EUROPEAN CONGRESS AND EXHIBITION ON ADVANCED MATERIALS AND PROCESSES - EUROMAT 2021

One of the biggest challenges in the design and operation of nuclear reactors is to achieve a deep understanding of changes in material properties under the extreme conditions of the reactor core. In this context, one of the key safety limiting factors is the impact of severe neutron radiation; a complex, multi-scale problem. In general, accumulation and evolution of atomic-scale defects can result in macro-structural changes, such as irradiation-induced growth. Our research focuses on the initial stages of radiation damage, namely neutron-induced displacement cascades in zirconium, whose alloys are widely used in fuel assemblies. We use existing empirical potentials to perform large scale molecular-dynamics (MD) simulations, quantify the distribution of resulting defect populations and analyse the evolution of the system. While there are many examples of such studies in the literature, there are reasons to continue such numerical experiments. Firstly, ever-increasing computational power has allowed us to obtain large, high-quality data-sets, designed with a carefully selected sample of the primary knock-on atom initial momenta. At the same time, it was possible to take into account electronic stopping and electron-ion interactions via the two-temperature model. Furthermore, thanks to the ready availability of machine-learning tools, we were able to use modern analysis methods such as multi-output Gaussian process regression and take into account correlations between essential features of defect populations. As a result, we have developed a generative statistical model to produce representative damage patterns at low cost. This can also be parametrised and expressed as a closed-form expression. Investigation of model parameters will allow us to further the understanding of damage phenomena by incrementally improving the language of description. Furthermore, it will provide an easy-to-apply methodology to generate good-quality inputs for other methods.