Search

International audience; Gears and bearings are more and more used in every industrial area mainly due to their strong reliability. Nevertheless as every mechanical transmission system, failures appear during time life. It induces critical damage, time cost for maintenance services to repair the fault potentially on duty. A wide part of work in the scientific community already provides a large quantity of features to follow health status of these systems (e.g., RMS, kurtosis, crest factor, FM0) in order to detect the fault as soon as possible. Since few years, methods developed in signal post-processing are coupled with Machine Learning (ML). ML allows ability to detect novelty or fault based on a trained algorithm. According to the literature [1], to identify the type of damage, a supervised algorithm is needed. Consequently an accurate diagnosis implies labelled data which are often difficult to obtain practically. The aim of this paper is to provide keys, based on our knowledge about features in Structural Health Monitoring (SHM), to get higher information level in classification by adding a qualitative analysis (type of damage) without label or information about the type of fault. Work carries on a measurement database. The assumption is made about two classes “healthy” / “faulty” using a supervised algorithm. The contribution of our work brings a new step in the default analysis by adding a probability for a defect case to be identified. Indeed, by combining some sensitive features selected for their relevance to describe a type of fault, a probability to have this particular default can be given. This classification is tested against three fault classes: bearing, gear generalized, gear localized. Results show that a probability for having bearing fault can be identified using this method contrary to the gear generalized and localized fault which are more complex to characterize. This new step enables to help maintenance services to focus more efficiently on the incriminated faulty part of the system, inducing a reduction of time to repair for maintenance services, a shorter out of order time leading to a significant productivity gain.

Machine Learning techniques open important ways of development of physical models in almost every field. Performances reached rely on two major pillars: the (physical) model and the data. When a huge amount of data is available, the intrinsic characteristics of the chosen algorithm become less critical. On the other hand, with limited amount of data, all the human knowledge on the system to be modelled becomes critical to exploit.One of the multiple applications of Machine Learning (ML) technics concerns the meta-models. Indeed, in this paper, we show how we can bypass a computation scheme by using clever regression models. The approach is performed on a system from which we want to know instantaneously the first natural frequencies without performing each time finite elements (FE) computation. We study the performance reached by studying the number of training cases required to teach the algorithm how to link inputs to outputs within a satisfying accuracy. Different algorithms are tested with very encouraging results as going into higher dimensional problem. The final aim of this study is to provide global guidelines for the most efficient Machine Teaching.

Colloque avec actes et comité de lecture. Internationale.; International audience; In this work, an original calculation procedure is implemented by combining a finite element method and taking into account the parametric excitations and their coupling (Spectral Iterative Method, developed by the Ecole Centrale de Lyon). The procedure is based on a modal approach developed in the frequency domain, particularly efficient to analyze systems having many degrees of freedom. In a first step, the static transmission errors and tooth stiffness's are calculated from the knowledge of the teeth macro and micro geometry. In a second step, these data are used to calculate the dynamic transmission errors, the teeth dynamic loads, and the dynamic response of the gearbox. A finite element model of the gearbox (including housings, shafts and bearings) is used to extract the modal basis necessary for this method. Measurements carried out on a specific gearbox test bench allow comparison of simulation and experimental results.The correlation between calculations and measurements is used to validate the relevance of these numerical approaches.

Colloque avec actes et comité de lecture. Internationale.; International audience; L’objectif de ce travail est de présenter une méthode d’optimisation robuste des corrections de dentures des engrenages d’une cascade de distribution de véhicule poids lourd afin de minimiser les fluctuations de l’Erreur Statique de Transmission (EST). Les fluctuations de l’EST sont une des principales sources de bruit rayonné par les systèmes à engrenages. L’EST a pour origine les déformations des dentures dépendant du couple transmis, ainsi que les écarts de micro-géométrie volontaires (corrections de denture) ou involontaires (défauts de fabrication). On cherche donc à optimiser les corrections de denture afin de minimiser cette source excitatrice pour une large gamme de couples de fonctionnement. La distribution est constituée de 2 cascades de trois pignons en série synchronisant le mouvement du vilebrequin à celui de l’arbre à cames. Les paramètres d’optimisation sont la profondeur de dépouille en tête de dent et le diamètre de début de dépouille de chaque pignon, ainsi que le bombé d’hélice cumulé de chaque engrènement, soit 8 paramètres par cascade. Enfin, les EST des deux cascades doivent être optimisées simultanément. L’optimisation est donc multi-objectifs. L’algorithme d’optimisation choisi est la méthode des essaims particulaires, méta-heuristique particulièrement adapté à ce type d’optimisation complexe à plusieurs paramètres. L’optimisation permet de sélectionner plusieurs solutions et de tester leur robustesse face aux tolérances de fabrication et de montage. Par rapport à la micro-géométrie standard, on obtient une diminution théorique de la valeur crête-à-crête de l’EST au moins égale à 2, dans la plage de couple de fonctionnement. Les jeux de pignons standards (corrections nominales) et optimisés ont été montés sur un moteur thermique et des mesures acoustiques ont été réalisées. Les résultats confirment un gain acoustique lié à l'optimisation. Ces travaux ont été réalisés dans le cadre du projet MABCA (Maitrise du bruit de chaîne cinématique), financé par l’Agence National de la Recherche.

