Search

International audience

ECML/PKDD - The European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, Würtzburg, ALLEMAGNE, 16-/09/2019 - 20/09/2019; The increase in the amount of data collected in the transport domain can greatly benefit mobility studies and help to create high value-added mobility services for passengers as well as regulation tools for operators. The research detailed in this paper is related to the development of an advanced machine learning approach with the aim of forecasting the passenger load of trains in public transport. Predicting the crowding level on public transport can indeed be useful for enriching the information available to passengers to enable them to better plan their daily trips. Moreover, operators will increasingly need to assess and predict network passenger load to improve train regulation processes and service quality levels. The main issues to address in this forecasting task are the variability in the train load series induced by the train schedule and the influence of several contextual factors, such as calendar information. We propose a neural network LSTM encoder-predictor combined with a contextual representation learning to address this problem. Experiments are conducted on a real dataset provided by the French railway company SNCF and collected over a period of one and a half years. The prediction performance provided by the proposed model are compared to those given by historical models and by traditional machine learning models. The obtained results have demonstrated the potential of the proposed LSTM encoder-predictor to address both one-step-ahead and multi-step forecasting and to outperform other models by maintaining robustness in the quality of the forecasts throughout the time horizon.

5th International Workshop and Symposium TransitData 2019, Paris, France, 08-/07/2019 - 10/07/2019; On the basis of data collected by counting sensors deployed on trains, this paper deals with a forecasting of passenger load in public transport taking into account train operation. Providing passengers with train load forecasting, in addition to the expected arrival time of the next train, can indeed be useful for a better planning of their journeys, which can prevent over-crowding situations in the trains [6] [7]. The proposed approach is built on both a hierarchy of recurrent neural networks [8] and representation learning [9] with the aim to explore the ability of such mobility data processing to simultaneously perform a forecasting task and highlight the impact of events on the public transport operation and demand. An event refers here to an unexpected passenger transport activity or to a modification in transport operation compared to those corresponding to normal conditions. Two kind of historical data are used, namely train load data and automatic vehicle location (AVL) data. This latter source contains all information related to the train operation (delay, time of arrival/departure of vehicles ...). The proposed methodology is applied on a railway transit network line operated by the French railway company SNCF in the suburban of Paris. The historical dataset used in the experiments covers the period from 2015 to 2016.

CAP 2019, Conférence sur l'Apprentissage Automatique, Toulouse, France, 03-/07/2019 - 05/07/2019; Les possibilités offertes en termes de collecte et de stockage de données permettent de renouveler les approches de modélisation dans le domaine du transport. L'exploitation croisée de différentes sources de données a pour vocation la création de services à forte valeur ajoutée pour l'usager. Les travaux détaillées dans cet article portent sur le développement de modèles de prévision a base de méthodes d'apprentissage notamment profond, pour la prévision court-terme de la charge (nombre de passagers) des trains. Cette prévision de l'affluence dans les trains peut servir à enrichir l'information voyageur à destination des usagers des transports collectifs qui peuvent ainsi mieux planifier leur déplacement. Elle peut également servir aux opérateurs de transport pour une régulation "à la demande" de l'offre de transport. La principale difficulté dans la prévision est liée à la variabilité intrinsèque des séries temporelles des charges à prédire, induite par l'influence de plusieurs paramètres dont ceux liés à l'exploitation (horaire, retard, type de mission...) et au contexte (information calendaire, grand évènement, météo,...). Nous proposons un modèle LSTM encodeur-prédicteur pour résoudre cette tâche de prévision. Plusieurs expérimentations sont menées sur des données réelles du réseau Transilien de la SNCF sur une durée d'un an et demi. Les résultats de prévision sont détaillées en vue de comparer les performances d'un tel modèle à plusieurs horizons temporels avec celles d'autres modèles plus classiques utilisées en prévision.

International audience; The management and selection of relevant information for a given speaking tour remains a problem for open-domain dialogue systems. For the latter, the possible interactions between a user and an agent are a priori infinite and indefinite. The possibility of an incorrect response from the agent to the user therefore remains high. For goal-oriented systems, the problem is considered solved, but in our experience no system shows remarkable robustness when evaluated in real situations. In this article, we present a state of the art of agent learning methods and different conversational agent models. In our opinion, one of the agent's avenues for improvement lies in his memory, because the latter (often represented by the triplet: current speaking time, dialogue history and knowledge base) is not yet modelled with enough precision. By providing the agent with a cognitively inspired memory model, we believe we can increase the performance of a goal-oriented dialogue system in real life situations, by using automatic learning algorithms with an antagonistic approach in support of a new memory model for the agent. ABSTRACT Goal-oriented dialogue systems: a recent overview and research prospects Determining which information is relevant for a given dialogue turn is still a problem for open-domain dialogue systems since the interaction between a user and the system is a priori infinite and indefinite. Thus, the possibility for error remains high for the dialog system. Although the problem is considered solved, to our experience, no system has shown outstanding performance when confronted to "real-world" situation. We review the state-of-the-art of machine learning approaches used for dialogue system development and present the various models of agent architecture. We claim that one of the main directions for improvement is the agent architecture. Often the agent's memory (represented by the triple: dialogue turn, dialogue history and knowledge base) is not modeled accurately enough. Thus we decided to investigate more cognitive oriented models of memory. We hypothesize that combining machine learning approaches in an adversarial setup with a new memory model for the agent would result in improved performances for goal-oriented dialogue systems. KEYWORDS: dialogue systems, learning by reinforcement, memory model, learning by antagonism, goal-oriented systems.; La gestion et la sélection des informations pertinentes pour un tour de parole donné restent un problème pour les systèmes de dialogue à domaine ouvert. Pour ces derniers, les interactions possibles entre un utilisateur et un agent sont a priori infinies et indéfinies. La possibilité d'une réponse erronée de l'agent à l'utilisateur demeure donc élevée. Pour les systèmes orientés-but, le problème est considéré comme résolu, mais d'après notre expérience aucun système ne montre une robustesse remarquable lorsqu'il est évalué en situation réelle. Dans cet article, nous dressons un état de l'art des méthodes d'apprentissage de l'agent et des différents modèles d'agent conversationnel. Selon nous, l'une des pistes d'amélioration de l'agent réside dans sa mémoire, car cette dernière (souvent représentée par le triplet : tour de parole courant, historique du dialogue et base de connaissances) n'est pas encore modélisée avec assez de précision. En dotant l'agent d'un modèle de mémoire d'inspiration cognitive, nous pensons pouvoir augmenter les performances d'un système de dialogue orienté-but en situation réelle, par l'emploi d'algorithmes d'apprentissage automatique avec une approche antagoniste en support d'un nouveau modèle de mémoire pour l'agent. ABSTRACT Goal-oriented dialog systems : a recent overview and research prospects Determining which information is relevant for a given dialog turn is still a problem for open-domain dialog systems since the interaction between a user and the system is a priori infinite and indefinite. Thus, the possibility for error remains high for the dialog system. Although the problem is considered solved, to our experience, no system has shown outstanding performance when confronted to "real-world" situation. We review the state-of-the-art of machine learning approaches used for dialog system development and present the various models of agent architecture. We claim that one of the main directions for improvement is the agent architecture. Often the agent's memory (represented by the triple : dialog turn, dialogue history and knowledge base) is not modeled accurately enough. Thus we decided to investigate more cognitive oriented models of memory. We hypothesize that combining machine learning approaches in an adversarial setup with a new memory model for the agent would result in improved performances for goal-oriented dialog systems. MOTS-CLÉS : systèmes de dialogue, apprentissage par renforcement, modèle de mémoire, appren-tissage par antagonisme, systèmes orientés-buts.

