Instrumentation and Control (I&C) Systems important to safety must conform to their requirements, where regulatory requirements are first class entities, written by national safety authorities and completed using a set of national recommendation guides or standards. The global domain knowledge is scattered, not formalized and traceability links and the organization within the domain are implicit. Bridges between different national practices are not developed, whereas the understanding of requirements and practices variability concerns becomes a significant industrial issue. The thesis sets up in an industrial context with EDF R&D and the CONNEXION project that gathered the French nuclear I&C industry. Its contributions are defined around the INCREMENT approach (Instrumentation aNd Control Regulatory Requirement Modeling Environment) that addresses the two first challenges previously introduced. In particular, they consist in: (1) the domain formalization itself by the proposal of a metamodel that allows a high level capitalization of a requirements corpus as well as its organization, (2) a tool-support basis to gather partial knowledge from the textual documents, manipulate such models that conform to the proposed metamodel, and Information retrieval techniques to support better requirements traceability, (3) the proposal of an original hybrid approach, mixing both metamodeling and information retrieval, and combine them in a mutual beneficial joint use. The metamodel and its tool support are used in the industrial context of the CONNEXION project. Where information retrieval techniques for requirements traceability suffer from large sets of false positives limitations, our hybrid approach allowed us to reduce this noise and the size of the candidate links research space by a mean of 65% without decreasing their global quality.