The aim of this work is to characterize the vibro-acoustic response of a truck timing gear cascade, and especially the whining noise induced by internal excitations generated by the meshing process and the hammering noise induced by vibro-impacts between gears under fluctuations of external forces coupled with internal excitations. The tooth corrections are optimized with a meta-heuristic method (particle swarm)in order to minimize the fluctuations of static transmission errors. A robustness analysis is performed taking into account the wide range of operating torque and the geometric dispersions induced by gear manufacturing tolerances. The whining noise is then modelled using a spectral method considering the coupling between the different internal excitations. The gain provided by the optimization of tooth corrections on the whining noise of the timing gear cascade can thus be estimated. In a second step, the hammering noise is analyzed using nonlinear models which include the gears' backlashes. A continuation method coupled with centered finite differences method is implemented, as well as a direct time integration method. The dynamic behaviour of the cascade induced by fluctuating external forces coupled with internal excitations is complex : the gear responses can be periodic, pseudoperiodic,or chaotic, with possibility of multiple contact losses, including collisions between reverse flanks when the engine torque fluctuation is high. Vibro-impact regimes can be observed over a wide range of operating conditions.; L'objectif de ce travail est la compréhension et la modélisation des phénomènes physiques qui gouvernent la réponse vibro-acoustique d'une cascade de distribution de poids lourd. On s'intéresse en particulier au bruit de sirènement d'une cascade de six engrenages, qui correspond à la réponse aux excitations internes générées par le processus d'engrènement et au bruit de claquement, qui correspond à la réponse aux vibro-impacts entre dentures sous charge induits par les fluctuations des efforts extérieurs couplées aux excitations internes. Dans un premier temps, les corrections de denture sont optimisées par une méthode méta-heuristique (essaims particulaires), afin de minimiser les fluctuations des erreurs statiques de transmission en prenant en compte la large plage de couples de fonctionnement et les dispersions géométriques induites par les tolérances de fabrication des engrenages. Le bruit de sirènement est alors modélisé à partir d'une méthode spectrale qui prend en compte le couplage entre les différentes excitations internes de la cascade de dentures. Le gain apporté par l'optimisation des corrections sur le bruit de sirènement de la cascade de dentures a pu être estimé. Dans un second temps, le bruit de claquement est analysé à partir de modèles non linéaires prenant en compte les jeux entre dentures. Une méthode de continuation couplée à une discrétisation par la méthode des différences centrées et une méthode d'intégration temporelle directe sont mises en oeuvre. Le comportement dynamique de la cascade induit par la fluctuation des efforts extérieurs couplée aux excitations internes est complexe : les réponses des différents pignons peuvent être périodiques, pseudo-périodiques ou chaotiques avec possibilité d'observer des pertes de contact multiples, et notamment des chocs entre les flancs rétros des dentures si l'acyclisme moteur est élevé. Les régimes de type vibro-impacts peuvent être établis sur une large plage de vitesses de fonctionnement.

International audience; Statistical energy analysis is well-known method in the field of high frequencies. This method is founded on an analogy with thermodynamics and is largely inspired from results of statistical mechanics. But nowadays, the method is limited to the prediction of vibrational energies by application of the first principle of thermodynamics that is an energy balance on individual subsystems. In this study, we propose to extend statistical energy analysis by introducing the concept of vibrational entropy. The explicit formula given the vibrational entropies in terms of energies and number of modes is given. The entropy created during the energy echanges between subsystems is also given. Some examples are described which illustrate the meaning of vibrational entropy and vibrational temperature.

International audience; Statistical energy analysis (SEA) is the most famous method intended to prediction of sound and vibration in high frequency range with random excitations. Although SEA is largely inspired from statistical mechanics and thermodynamics, it is up to now limited to the application of the first principle of thermodynamics. In this study, we introduced the related entropy concept and it is shown that an entropy balance iAn illustration of the approach is proposed.