International audience; The management and selection of relevant information for a given speaking tour remains a problem for open-domain dialogue systems. For the latter, the possible interactions between a user and an agent are a priori infinite and indefinite. The possibility of an incorrect response from the agent to the user therefore remains high. For goal-oriented systems, the problem is considered solved, but in our experience no system shows remarkable robustness when evaluated in real situations. In this article, we present a state of the art of agent learning methods and different conversational agent models. In our opinion, one of the agent's avenues for improvement lies in his memory, because the latter (often represented by the triplet: current speaking time, dialogue history and knowledge base) is not yet modelled with enough precision. By providing the agent with a cognitively inspired memory model, we believe we can increase the performance of a goal-oriented dialogue system in real life situations, by using automatic learning algorithms with an antagonistic approach in support of a new memory model for the agent. ABSTRACT Goal-oriented dialogue systems: a recent overview and research prospects Determining which information is relevant for a given dialogue turn is still a problem for open-domain dialogue systems since the interaction between a user and the system is a priori infinite and indefinite. Thus, the possibility for error remains high for the dialog system. Although the problem is considered solved, to our experience, no system has shown outstanding performance when confronted to "real-world" situation. We review the state-of-the-art of machine learning approaches used for dialogue system development and present the various models of agent architecture. We claim that one of the main directions for improvement is the agent architecture. Often the agent's memory (represented by the triple: dialogue turn, dialogue history and knowledge base) is not modeled accurately enough. Thus we decided to investigate more cognitive oriented models of memory. We hypothesize that combining machine learning approaches in an adversarial setup with a new memory model for the agent would result in improved performances for goal-oriented dialogue systems. KEYWORDS: dialogue systems, learning by reinforcement, memory model, learning by antagonism, goal-oriented systems.; La gestion et la sélection des informations pertinentes pour un tour de parole donné restent un problème pour les systèmes de dialogue à domaine ouvert. Pour ces derniers, les interactions possibles entre un utilisateur et un agent sont a priori infinies et indéfinies. La possibilité d'une réponse erronée de l'agent à l'utilisateur demeure donc élevée. Pour les systèmes orientés-but, le problème est considéré comme résolu, mais d'après notre expérience aucun système ne montre une robustesse remarquable lorsqu'il est évalué en situation réelle. Dans cet article, nous dressons un état de l'art des méthodes d'apprentissage de l'agent et des différents modèles d'agent conversationnel. Selon nous, l'une des pistes d'amélioration de l'agent réside dans sa mémoire, car cette dernière (souvent représentée par le triplet : tour de parole courant, historique du dialogue et base de connaissances) n'est pas encore modélisée avec assez de précision. En dotant l'agent d'un modèle de mémoire d'inspiration cognitive, nous pensons pouvoir augmenter les performances d'un système de dialogue orienté-but en situation réelle, par l'emploi d'algorithmes d'apprentissage automatique avec une approche antagoniste en support d'un nouveau modèle de mémoire pour l'agent. ABSTRACT Goal-oriented dialog systems : a recent overview and research prospects Determining which information is relevant for a given dialog turn is still a problem for open-domain dialog systems since the interaction between a user and the system is a priori infinite and indefinite. Thus, the possibility for error remains high for the dialog system. Although the problem is considered solved, to our experience, no system has shown outstanding performance when confronted to "real-world" situation. We review the state-of-the-art of machine learning approaches used for dialog system development and present the various models of agent architecture. We claim that one of the main directions for improvement is the agent architecture. Often the agent's memory (represented by the triple : dialog turn, dialogue history and knowledge base) is not modeled accurately enough. Thus we decided to investigate more cognitive oriented models of memory. We hypothesize that combining machine learning approaches in an adversarial setup with a new memory model for the agent would result in improved performances for goal-oriented dialog systems. MOTS-CLÉS : systèmes de dialogue, apprentissage par renforcement, modèle de mémoire, appren-tissage par antagonisme, systèmes orientés-buts.

48th International Congress and Exposition on Noise Control Engineering (Inter-Noise 2019), MADRID, ESPAGNE, 16-/06/2019 - 19/06/2019; Noise emission is a major concern for residents living along railway lines. For a large range of conventional speeds, railway noise is dominated by rolling noise. The contribution of the track to rolling noise is mainly driven by the level of rail roughness and the Track Decay Rate (TDR). The MEEQUAI project (on-board measurement of the acoustic quality of the infrastructure) aims to develop a system capable of measuring these characteristics using on-board instrumentation. A state-of-the-art on the subject shows that indirect measurement from vibro-acoustic sensors like axle-box accelerometers or microphones in the bogie area seems possible, but that a number of improvements could be made to the existing methods, especially in estimating the transfer functions between the e_ective roughness and the signals provided by the sensors. The idea is to combine modelling and measurements to optimize the estimation of the transfer functions and the location of sensors while taking into account the variability of circulated tracks. Numerical analyses of transfer functions based on axle/track vibro-acoustic models are performed. Based on the simulations results, several improvements are proposed, with a view to future tests in rolling conditions.

International audience; Despite the importance of new concept development (NCD) literature, the variety and evolution of concepts generated in the fuzzy front end and that lead to the concepts that will be design briefs for new product, technology and radical innovation projects is still misunderstood. In this paper, we propose to address this issue of multiplicity and coherence of emerging concept descriptions in the fuzzy front end at three levels of analysis — cognitive, managerial and strategic — in order to describe the dynamics of conceptual works. We rely on a longitudinal study (8 years) of the innovation capability management in a large established firm, SNCF, the French railroad company. Our main results are i) to give a typology of emerging concept formulations; ii) to elicit a structured process of building a “desirable unknown” for both the firm and involved individuals, that gather three dimensions: cognitive generative power, collaborative attractiveness for new organizations experimentation, and strategic positioning renewal of the firm in quickly evolving environments; and iii) to explicit specific patterns of emerging concept that can improve the performance of design briefs. We thereby contribute to guide practitioners involved in NCD work to reach their innovation goals.

