The aim of this research work is to propose a method allowing innovation integration in early design stages and supporting architecture design of complex systems that have significant implications for the rest of overall system life-cycle. Focusing on system architectures generation support, this method proposes to use Bayesian networks combined with Constraint Satisfaction Problem (CSP) techniques in order to semi-automatically generate and evaluate complex systems architectures. Bayesian network model is used to represent the design problem in terms of decision variables, constraints and performances. Furthermore, an architecture generation algorithm is proposed to generate feasible solutions and to cluster them with regard to a given confidence level threshold. This confidence level is representing the estimation of the uncertainty on the overall system. Estimation of architecture performances are also calculated within the Bayesian network. Once the system architectures are generated, a CSP model optimises the component placement regarding placement constraints and optimisation objectives defined by designers. Software has been developed for the purpose of problem modelling and solutions visualisation. Two industrial implementations yielded in a generation of a high number of architecture solutions. In order to test the feasibility of architecture selection in an industrial environment, a study was conducted integrating four system designers. This study underlined the difficulties in defining architecture selection criteria and provides recommendations for the future system architecture selection support.
