PROBABILISTIC ASSESSMENTS OF THE REACTOR PRESSURE VESSEL STRUCTURAL INTEGRITY: DIRECT COUPLING BETWEEN PROBABILISTIC AND FINITE-ELEMENT CODES TO MODEL SENSITIVITY TO KEY THERMO-HYDRAULIC VARIABILITY

摘要

The structural integrity assessment of a nuclear Reactor Pressure Vessel (RPV) during accidental conditions such as loss-of-coolant accident (LOCA) is a major safety concern. Besides conventional deterministic calculations to justify as a nuclear operator the RPV integrity, Electricite de France (EDF) carries out probabilistic analyses. Probabilistic analyses become most interesting when some key variables, albeit conventionally taken at conservative values, can be modelled more accurately through statistical variability. In the context of low failure probabilities, this requires however a specific coupling effort between a specific probabilistic analysis method (e.g. Form-Sorm method) and the thermo-mechanical model to be reasonable in computing time. In this paper, the variability of a key variable - the mid-transient cooling temperature, tied to a climate-dependent tank - has been modelled, in some flaw configurations (axial sub-clad) for a French vessel. In a first step, a simplified analytical approach was carried out to assess its sensitivity upon the thermo-mechanical phenomena; hence, a direct coupling had to be implemented to allow a probabilistic calculation on the finite-element mechanical model, taking also into account a failure event properly defined through minimisation of the instantaneous failure margin during the transient. Comparison with the previous (indirectly-coupled) studies and the simplified analytical approach is drawn, demonstrating the interest of this new modelling effort to understand and order the sensitivity of the probability of crack initiation to the key variables. While being noticeable in the cases studied, sensitivity to the safety injection temperature variability proves to be less than the choice of the toughness model. Finally, regularity of the thermo-mechanical model is evidenced by the coupling exercise, suggesting that a modified response-surface based method could replace direct coupling for further investigation.