Application of DeepONet to predict transient drop motion of the control rod in real-time

Abstract

This paper proposes a surrogate model using a deep operator neural network (DeepONet) to predict transient drop motion of CR in real-time. The DeepONet model is trained using synthetic data obtained from the experimentally verified flexible multibody dynamic model, which considers large deformation, fluid-structure interaction, and friction. In the DeepONet architecture, the branch network processes the deformation profile of the fuel assembly (FA), while the trunk network handles the three actuator strokes as structural parameters along with time, enabling the model to predict the displacement and velocity of the CR. The accuracy of DeepONet is evaluated by calculating the mean absolute error of the normalized output, and the model's efficiency is assessed by comparing the computational time of the flexible multibody dynamics model with that of the DeepONet. Validation results demonstrate strong agreement between the DeepONet predictions and those of the flexible multibody dynamics model, achieving 99.39 % accuracy for displacement and 98.72 % accuracy for velocity under various FA deformations. Furthermore, the DeepONet provides a computational speedup of over 1500 times, with average prediction times of 321.4 ms. The proposed surrogate model offers a reliable solution for predicting CR drop motion with high accuracy and computational efficiency, enhancing nuclear reactor safety.