Machine learning augmented high throughput formulations

• Main Author: Shi, Shi Jun
• Other Authors: Kedar Hippalgaonkar
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Materials
• Online Access: https://hdl.handle.net/10356/166921
• Institution: Nanyang Technological University

Material discovery holds the key to technological advancement as materials’ properties dictate their potential applications. However, conventional methods targeted at discovering new materials can be time-consuming and labour-intensive, which hinders technological advancements. There are some ways to accelerate the pace of material discovery. Firstly, there is high throughput experimentation which introduces the use of automated systems to expeditiously prepare and test large numbers of samples. An example will be the use of Opentrons robot to prepare large numbers of samples that otherwise would have taken much longer time for scientists to manually prepare them. It reduces labour cost and free up more time for scientists to work on other project synchronously. Secondly, there are computational modelling and machine learning. As higher computational power becomes readily available, researchers are now tapping on the use of algorithms and simulations to predict the properties of materials. Machine learning algorithms can be trained to recognise patterns and relationships between material properties and its chemical composition. This will reduce wastage and help scientists identify potential candidates for further studies. In addition, collaboration and data sharing among various organisations and researchers can also help accelerate the process of material discovery. While there are many other ways to accelerate the pace of material discovery, this project seeks to focus on the use of high throughput method as well as machine learning. This paper seeks to detail the use of machine learning alongside high throughput methods to identity the relationship between a formulation’s chemical compositions and its chemical stability. High throughput method will help to prepare and test large numbers of samples while machine learning will seek to optimise a computational model that continuously identify next best candidates to sample until we found a chemically stable formulation.