Abstract : In the increased competition that characterizes today's society, the mastery of time of study and quality is going through a process of instrumentation of the design process. The acceleration of product renewal and implementation of innovative technology, the introduction of new constraints related to sustainable development and definition of design objectives at the system level challenge old strategies for design process. Faced with the complexity of new concepts, instrumentation of the design process is done through the use of optimization techniques and modeling tools within rational strategies.
An optimal design is a result of a good fit between the models, optimization algorithms, the mathematical formulations and the design approaches. This dissertation presents each of these elements and highlights their interactions. The formulations are many and the designer has to choose the one that suits its application. The transformations can switch from one to another and provide greater flexibility in the design process. The models most commonly used for the design of electrical devices are classified into three categories: analytical models, finite element models, and the semi-numerical models. Their properties are complementary and no class dominates another, but each is an optimal compromise between accuracy and calculation time. Finally, the characteristics of the algorithms are often complementary and hybridization is an effective solution to reduce the time and increase the accuracy.
Optimization runs with several algorithms and models confirm the strong interactions between the choice of a model, an algorithm and a formulation. Beyond the intrinsic characteristics of algorithms, implementation for the optimization of electrical devices remains a litmus test and brings a few surprises. Proficiency in a variety of models, algorithms and formulas guarantee the designer in order to arrive at an optimal device.
The outlook for medium-term research is related to deterministic global optimization algorithms with mixed variable such as models and algorithms able of making choices in structural and materials. Long-term prospects focused on systemic multi-level and multi-scale optimization.