and can be surface-modified through abundant hydroxyl groups, they provide a versatile platform for designing green, bio-based polymer composites with tailored interfacial and mechanical properties. This thesis develops and investigates CNC-stabilized waterborne non-isocyanate polyurethane (WNIPU) and polyacrylate latexes with the goal of achieving high colloidal stability and excellent material performances. The work focuses on optimizing latex synthesis, engineering CNC surface functionalities, elucidating CNCs-polymer matrix interactions, and assessing the structural and mechanical properties of the resulting latexes and composites. In the first part, pristine CNCs were utilized to prepare WNIPUs through CNC-stabilized suspension polymerization, eliminating hazardous isocyanates, volatile organic compounds (VOCs), and external surfactants. In paper I, CNC-mediated Pickering stabilization of WNIPU latexes was successfully achieved for the first time, with systematic optimization of CNCs concentration, monomer composition and reaction parameters. The obtained WNIPUs also exhibited florescence, indicating potential for sensing or anti-counterfeiting applications. Paper II demonstrated the dual role of CNCs as both stabilizer and reinforcing agent for the WNIPU latex. The incorporation of 17 wt.% CNCs increased probe tack adhesion strength by 680% and lap-shear strength by 340%, enabled by their uniform dispersion within the WNIPU matrix. The second part of the thesis focused on functionalized CNCs as advanced stabilizers for waterborne latexes. In Paper III, CNCs were covalently modified with octylamine (oCNCs) to increase surface hydrophobicity. The oCNCs-stabilized WNIPU showed improved stabilization efficiency and smaller sizes in monomer emulsion droplets. The resulting WNIPU/oCNC composites showed up to 34% and 57% increases in probe tack adhesion strength and work of adhesion, respectively, compared to pristine CNC-stabilized WINPU. Paper IV introduced a non-covalent surface modification using charge coupling method, which rendered CNCs sufficiently hydrophobic to stabilize polyacrylate latexes with solid contents up to 20 wt.%. The resulting polyacrylate/CNC composites demonstrated coating performance comparable to surfactant-stabilized references. Overall, this thesis establishes CNC-stabilized WNIPU latexes as a new class of Pickering polymer systems and extends the concept to polyacrylate matrices. It elucidates how CNC surface chemistry governs dispersion stability, interfacial stabilization, and composite performance These findings provide a basis for designing scalable, surfactant-free, and sustainable waterborne polymer systems, advancing the use of functionalized nanocellulose in environmentally friendly coatings and adhesives.
ED Entreprise Economie Société
understanding the dynamics of business angels activity: an institutional perspective
by Muzna RAHMAN (IRGO - Institut de Recherche en Gestion des Organisations)
The defense will take place at 10h00 - C002 35 avenue Abadie - 33072 Bordeaux cedex
in front of the jury composed of
- Pascal GRANDIN - Professeur - University of Lille - Examinateur
- Jessica FOUILLOUX - Professeure - University of Bordeaux - Examinateur
- Céline BARRÉDY - Professeure - Université Paris Nanterre - Rapporteur
- Antoine RENUCCI - Professeur - University of Paris 1 Panthéon-Sorbonne - Rapporteur
- JERRY COAKLEY - Professeur - University of Essex - Examinateur
Business Angels (BAs) have a unique capacity to fill the funding gaps left between founders' money, family, friends, foolhardy (3Fs), and venture capital for innovative ventures. They also offer new ventures with additional managerial skills and knowledge. Recognising the crucial role of BAs in the entrepreneurial ecosystem, nations are striving to develop BA markets in order to support startups that significantly contribute to economic growth and success. The key factors responsible for BA investment development have not yet been fully explored. This thesis employed both qualitative and quantitative approaches to build an understanding of BAs' contributions in startups, as well as to explore the key determinants of BA investments across countries and borders. A meta-synthesis review reveals that BA perform managerial functions and roles that nurture investee firms. Empirical results obtained using factor analysis and structural equation modelling based on cross-sectional BA investment data indicate that the institutional factor is significant for the development of BA domestic markets. This finding is reinforced by panel data on BAs' investments across European countries, where the Panel Corrected Standard Error (PCSE) regression reaffirms the central role of institutional quality while also highlighting that cultural values and demand-creating factors cannot be negated. Finally, at the country-dyad level, analysis of cross-border BA deals in developed and developing countries reveals that institutional and cultural distances affect BAs' cross-border investment outflows; however, the significance of each factor is context-dependent. Collectively, findings suggest that countries need to enhance their institutional arrangements and foster institutional harmony to develop BA investment activities, as well as their non-financial contributions. These research findings enhance understanding and contribute to the development of theory, practice, and policy, providing essential implications.
ED Sciences de la Vie et de la Santé
Challenges and strategies for the production, solubilisation and purification ofmembrane proteins : EmrA and EmrB study of the MFS EmrAB-TolC efflux pump of the Escherichia coli bacterium
by Marine CROUVIZIER (Institut de Chimie & de Biologie des Membranes & des Nano-objets)
The defense will take place at 14h00 - Amphi F ENSEIRB-MATMECA 1 avenue du Dr Albert Schweitzer 33402 Talence
in front of the jury composed of
- Laetitia DAURY - Maîtresse de conférences - Université de Bordeaux - Directeur de these
- Muriel MASI - Maîtresse de conférences - Aix-Marseille Université - Rapporteur
- Martin PICARD - Directeur de recherche - Université Paris Cité - Rapporteur
- Stéphen MANON - Directeur de recherche - Université de Bordeaux - Examinateur
The discovery of penicillin in 1928 by Alexander Fleming marked the beginning of the golden age of antibiotics, which have since saved millions of lives. However, the emergence of bacterial resistance quickly compromised their effectiveness, making antimicrobial resistance a major public health and economic concern. Throughout evolution, bacteria have developed various mechanisms to survive under this selective pressure. One of them is active efflux, which expels a wide range of substrates, including antibiotics, into the extracellular medium before they can exert their effect. By lowering the intracellular concentration of antibiotics, this mechanism compromises their activity and contributes to the phenomenon of multidrug resistance. In Gram negative bacteria, tripartite efflux pumps span the entire cell envelope. They are composed of an energy-dependant inner membrane transporter, an outer membrane channel protein, and a periplasmic adaptor protein that bridges the transporter to the channel protein. Investigating this mechanism is crucial for a better understanding of its function and for the development of strategies to block it, thereby restoring bacterial susceptibility to antibiotics. This PhD project focuses on the EmrAB-TolC tripartite efflux pump of Escherichia coli, which belongs to the Major Facilitator Superfamily (MFS). This system is involved in resistance to nalidixic acid and, more recently, to nitroxoline. EmrB is the inner membrane transporter that uses the proton gradient to drive substrate export, TolC is the outer membrane channel protein, and EmrA is the periplasmic adaptor protein. Unlike Resistance-Nodulation-Cell-Division (RND) pumps, MFS pumps remain poorly characterised, with little structural data available on EmrA and EmrB. The first objective of this work was to investigate EmrA. Given the limited information on this protein, a cell-free expression strategy was employed. The addition of the DDM detergent to the reaction mixture enabled solubilisation of the produced EmrA protein. The purified protein exhibited α-helical secondary structures, as determined by circular dichroism. The second objective concerned EmrB, with the optimisation of an overexpression and purification protocol to obtain sufficient quantities of the protein in its native state for structural analysis by cryo-electron microscopy. Since initial extraction with the DDM detergent failed to yield stable and sufficient protein, different polymer-based strategies were tested, including four SMAs, four DIBMAs and CyclApol. SMA and CyclApol proved effective in extracting EmrB. CycApol was ultimately chosen for EmrB purification, firstly because it preserves the native membrane environment around EmrB and secondly because it represents a recent and innovative polymer that circumvents limitations encountered with other types. Using this approach, EmrB was successfully extracted and purified in sufficient quantity and in its native state. Initial cryo-electron microscopy micrographs enabled the generation of 2D classes, which were used to reconstruct a first ab initio model of EmrB in CylApol. These results pave the way for further structural characterisation of the transporter and, in the long term, for a better understanding of transport mechanisms in the MFS family, with the ultimate goal of combating antimicrobial resistance.
