Prediction of Enthalpy of Mixing of Binary Alloys Based on Machine Learning and CALPHAD Assessments

Prediction of Enthalpy of Mixing of Binary Alloys Based on Machine Learning and CALPHAD Assessments

The enthalpy of mixing, a critical thermodynamic property in the liquid phase reflecting element interaction strength and pivotal for studying phase equilibria, can now be predicted efficiently using machine learning. This study proposes a model combining machine learning with the Calculation of Pha...

Full description

Saved in:
Bibliographic Details
Main Authors: Shuangying Huang, Guangyu Wang, Zhanmin Cao
Format: Article
Language:English
Published: MDPI AG 2025-04-01
Series:Metals
Subjects:
machine learning
CALPHAD
enthalpy of mixing
thermodynamic properties
binary alloys
Online Access:https://www.mdpi.com/2075-4701/15/5/480
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The enthalpy of mixing, a critical thermodynamic property in the liquid phase reflecting element interaction strength and pivotal for studying phase equilibria, can now be predicted efficiently using machine learning. This study proposes a model combining machine learning with the Calculation of Phase Diagram (CALPHAD) to predict the enthalpy of mixing. We obtained data for 583 binary alloy systems from the SGTE database, ensuring experimental constraints for accuracy. Using pure element properties and Miedema’s model parameters as descriptors, we trained and evaluated four machine learning algorithms, finding LightGBM to perform best (R<sup>2</sup> = 92.2%, MAE = 3.5 kJ/mol). The model performance was further optimized through Recursive Feature Elimination (RFE) and Maximal Information Coefficient (MIC) feature selection methods. Shapley Additive Explanations reveals that the primary factors affecting the mixing enthalpy, such as atomic radius and electronegativity, align with the key parameters of the Miedema model, thereby confirming the physical interpretability of our data-driven approach. This work offers an accelerated method for predicting complex multi-component system thermodynamics. Future research will focus on collecting more high-quality data to enhance model accuracy and generalization.
ISSN:2075-4701

Similar Items