Aluminium Alloy Design Using Machine Learning

Aluminium alloys play a pivotal role in modern engineering and manufacturing, owing to their unique combination of strength-to-weight ratio. The design of aluminium alloys often encounters a trade-off between strength and ductility, making it challenging to achieve desired properties. Machine learning has emerged as a crucial tool in resolving this dilemma, enabling the precise balancing of these properties to design new alloys. The thesis combines unsupervised machine learning, multi-target regression, and genetic algorithms to advance aluminium alloy classification, mechanical property prediction, and design. In the first study, an unsupervised machine learning approach, specifically iterative label spreading (ILS), is used to identify eight distinct classes of aluminium alloys from a comprehensive dataset. This classification enhances understanding and provides a framework for subsequent studies. The second study leverages this classification to improve the prediction of mechanical properties of aluminium alloys. The results demonstrate that more accurate and interpretable predictions can be achieved by training individual regressors on each identified class than by using traditional models. Building on these findings, the third study used a genetic algorithm, integrating data-driven classes for inverse design. This approach combines class-specific regressors with multi-objective optimisation using genetic algorithms to identify the optimal trade-off between strength and ductility in aluminium alloys. Lastly, the fourth study proposes a multi-target regression framework for inverse alloy design, using a random forest regressor and classifier to predict alloy concentrations and processing conditions. This study accelerates the design process and validates its effectiveness against existing literature. The thesis presents a comprehensive and innovative approach to aluminium alloy development, using machine learning techniques to enhance the design of alloys and demonstrating significant advancements in materials science.

Author dc:creator

• Bhat, Ninad

Language dc:language.iso

en_AU

Handle dc:identifier.uri

http://hdl.handle.net/1885/317323

OAI identifier oai:identifier

oai:openresearch-repository.anu.edu.au:1885/317323