In English

Development of a Clad Stress Predictor for PCI Surveillance using Neural Networks

Otto Gärdin
Göteborg : Chalmers tekniska högskola, 2016. 75 s.
[Examensarbete på avancerad nivå]

Westinghouse has recently proposed a new methodology to estimate the risk of PCI failure in a reactor, linking the probability for PCI failures to the cladding hoop stress. However, the current thermo-mechanical performance tool, STAV7, is not intended for on-line surveillance, making it too time-consuming to be used for this application. Instead a new approach is attempted, by using a machine learning technique called Artificial Neural Networks. Here, models for clad stress calculations are trained in order to reproduce as close as possible the results from a large number of STAV7 simulations. In order to create a functioning model, five different phases during the clad stress evolution have been identified. There are three phases during which the power remains constant and the stress evolves over time: Initial reactor power (prior to any power variations), Relaxation (following power increments) and deconditioning (following power reductions). In addition to these, there are two phases during which the power level changes instantaneously: Power increases and power decreases. It has been demonstrated that it is possible to use neural networks to reproduce the STAV7 clad stress results with high accuracy for the different phases. The calculations are fast enough to be used in a core monitoring system, although more validation, and potentially training, is needed before the networks can be used for reliable application to real operation cases. Please note that this is the public version of the master thesis report and that certain information has been omitted. For access to the full version contact Westinghouse Electric Sweden AB.

Publikationen registrerades 2017-05-16. Den ändrades senast 2017-05-16

CPL ID: 249359

Detta är en tjänst från Chalmers bibliotek