A machine learning approach to predicting the melting points of liquid metal alloys for consumer electronics cooling

Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

School of Chemical Engineering | Master's thesis

Authors

Date

2025-09-29

Department

Major/Subject

Functional Materials for Global Challenges

Mcode

Degree programme

Master's Programme in Advanced Materials for Innovation and Sustainability

Language

en

Pages

47

Series

Abstract

This thesis investigates liquid metal phase change materials for cooling consumer electronics. A phase change material stores heat while melting and releases it while solidifying, which helps keep device temperatures in the safe 45–55℃. The study compiles liquidus data from peer reviewed papers and phase diagram databases. The final physics-expanded table contains 22,231 compositions and uses an 85/15 split by alloy system for validation. The method trains a Transformer model to predict the liquidus temperature from elemental fractions and simple physicochemical features. The screening then enumerates 1,317,840 unique compositions by combining 19 elements on a 0.1 at.%grid. The workflow ranks candidates by absolute deviation from 50℃ and forms a shortlist within 45 to 55℃. The lab prepared several shortlisted alloys for tests. The paper presents one representative DSC result. Differential scanning calorimetry measured the melting behavior, showing an onset at about 45.6℃ and a main peak at about 49.3℃ for the shortlisted alloy. These findings show that the workflow contracts a very large search space and supports safe thermal design in compact devices.

Description

Supervisor

Lehtivuori, Heli

Thesis advisor

Khachatryan, Hayk

Keywords

liquid metals, phase change materials, low melting temperature alloys, machine learning, transformer model, thermal management

Other note

Citation

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-202510228523

Collections

[dipl] Kemian tekniikan korkeakoulu / CHEM