Application of machine learning for quality data generation and materials modelling
File(s)
Author(s)
Dear, James
Type
Thesis
Abstract
This PhD thesis focuses on the application of machine learning (ML) techniques to thermomechanical testing, with a particular emphasis on improving strain rate control and determining constitutive properties. The integration of ML techniques into thermomechanical testing, particularly in uniaxial tensile tests, presents a novel approach to overcoming inaccuracies commonly encountered in traditional methods of Stroke-based Strain Testing (SST). The ML-based approach developed in this study provides for near-constant strain rate control, significantly reducing errors associated with non-constant strain rate and flow stress measurements. The research employs aluminium alloy AA6082 as the primary material, where deformation behaviours under varying temperatures and strain rates were analysed. A comparison between conventional testing approaches and machine learning approach demonstrated a marked improvement in strain rate control, reducing the Mean Absolute Percentage Error (MAPE) by over 90%. Additionally, this study examines the effectiveness of several constitutive models, including the Norton-Hoff, Johnson-Cook, and ODE (Ordinary Differential Equation) based unified viscoplastic equations, which were calibrated against experimental data to enhance their reliability in metal forming simulations. This innovative data-driven deep learning method provides an effective tool for controlling strain rate under a range of deformation conditions, allowing the accurate determination of thermomechanical properties of sheet metals.
Version
Open Access
Date Issued
2024-10-01
Date Awarded
01/03/2025
License URL
Advisor
Lin, Jianguo
Shi, Zhusheng
Publisher Department
Department of Mechanical Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)