Methodological developments for non-invasive intercostal treatment using high-intensity focused ultrasound guided by magnetic resonance imaging
by Fanny DABRIN (Centre de Résonnance Magnétique des Systèmes Biologiques)
The defense will take place at 13h30 - Amphithéâtre Bailey IHU Liryc, Avenue du Haut Lévêque 33604 Pessac cedex
in front of the jury composed of
- Bruno QUESSON - Directeur de recherche - Université de Bordeaux - Directeur de these
- Jean-François AUBRY - Directeur de recherche - Laboratoire physique pour la médécine - Rapporteur
- David MELODELIMA - Directeur de recherche - Labtau - Rapporteur
- Monique BERNARD - Directrice de recherche - Centre de Résonance Magnétique Biologique et Médicale - Examinateur
High-intensity focused ultrasound (HIFU) has the ability to deliver ultrasound energy locally and non-invasively into biological tissue. It is possible to exploit the mechanical and/or thermal effects depending on the ultrasound parameters. Magnetic resonance imaging (MRI) allows treatment planning and real-time monitoring. However, the rib cage interferes with the propagation of ultrasound and can lead to significant heating of the bones, which complicates the application of these technologies for organs behind the thoracic cage. In this work, an experimental MRI-compatible HIFU device allowing intercostal access was used and characterized. To compensate for the loss of focus in a heterogeneous medium, a MRI-guided adaptive focusing technique was implemented and validated. This technique has been greatly accelerated compared to the state of the art and allows for the correction of focusing aberrations in several locations based on a single calibration measurement performed in a few minutes. These technological developments are a first step towards the development of non-invasive intercostal treatments with quantitative real-time MRI monitoring.
Development of a gingival graft using 3D bioprinting
by Camille DECHELETTE (Bioingénierie tissulaire)
The defense will take place at 9h00 - Amphithéâtre BBS Bâtiment Bordeaux Biologie Santé 2 Rue Dr Hoffmann Martinot 33000 Bordeaux
in front of the jury composed of
- Adrien NAVEAU - Professeur des universités - praticien hospitalier - Université de Bordeaux - Directeur de these
- Tchilalo BOUKPESSI - Professeure des universités - praticienne hospitalière - Université Paris Cité - Rapporteur
- Caroline GORIN - Maîtresse de conférences - praticienne hospitalière - Université Paris Cité - Examinateur
- Maxime DUCRET - Professeur des universités - praticien hospitalier - Université Claude Bernard Lyon I - Rapporteur
- Jérôme SOHIER - Directeur de recherche - Université Claude Bernard Lyon I - Examinateur
The aim of this PhD project was to develop a three-dimensional bioprinted gingival graft using a co-culture of human cells, based on a thorough characterization of the native gingival tissue mechanics and the optimization of a suitable bioink for tissue engineering applications. In the first part, the viscoelastic properties of human and porcine palatal mucosa were assessed under physiological conditions using shear rheology. Histological analyses were performed to correlate tissue composition with the mechanical behavior. Comparable shear moduli were obtained for both species, confirming the relevance of porcine mucosa as a model for human gingiva. These results provided essential reference values for the design of biomimetic matrices in gingival tissue engineering. In the second part, a cellularized bioink was developed from methacrylated collagen (CollMA) and methacrylated hyaluronic acid (HAMA). Several formulations were compared regarding their mechanical properties, temporal stability, and ability to support cell growth. The combination of 0.5% CollMA and 2.5% HAMA showed the best compromise between stiffness, stability, and biocompatibility. Hydrogels were seeded with human gingival fibroblasts (hGF) and endothelial cells (HUVEC) to promote the formation of pre-vascular networks. The addition of pro-angiogenic peptides (SVV and QK) was tested to enhance cell viability and vascular network organization. The SVV peptide alone improved cell metabolic activity and promoted the formation of stable vascular-like structures within the matrix. Finally, the crosstalk between hGF and HUVEC was investigated in both two-dimensional co-culture and within the optimized bioprinted construct. The construct was implanted subcutaneously in mice for 28 days to evaluate its biocompatibility and remodeling potential. Histological analysis revealed progressive integration of the graft, matrix deposition, and the presence of vascular structures, indicating physiological remodeling. Altogether, this work established the mechanical, biological, and technological foundations required for the fabrication of a bioprinted gingival substitute. The proposed model represents an innovative, reproducible, and clinically relevant strategy for oral soft tissue reconstruction.
