Using Phase Field Modelling and Machine Learning to Develop a Microstructure Design Platform for Hydraulic Turbine Steels

Materials fatigue has a detrimental impact upon the lifetime and cost-effectiveness of turbine equipment in Hydro-Québec’s energy production division. Metal fatigue, in turn, is linked to it internal microstructure, that developed in a metal during its fabrication or repair through welding, as is often done to damaged hydroelectric turbine blades that break during operation. The ability to efficiently and accurately predict the evolution of said microstructure is thus key to optimizing turbine blade performance. This project will develop and ultimately combine two state-of-the art computational approaches to develop a tool that can id Hydro-Québec to predictively design repair strategies of optimal turbine performance. The first will be a multi-physics model, called a phase field (PF) model, that can predict realistic solidification microstructures in steel. The second is a machine learning (ML) algorithm that will be trained from PF data to rapidly predict the evolution of microstructure from partial knowledge of initial conditions and processing environment. The first model will be used to investigate strategies for optimal microstructure design in turbine blade welding of austenitic and ferritic steels. The machine learning algorithm will be used to improve the PF model’s predictive potential over longer intervals of welding time.

Faculty Supervisor:

Nikolas Provatas;Kirk H Bevan

Student:

Partner:

Institut de Recherche Hydro-Québec

Discipline:

Physics

Sector:

Professional, scientific and technical services; Utilities

University:

McGill University

Program: