Probabilistic calculations are often used to evaluate reliability in nuclear industry. One of their main difficulties is that failure probabilities are, in this domain, very low and therefore their computations are very long. The speed of the calculations depends on the probabilistic algorithm and the complexity of the physical problem (usually modeled by a finite element analysis). The optimization of the probabilistic algorithms benefits from a wealth of literature but the physical problem is often very simplified by a lot of approximations. This paper develops a methodology to avoid some approximations. The geometry of the problem is often brought back to a 1D or 2D problem. Here, large 3D mesh can still be used thanks to transfer functions. This requires the linearity of the problem and especially a constant heat transfer coefficient for a thermo-elastic analysis. This limitation has been removed.
This article’s focus is on methodology but qualitative results of a probabilistic brittle fracture application of a reactor pressure vessel (RPV) in ferritic steel are given. Other kinds of analysis can benefit from similar methodology.