Study of the interrelationships between TOR, CWI, calcineurin and HOG signalling pathways and their roles in resistance to lipopeptide antifungals (echinocandins) in the yeast pathogen Candida
by Maxime LEFRANC (Microbiologie fondamentale et Pathogénicité)
The defense will take place at 14h00 - Salle de conférence Bâtiment CARF : Centre d'Appui à la Recherche et de Formation 146 rue Léo Saignat 33000 Bordeaux
in front of the jury composed of
- Thierry NOEL - Professeur des universités - Université de Bordeaux - Directeur de these
- Virginie BELLET - Maîtresse de conférences - Université de Montpellier - Rapporteur
- Thierry FONTAINE - Directeur de recherche - Institut Pasteur - Rapporteur
- Gwenaël RUPRICH-ROBERT - Maîtresse de conférences - Université Paris Cité - Examinateur
- Florent MORIO - Professeur des universités - praticien hospitalier - Université de Nantes - Examinateur
Echinocandins (caspofungin, micafungin, anidulafungin) are the first class of antifungals to target the fungal cell wall. These non-competitive inhibitors of β-(1,3)-D-glucan synthase alter the integrity of the cell wall, leading to osmotic lysis of fungal cells. Their excellent tolerance profile, low number of drug interactions and marked fungicidal activity against Candida spp. have established them as the first-line treatment for invasive candidiasis in most patients. Their use has therefore increased significantly in recent years, and resistance associated with the acquisition of point mutations in the ‘hot-spot' regions of the FKS genes, which code for β-1,3-D-glucan synthase, has been reported. These mutations decrease the affinity of echinocandins for their target and also cause a change in the parietal composition, with a decrease in β-1,3-D-glucan compensated by an increase in chitin. Chitin synthesis is regulated directly and indirectly by several signalling pathways, including the CWI (Cell Wall Integrity), HOG (High Osmolarity Glycerol), calcineurin and TOR (Target Of Rapamycin) pathways. The aim of this study was to explore the involvement of these signalling pathways in caspofungin resistance in our study model, Candida lusitaniae. The main FKS1 resistance mutations, as well as an FKS1 mutation described for the first time by our laboratory, identified in clinical isolates of Candida spp., were introduced into the FKS1 gene of C. lusitaniae in an isogenic background. These mutations not only confer in vitro resistance to all echinocandins, but also lead to constitutive activation of the CWI pathway. This activation results in overexpression of the MKC1 gene, which encodes the main kinase of the CWI pathway, Mkc1p kinase, found in greater abundance in phosphorylated form in mutant strains even in the absence of caspofungin. Deletion of MKC1 partially or completely restores susceptibility to caspofungin in FKS1 mutant strains, demonstrating for the first time the central role of the CWI pathway in echinocandin resistance. Similarly, deletion of CNB, which encodes a regulatory subunit of calcineurin, increases susceptibility to caspofungin in FKS1 mutant strains. Conversely, deletion of HOG1, which encodes a kinase in the HOG pathway, has no effect on susceptibility to caspofungin. Pharmacological inhibition of the calcineurin and TOR pathways (by tacrolimus and sirolimus, respectively) also restores susceptibility to caspofungin, with a synergistic effect when these inhibitors are combined with caspofungin, even in strains resistant to this antifungal agent. These synergies, which are largely dependent on the Fpr1p protein, a cellular receptor common to tacrolimus and sirolimus, confirm the involvement of the calcineurin and TOR pathways in caspofungin resistance. In summary, we demonstrate in this study that the FKS1 mutations studied phenotypically mimic the effects of echinocandin treatment: decreased β-1,3-D-glucan synthesis, parietal stress and constitutive activation of rescue signalling pathways. These mutations therefore appear to be both affinity mutations, altering drug-enzyme binding, and loss-of-function mutations, compromising the intrinsic catalytic activity of β-1,3-D-glucan synthase even in the absence of echinocandin. The inhibition of the CWI, calcineurin or TOR pathways, which potentiates the antifungal effect of caspofungin even in the presence of resistance mutations, suggests a complex regulation of echinocandin resistance in yeasts of the genus Candida.
