This tool can be used to train neural networks to predict properties of high entropy alloys. High entropy alloys are metal alloys with 4 or more metals present in significant percentages that are of interest due to their mechanical properties and tailorability to specific applications. However, the range of local atomic environments in HEAs makes it difficult to predict properties of atoms in the alloys. Therefore, we present neural networks to predict properties of the CoCrCuFeNi-family of alloys, including the properties of relaxed vacancy formation energy, cohesive energy, pressure, and volume. Inputs to the model are bispectrum coefficients and central atom descriptors. We test the models ability to predict on the system it was trained on and see how well it can predict properties of systems with different compositions than the training system.

Python, Keras, Tensorflow, Pymatgen

Researchers should cite this work as follows:

• Mackinzie S Farnell, Zachary D McClure, Alejandro Strachan (2021), "Machine learning for high entropy atomic properties," https://nanohub.org/resources/mlatomprop. (DOI: 10.21981/FMF8-AM68).

  BibTex | EndNote

1. high entropy alloys
2. machine learning
3. neural networks