A combination of first-principles calculations with machine learning, molecular dynamic simulations to study the internal corrosion behavior of copper
Title: A combination of first-principles calculations with machine learning, molecular dynamic simulations to study the internal corrosion behavior of copper
DNr: NAISS 2024/5-61
Project Type: NAISS Medium Compute
Principal Investigator: Jinshan Pan <jinshanp@kth.se>
Affiliation: Kungliga Tekniska högskolan
Duration: 2024-03-01 – 2025-03-01
Classification: 10407 10403 20506
Homepage: https://www.kth.se/profile/jinshanp/


Corrosion of high-level nuclear waste containers can occur within a deep geologic repository. The Swedish concept of container design involves a copper canister as the corrosion barrier. The biggest threat to the canister is corrosion, which may result in environmental disaster. The corrosion of Cu canister in ground water has been debated for many years and gained huge attention recently. Oxidation by O2 and H2O leads to degradation of metals. Halide ions, especially Cl-, may cause initiation of dangerous pitting corrosion. Corrosion of Cu in ground water has been extensively studied, focusing on the processes occurring on Cu surface. However, H atoms can enter into metals leading to hydrogen embrittlement. O and S interact with metals and may also enter into metals, causing internal corrosion or promoting embrittlement and cracking. DFT calculations have been done to assess thermodynamics of simplified Cu corrosion systems. However, the interactions in the system are very complicated. Further studies using state-of-the-art experimental and computational methods are needed to reveal fundamental processes of internal Cu corrosion. By using state-of-the-art techniques, we have observed degradation of Cu microstructure and penetration of corrosive species (H, S, O, and Cl) into Cu matrix during exposure to simulated ground water. Our results indicate that Cl, S, H, O accelerate internal corrosion of Cu along grain boundaries, especially at elevated temperature. However, there is a lack of fundamental understanding of the mechanism for internal Cu corrosion, which has a large societal impact, i.e., the safety of nuclear waste management. Recently, we have employed DFT calculations to study corrosion of Ni superalloys, and also Molecular dynamic (MD) simulation combining with machine learning to investigate corrosion of Al alloys. To understand internal corrosion of Cu in ground water, we plan to use a combination of DFT calculations, machine learning, and MD simulation to provide fundamental insights into these interactions. There is a lack of the force field potential file used for MD simulation to describe the interactions between Cu, O, S, H, and Cl species. In this proposed project, we will perform DFT calculations to obtain 350,000 configurations for Cu-O-S-H-Cl systems, involving pristine structures, defective structures, deformed structures, etc. We will derive parameters such as energies, lattice parameters and forces from DFT calculations. Furthermore, we will utilize the derived parameters to train a force field potential used for MD simulation, by using machine learning. A large number of DFT calculations is needed to ensure the accuracy of obtained potential file. Subsequently, we will perform MD simulation with the obtained potential to simulate corrosion of Cu in the presence of O, S, H, and Cl, including atomic transport, deformation and dissolution of Cu, formation of particles, effect of temperature, and other properties. We aim to achieve a fundamental understanding of the corrosion mechanism at the atomic scale, which will provide guidance for protection of Cu corrosion, and ensure the safety of service life of high-level nuclear waste containers.