Abstract

In this case study, we illustrate the use of a data-driven degradation model in a nuclear-specific application called void swelling. Void swelling is a complex, radiation-induced degradation mechanism that changes the dimensions of materials and damages the structural integrity. Accurate modeling and prediction of void swelling processes is crucial in nuclear power plant (NPP) management and maintenance planning by providing a guideline on the future state of the materials subject to reactor irradiation. Using a Bayesian hierarchical piecewise linear regression framework with a real-world void swelling dataset, we address the following three research questions: (1) How can we construct a data-driven degradation model such that its predictions satisfy the physical properties of void swelling? (2) How can we measure the joint effect of multiple experimental factors on the swelling process? (3) How can we accurately predict the future swelling status under limited data availability? The results on a real-world void swelling dataset not only improve our understanding of the swelling process but also provide a useful reference for nuclear practitioners and degradation researchers.

*Supplemental material accessed online through Taylor & Francis.