This communication aims at presenting a clustering methodology applied to a complex project consisting of the delivery of three interdependent sub-systems. This enables small and complementary task forces to be constituted, enhancing the communication and coordination on transverse issues related to the complexity of the whole system. The problem is to gather and exploit data for such systems, with numerous and heterogeneous risks of different domains (product, process, organization). The method consists in regrouping actors through the clustering of the risks they own. The result is a highlight on important and transverse risk interdependencies, within and between projects. These should not be neglected in order to avoid potential severe issues, whether during the project or during the exploitation of its deliverable. An application on a real program of plant implementation in the CEA-DAM is presented, with a sensitivity analysis of the clustering results to the inputs and chosen configurations of the problem.