No abstract

Engine torque estimation has important applications in the automotive industry: for example, automatically setting gears, optimizing engine performance, reducing emissions and designing drivelines.A methodology is described for the on-line calculation of torque values from the gear, the accelerator pedal position and the engine rotational speed. It is based on the availability of input-torque experimental signals that are preprocessed (resampled, filtered and segmented) and then learned by a statistical machine-learning method.Four methods, spanning the main learning principles, are reviewed in a unified framework and compared using the torque estimation problem: linear least squares, linear and non-linear neural networks and support vector machines. It is found that a non-linear model structure is necessary for accurate torque estimation. The most efficient torque model built is a non-linear neural network that achieves about 2% test normalized mean square error in nominal conditions.

In the automotive industry, temporal, financial and human constraints require continuous improvements in the design process of new vehicles, by delivering relevant specifications and providing reliability and robustness in design. In order to analyze factors like behaviors of drivers and types of roads and guarantee the reliability of car components, measurements of forces from wheels are stored when the vehicle is tested on tracks and used by customers. The measurements represent the time history of multi-dimensional forces on the four wheels in the longitudinal, vertical and transversal directions. They are applied on structures (suspensions or motoring for instance) during the design life of the vehicles. The context of this paper is the fatigue analysis of multi-input loadings. The study will be focused on random and possibly correlated multi-input processes, representing multidimensional forces. The goal of this paper is to present an approach to generate simple multi-input loadings equivalent to measurements in terms of damage. The simple loadings have to be equivalent for any arbitrary structure, satisfying the reliability requirements imposed by the car manufacturer.

No abstract

L'industrie automobile utilise aujourd'hui les enquêtes en clientèle, notamment, pour le dimensionnement en fiabilité de ses composants. Le problème type du classement par masses de véhicules à partir de mesures d'accélérations et de vitesses est traité. Une méthode de discrimination optimale pour ce problème est construite en considérant 4 niveaux : choix de l'espace de recherche (sélection de variables par algorithmes évolutionnaires ou par heuristiques, analyse discriminante), choix du critère de discrimination (critère Bayésien ou de marge large), choix de la complexité et choix des paramètres des algorithmes. ABSTRACT. Customer surveys have recently been used in the automotive industry in order to statistically characterize loadings. The objective is reliable design of car components. The standard problem of classifying cars based on acceleration and speed measures is addressed here. An optimal discrimination strategy is methodically devised by considering 4 decision levels: the choice of the search space (variables selection through heuristics or evolutionary algorithms, discriminant analysis), the choice of the discrimination criterion (probabilistic or maximum margin), the choice of the classification algorithm complexity and the tuning of other algorithm parameters. MOTS-CLÉS : identification en dynamique, sélection de variables, analyse discriminante, machines à support vectoriel, classification, automobile, méthode inverse en dynamique.