Lithium-ion batteries are getting larger due to the expansion of transportation and mass storage markets and they can now contain up to thousands of cells. However, a sole damaged cell can significantly impact the whole battery pack efficiency. Thus, the diagnosis of a single cell remains critical for those systems. Many methods exist in which the cell is considered homogeneous. We recently developed a heterogeneous equivalent circuit model that considers an internal resistance distribution to better represent a real single cell behaviour. This internal resistance distribution (IRD) may bring valuable information about a single cell internal quality, but only if it is determined with a sufficient accuracy. In this paper, we propose an algorithm that allows a responsive determination of the IRD. The results are compared to previous determination methods. This IRD, which is determined thanks to the preliminary construction of a homogenous model and a single discharge, is also valid for other operating conditions. The determination of a cell IRD can be used as a non-invasive diagnosis tool to track the internal degradations of a cell. The IRD of two different cell, one aged and one new are then compared, the IRD of the aged one being on average larger. This proves the relevance of the determination method and its use to get an insight into a cell. Differences in shape between the aged and the new cell IRD are discussed, as well as criteria that seems interesting. Although this work is developed for a single cell, its initial goal is to be used to detect a damaged cell inside a battery pack and may thus be applied to several cells connected in parallel.

JTAV 2019, Journées Techniques Acoustique et Vibrations, Strasbourg, FRANCE, 15-/05/2019 - 16/05/2019; Pour les riverains des infrastructures ferroviaires, les émissions acoustiques constituent une préoccupation de premier plan. Parmi les sources incriminées, le bruit de roulement est prépondérant pour une large gamme de vitesses d'exploitation. Pour ce qui est de l'infrastructure, le niveau de bruit de roulement est principalement déterminé par deux paramètres : la rugosité des rails et le taux de décroissance des ondes vibratoires de la voie (TDR). Le projet MEEQUAI (MEsure Embarquée de la QUalité Acoustique de l'Infrastructure) vise à développer un système de mesure de la rugosité des rails et du taux de décroissance à l'aide de microphones et d'accéléromètres embarqués sur un train. Ce système a pour objectif principal d'améliorer la connaissance des données d'entrée pour la réalisation de cartographies acoustiques. Cependant, il ouvrira également de nouvelles opportunités pour le suivi dans le temps de la qualité acoustique des infrastructures ferroviaires ainsi que pour la maintenance des points singuliers tels que les joints de rail et les appareils de voie.Dans cette présentation, les mécanismes de génération du bruit de roulement sont présentés, notamment à travers le rôle de la rugosité et du taux de décroissance. Les principaux systèmes de mesure de ces paramètres sont passés en revue et les limites des méthodes indirectes examinées. Enfin, les résultats des premières études numériques et expérimentales réalisées dans le projet sont exposés. Ils valident les premières briques de l'approche retenue dans MEEQUAI.

Ballast grains in railway tracks progressively wear due to the efforts exerted by the continuous passage of trains and the periodic maintenance operations. The initial sharp edges and vertices of the grains tend to become smoother and the surface texture is removed. This change in morphology plays a key role on the proper behaviour of the ballast layer and, consequently, on the frequency maintenance required on the track. The object of this paper is to improve the understanding of the wear process by tracking the morphology of a sample of grains submitted to an accelerated ageing using the Micro-Deval standard test. To this goal, X-ray Computed Tomography is used to scan a sample of grains at different states of wear and the resulting images are compared using 3D image analysis. A description of morphology evolution at different scales is provided using scalar parameters and spherical harmonic analysis, proving that the general form is not significantly changed during a standard Micro-Deval test. Thus a detailed analysis at the asperity level is performed, showing the key role of the edge broadening and vertex smoothing phenomena on ballast wear.

International audience; Broadband acoustic attenuation produced by a three dimensional (3D) locally resonant sonic crystal (LRSC), exploiting both the multiple coupled resonances and the Bragg band gaps, is numerically and experimentally reported in this work. The LRSC is made of square cross-section scatterers arranged on a square lattice and periodically incorporating both quarter-wavelength and Helmholtz resonators along their heights. Local resonators of different types are combined with the periodicity of the system generating multiple coupled resonances at low frequencies and opening Bragg band gaps respectively. This twofold coupling produces a strong broadband attenuation: a large insertion loss (IL), with an average value of 16.8 dB, covering three and a half octaves from 350 Hz to 5000 Hz with a LRSC of 30 cm width. This frequency band corresponds to one of the several railway noise sources (rolling noise, traction auxiliaries , etc.). A simplified 2D LRSC is finally analyzed numerically in a real train-track configuration, showing the efficiency of the proposed design to attenuate the railway rolling noise.

International audience

International audience; Calcite precipitation causes clogging of barbicans and water gutters in drainage systems of railway tunnels, posing serious problems to railway tracks and subsequently to train circulation. In this paper, a number of tunnels, mostly in France, with calcite precipitation problems are identified. A first analysis of the problematic tunnels in France based on a geological map evidences the key role of the surroundings materials in the calcification phenomenon – it occurs in sedimentary rocks. For further analysis, an empirical index is defined allowing the tunnel calcification level (C-L) to be quantified and the main influencing factors to be analyzed. Results show that calcite precipitation depends not only on the hydraulic conductivity and CaCO3 content of surrounding rocks, but also on the nature of lining materials and the geometry of tunnels. As far as the potential mitigation solutions are concerned, removable geotextiles are proposed to be placed on the surface of gutters and replaced regularly. This new method is believed to be of high efficiency and low cost. It can be easily verified in the field conditions.

National audience

National audience; La finalité de notre travail est la détection des anomalies dans les traces de fonctionnement de l'infrastructure de communication du Système d'Information (SI) de la SNCF. Deux techniques récentes et indépendantes semblentparticulì erement appropriées dans notre cas. Il s'agit d'une part du stockage et de l'indexation de séries temporelles dans un arbre appelé arbre iSAX, et d'autre part d'un score de détection d'anomalie nommé CFOF dont la robustesse au phénomène de concentration en haute dimension a ´ etéetéétablie de façon formelle. Dans cet article nous montrons qu'il est possible d'utiliser la structuration des in-formations dans l'arbre iSAX pour déterminer rapidement une approximation du score CFOF. La valeur obtenue est proche du score exact sur des données synthétiques et réelles. Les premiers retours d'expertises in-diquent que la méthode semble pertinente pour le déclenchement d'alarmes sur les données issues de trace d'activité du SI de la SNCF.

