HAL Portal UTC Université de Technologie de Compiègne
Not a member yet
11652 research outputs found
Sort by
Detecting visual anomalies in an industrial environment: Unsupervised methods put to the test on the AutoVI dataset
No full textInternational audienceThe methods for unsupervised visual inspection use algorithms that are developed, trained and evaluated on publicly available datasets. However, these datasets do not reflect genuine industrial conditions, and thus current methods are not evaluated in real-world industrial production contexts. To answer this shortcoming, we introduce AutoVI, an industrial dataset of visual defects that can be encountered on automotive assembly lines. This dataset, comprising six inspection tasks, was designed as a benchmark to assess the performance of defect detection methods under realistic acquisition conditions. We analyze the performance of current stateof-the-art methods and discuss the difficulties specifically encountered in the industrial context. Our results show that current methods leave considerable room for improvement. We make AutoVI publicly available to develop unsupervised detection methods that will be better suited to real industrial tasks
Limit theorems for nonparametric conditional <i>U</i>-statistics smoothed by asymmetric kernels
No full textInternational audience-statistics represent a fundamental class of statistics used to model quantities derived from responses of multiple subjects. These statistics extend the concept of the empirical mean of a -variate random variable by considering sums over all distinct -tuples of observations of . Within this realm, W. Stute [134] introduced conditional -statistics, a generalization of the Nadaraya-Watson estimators for regression functions, and demonstrated their strong point-wise consistency. This paper presented a first theoretical examination of the Dirichlet kernel estimator for conditional -statistics on the -dimensional simplex. This estimator, being an extension of the univariate beta kernel estimator, effectively addressed boundary biases. Our analysis established its asymptotic normality and uniform strong consistency. Additionally, we introduced a beta kernel estimator specifically tailored for conditional -statistics, demonstrating both weak and strong uniform convergence. Our investigation considered the expansion of compact sets and various sequences of smoothing parameters. For the first time, we examined conditional -statistics based on mixed categorical and continuous regressors. We presented new findings on conditional -statistics smoothed by multivariate Bernstein kernels, previously unexplored in the literature. These results are derived under sufficiently broad conditions on the underlying distributions. The main ingredients used in our proof were truncation methods and sharp exponential inequalities tailored to the -statistics in connection with the empirical processes theory. Our theoretical advancements significantly contributed to the field of asymmetric kernel estimation, with potential applications in areas such as discrimination problems, -sample conditional -statistics, and the Kendall rank correlation coefficient. Finally, we conducted some simulations to demonstrate the small sample performances of the estimators.</p
Determination of the different phases of the horse swim cycle and correlation to the limbs kinematic obtained by underwater 3D motion capture analysis: preliminary results
No full textInternational audienc
A variable kinematic multi-field model for lamb wave propagation analysis in smart composite panels
No full textInternational audienceThe study explores the propagation of symmetric and antisymmetric fundamental Lamb waves in laminated composite strips, employing multi-field piezo-elastic plate models. The findings indicate the necessity of employing higher-order kinematic models to enhance the accuracy of wave propagation assessments. To further optimize computational efficiency, node-dependent kinematic models are utilized. The research delves into the tradeoff between reducing computational costs and maintaining result accuracy demonstrating a reduction of up to 60% in computational expenses in simple models while keeping the error below 1%. Two-dimensional wave propagation in laminated composite plates was further considered, expanding the scope due to material orthotropy
Combination of dependent and partially reliable Gaussian random fuzzy numbers
No full textInternational audienceGaussian random fuzzy numbers are random fuzzy sets generalizing Gaussian random variables and possibility distributions. They define belief functions on the real line that can be conveniently combined by the product-intersection rule under the independence assumption. In this paper, we introduce various extensions of this rule to account for dependence and partial reliability of the pieces of evidence. We first provide formulas for the combination of an arbitrary number of Gaussian random fuzzy numbers whose dependence is described by a correlation matrix, and we introduce a minimum-conflict combination operation. To account for partially reliable evidence, we then introduce two discounting operations called possibilistic and evidential discounting, as well as several combination operators based on different assumptions, each one parameterized by a correlation matrix and a vector of discounting coefficients. We demonstrate the application of these operators to the combination of predictions with different sets of inputs in machine learning, and show that performance can be enhanced by optimizing the parameters of the combination operators
A data analysis pipeline integrating ion mobility and high-resolution mass spectrometry for non-target screening in environmental studies
