A primitive machine learning tool for the mechanical property prediction of multiple principal element alloys

TL;DR

This paper presents a basic machine learning tool that predicts multiple mechanical properties of multi-principal element alloys using curated data, a composition parser, and an interactive visualization interface, serving as an initial research workflow.

Contribution

It introduces a primitive ML model and a parser tool for MPEA composition data, along with an interactive visualization interface, providing a foundational workflow for future research.

Findings

ML models can predict MPEA mechanical properties

A parser converts alloy compositions into ML input format

An interactive interface visualizes model predictions

Abstract

Multi-principal element alloys (MPEAs) are produced by combining metallic elements in what is a diverse range of proportions. MPEAs reported to date have revealed promising performance due to their exceptional mechanical properties. Training a machine learning (ML) model on known performance data is a reasonable method to rationalise the complexity of composition dependent mechanical properties of MPEAs. This study utilises data from a specifically curated dataset, that contains information regarding six mechanical properties of MPEAs. A parser tool was introduced to convert chemical composition of alloys into the input format of the ML models, and a number of ML models were applied. Finally, Gradio was used to visualise the ML model predictions and to create a user-interactive interface. The ML model presented is an initial primitive model (as it does not factor in aspects such as MPEA production and processing route), however serves as a an initial user tool, whilst also providing a workflow for other researchers.