HAL Portal UTC Université de Technologie de Compiègne
Not a member yet
11652 research outputs found
Sort by
Estimation of acoustic velocity in steady turbulent flow using compressed sensing on particle image velocity measurements
No full textInternational audienceMeasuring the acoustic velocity is crucial for expressing certain physical quantities like acoustic intensity or liner impedance. However, measuring acoustic velocity, using non-intrusive techniques, in the presence of turbulent flow is quite challenging. A signal processing technique is proposed to estimate the acoustic velocity from the velocity maps obtained using a Particle Image Velocimetry (PIV) system. At the Roberval laboratory, the PIV system is used to measure the flow velocity inside a rectangular-section aeroacoustic test bench, in which various obstacles can be installed. The acoustic velocity can then be extracted through signal processing. The proposed signal processing technique allows the estimation of the acoustic velocity even if Shannon's criterion is not respected (sampling frequency is much lower than the acoustic frequency) and the turbulent velocity fluctuations are order of magnitudes higher than the acoustic velocity. The method is based on a compressed sensing technique, which inherently includes two assumptions: acoustics can be decomposed according to a Fourier basis, and the sampling frequency varies randomly between two predefined limits. Loudspeakers are used to generate the acoustic velocity field inside the duct that has a steady turbulent flow with Reynold's number of about 20000. The flow velocity fields are measured for one acoustic frequency which is 200 Hz, lower than the first cut-off frequency of the duct (i.e. 850 Hz), using the PIV system with a random sampling frequency that varies between 4 and 7 Hz. The acoustic velocity maps (including amplitude and phase information) extracted from the proposed signal processing method show a good agreement with the classical intrusive microphone's measurement technique
HSMM piloté par les observations pour l’estimation de la dynamique des adventices.
No full textInternational audienceLes adventices sont des plantes qui poussent spontanément dans les parcelles agricoles et qui entrent en compétition avec les cultures. Leur dynamique repose sur la colonisation et la dormance. La banque de graines n'étant jamais observée de manière naturelle, une modélisation de cette dynamique a été proposée dans le cadre des Hidden Markov Models (HMM). Ce modèle, appelé Observation Driven-HMM (OD-HMM) étend les HMM au cas où les probabilités de transition dépendent de l'observation courante pour tenir compte des nouvelles graines produites qui entrent dans la banque de graines. Cependant, pour plus de réalisme sur la distribution de la survie de la banque de graines, le cadre naturel serait celui des Hidden Semi-Markov Models (HSMM). Néanmoins la notion de durée de séjour dans l'état caché n'est plus adaptée dès lors que l'observation influence la chaîne cachée à chaque instant. En nous appuyant sur les deux cadres OD-HMM et HSMM, nous proposons un nouveau modèle général : l'OD-HSMM, permettant à la fois de tenir compte d'une influence des données sur la chaîne cachée et de s'affranchir de la loi du temps de séjour géométrique. Nous en présentons une version paramétrique à partir des paramètres clés de la dynamique d'une espèce adventice et nous discutons différentes approches pour leur estimation
An evidential time-to-event prediction model based on Gaussian random fuzzy numbers
No full textInternational audienceWe introduce an evidential model for time-to-event prediction with censored data. In this model, uncertainty on event time is quantified by Gaussian random fuzzy numbers, a newly introduced family of random fuzzy subsets of the real line with associated belief functions, generalizing both Gaussian random variables and Gaussian possibility distributions. Our approach makes minimal assumptions about the underlying time-to-event distribution. The model is fit by minimizing a generalized negative log-likelihood function that accounts for both normal and censored data. Comparative experiments on two real-world datasets demonstrate the very good performance of our model as compared to the state-of-the-art
Enhancing the Intelligibility of Boolean Decision Trees with Concise and Reliable Probabilistic Explanations
No full textInternational audienceThis work deals with explainable artificial intelligence (XAI),specifically focusing on improving the intelligibility of decision trees throughreliable and concise probabilistic explanations. Decision trees are popular because they are considered highly interpretable. Due to cognitivelimitations, abductive explanations can be too large to be interpretableby human users. When this happens, decision trees are far from beingeasily interpretable. In this context, our goal is to enhance the intelligibility of decision trees by using probabilistic explanations. Drawinginspiration from previous work on approximating probabilistic explanations, we propose a greedy algorithm that enables us to derive conciseand reliable probabilistic explanations for decision trees. We provide a detailed description of this algorithm and compare it to the state-of-the-artSAT encoding, emphasizing the gains in intelligibility and highlightingits empirical effectiveness
Détection d’anomalies robuste basée sur l’apprentissage profond semi-supervisé dans l’industrie du futur
