Recent RIA and LOCA Analyses Performed at VTT Using Fuel Performance Codes Scanair and Fraptran-Genflo

VTT acquires and maintains independent calculation tools to be applied to fuel performance analyses and safety evaluations under changing circumstances. For transients and accidents analyses, two fuel performance codes are used. Under a collaborative arrangement with the French IRSN, the SCANAIR code is being validated and applied to RIA studies. Amended versions of the US NRC originated FRAPTRAN codes are used in parallel for LOCA analyses. The latter may be used in active combination with the general fluid model GENFLO for advanced thermal hydraulic boundary conditions. This paper summarises the latest applications of SCANAIR and FRAPTRAN-GENFLO codes at VTT. The SCANAIR code is intended for fuel behaviour analyses during an RIA type transient in PWR. An application concerning VVER fuel response in a control rod ejection accident is analysed, with code-to-code comparisons with neutronics code results. The significance of power peaking to the peripheral regions of the fuel pellet with increasing burnup is addressed. Extending the code's application field to BWR fuel is under way. Adequacy of the code to model a BWR rod drop accident starting from cold zero power with stagnant coolant is examined and reviewed. The coupled FRAPTRAN-GENFLO code is introduced as the fuel rod model in a completely new statistical fuel failure analysis procedure under development. The safety regulations in Finland limit the number of rods that fail in any accident to 10% of all the rods. So far there has not been an independent calculation tool dedicated to ascertain that. The statistical best-estimate procedure now developed relies on what is known as the Wilks' formula, a result of nonparametric statistics. Also in the method, neural networks are introduced as a novel way to reduce the number of fuel code simulations. A neural network is first trained with the results of stacked fuel performance code calculations, and then it is used as a substitute for the analysis code. Neural networks should provide superior flexibility over, e.g. the more conventional response surface method. The system has been successfully tested with a small-scale analysis of a LOCA scenario, and it is now ready to be applied to full reactor scale testing and validation. (author)