Proceedings of the ICA 2019 and EAA Euroregio : 23rd International Congress on Acoustics, integrating 4th EAA Euroregio 2019 : 9-13 September 2019, Aachen, Germany / proceedings editor: Martin Ochmann, Michael Vorländer, Janina Fels 23rd International Congress on Acoustics, integrating 4th EAA Euroregio 2019, ICA 2019, Aachen, Germany, 9 Sep 2019 - 13 Sep 2019; Aachen (2019)., Published by Aachen

Air pollution has become a major issue since it is responsible for several adverse health effects. Among pollutants, exposure to particle matter (PM) including ultrafine (UFP) and nanoparticles has been defined as particularly hazardous. It was shown that PM exposure increases the risks of cardiovascular and respiratory diseases. Particles size is strongly linked to their dangerousness. The smaller they are, the higher their deposition rate and their ability to penetrate respiratory and blood system are. Former studies carried out in underground train stations across the world revealed that particle concentration rates can reach higher levels than outdoor.Thus, commuting is one of the main contribution to daily personal exposure to particulate pollutants This work aims to increase our knowledge about particle dynamics issued from trains brake discs in the context of an underground station. Then, the influence of the flow surrounding a train passing a station at low speed is investigated. To assess the particle dynamics for these three specific size ranges, three-dimensional numerical simulations are conducted using the commercial code Fluent 19.2. The Eulerian-Lagrangian approach is used allowing a fine tracking of particles in the flow around the rolling stock. The turbulent mean flow is computed using Unsteady Reynolds-Averaged Navier-Stokes (URANS) through the RNG k-µ turbulence model. Particles dispersion from disks brakes to the station is computed with the Discrete Phase Model. Our methodology is described in the first part of this paper, followed by the results of flow computation. Train-induced flow influence on brake particle dispersion is discussed in the third section. This paper ends with our conclusions about regions of interested highlighted by the present study and perspective for further studies to assess properly particle dispersion in underground stations., La pollution de l'air est devenue un problème majeur puisqu'elle est responsable de plusieurs effets néfastes sur la santé. Parmi les polluants, l'exposition aux particules (PM), y compris les particules ultrafines (UFP) et les nanoparticules, a été définie comme particulièrement dangereuse. Il a été démontré que l'exposition aux particules augmente les risques de maladies cardiovasculaires et respiratoires. La taille des particules est fortement liée à leur dangerosité. Plus ils sont petits, plus leur taux de dépôt et leur capacité à pénétrer les voies respiratoires et le système sanguin sont élevés. Des études antérieures menées dans des gares souterraines à travers le monde ont révélé que les taux de concentration de particules peuvent atteindre des niveaux plus élevés qu'à l'extérieur, de sorte que les trajets domicile-travail constituent l'une des principales contributions à l'exposition quotidienne des personnes aux particules polluantes. Ensuite, l'influence de l'écoulement autour d'un train qui passe à faible vitesse devant une gare est étudiée. Pour évaluer la dynamique des particules pour ces trois plages de tailles spécifiques, des simulations numériques tridimensionnelles sont effectuées à l'aide du code commercial Fluent 19.2. L'approche eulérienne-lagrangienne permet un suivi fin des particules dans le flux autour du matériel roulant. Le débit moyen turbulent est calculé à l'aide du modèle de turbulence Unsteady Reynolds-Averaged Navier-Stokes (URANS) à l'aide du modèle RNG k-µ. La dispersion des particules des freins à disque vers la station est calculée à l'aide du modèle de phase discrète. Notre méthodologie est décrite dans la première partie de ce document, suivie des résultats du calcul des flux. L'influence du débit induit par le train sur la dispersion des particules de freinage est abordée dans la troisième section. Le présent document se termine par nos conclusions sur les régions d'intérêt mises en évidence par la présente étude et les perspectives d'autres études visant à évaluer correctement la dispersion des particules dans les stations souterraines.

International audience; This paper presents a Bayesian calibration method for a simulation-based model with stochastic functional input and output. The originality of the method lies in an adaptation involving the representation of the likelihood function by a Gaussian process surrogate model, to cope with the high computational cost of the simulation, while avoiding the surrogate modeling of the functional output. The adaptation focuses on taking into account the uncertainty introduced by the use of a surrogate model when estimating the parameters posterior probability distribution by MCMC. To this end, trajectories of the random surrogate model of the likelihood function are drawn and injected in the MCMC algorithm. An application on a train suspension monitoring case is presented.

Little research has examined the behavior of public transport users in response to a service disruption that has just occurred. This article aims to identify the different ways in which suburban train passengers cope with service interruptions or delays and to identify the factors involved in their decision-making process. We conducted a study mixing two methods: a revealed-preference questionnaire that asked 185 participants about their choices during the last major disruption they encountered and a diary study in which participants described all disruptions they experienced during a two-week period. Eighty disruptions were reported in detail by 38 users. We analyze our results using multiple correspondence analysis and ascending hierarchical clustering to construct eight suburban train passenger behavioral profiles. Additionally, we compare different cases of disruption (in a multiple-case study). We identify three categories of factors affecting suburban train user behavior: individual-specific factors, journey-specific factors and information-specific factors. The findings show that user expertise, car availability, perception of service recovery time, opinions on passenger information services, available transport services, time constraints, and the moment and place at which communication about the disruption is received influence user behavior.; Peu d'études ont examiné le comportement des usagers des transports publics face à une interruption de service qui vient de se produire. Cet article vise à identifier les différentes manières dont les voyageurs des trains de banlieue font face aux interruptions ou aux retards de service et à identifier les facteurs impliqués dans leur processus décisionnel. Nous avons mené une étude combinant deux méthodes : un questionnaire à préférence révélée qui interrogeait 185 participants sur leurs choix lors de la dernière perturbation majeure qu'ils ont rencontrée et une étude grâce à un journal de bord dans laquelle les participants décrivaient toutes les perturbations qu'ils ont vécues pendant une période de deux semaines. Quatre-vingts perturbations ont été signalées en détail par 38 utilisateurs. Nous analysons nos résultats à l'aide d'une analyse en correspondances multiples et d'une classification ascendante hiérarchique pour construire huit profils comportementaux de passagers de trains de banlieue. De plus, nous comparons différents cas de perturbation (dans une étude de cas multiples). Nous identifions trois catégories de facteurs affectant le comportement des usagers des trains de banlieue : les facteurs individuels, les facteurs spécifiques au trajet et les facteurs spécifiques à l'information. Les résultats montrent que l'expertise des utilisateurs, la disponibilité des véhicules, la perception du temps avant le rétablissement du service, les opinions sur les services d'information aux passagers, les services de transport disponibles, les contraintes de temps et le moment et le lieu où la communication concernant la perturbation est reçue influencent le comportement des utilisateurs.