ED Sciences Physiques et de l'Ingénieur
Identification and decision by coupling deep learning methods and knowledge models for predictive maintenance
by Mathias LECROART (Laboratoire de l'Intégration du Matériau au Système)
The defense will take place at 14h00 - Amphi J.-P. DOM IMS - Laboratoire de l'Intégration du Matériau au Système 351 Cours de la Libération, 33405 Talence Cedex, France
in front of the jury composed of
- Audrey GIREMUS - Professeure des universités - Université de Bordeaux - Directeur de these
- Marie CHABERT - Professeure des universités - Toulouse INP - Rapporteur
- Antoine PICOT - Maître de conférences - Toulouse INP - Rapporteur
- Gaël PAGES - Chargé de recherche - ISAE-SUPAERO - Examinateur
- Emmanuelle ABISSET-CHAVANNE - Professeure des universités - Arts et Métiers ParisTech - Examinateur
- Christophe FARGES - Maître de conférences - Université de Bordeaux - Examinateur
This thesis addresses the estimation of the Remaining Useful Life (RUL) of a proton exchange membrane fuel cell by combining physical modeling with artificial intelligence. The objective is to develop reliable and robust predictive maintenance methods. A nonlinear degradation model has been developed to represent the evolution of the fuel cell's performance over time. The proposed approach is hybrid: it integrates a robust Kalman filter with recurrent neural networks to estimate states whose dynamics are not explicitly known. This combination leverages both physical knowledge and the iterative structure of the model, while exploiting the predictive power of sequential learning methods. Particular attention is given to numerical stability issues and computational costs typically associated with classical methods. Different hybridization strategies, loose and tight, have been investigated in order to balance accuracy, robustness, and efficiency. The use of robust functions and semi-supervised learning compensates for the lack of direct access to internal states. Experimental results are obtained from a dataset generated through accelerated aging tests. The contribution offers a trade-off between RUL prediction accuracy and computational cost. Comparisons with alternative approaches, purely model-based, purely data-driven, or loosely coupled, confirm the superiority of the proposed method. In particular, it improves prediction accuracy while mitigating training instabilities.