No full textAnomaly detection (AD) in smart manufacturing systems involves identifying abnormal data points or events from a large volume of data from various domains such as engine temperature, bearing vibration, and product images. AD methods using deep learning are gaining attention for their ability to automatically extract key features from large datasets. However, their performance can be limited in practice due to data noise and a lack of knowledge about anomalies. In this thesis, we propose various solutions to enhance the performance of deep learning-based AD methods and explore their applications across different data domains, including images and time series data. First, we demonstrate that when trained through a semi-supervised process (i.e., using only normal data), anomaly detection methods can experience reduced performance if the training data contains noise. We propose a method to separate noise from the training data while training the model simultaneously. We experimented with this approach on time series data, such as temperature and vibration measurements. The results show that this method outperforms current state-of-the-art models, especially for noisy data. For spatially dependent data, such as images, we introduce a novel augmentation technique aimed at addressing logical anomalies, which involve incorrect combinations of normal objects. This technique generates anomalies by modifying the vectors that encapsulate the general information of the image. The results suggest that integrating this augmentation technique can enhance the performance of the AD model without impacting computation time or model complexity.La Détection d'Anomalies (DA) dans les systèmes de fabrication intelligents consiste à identifier des points de données ou des événements anormaux à partir d'un grand volume de données provenant de divers domaines tels que la température du moteur, les vibrations des roulements et les images de produits. Les méthodes de DA utilisant l'apprentissage profond attirent l'attention en raison de leur capacité à extraire automatiquement les caractéristiques clés de grands ensembles de données. Cependant, leurs performances peuvent être limitées dans la pratique en raison du bruit des données et d'un manque de connaissances sur les anomalies. Dans cette thèse, tout d'abord, nous démontrons que lorsqu'ils sont entraînés par un processus semi-supervisé (en utilisant uniquement des données normales), les méthodes de DA peuvent voir leurs performances réduites si les données d'entraînement contiennent du bruit. Nous proposons une méthode pour séparer le bruit des données d'entraînement tout en entraînant le modèle simultanément. Nous avons expérimenté cette approche sur des séries temporelles. Les résultats montrent que cette méthode surpasse les modèles actuels de pointe, en particulier pour les données bruitées. Pour les données d'images, nous introduisons une nouvelle technique d'augmentation visant à traiter les anomalies logiques, qui impliquent des combinaisons incorrectes d'objets normaux. Cette technique génère des anomalies en modifiant les vecteurs qui encapsulent les informations générales de l'image. Les résultats suggèrent que l'intégration de cette technique peut améliorer les performances du modèle de DA sans affecter le temps de calcul ou la complexité du modèle
Accelerate K-Mode Algorithms Using The Triangle Inequality
No full textInternational audienceClustering is an unsupervised machine learning task that aims to discover natural groups in the given dataset. K-mode algorithms, which are adaptions of K-mean algorithms for continuous data, are among the most popular algorithms for discovering clusters in categorical data. In this paper, we present some results on how to accelerate them using the triangle inequality, while still always computing exactly the same result as the original K-mode algorithms. We also provide some empirical evidence to illustrate the potential gains provided by leveraging the triangle inequality. Finally, we envision future work aimed at providing a comprehensive understanding of the use of triangle inequality in accelerating other clustering algorithms for categorical data.</div
(Re)thinking Repairs in the Longue Durée
No full textInternational audienceThe history of repairs underscores the complexity of time through the practices of reusing and adapting objects. While this research field primarily focuses on the contemporary period, its close ties to present-day life risk obscuring the historical depth of these practices and the stakes they hold in various societies. How can we incorporate a truly historical perspective into our understanding of repairs? This article, a product of collective reflection, aims to broaden the analysis of repairs by embracing the concept of longue durée, a term coined by Fernand Braudel to describe an extensive span of historical time. Rather than striving to construct a grand narrative, this article addresses the challenge of acknowledging the diverse contexts and the discrepancies in how individuals interact with the lifespan of materials, artifacts, and their functionalities
Experimental assessment of high-energy high nickel-content NMC lithium-ion battery cycle life at cold temperatures
