We have located in the context of simulation of industrial systems, whish are complex and distributed in operational, informational and decisional terms. In this context, we consider the problems of formalization, modularity, centralization and highlighting of flows and subsystems. Indeed, the evolution of the industrial context pushes the companies to evolve to more and more decentralized systems (virtual enterprises, consortium, decentralization...). Existing methodologies and simulation tools make difficult to have an optimal support of these new systems. Indeed, there remains today very difficult to model and simulate the behavior of systems such as the consortia. After highlighted these problems, we propose within the framework of our research a methodological approach adapted to the strongly distributed industrial systems. This approach is based on the multi-agent systems but remains independent of any platform or simulation tool. We propose a life cycle and a first definition of the most significant phases : specification of a simulation model, design of a multi-agent model and implementation of a data-processing model. The concepts which we use (multi-agent systems, systemic...) enable use to answer the various problems arising from the complex and distributed systems of production.