Nonlinear response of local defect resonance excited by Lamb waves as a method for delamination assessment in composites
by Biaou Jean-Baptiste KOUCHORO (I2M - Institut de Mécanique et d'Ingénierie de Bordeaux)
The defense will take place at 14h00 - Amphi 1 351 cours de la Libération CS 10004 Bâtiment A9 33405 Talence Cedex,
in front of the jury composed of
- Anissa MEZIANE - Professeure des universités - Université de Bordeaux - CoDirecteur de these
- Philippe MICHEAU - Professeur - Université de Sherbrooke - CoDirecteur de these
- Emmanuel MOULIN - Professeur des universités - Université Polytechnique Hauts de France - Examinateur
- Pierre MARECHAL - Maître de conférences - Université du Havre - Rapporteur
- Pierre BELANGER - Professeur - ETS Montréal - Rapporteur
- Patrice MASSON - Professeur - Université de Sherbrooke - Examinateur
The monitoring of composite structural integrity, and in particular the detection of internal defects such as delaminations, is a major challenge for many industrial sectors. Despite recent advances, several bottlenecks remain: the difficulty of characterizing hardly accessible defects, the low sensitivity of linear approaches to small damaged areas, and the need to strengthen methodologies capable of exploiting nonlinear responses. In this context, this thesis aims to develop nonlinear vibrational diagnostic methods based on the identification of local resonances. A first research axis investigated model defects of the flat-bottom hole type through numerical simulations, in order to analyze the influence of parameters such as defect size and depth position within the plate on resonance frequencies. An experimental protocol using vibrometry was implemented; the measured fields were reduced through proper orthogonal decomposition, and the dynamic behavior was identified using a Lur'e-type model, incorporating a cubic nonlinearity (Duffing-type) to describe the response from a dominant singular value. A configuration involving a fatigue-degraded flat-bottom hole further highlighted the need to account for more complex modal couplings. Finally, the methodology was applied to a composite plate containing a delamination, to assess its robustness under configurations close to real conditions. The critical role of defect geometry and localization, as well as a threshold effect potentially associated with the activation of contact nonlinearities, was demonstrated. The contributions of this work concern the understanding, in the linear regime, of the phenomenon of local resonance and its influencing parameters, the development of a methodology for identifying nonlinear behavior from a dominant singular value, and the extension of this approach to more complex defects, revealing richer nonlinear regimes and opening perspectives for damage detection based on both linear and nonlinear dynamic analysis.
Valorization of by-products of the wood industry in eco-construction in the tropical region of the Congo Basin: Taking into account thermal comfort and energy performance.
by Horly LUZOLO NSUMBU (I2M - Institut de Mécanique et d'Ingénierie de Bordeaux)
The defense will take place at 10h00 - Amphithéâtre 2 351 cours de la Libération – CS 10004, 33405 Talence Cedex Bâtiment A9a
in front of the jury composed of
- Régis POMMIER - Chargé de recherche - Université de Bordeaux - Directeur de these
- Denis BRUNEAU - Professeur des ENSA - Ecole Nationale Supérieure d'Architecture et de Paysage de Bordeaux - CoDirecteur de these
- Bertrand CHARRIER - Professeur des universités - Université de Pau et des Pays de l'Adour - Examinateur
- Rostand MOUTOU PITTI - Maître de conférences - Université Clermont Auvergne - Rapporteur
- Pierre BLANCHET - Professeur des universités - Université Laval - Rapporteur
- Alain SEMPEY - Maître de conférences - Université de Bordeaux - Examinateur
This thesis is part of a research approach dedicated to the valorization of wood industry by-products in the context of sustainable construction in the tropical region of the Congo Basin. In this area, where forest resources are abundant but locally sourced materials remain underused, the work aims to demonstrate the potential of wood residues particularly wood chips and sawdust as bio-based construction materials capable of improving both thermal comfort and energy performance of buildings. The first part provides a comprehensive assessment of the Congo Basin region, focusing on the forest–wood sector and the building industry. This analysis highlights the climatic and energy challenges that call for a rethinking of construction practices around thermal resilience and climate adaptation. Cultural barriers still limit the adoption of local ecological solutions. A thorough review of valorization techniques and thermophysical properties of bio-based materials helped identify promising material combinations suited to humid tropical climates. Experimentally, several formulations using wood by-products were developed and characterized: wood chips with raw earth, with cement, and with a lime–earth blend. The samples exhibited bulk