No full textInternational audienceAn essential element for the diffusion and environmental impact assessment of electrochemical storage systems is their lifetime. This lifetime also impacts the overall cost of the equipment in which the storage system is used. At the same time, a number of applications, e.g. electric vehicles, do not allow a temperature control of the batteries, although the performances of the latter are very strongly linked to them. The phenomena investigated, such as aging, behavior at low temperature and rapid charge for low temperature, are complex and difficult to model. Only the establishment of databases representative of the conditions of use can lead to elements of response. The exploitation of these databases by artificial intelligence tools is certainly relevant but requires a learning on data representative of the conditions of use and of the phenomena to be evaluated.The work presented in this paper is in this context and is intended to model the influence of the rapid charge at low temperature (0 °C, −10 °C, −20 °C) on the lifetime (initial capacity loss) of lithium-Ion NMC elements. The problem with data-based approaches is the large number of tests required and the development in relevant experimental designs. Faced to this challenge, a consortium of French laboratories and manufacturers has been working together since the 2000s to carry out this type of work. The paper proposes to present a test campaign and summarizes the main results to assess the service life (loss of capacity) for a fast-charging application at low temperature and position them against the manufacturer data established at 25 °C. The impact of two partial discharge processes (50 %) will also be assessed. In a last part, we show that for low temperatures, even for the nominal charge current the life time is reduced to few dozens of cycles and a mechanical destruction of the cell with no activation of the current interrupt device is achieved. This study demonstrates and quantifies the very important impact of low-temperature charging processes on the lifetime of high energy lithium-ion NMC batteries
Prioritizing brownfields transformation towards green infrastructure. The case of Rouen metropolis
No full textInternational audienceAs the global population increases, cities spatially expand, leading to notable environmental challenges and growing demand for improved living conditions for residents. In response to these concerns, reuse of developed lands offers a potential solution. In Rouen metropolis, more than 400 brownfields were identified, which is subject to regenerate brownfield measures. Moving beyond random allocation, this study aimed to provide decision-makers and researchers with a practical tool for prioritizing brownfields to transform into urban green infrastructure. Using a GIS-based multi-criteria evaluation method, sites’ suitability and urban ecological demands were considered, and the coupling coordination degree between them was calculated. This approach is in line with the principles of sustainable development, by addressing a range of environmental challenges and delivering considerable benefits to society and the local area. The presented GIS-based analysis technique can be used as part of a spatial planning process, applicable to any city. It provides support to identify suitable areas for green infrastructure expansion.Avec l’augmentation de la population mondiale, les villes s’étendent, ce qui entraîne des problèmes environnementaux importants et une demande croissante d’amélioration des conditions de vie des habitants. En réponse à ces préoccupations, la réutilisation des terrains aménagés offre une solution potentielle. Dans la métropole rouennaise, plus de 400 friches industrielles ont été identifiées et font l’objet de mesures de régénération. Au-delà de l’attribution aléatoire, cette étude vise à fournir aux décideurs et aux chercheurs un outil pratique pour hiérarchiser les friches industrielles à transformer en infrastructures vertes urbaines. À l’aide d’une méthode d’évaluation multicritères basée sur un SIG, l’adéquation des sites et les exigences écologiques urbaines ont été prises en compte, et le degré de coordination entre les deux a été calculé. Cette approche est conforme aux principes du développement durable, car elle permet de relever une série de défis environnementaux et d’offrir des avantages considérables à la société et à la région locale. La technique d’analyse basée sur le SIG présentée peut être utilisée dans le cadre d’un processus d’aménagement du territoire applicable à n’importe quelle ville. Elle permet d’identifier les zones propices à l’expansion de l’infrastructure verte
Gaussian-Smoothed Sliced Probability Divergences
No full textInternational audienceGaussian smoothed sliced Wasserstein distance has been recently introduced for comparing probability distributions, while preserving privacy on the data. It has been shown that it provides performances similar to its non-smoothed (non-private) counterpart. However, the computationaland statistical properties of such a metric have not yet been well-established. This work investigates the theoretical properties of this distance as well as those of generalized versions denoted as Gaussian-smoothed sliced divergences. We first show that smoothing and slicing preserve the metric property and the weak topology. To study the sample complexity of such divergences, we then introduce the double empirical distribution for the smoothed-projected . The distribution is a result of a double sampling process: one from sampling according to the origin distribution and the second according to the convolution of the projection of on the unit sphere and the Gaussian smoothing. We particularly focus on the Gaussian smoothed sliced Wasserstein distance and prove that it converges with a rate . We also derive other properties, including continuity, of different divergences with respect to the smoothing parameter. We support our theoretical findings with empirical studies in the context of privacy-preserving domain adaptation