A Hybrid Search Algorithm in a Multi-Agent System Environment for Multicriteria Optimization of Products Design

This study is related to the application of Artificial Intelligence approaches for the design of complex systems. The purpose is to propose methods and tools in order to help designers to optimize and to evaluate design parameters according to technical specifications during the embodiment design phase. For this purpose, multi-agent systems are interesting because of their ability to virtually duplicate the process followed by designers' teams. Because of the high number of parameters and possible combinations, a hybrid search approach based on metaheuristic mechanisms is proposed for optimization. More particularly when the task is a multiple objective combinatorial optimization and preference order cannot be defined, the objective functions of the criteria to optimize cannot be weighted and optimization cannot be resumed to a single-objective one. We specified a hybrid algorithm deriving the best (not dominated) solutions set: the Pareto front, from the possible solutions set. Self-Organizing Maps are then used to analyze and evaluate the obtained front. Our approach is illustrated in the case of the design of a 2-Degrees Of Freedom robot.