Software Product Line Engineering (SPLE) is a development approach used for handling large amounts of variants in software systems. The idea behind this approach is to exploit reusability of various similar and diverse products. Reusable products have various commonalities and differences that can be exploited, and to do so, developers need to define those differences (i.e., variabilities) within them. Variabilities can occur at different abstraction levels, through whole product lifecycle and developer need to handle it through the whole process. To address this problem at the architectural and requirement level, we used pure::variants, a leading variant-management commercial tool, to model variability within requirements in the railway domain. With this tool, we explicitly define a process on how to design a variability model that could be used to model several aspects of requirement variability, which can be reused again in the future, for the requirement engineering. We propose an approach for engineers to automatically generate models from requirement documents and then with the use of pure::variants functions, create various aspects which are then further transformed into feature models. Finally, the results of this transformation made possible the identification between core and variant features presented in these requirements making it easy to define what parts of the software are project specific and what are common for all generated models. Our results indicate that variability modelling using the pure::variants tool is applicable for requirement variant handling in the railway domain.

La tendance actuelle des équipements électriques ferroviaires est une concentration de matériels dans des volumes de plus en plus faibles et pour des puissances de plus en plus élevées. Ce progrès est permis par des composants à fort rendement (IGBT) qui ont la caractéristique de générer des pertes harmoniques importantes dans les composants magnétiques, provoquant du bruit à forte composante tonale. En conséquence, les cahiers des charges deviennent de plus en plus drastiques et augmente la difficulté de conception optimale des transformateurs. C’est pourquoi il est nécessaire de développer des modèles multi-physiques afin d’intégrer les différents phénomènes tels que la thermique, le bruit ou les effets de la PWM. Cette thèse porte plus précisément sur le dimensionnement optimal des transformateurs et inductances présent dans les convertisseurs auxiliaires ferroviaires. L’intérêt de la mise en place une boucle d’optimisation sur un modèle multi-physique est montrée. Pour cela il est important de développer des modèles comportent un bon compromis entre le temps de calcul et la précision. Dans cette thèse seule les composants magnétiques refroidis par une ventilation forcée sont étudiés. Un modèle thermique 3D semi-analytique couplé avec un modèle électromagnétique est utilisé. Le choix de l’optimiseur s’est porté sur l’algorithme NSGA-II permettant d’effectuer des optimisations multi-objectifs (poids et pertes) en incluant des contraintes thermiques. Pour finir une étude de sensibilité est effectuée grâce à une méthode de plan d’expérience afin de juger de la robustesse des solutions optimales / Nowadays, power converters in railway domain are more and more compact and powerful. This progress is due to the use of fast efficient components, working at high frequency like IGBT. But this evolution generates many harmonics losses in different components as inductors or transformers, and complicates their design. In addition, for the design stage, acoustics comfort is an increasingly important factor. Hence, it is necessary to develop multi-physic models in order to integrate different phenomena as the PWM effects, the temperature, and the noise. These models are coupled to an optimization tool in order to define the design rules of passive components: inductors or transformers coupled with inductor for the railway applicationIt’s necessary to develop fast models with a good compromise between the accuracy of field calculations. In this thesis, only transformers and inductors with an air forced cooling are studied. A 3D thermal model based on a nodal network linked to electromagnetic model is used. In order to optimize this structure, NSGA-II algorithm is chosen. Some global optimizations, with respect to several objectives (weight and losses), including thermal constraint are finally presented. Experimental design method is applied to know how these response functions behave in the neighborhood of the optimal point