The aim of this thesis is to examine the process used to develop and revise a risk assessment methodology, and how it can be improved in line with new knowledge regarding the phenomenon in question. This entails looking at how, based on scientific knowledge regarding a constantly-changing natural phenomenon, expertise is used to assess the risk at a specific moment in time (t), and how such an assessment is used as the basis for concrete recommendations that can be applied at facilities. This process will be studied throughout the history of the civil nuclear industry, from the 1960s to the present, specifically focusing on the Fukushima accident and its consequences. We will also examine the ways in which this process is and has been influenced by the political, economic, cultural and social context. Then, by comparing how this process has evolved in France with that in other countries with nuclear programs, we will identify the specific features of the case in France. Last, this thesis aims a) to demonstrate the impact of the Fukushima accident on the process of continuously revising expert assessments, and b) to clarify the influences underlying the process.

We should then be able to identify the characteristics of the post-accident context that have led to improvements in the corpus of knowledge and expertise in nuclear safety. The results of this research should provide IRSN with a new perspective on the characteristics of the post-accident context and of the "continuous improvement" process relative to safety, enabling it to better assess the limits of the latter and thus define ways in which it can be enhanced. By defining the characteristics of the process implemented to improve the corpus and expertise in the field of nuclear safety in this way, we also hope to foster discussion among the experts of their practices, enabling them to better identify what is often taken as "normal practice" within an institution: the organization of expertise and information-sharing between generalists and specialists, safety paradigms, and models, etc.