densities from 600 to 960 kg/m³, thermal conductivities (λ) between 0.13 and 0.28 W/(m·K), and specific heat capacities ranging from 800 to 1800 J/(kg·K). Loose wood chips also showed valuable physical and thermal properties (0.07 W/(m·K), and 1750 J/(kg·K)) comparable to those of conventional insulating materials. These results confirm the relevance of wood by-products for eco-friendly construction under tropical conditions. An experimental building made of reclaimed wood was then constructed and instrumented to assess its hygrothermal behavior and indoor comfort. Field measurements enabled a detailed analysis of the building's dynamic thermal response and the influence of wall infill materials. The data collected were used to calibrate and validate a numerical model under Pléiades, which simulated the building's thermal and energy performance. Parametric analyses were then performed to assess several passive design strategies: use of bio-based wall infills, enhancement of natural ventilation, and application of reflective roofing (cool roof). Simulation results revealed a significant decrease in indoor discomfort degree-hours (up to 60 %) and thermal discomfort rate (over 40 %), along with a marked reduction in cooling energy demand. These outcomes confirm the efficiency of passive measures in improving indoor comfort and energy performance in tropical buildings. Finally, the thesis emphasizes the importance of environmental awareness and dissemination among local communities and construction stakeholders to overcome cultural perceptions that associate wood with fragility and impermanence. It also highlights the key role of numerical modeling and dynamic simulation as decision-making tools for optimizing, predicting, and improving real building performance. In conclusion, this work demonstrates the technical feasibility of valorizing wood by-products in sustainable tropical construction, while opening new perspectives for the development of innovative bio-based materials and the integration of passive design strategies into architectural practice. It therefore constitute one of the milestones laid for building an eco-construction sector based on the circular economy around the wood industry in the Congo Basin region.
ED Sociétés, Politique, Santé Publique
CONSTITUTION AND SOCIALIZATION OF THE BODY IN FRENCH NURSING EDUCA TION Ethical Stakes in the Learning of Care and the Deviant Sense of Self
by Camille BEAUDOIN (Laboratoire Cultures et Diffusion des Savoirs)
The defense will take place at 14h00 - a définir Université de Bordeaux Campus Victoire - Bât. F 1er étage 3 ter place de la Victoire 33000 BORDEAUX
in front of the jury composed of
- Marie-Pierre CHOPIN - Professeure des universités - Université de Bordeaux - Directeur de these
- Eliane ROTHIER BAUTZER - Professeur des universités - université Paris Cité - Rapporteur
- Hervé BRETON - Professeur des universités - université de tours - Examinateur
- Julie PINSOLLE - Maîtresse de conférences - université de Bordeaux - Examinateur
- Nicolas VONARX - Professeur des universités - praticien hospitalier - université de Laval à Québec - Rapporteur
This doctoral thesis examines the transmission of care within a training context that combines academic education and professional practice. The main objective is to analyze how the modes of transmission of care practices contribute to the making of nurses' bodies. The central research question focuses on the process of transmitting "caring," an activity recognized as fundamental in the nursing profession, yet potentially weakened by the current context of care provision—characterized by the standardization of practices, the objectification of knowledge, and the rationalization of care. Our hypothesis is that this context may hinder the learning of caring by pushing the professionals' sensitive and embodied experiences into the background. The overall methodology is inspired by curriculum sociology and examines three levels of initial nursing education in France: the prescribed curriculum, through a lexicographical analysis of training guidelines; the enacted curriculum, through ethnographic observation of a team of six third-year educators in a nursing school (IFSI) and five professional contributors involved in the Palliative Care teaching unit; and finally, the hidden curriculum, by analyzing the reflective writings of five students during their clinical placements, using two qualitative methods: Interpretative Phenomenological Analysis (IPA) and Foucauldian Discourse Analysis. The main contribution lies in how this research connects the disciplining of bodies with the formation of the self. We conceptualize this as a "deviant sense of self," produced by the convergence of three phenomena: pedagogical standardization, the marginalization of the sensitive dimensions of care in both official guidelines and professionalizing activities, and the self-discipline of bodies. This study contributes to a deeper understanding of the interactions between training, the socialization of the body, and the ethics of care.