The design of systems architectures often involve a combinatorial design-space made of technological and architectural choices. A complete or large exploration of this design space requires the use of a method to generate and evaluate design alternatives. This paper proposes an innovative approach for the design-space exploration of systems architectures. The SAMOA (System Architecture Model-based OptimizAtion) tool associated to the method is also introduced. The method permits to create a large number of various system architectures combining a set of possible components to address given system functions. The method relies on models that are used to represent the problem and the solutions and to evaluate architecture performances. An algorithm first synthesizes design alternatives (a physical architecture associated to a functional allocation) based on the functional architecture of the system, the system interfaces, a library of available components and user-defined design rules. Chains of components are sequentially added to an initially empty architecture until all functions are fulfilled. The design rules permit to guarantee the viability and validity of the chains of components and, consequently, of the generated architectures. The design space exploration is then performed in a smart way through the use of an evolutionary algorithm, the evolution mechanisms of which are specific to system architecting. Evaluation modules permit to assess the performances of alternatives based on the structure of the architecture model and the data embedded in the component models. These performances are used to select the best generated architectures considering constraints and quality metrics. This selection is based on the Pareto-dominance-based NSGA-II algorithm or, alternatively, on an interactive preference-based algorithm. Iterating over this evolution-evaluation-selection process permits to increase the quality of solutions and, thus, to highlight the regions of interest of the design-space which can be used as a base for further manual investigations. By using this method, the system designers have a larger confidence in the optimality of the adopted architecture than using a classical derivative approach as many more solutions are evaluated. Also, the method permits to quickly evaluate the trade-offs between the different considered criteria. Finally, the method can also be used to evaluate the impact of a technology on the system performances not only by a substituting a technology by another but also by adapting the architecture of the system.
Turing patterns are common but not robust