This document reports the execution of the track area classification process developed during the work package « WP4 Track Survey and Track Classification » along with the main outcomes and results obtained during this process. The goal of WP4 is to set up a standard measurement and classification system that is able to provide a prediction of the expected presence of GNSS hazard causes and degraded performance to be expected at a given track area, in order to enable the evaluation if the specific environment is suitable or not to place virtual balises or on the contrary should be discarded beforehand. In this document, we detail the results of applying the classification standard process, by means of the developed software Toolset to the measurement data collected in the three measurements campaigns during ERSAT GGC project. This execution report addressed the main important aspects that need to be carried out for the correct application of the classification process and the results obtained with the data collected.

This is the deliverable D4.5 Process Execution Report of the ERSAT GGC project. This document reports the execution of the track area classification process developed during the work package WP4 Track Survey and Track Classification along with the main outcomes and results obtained during this process. The goal of WP4 is to set up a standard measurement and classification system that is able to provide a prediction of the expected presence of GNSS hazard causes and degraded performance to be expected at a given track area, in order to enable the evaluation if the specific environment is suitable or not to place virtual balises or on the contrary should be discarded beforehand. In this document, we detail the results of applying the classification standard process, by means of the developed software Toolset to the measurement data collected in the three measurements campaigns during ERSAT GGC project. This execution report addressed the main important aspects that need to be carried out for the correct application of the classification process and the results obtained with the data collected.

International audience; This paper presents a novel method for the state health monitoring of high-speed train suspensions from in-line acceleration measurements by embedded sensors, for maintenance purposes. We propose a model-based method relying on a multibody simulation code. It performs the simultaneous identification of several suspension mechanical parameters. It is adapted to the introduction of uncertainties in the system and to the exploitation of numerous high-dimensional measurements. The novel method consists in a Bayesian calibration approach using a Gaussian process surro-gate model of the likelihood function. The method has been validated on numerical experiments. We demonstrate its ability to detect evolutions of the health state of suspension elements. It has then been tested on actual acceleration measurements to study the time evolution of the suspension parameters.

International audience; The ROADEF/EURO challenge is a contest jointly organized by the French Operational Research and Decision Aid society (ROADEF) and the European Operational Research society (EURO). The contest has appeared on a regular basis since 1999 and always concerns an applied optimization problem proposed by an industrial partner. The 2014 edition of the ROADEF/EURO challenge was led by the Innovation & Research department of SNCF, a global leader in passenger and freight transport services, and infrastructure manager of the French railway network. The objective of the challenge was to find the best way to store and move trains on large railway sites, between their arrivals and departures. Since trains never vanish and traffic continues to increase, in recent years some stations have been having real congestion issues. Train management in large railway sites is of high interest for SNCF, which is why it was submitted to the operations research community as the industrial problem for the 2014 edition of the ROADEF/EURO challenge. This paper introduces the special section of the Annals of Operations Research volume devoted to the ROADEF/EURO challenge 2014, as well as the methods of the finalist teams and their results.

International audience; Thanks to its easy implementation and robust performance, beamforming is applied for source localisation in several fields. Its effectiveness depends greatly on the array sensor configuration. This paper introduces a criterion to improve the array beampattern and increase the accuracy of sound source localisation. The beamwidth and the maximum sidelobe level are used to quantify the spatial variation of the beampattern through a new criterion. This criterion is shown to be useful, especially for the localisation of moving sources. A genetic algorithm is proposed for the optimisation of microphone placement. Statistical analysis of the optimised arrays provides original results on the algorithm performance and on the optimal microphone placement. An optimised array is tested to localise the sound sources of a high speed train. The results show an accurate separation.

International audience; Prior to renewal or maintenance works on a railway track it is required to perform a proper investigation of the stiffness thickness of each material constituting the track-bed. In practice, different techniques are used for this purpose. In this study, a comparison of two methods of determining the stiffness of track-bed materials (dynamic penetration dynamic plate load) is made for a representative French conventional railway line, aiming at optimising the use of different geotechnical auscultation techniques for the railway applications. Firstly, results from geo-endoscopic tests are analysed to define the thickness nature of the different materials found in the track. Then, dynamic penetration tests (PANDA tests) are performed to evaluate the stiffness of the different layers. Statistical distribution of soil stiffness is analysed for each layer. In addition, the elastic moduli of different materials are estimated from their stiffness using empirical equations. Secondly, dynamic plate load tests using a light weight deflectometer device (LWD) are carried out on surfaces (on tracks service paths) in order to estimate the dynamic static moduli of ballast subgrade. A statistical analysis of the obtained results shows a low dispersion rate a satisfactory repeatability. The static moduli estimated from LWD tests are found to be consistent with the elastic modulus estimated from PANDA tests, showing that different auscultation methods give complementary information about the mechanical properties of the materials constituting the conventional tracks. (C) 2016 The Japanese Geotechnical Society. Production hosting by Elsevier B.V. All rights reserved.

International audience; The influence of the track geometry on the dynamic response of the train is of great concern for the railway companies, because they have to guarantee the safety of the train passengers in ensuring the stability of the train. In this paper, the long-term evolution of the dynamic response of the train on a stretch of the railway track is studied with respect to the long-term evolution of the track geometry. The characterization of the long-term evolution of the train response allows the railway companies to start off maintenance operations of the track at the best moment. The study is performed using measurements of the track geometry, which are carried out very regularly by a measuring train. A stochastic model of the studied stretch of track is created in order to take into account the measurement uncertainties in the track geometry. The dynamic response of the train is simulated with a multibody software. A noise is added in output of the simulation to consider the uncertainties in the computational model of the train dynamics. Indicators on the dynamic response of the train are defined, allowing to visualize the long-term evolution of the stability and the comfort of the train, when the track geometry deteriorates.

International audience; This paper analyzes the dynamic behavior of a randomly-heterogeneous continuum model of a ballasted railway track through the numerical construction of dispersion curves along the axis of the track. At intermediate frequencies (50-100 Hz), the dispersion curves display a strong decrease of the group velocity. At higher frequencies (above 100 Hz), the wave are strongly localized and cannot propagate along the track. This behavior is coherent with experimental observations reported in the literature on ballasted railway tracks and other granular media. Additional simulations are performed to prove that the localization is not a consequence of the geometry of the track or the impedance mismatch between the ballast and the underlying soil, but is indeed Anderson localization due to the heterogeneity of the ballast.

