HAL Portal UTC Université de Technologie de Compiègne
Not a member yet
11652 research outputs found
Sort by
Améliorer l'intelligibilité des arbres de décision avec des explications probabilistes concises et fiables
No full textThis work deals with explainable artificial intelligence (XAI), specifically focusing on improving the intelligibility of decision trees through reliable and concise probabilistic explanations. Decision trees are popular because they are considered highly interpretable. Due to cognitive limitations, abductive explanations can be too large to be interpretable by human users. When this happens, decision trees are far from being easily interpretable. In this context, our goal is to enhance the intelligibility of decision trees by using probabilistic explanations. Drawing inspiration from previous work on approximating probabilistic explanations, we propose a greedy algorithm that enables us to derive concise and reliable probabilistic explanations for decision trees. We provide a detailed description of this algorithm and compare it to the state-of-the-art SAT encoding. In the order to highlight the gains in intelligibility while emphasizing its empirical effectiveness, we will conduct in-depth experiments on binary decision trees as well as on cases of multi-class classification. We expect significant gains in intelligibility. Finally, to demonstrate the usefulness of such an approach in a practical context, we chose to carry out additional experiments focused on text classification, in particular the detection of emotions in tweets. Our objective is to determine the set of words explaining the emotion predicted by the decision tree
Organoids, organs-on-chips, complex in vitro model: Definitions, applications, validation, ethics
No full textInternational audienceOver the past decade, new in vitro biological models have emerged which can reproduce certain characteristics of human physiology and pathologies. From organoids to organs-on-chips, these new technologies are currently revolutionizing the entire chain of research and development in pharmacology. All stakeholders are thus involved, from academic laboratories to pharmaceutical companies, start-ups, and assessment agencies. By providing better predictability, these new human biomimetic models also help address societal and ethical challenges regarding the place and use of animals in biomedical research. In this context, participants at the Ateliers de Giens (Giens Workshops) roundtable wished to examine the issues related to these new technologies with their various expertise and, given the stakes involved in the medicine and pharmacology of the future, formulate several recommendations aimed at strengthening the national structuring of the academic and industrial sector, therefore accelerating the development and validation of these new models. (C) 2024 Published by Elsevier Masson SAS on behalf of Societe francaise de pharmacologie et de therapeutique
Automatic checking and correction of localization signs in HD maps with onboard perception systems
No full textInternational audienceNavigation maps are subject to various errors coming from the mapping process, or arising over time from environment changes. To check the quality of map data before using it, in particular in the localization layer of a High-Definition (HD) map, the information given by the sensors can be processed in real-time while the vehicle is moving during a first passage. Two methods are proposed and compared to address this problem and experimental results are reported
Vision-based algorithm for autonomous aerial landing
No full textInternational audienceThe landing phase is a critical stage in autonomous aerial landing, especially when the aerial vehicle lands in a moving platform, as ground vehicles. In this paper, a solution combining the information from the onboard camera of the drone with an observer is used to estimate and predict the future position of the landing platform. This landing estimation is used in the control algorithm, based on quaternions, for generating and tracking a landing trajectory. The proposed solution is then validated in real-time experiments (two scenarios) to demonstrate the well performance and efficiency of the closed-loop system. Main graphs from these experiments are reported in this paper. Moreover, as this work aims to set the base for future developments, existing limitations from this work are discussed in the last section.</div
Optimal Energy Trajectory Generation Based on Pitch-Dependent Mutual Inductance Model for In-Flight Inductive Power Transfer of Drones
No full textInternational audienceTo address the challenge of limited flight duration in drones, research into in-flight inductive power transfer has emerged as a crucial solution. In this regard, optimizing the flight trajectory is vital for maximizing energy reception during a single pass over the transfer coils. This paper proposes a strategy that incorporates trajectories optimized on the basis of pitch angle. The optimal trajectories are derived analytically using a pitch dependent mutual inductance model, and implemented on the drone's motion controller as the reference input for each cascaded control loop. The proposed method has been validated through simulations and experiments. Implementation of two proposed trajectories in the experiments reveals that the drone's receiving power could be improved by 13.4 % maximum compared to the conventional method
Analyse de l'effet des singularités de placement fibre automatisé par essais multi-instrumentés sur éprouvettes lisses et trouées en compression
No full textInternational audienceThis study analyses the impact of singularities caused by the automated fiber placement manufacturing process on the compression behavior of composite coupons. Compression tests are performed on plain and open-hole coupons with standard aeronautical stacking layups, featuring various configurations of gaps and overlaps. A rich multi-instrumentation is used for monitoring and better understanding the effects of singularities on the damage and failure scenario. Micro-computed X-ray tomography (CT-scan) observations performed at different loading steps allow the monitoring of damage within the core of the coupons. The failure scenario is characterized using acoustic emissions sensors and computed tomography observations. The instrumentation helps to understand the effective local loading state at the coupon-scale and highlights the necessity of numerical simulations for better understanding the effect of singularities