No full textInternational audienceNon-target analysis (NTA) based on high-resolution mass spectrometry (HRMS) allows advances in environmental analysis through the identification of unknown contaminants. Recently, ion mobility spectrometry (IMS) coupled to HRMS has gained some interest to obtain an additional separation dimension in addition to gas or liquid chromatography. Nonetheless, the analysis of the high-dimensional data generated by these techniques is often hindered by challenges related to proprietary software, including misalignment, restrictive data formats , and complications in developing customisable analytical workflows. Some efforts are being made to provide open databases, softwares and workflows but they are not always compatible with data acquired with ion mobility and the most commonly used open data format (mzML), produces large files, especially when including ion mobility (several Gb), thus requiring extensive storage space and computing resources. The goal of this study is to develop a pipeline for analysis of HRMS data including ion mobility, based on the advantages of recent open data formats and softwares, and to apply it to the detection of contaminants in complex environmental mixtures.The environmental data used in this study was acquired on a UPLC-IMS-QTOF system (Waters Vion) in Data-Independent Analysis mode (HDMSE). On this system, data can be retrieved from the UNIFI software using an API. Asynchronous HTTP requests were implemented to speed up the retrieval of large binary data streams. The collected data was then converted to tabular data and saved in the Apache Parquet format , an open-source file format providing efficient compression and fast retrieval of column-oriented data, compatible with many programming languages and data analysis packages. This makes it really fast to import and use into different analytical pipelines, e.g. in DEIMoS , a recent Python-based package allowing for efficient processing of multi-dimensional HRMS data, compatible with IMS. It provides efficient algorithms for feature detection, alignment and MS/MS spectral deconvolution. Comparisons were made between the data formats used in this pipeline with other open formats typically obtained with the ProteoWizard(MsConvert) software . The speed of conversion, data retrieval and storage space were evaluated for 10 environmental samples.The developed pipeline and the Parquet format demonstrated superior efficiency in terms of data collection and saving speed compared to other formats. Storage size was also significantly reduced with the Parquet format , with ~100-200 Mo for typical environmental samples, while mzML files reached 5-6 Go. The data filtering (e.g., MS1 vs MS2) and visualization is also easily performed with commonly used packages in the R or Python environments . The DEIMoS package's peak detection and alignment algorithms displayed a significantly faster processing speed than proprietary software. The data processed with the DEIMoS package will be further used with other open-source packages such as patRoon to enable extensive interrogation of a number of spectral libraries. The conversion package (developed in R) will be released on GitHub for public access and use, along with a data visualization application (R Shiny).The pipeline developed in this study allows substantial gains in efficiency in terms of data collection, storage and processing when dealing with large datasets of environmental samples. Current efforts are made on linking the data obtained with this pipeline with other open-source tools and workflows used for NTA of environmental samples (patRoon ). Based on this pipeline, results of contaminants detection in surface waters impacted by urban wet weather discharges will be presented. The processed data will also be used for machine-learning modeling in order to link HRMS signals with the measured ecotoxicity of environmental mixtures
Medical device development to quantify lung stiffness in smokers using magnetic resonance elastography
No full textInternational audienceTobacco smoking is responsible for 80 % of lung cancer in Europe [1] and in 2040 the European commission aims to achieve a tobacco-free Europe. Smokers develop symptoms as a function of tobacco use estimated by the number of pack-years [2]. Current medical test is Computed Tomography (CT) to evaluate the morphological lung changes. However, CT uses X-rays, limiting regular follow up of patients. Based on our experience in fibrotic tissues [3], we have developed magnetic resonance elastography (MRE) to quantify themechanical properties of smoker lung. Of note, tobacco-related pathologies are the most preventable diseases and thus early detection and regular monitoring methods are essential. Our goal is to provide personalized cartography of lung stiffness which would 1) become a biomarker to prevent future lung disease and 2) trigger an awareness of the patient regarding the state of their lung function through a visual color scale
Formaliser pour mieux partager : bilan du travail de modélisation pour la BDD du répertoire de la Comédie Italienne de Paris (XVIII siècle)
No full textInternational audienceLa création d’une base de données relationnelle pour le répertoire de la Comédie Italienne de Paris au XVIII siècle, s’insère dans un projet doctoral, débuté en septembre 2022 pour une durée de trois ans. L’objectif est d’étudier l’évolution du répertoire de manière chronologique - avec la génération de listes, graphes et autres visualisations - à partir des répertoires existants. Après avoir commencé la saisie des données,nous pouvons dresser un bilan du travail effectué lors de la formalisation des données, par rapport aux objectifs du projet ainsi qu’à des préoccupations écologiques, selon la perspective du minimal computing. L’article présente les bénéfices et les problèmes que nous avons rencontré dans le processus de définition du modèle entité-association pour la base de données
Predictive Degradation Modelling Using Artificial Intelligence: Milling Machine Case Study
No full textInternational audiencePredictive degradation modelling using artificial intelligence involves employing artificial intelligence techniques to anticipate the deterioration or aging of systems, equipment, or materials over time. This approach is particularly valuable in various industries such as manufacturing, healthcare, energy, and transportation, where the timely prediction of degradation can enable proactive maintenance, reduce downtime, and enhance overall system reliability. In degradation modelling, the first hitting/passage time refers to the moment when degradation stochastic process or degradation random variable reaches a predetermined threshold or specific value for the first time, which is crucial in predicting the remaining useful life and making informed decisions regarding maintenance schedules and asset management strategies. Units that fail before reaching a degradation threshold often indicates premature failures, which is a significant concern in reliability analysis as it suggests that the units did not degrade as expected and failed earlier than anticipated. To address this challenge in this article, the first hitting degradation value is introduced to be modelled through artificial intelligence technics. Furthermore, the milling machine degradation data is used to model the machine status using LSTM model, and the degradation trend is predicted using sequential models to forecast the machine status