International audience

IWSHM-RS 2018, 2nd International Workshop on Structural Health Monitoring for Railway Systems, Qingdao, CHINE, 17-/10/2018 - 19/10/2018; Track circuits play a major role in railway signaling. In some exceptional conditions, poor rail/wheel contact conditions may lead to a non-detection of the train on the zone. A presentation of the principle of detection by track circuits is proposed to introduce the existing detection criterion. The aim of the paper is first to present new detection approaches based on signal processing on an experiment with a dedicated train running on a track equipped with a track circuit. The second objective is to present a strategy to test new detection criteria on commercial zones over a long period of time (a few months) with the help of the PEGASE acquisition board. PEGASE has been developed by IFSTTAR and the presented work is the result of SNCF/IFSTTAR collaboration.

International audience

International audience

International audience

International audience; Train suspension elements ensure its stability and play a key role in the ride safety and passengers comfort. They undergo damage throughout their lifetime, which may influence the train dynamic behavior. Consequently, they require regular maintenance, usually based on visual inspection or mileage criteria. However, a better knowledge of the actual health state of the suspensions would allow for providing maintenance closer to the real needs.This work deals with the development of a remote diagnosis method for high-speed train suspensions, which consists in the inverse identification of the suspension mechanical parameters from in-service measurements of the train dynamic behavior by embedded accelerometers.The excitation source of a rolling train is the track geometric irregularities, which consist of small displacements of the rails relatively to the theoretical track design. Track geometry also undergoes damage because of railway traffic. Consequently, the irregularities evolve through time. Because the train dynamic behavior is very dependent on them, sole acceleration measurements are not sufficient to correctly identify the suspensions mechanical parameters. Measurements of the track irregularities must be taken into account along with the corresponding measurements of the train dynamic behavior. This implies the use of a train dynamics software, in order to simulate the train dynamic behavior on a specific track geometry. For this work, we relied on the commercial multibody code Vampire.The studied vehicle is a French TGV Réseau. Accelerometers are located at the connections between carbodies, above and on the shared bogie. For each connection, carbody and bogie vertical and lateral accelerations are measured. The various acceleration signals are studied in the frequency domain. Seven mechanical parameters of various suspension types are simultaneously identified: dampers, airsprings, elastomer stiffnesses…Measurements are performed without interruption during the ride. Consequently, for a single inverse identification, joint measurements of the track geometric irregularities and of the train dynamic behavior on several hundreds of kilometers of track are generally available.The large quantity of data as well as the uncertain nature of the different physical quantities of interest encourage a statistical approach of the problem. The inverse identification is performed thanks to a Bayesian calibration procedure. The principle of Bayesian calibration is to update the initial knowledge about the system parameters using measurements of the system output. This procedure provides the distribution of probable values of the parameters. Such information allows for estimating of the accuracy of the inverse identification, through the computation of confidence intervals for instance.Because it requires simulation runs on the hundreds of kilometers of track for numerous values of the parameters, the classical Bayesian calibration procedure would be computationally unaffordable. An adaptation of the procedure relying on the approximation of the expensive likelihood function by a Gaussian process surrogate model has been developed to address this numerical cost issue. The impact of the use of a random surrogate model has been studied, in particular the influence of the surrogate model uncertainty, which represents the error inherent in the approximation of the likelihood function. The inverse identification procedure has first been validated on a numerical experiment. The principle of a numerical experiment is to generate artificial acceleration signals thanks to simulation, using a vehicle model with known degraded suspension parameters. They can then be used as if they were measurements to perform a mock identification. Since the parameters values are known, the quality of the identification can be measured. In such a case, the inverse identification displayed very satisfying results, with identification errors below 5% of the admissible interval for every parameter.The inverse identification procedure has then been tested on actual measurements of the train dynamic behavior. A significant evolution can be observed from the parameters nominal value. Since the real value of the suspension parameters remains unknown, no comparison could be performed in this case.The influence of the surrogate model uncertainty is also emphasized by this study. Indeed, when it is taken into account in the identification procedure, the size of the confidence intervals for the identified parameters significantly increases. This means that the accuracy of the identification tends to be overestimated if the surrogate model uncertainty is ignored.Transportation systems are nowadays more and more equipped with various kinds of sensors that allow for monitoring its different component. They make remote diagnosis method possible, which can be a precious tool for maintenance optimization. For train suspensions, embedding sensors remains difficult because of the variety and number of suspension elements. The advantage of an approach relying on accelerometers is to provide monitoring with a limited number of sensors. It however requires investing bigger efforts on data processing and on the identification method. Indeed, the expected information, the suspensions state, is not directly accessible in the measurements.We developed a statistical inverse identification method using measurements of the train dynamic behavior by embedded accelerometers. It involves train dynamics simulations in order to take into account track geometric irregularities measurements. The method shows very promising results on numerical experiments as well as on actual measurements. A subsequent step now consists in developing criteria on the suspension parameters to trigger maintenance operations.Concerning the mathematical aspects of the method, it is based on a Bayesian calibration procedure, which allow for estimating the identification accuracy. It also uses Gaussian process surrogate modeling in order to reduce computational costs. By applying the identification procedure on measurements performed at different time steps (with a time gap of several months), the evolution of the suspensions parameters with time and thus the gradual degradation of the suspension elements could be studied.

International audience; After the success of the first two international IEEE VTS Motor Vehicles Challenges launched in 2016 at Hangzhou, China, and in 2017 at Belfort, France during the IEEE Vehicle Power and Propulsion Conference (VPPC), this paper proposes the technical framework of the third challenge. For the edition of 2019 the challenge is focused on a dual-mode locomotive. The locomotive is powered either from a DC overhead line through a pantograph or using an on-board fuel cell with battery and supercapacitors. By taking into account the operational costs the challenge issue is to propose an Energy Management Strategy (EMS) to minimize the consumption of the electricity network and hydrogen; and to increase the lifetime of the fuel cell and energy storage system. The locomotive model will be provided to the participants using Matlab-Simulink software. Both industrial and academic teams are welcomed to propose their own EMS. The participants with top scoring will be invited to present their results in a special session at the 2019 IEEE VPPC.

International audience; La réorganisation du trafic ferroviaire en cas de panne sur infrastructure électrique est un processus de décision important pour une bonne exploitation du réseau. Le système étudié est complexe et son analyse repose sur des simulations coûteuses, ce qui rend difficile la gestion des incidents. Cet article propose d'utiliser l'analyse de sensibilité afin d'évaluer l'influence des différentes variables d'ajustement du trafic (espacement entre trains, vitesses de circulation...) sur la grandeur critique qu'est la tension d'alimentation des trains. Deux méthodes complémentaires sont mises en œuvre : l'analyse de sensibilité globale avec le calcul d'indices de sensibilité de Sobol, puis l'analyse régional avec du filtrage de Monte-Carlo associé à des comparaisons de densités de probabilité cumulées. L'approche proposée est illustrée sur un cas simple, pour lequel l'espacement entre trains se révèle avoir plus d'influence que la réduction de vitesse.