Modélisation de batteries sodium-ion pour le diagnostic embarqué de leurs états de charge et de santé
No full textThe thesis presents a physics-based model for sodium-ion cells of the NVPF/HC type. The model takes into account the complexities involved in modeling these cells, notably due to their nonlinearities in current, temperature and state of charge (SoC). The model structure is an equivalent circuit enhanced by concepts used in single particle models (SPM) to form an ECM-SPe structure, suitable for diagnosis of the SoC and the state nof health (SoH) of the cell. The ECM-SPe structure separates the effects of solid-phase and liquid-phase diffusion. It is shown that classical modeling of solid-phase diffusion effects by impedance leads to numerous difficulties. As a replacement, an innovative approach is proposed using shifts in SoC of the open circuit voltage (OCV). These shifts are directly related to electrochemical phenomena, enabling the implementation of physical nonlinearity models. The ECM-SPe structure characterizes various electrochemical phenomena as voltage contributions: ohmic or static overvoltage (Vs), surface overvoltage (Vsurf) which includes charge transfers and the solid electrolyte interphase (SEI) overvoltages, liquid phase diffusion overvoltage (Vld) and the surface equilibrium voltage (SEV). Analytical models for current, temperature and SoC nonlinearities are proposed for each voltage contribution. These nonlinearity models are based on physical and empirical equations to characterize the model parameters without the use of lookup tables. The NVPF/HC model consists of the ECM-SPe structure with the nonlinearity models for the voltage contributions. The PhD details experimental protocols for parameter determination, which include constant current tests, pulse tests, and impedance spectroscopies, performed under various currents, temperatures and SoC. The performance of the model is validated by comparisons with experimental data, demonstrating promising results. In addition, the NVPF/HC model can be used as a diagnostic tool for charge transfers and SEI by separating their nonlinearities in current, temperature and SoC. The NVPF/HC model shows its reliability for estimating the cell voltage in variablecurrent applications, with good performance at currents up to ±5C and temperatures above 5 ◦C. The study also highlights the need to develop distinct characterization tests for solid and liquid diffusion contributions.La thèse présente un modèle basé sur la physique pour les cellules sodium-ion de type NVPF/HC. Le modèle tient compte des complexités de la modélisation de ces cellules, notamment à cause de leurs non-linéarités liées au courant, à la température et à l'état de charge (State of charge : SoC). La structure du modèle est un circuit équivalent amélioré par des concepts utilisés dans les modèles à particules uniques (Single particle model : SPM) pour former une structure ECM-SPe, adaptée aux diagnostics de SoC et d'état de santé (State of health : SoH). La structure ECM-SPe sépare les effets de la diffusion en phase solide et en phase liquide. Il est montré que la modélisation classique des effets de diffusion en phase solide par une impédance entraîne de nombreuses difficultés. En remplacement, une approche novatrice est proposée en utilisant des décalages en SoC de la tension en circuit ouvert (Open circuit voltage : OCV). Ces décalages sont en lien direct avec les phénomènes électrochimiques, ce qui permet la mise en place de modèles de non-linéarités physiques. La structure ECM-SPe caractérise divers phénomènes électrochimiques sous forme de contributions de tension : surtension ohmique ou statique (Vs), surtension de surface (Vsurf) qui comprend les surtensions des transferts de charge et de l'interphase solide-électrolyte (Solid-electrolyte interphase : SEI), surtension de diffusion en phase liquide (Vld) et la tension d'équilibre de surface (Surface equilibrium voltage : SEV). Des modèles analytiques pour les non-linéarités en courant, en température et en SoC sont proposés pour chaque contribution de tension. Ces modèles des non-linéarités sont basés sur des équations théoriques et empiriques pour caractériser les paramètres du modèle sans recours à des tables d'interpolation. Le modèle NVPF/HC est l'ensemble de la structure ECM-SPe avec les modèles des non-linéarités des contributions en tension. La thèse détaille les protocoles expérimentaux pour la détermination des paramètres, ce qui comprend des tests à courant constant, des créneaux de courant, et des spectroscopies d'impédance, réalisés sous divers courants, températures et SoC. La performance du modèle est validée par des comparaisons avec des données expérimentales, démontrant des résultats prometteurs. En plus, le modèle NVPF/HC peut servir comme un outil de diagnostic des transferts de charge et de la SEI en séparant leurs non-linéarités en courant, température et en SoC. Le modèle montre sa fiabilité pour estimer la tension de la cellule dans des applications à courant variable, avec de bonnes performances à des courants allant jusqu'à ±5C et des températures au-dessus de 5 °C. L'étude souligne également la nécessité de développer des tests de caractérisation distincts pour les contributions de diffusion solide et liquide
Multiscale mechanical constitutive equation of the passive properties of murine skeletal muscle: application to the influence of the Klf10 gene on the extra cellular matrix
No full textValorization of silk biomass for the development of amphiphilic lipopeptides
No full textInternational audienc
Damage assessment of smart polymer-matrix sandwich composite using multi-physics responses of in-situ piezoelectric transducers
No full textInternational audienceThis study focuses on assessing damage in polymer matrix composite sandwiches, crucial for structural integrity in aerospace, automotive and others industries. These composites, known for their high strength-to-weight ratio, are prone to delamination, debonding, and cracking. Early detection and precise evaluation of such damage can allow timely repairs and prevent catastrophic failures. The research integrates piezoelectric transducers, specifically piezo-copolymers (CP) and piezo-ceramics (PZT), into the structure for real-time health monitoring. Using four-point bending tests, damage is assessed through electrical capacitance, acoustic emission, and ultrasonic testing. The innovation lies in combining different transducers with multi-method diagnostics for enhanced accuracy and performance