Abstract by Felicity Nielson
Physics and Astronomy
Solving Hydrogen Embrittlement on Steel Grain Boundaries
Hydrogen Embrittlement describes the degradation of steel due to harmful hydrogen invasion. It is a phenomenon that has plagued the steel industry since its inception and kept a new steel, one third the weight of current steel though just as strong, from being usable. Hydrogen will deleteriously adsorb onto the defects of steel’s crystalline structure called grain boundaries (GBs), leading to early failure. Many methods have been made and models built to attempt to tackle this centuries old problem, but because the variety of GBs differ from one steel type to another (of which there are thousands), there has yet to be a comprehensive solution to the GB problem. We present the continuation of a new methodology using machine learning as a tool to create a statistical distribution of GB properties. Our methodology employs decision trees and Local Environment Representations that not only allow us to predict GB properties, but also to understand the physics behind the predictions.The methodology is generalizable to any GB configuration. We present our methodology tested on the Imeall alpha-Fe GB database.