International audience; The present work addresses traffic rescheduling in case of electric infrastructure failure. The power available for train traction is restricted and the traffic must be reorganized according to this constraint. The system behaviour is computed using a dynamic multi-physics railway simulator which gives physi- cal quantities such as the train speed profiles, voltage along the catenary lines and temperatures. The rescheduling problem relies on this non-linear model, with a large number of continuous and discrete variables, constraints on dynamic outputs (typically voltage limits) and a high computation cost. We propose a rescheduling process based on sensitivity analysis in order to analyse the behaviour of this complex system and obtain information about the adjustment operations needed in order to reschedule the traffic in an optimal way. Our approach is based on statistics, with predefined variation ranges of the input parameters. In a first stage, variance decomposition-based sensitivity analysis (generalized Sobol indexes) is used for prioritization and fixing factors; then regional sensitivity analysis is used for factor mapping. The proposed approach has been tested on a simple case, with a nominal traffic running on a single-track line. The considered incident is the loss of a feeding power substation. The variables to be adjusted are the time interval between departure times and speed reduction in the vicinity of the faulty substation. The results show that increasing the time interval between trains is the most influential vari- able. Pareto-optimal fronts are also built in order to perform multi-criteria analysis according to travel- ling time, train delays and traction energy.

International audience; Despite the importance of conceptual formulations for exploratory projects, the variety and evolution of concepts generated in an exploratory project to describe the innovative purpose across the whole innovation process is still misunderstood. In this paper, we propose to address this issue of multiplicity and coherence of concept formulation in exploratory projects at three levels - cognitive, managerial and strategic - in order to describe the dynamics of conceptual works. We rely on a on a longitudinal study (7 years) of the innovation capability management in a large established firm, SNCF, the French railroad company. Our main results are i) to give a typology of seven concept formulations; ii) to show that these are interdependent and part of a structured process of building a "desirable unknown" to impact three dimensions: cognitive generative power, collaborative attractiveness for new organizations experimentation, and strategic positioning renewal of the firm in quickly evolving environments; and iii) to explicit specific conceptual formulation patterns that can improve their performance. We thereby contribute to guide practitioners on building conceptual formulations to reach their innovation goals.

International audience; The present work addresses train traffic rescheduling that is needed when an electric incident limits the power available for train traction. This is a difficult process of prime importance and we propose a decision-support tool to help the operator. The railway network is a complex multi-physics dynamic system, with many operational constraints, and its simulation is expensive. This makes the management of an incident difficult. The proposed approached applies regional sensitivity in order to study the influence of the different adjustment variables (train delays, speed references ...) on traffic quality indicators and operational constraints (catenary voltage, for example). The analysis is divided in two parts: Monte Carlo filtering for factor mapping (qualitative analysis), and two samples Kolmogorov-Smirnov test for prioritization and fixing factors (quantitative analysis). The results provide information about the behavior of the system and the influence of the different traffic adjustment variables, and help reorganizing the train traffic in an optimal way. As a result of the analysis, we obtain a set of feasible solutions that are organized according to different performance criteria. Pareto-optimal front are plotted in order to guide the decisionmaker. The proposed approach is illustrated for a simple case representative of suburban traffic.

JNGG 2018, Journées Nationales de Géotechnique et de Géologie de l'Ingénieur, CHAMPS SUR MARNE, FRANCE, 13-/06/2018 - 15/06/2018; Since 2015, a consortium comprising six complementary partners has been formed within the project ANR SSHEAR ('Soils, Structures and Hydraulics: Expertise and Applied Research') on the understanding and control of scour phenomena. After a presentation of the project, a first assessment of the achieved results is proposed: experimental developments; numerical modelling; feedback in the field.; Depuis 2015, le projet ANR SSHEAR (« Sols, Structures et Hydraulique : Expertise et Recherche Appliquée ») met en synergie les compétences de 6 partenaires pour avancer sur la compréhension et la maîtrise des risques d'affouillement. Après une présentation du projet, un premier bilan des résultats obtenus est proposé : développements expérimentaux, modélisation numérique, observations de terrain.

To study the behaviour of an axially loaded grouted rock anchor, a series of pull-out tests was performed within a strong and little fractured limestone rock mass. Optical fibre measurements allow the efficient grouted length to be evaluated. By comparing the test results, we highlighted the importance of the diameters (rebar, boring) in the tensile behaviour of the anchors., Afin d'étudier le comportement sous sollicitation axiale d'un ancrage passif scellé au rocher, une campagne d'essais d'arrachement a été réalisée au sein d'un massif calcaire résistant et peu fracturé. La mesure par fibre optique a permis d'identifier la longueur de scellement efficace. La comparaison des résultats met en évidence l'importance des diamètres (barre, forage) sur le comportement de l'ancrage.

International audience; Rebuilding society through the creation of innovation communities. The case of SpotLAB workshops on new mobilities in French regionsInnovation is a key to territorial development, but the management to achieve it remains a challenge. In line with works on the building of collaborative innovation capabilities through design, our intervention research introduces SpotLAB, an original device inspired by KCP workshops. Three proposals for a territorial innovation model appeared: the conditions for the emergence of a territorial community of innovation, its engineering (polycentric laboratory-network), and ephemeral institutionalization.; L’innovation est une clé du développement territorial néanmoins la nature du pilotage pour y parvenir demeure un challenge. Dans la lignée de travaux sur la restauration de capacités collaboratives d’innovation par la conception, cette recherche-intervention présente les SpotLAB, dispositif original inspiré des ateliers KCP. Il en ressort trois propositions pour un modèle territorial d’innovation : les conditions d’émergence d’une communauté territoriale d’innovation, son ingénierie (laboratoire-réseau polycentrique), et l’institutionnalisation éphémère.

International audience; The objective of the work presented here is to perform the Bayesian calibration of parameters describing the mechanical properties of high-speed train suspensions for maintenance purposes. This calibration requires joint measurements of the track geometric irregularities and of the train dynamic response. The calculation of the likelihood function relies on simulation, which makes it computationally expensive. Therefore, the likelihood function is represented by a Kriging metamodel. We present a calibration method that allows for taking into account the uncertainty introduced by the use of this metamodel.

