Artificial neural networks for nuclear power: Analysis of boiling water reactor control rod patterns in the startup range

The start-up phase of boiling water reactors poses a safety challenge due to the risk of control rod malfunction. During this phase, control rods are withdrawn sequentially to achieve reactor criticality, and then the nominal power. However, a failed control rod may unexpectedly drop, causing a surge of positive reactivity that could trigger a prompt criticality accident, presenting serious safety risks. The reactivity worth of a dropped rod depends on the positions of other rods, making it crucial to avoid dangerous configurations. Evaluating all possible rod arrangements ($\sim 10^{10}$ and more), is infeasible, and only a small fraction of these configurations ($\sim 0.001\text{\%}$) represents safety concerns. To address this, we propose a novel approach using artificial neural networks to predict the reactivity of control rod configurations. This method enables the rapid and efficient identification of limiting patterns, providing a practical solution to enhance reactor safety during start-up operations.