Development of artificial neural networks to analyze micrographs of zirconium-based alloys and hydrides for nuclear power applications

点击翻译按钮获取中文摘要

Nuclear power provides 60% of Ontario's electricity. This energy source is carbon-free and has a small environment footprint. In order to ensure safe, reliable, and economical operation of our reactors, the structural materials of which they are made are regularly inspected. Notably, as the neutrons flowing through the reactor hit the pressure tubes made of zirconium alloys, they displace atoms out of their normal positions, creating nanoscale microstructure that is not observed in other conditions. Also, since the hot pressure tubes are in contact with water, they tend to pick-up hydrogen, and form zirconium hydrides, which also changes the material's mechanical properties. Characterizing radiation-induced microstructures and hydrides necessitates very high resolution instruments, including transmission electron microscopes, which have a resolution of a couple of nanometers (i.e. a few atoms wide). However, analyzing the micrographs produced by these instruments is a tedious, time-consuming task. Currently, the Canadian Nuclear Laboratories (CNL) perform this characterization for the benefit of its clients, including Canadian utilities. Skilled scientists spend countless hours painstakingly analysing these images manually, since the images are often too noisy for standard image filters to be of much help. This Alliance industrial partnership aims at developing neural networks to largely automate the image analysis, and let scientist focus on solving other important challenges in the field of materials science.