Many old railway lines in Europe need to be reinforced as they no longer meet the requirements of modern traffic. This paper presents the results of field tests carried out, as part of the French RUFEX research project (Reinforcement of railway tracks), to study the feasibility of the installation of 600 mm diameter vertical soil-cement columns under existing railway lines without removing the tracks or cementing the ballast. These columns were installed under an existing track using a spreadable mixing tool. The execution procedure is described and the quality / homogeneity of the mixed material is studied with excavated columns after curing periods of 28 and 180 days and several samples were taken for laboratory testing. The tests included unconfined compressive strength tests, splitting tensile strength tests, porosity and ultrasonic wave velocity measurements. The results show an increase in strength of 30% between 28 and 180 days. A linear correlation between compressive and tensile strengths is proposed. Besides, data obtained from ultrasonic wave velocity measurements suggests that this test can be used for the estimation of index parameters such as density and porosity but also for preliminary assessments of static stiffness and strength by correlation with the dynamic elastic modulus.

International audience

International audience; Rolling contact fatigue cracks in rails undergo non-proportional mixed-mode I+II+III, in variable proportionsalong their front. In order to determine the crack growth thresholds and kinetics in mixed-mode II/III, asymmetricfour point bending tests are run on a rail steel with different angles between the crack front and theshearing load, so as to vary the mode mixity ratio. For sufficiently high loading ranges, these tests give rise tocoplanar shear-mode crack growth. The effective stress intensity factors (SIFs) are derived by an inverse methodfrom the measured crack face relative displacements. It appears to be 10–70% lower than the nominal SIFs andto allow a reasonable correlation of the measured crack growth rates. The local application, ahead of the crackfront, of shear-driven or tension-driven fatigue damage models – after 3D elastic-plastic computations of localstress and strain ranges – allows a successful prediction of crack fronts paths and growth rates.

International audience; The segmentation of large scale power grids into zones is crucial for control room operators when managing the grid complexity near real time. In this paper we propose a new method in two steps which is able to automatically do this segmentation, while taking into account the real timecontext, in order to help them handle shifting dynamics. Our method relies on a ”guided” machine learning approach. As a first step, we define and compute a task specific ”Influence Graph” in a guided manner. We indeed simulate on a grid state chosen interventions, representative of our task of interest (managing active power flows in our case). For visualization and interpretation, we then build a higher representation of the gridrelevant to this task by applying the graph community detection algorithm Infomap on this Influence Graph. To illustrate ourmethod and demonstrate its practical interest, we apply it on commonly used systems, the IEEE-14 and IEEE-118. We show promising and original interpretable results, especially on the previously well studied RTS-96 system for grid segmentation. We eventually share initial investigation and results on a large- scale system, the French power grid, whose segmentation had a surprising resemblance with RTE’s historical partitioning

In the context of direct-current electrified railway systems, power supply quality issues are a major concern for the railway network operators. Indeed, after signaling systems and track equipment, voltage issues may be the most challenging factor for the capacity of a direct-current railway line. These issues and the high cost of new substations motivate the implementation of wayside energy storage systems to support the system. Since an appropriate sizing methodology is a key step towards an energy storage system implementation in a railway line, in this paper a sizing methodology is presented in detail. As hypothesis, the sizing methodology is defined such that the storage system is able to support the railway line under projected conditions of the rolling stock traffic, along with other technical criteria relevant to the railway operator. A real-time simulation oriented, direct-current railway network modeling approach is proposed and exploited in this sizing methodology. Using this modeling method, the optimal energy storage sizing formulation is described. The objective function to minimize is the trade-off between energy storage capacity and charging power. The proposed sizing method is applied to a real railway line with known power supply quality issues. This case study is introduced by analyzing real voltage measurements on two key sites of the line over several days. Then, the sizing methodology is applied and the results are discussed for the study case, defining the minimal technical requirements to install. The obtained minimal technical requirements for the storage system are considered by the operator to take a cost-effective decision in contrast with the reinforcement of the line with new substations.

International audience; The multidimensional Gaussian kernel-density estimation (G-KDE) is a powerful tool to identify the distribution of random vectors when the maximal information is a set of independent realizations. For these methods, a key issue is the choice of the kernel and the optimization of the bandwidth matrix. To optimize these kernel representations, two adaptations of the classical G-KDE are presented. First, it is proposed to add constraints on the mean and the covariance matrix in the G-KDE formalism. Secondly, it is suggested to separate in different groups the components of the random vector of interest that could reasonably be considered as independent. This block by block decomposition is carried out by looking for the maximum of a cross-validation likelihood quantity that is associated with the block formation. This leads to a tensorized version of the classical G-KDE. Finally, it is shown on a series of examples how these two adaptations can improve the nonparametric representations of the densities of random vectors, especially when the number of available realizations is relatively low compared to their dimensions.

International audience; The work presented here deals with the development of a state health monitoring method for high-speed train suspensions using in-service measurements by embedded accelerometers. Mathematically, it consists in solving a statistical inverse problem.A rolling train is a dynamic system excited by the track geometric irregularities. They consist of small displacements of the rails relatively to the theoretical track design. The suspension elements play a key role for the ride safety and comfort. The train dynamic response being dependent on the suspensions mechanical characteristics, information about the suspensions state can be inferred from acceleration measurements in the train. This information would allow for providing a more efficient maintenance.Track geometry is subject to damage caused by railway traffic and to maintenance operations. Consequently, it evolves through time. Because of the high sensitivity of the train dynamic response to the track geometric irregularities, their evolution must be taken into account through the use of train dynamics simulation.Because the system input (the track geometric irregularities) and output (the train dynamic response) are stochastic quantities, the inverse problem is solved in the Bayesian framework. The monitoring method thus consists in performing a Bayesian calibration of a simulation-based model using joint measurements of the system input and output. Its objective is to identify the posterior distribution of the model parameters describing the suspensions mechanical characteristics.Classical Bayesian calibration implies the computation of a likelihood function using a stochastic model and experimental data. This likelihood function is then used to estimate the posterior distribution of the model parameters. This step can be performed by Markov Chain Monte Carlo (MCMC) algorithms, which require numerous calls to the likelihood function. If the latter is expensive to compute, it may result in unaffordable computational costs, which is the case here. To address this issue, we propose to rely on surrogate models. They are usually used to provide an algebraic approximation of the system output. However, in the present case, the output is functional, which makes a surrogate model difficult to build. Instead, we propose a calibration method based on a Gaussian surrogate model of the scalar likelihood function. We present how such a random surrogate model can be used to estimate the model parameters distribution, how the new uncertainty it introduces can be taken into account to correctly evaluate the calibration accuracy, and the results of the method applied to our railway monitoring case.