Nano-porosity effects on corrosion rate of Zr alloys using nanoscale microscopy coupled to machine learning. (November 2022)

• Record Type: Journal Article
• Authors: Zhang, Hongliang; Kim, Taeho; Swarts, Jadin; Yu, Zefeng; Su, Ranran; Liu, Longfei; Howland, William; Lucadamo, Gene; Couet, Adrien
• Is Part Of: Corrosion science. Volume 208(2022)
• Publication Date: 2022-11
• Subjects: Corrosion -- Zr alloys -- Machine-learning -- Porosity; Corrosion and anti-corrosives -- Periodicals; 620.11223
• DOI: 10.1016/j.corsci.2022.110660
• Languages: English
• ISSNs: 0010-938X

Abstract

Abstract: Porosity throughout the zirconium oxide films plays a crucial role in the corrosion behavior and oxidation kinetics of Zr alloys, as it likely provides pathways for oxidizing and hydriding species through the oxide. In addition, it is known that substrate texture can affect the corrosion rate, although the detailed mechanisms are unclear. In this work, Zircaloy-4 was oxidized for up to 120 days at 260 and 360 °C in autoclave. Nano porosity was characterized by transmission electron microscope and precisely quantified by both manual counting as well as a newly developed, grayscale-value-difference-based, machine learning method. The oxide/substrate crystal structure, including suboxide and oxide phases, was investigated by ASTAR and directly correlated to the local pore density. The pore density shows a complex evolution as functions of oxide and metal grain orientations, oxide depth, temperature, and exposure time. Corrosion rate, substrate texture, and oxide texture are shown to affect pore density throughout the pre-transition zirconium oxide films. Two oxide growth modes (lattice-match and stress-driven) were studied, and the latter one was found to have much better resistance to corrosion and pore formation. The development of ML-based method enables a high-throughput measurement of nanopores in ZrO2, enabling rapid and systematic quantification of pore density in multiple oxides formed at different exposure times, temperatures, and substrate orientations. Highlights: A new, grayscale-value-difference-based, machine-learning method was trained to recognize the pores in TEM images in the corroded Zircaloy-4. Porosity is precisely quantified by both manual counting and machine learning counting in Zircaloy-4 oxidized for up to 120 days at 260 and 360 °C in an autoclave. The oxide/substrate structure, including suboxide and oxide phases, was investigated by ASTAR and directly correlated to the local pore density. The relation between pore density and corrosion conditions (grain orientations, oxide depth, temperature, and exposure time) was investigated. The development of ML-based method enables a high-throughput measurement of nanopores in ZrO2 .