Temporal Convolutional Neural Nets as a Surrogate for Fuel Performance Codes

Westinghouse Electric Sweden AB and other nuclear fuel vendors use fuel performance codes [1] to demonstrate that fuel rods sustain regular operation and transient events without damage. However, the execution time of a typical fuel rod simulation ranges from tens of seconds to minutes which can be impractical in certain applications. One such application is when it is desirable to quickly forecast the behavior of all rods in an entire core.

A surrogate model can be applied to speed up such applications and must predict various time-dependent outputs (e.g., temperature, pressure, strain, and stresses, etc.) as a function of a time-dependent heat generation rate. Several different classes of artificial neural networks for temporal sequence modeling exist for this purpose. For example, ref. [2] presents the use of Recurrent Neural Networks (RNNs) for predicting clad strain and stress, but with moderate success in performance. Reference [3] offers temporal convolutional networks (TCNs) as an alternative to RNNs and concludes that TCNs are “a natural starting point for sequence modeling”. In addition, a recently conducted study [4] presents TCNs as a promising candidate to predict cladding oxidation. Based on this, we offer a Master’s Thesis proposal to evaluate TCNs as surrogate models for a complete fuel performance code.

The student will conduct this diploma work at the Department of Physics and Astronomy, Division of Applied Nuclear Physics, collaborating with Westinghouse Electric Sweden AB.

For more information, contact:
gustav.robertson@physics.uu.se

For more information about Westinghouse Electric Sweden AB, visit:
https://www.westinghousenuclear.com/sweden/

References

[1] P. Van Uffelen, J. Hales, W. Li, G. Rossiter, and R. Williamson, “A review of fuel performance modelling”, J. Nucl. Mater., vol. 516, pp. 373–412, 2019.
[2] O. Gärdin, “Development of a Clad Stress Predictor for PCI Surveillance using Neural Networks”, p. 75.
[3] S. Bai, J. Z. Kolter, and V. Koltun, “An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling”, ArXiv180301271 Cs, Apr. 2018, Accessed: Aug. 05, 2021. [Online]. Available: http://arxiv.org/abs/1803.01271
[4] V. Nerlander, “Temporal Convolutional Networks in Lieu of Fuel Performance Codes: Conceptual Study Using a Cladding Oxidation Model”, Advanced Project Work in Energy Systems Engineering, 2021. Accessed: Oct. 16, 2021. [Online]. Available: http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-455904