TRistructural ISOtropic nuclear fuel, otherwise known as TRISO, is a type of ceramic coated nuclear fuel that is under consideration for the next generation of nuclear reactors. TRISO particles consist of a central fuel kernel, a porous carbon ‘buffer’ layer and a SiC layer that is sandwiched between two pyrolytic carbon (PyC) layers. These layers add to the structural integrity of the fuel, making it more robust and resistant than conventional fuel rods, and improve the safety of the fuel overall. This work focuses on investigating the porosity distribution of the carbon-based buffer and PyC layers.

While TRISO fuel was first developed in the 1960s, as part of the OECD High Temperature Reactor Dragon project in the UK, it has evolved over the years before arriving to this current iteration. The Gen IV reactors, such as the very high temperature and molten salt reactors, are two of the current candidates for TRISO application.

To ensure optimal fuel performance, there are many aspects to be considered, one of these is the heat flow from the fuel kernel through the coatings. Many models use standard material characteristics such as density, however this may not reflect any porosity distribution seen within the carbon-based coatings. Understanding the porosity and its distribution across a layer is vital as it will have an impact on modelling results. For example, it may cause reduced heat flow or even hot spots, reducing the fuel performance.

A technique that can be used to investigate the microstructure of materials is plasma focussed ion beam scanning electron microscopy (PFIB-SEM). Unlike conventional FIB-SEM, PFIB can run at much higher currents, allowing for much faster milling speeds. This allows for larger areas to be milled and imaged, leading to a more representative data sets of the material under investigation. In this work, the PFIB capabilities have allowed for milling across large sections of the TRISO particle, covering the entire length of the buffer and inner PyC (IPyC) coating and the IPyC-SiC interface. For this work, the segmentation focuses on the buffer and buffer-IPyC interface.

By carrying out PFIB-SEM on TRISO coatings, the images collected can be reconstructed in 3D space, and AI deep learning segmentation can be applied to these image stacks. By carrying out this deep learning, models can be trained to examine the difference between pores and the coating material. These trained models can then be applied to large stacks of images with minimal manual segmentation.

Once this segmentation is complete, any porosity distribution across the coatings can be evaluated, and fed into thermal modelling to better represent the microstructure and to infer thermal conductivity as a function of density. By better representing the coating structure, the thermal flow models will enable the generation of more realistic heat flow throughout the particle. With this information we will produce more accurate models of TRISO fuel performance underpinning its safe operation in reactors.