Evaluating generative models for inverse design of high-entropy refractory alloys

Generative modeling is an innovative new method to design functional materials. In this article, we quantify the performance of several generative modeling architectures – autoencoder, variational autoencoder, and generative adversarial network – to design novel materials. These are compared to rational design method by case study of refractory high-entropy alloys for ultra-high-temperature applications. Furthermore, we apply a series of validation methods to evaluate the models' effectiveness and express the current difficulty. Overall, cAE is able to create versatile compositions and keep at lower similarity; cVAE is capable of generating compositions with high accuracy; WcGAN has ability of creating novel compositions compared to training data. However, since the mode collapse phenomenon occurred in our results, inverse design still cannot totally replace rational design method at this stage.

Files

Metadata

Work Title Evaluating generative models for inverse design of high-entropy refractory alloys
Access
Open Access
Creators
  1. Yen Cheng Tung
  2. Arindam Debnath
  3. Wesley F. Reinhart
Keyword
  1. Machine learning
  2. Deep learning
  3. Generative model
  4. High-Entropy Alloys
License No Copyright - U.S.
Work Type Research Paper
Publication Date July 2022
Deposited October 10, 2023

Versions

Analytics

Collections

Work History

Version 1
published

  • Created
  • Updated
  • Updated Acknowledgments Show Changes
    Acknowledgments
    • Arindam Debnath, Wesley F. Reinhart
  • Added Creator Yen Cheng Tung
  • Added Tung_YenCheng_2022.pdf
  • Updated License Show Changes
    License
    • https://rightsstatements.org/page/NoC-US/1.0/
  • Published
  • Updated
  • Updated Acknowledgments Show Changes
    Acknowledgments
    • Arindam Debnath, Wesley F. Reinhart
  • Added Creator Arindam Debnath
  • Added Creator Wesley F. Reinhart
  • Updated