Machine learning driven property prediction in MPEAs

Multi-phase MPEAs allow great flexibility in tailoring the microstructure to obtain targeted properties. At IDEAs Lab, we build machine learning models that are capable of predicting both linear and non-linear changes in mechanical properties of MPEAs that are associated with phase transitions and ordering tendencies. We validate these models using first-principle calculations (in collaboration with Ames Lab, USA) and experimental observations in a variety of systems. These data-driven models enables high-throughput exploration of the wide compositional spaces in MPEAs to enable a targeted alloy design approach.

Publications:

1. D Beniwal et al., Distilling physical origins of hardness in multi-principal element alloys directly from ensemble neural network models, npj Computational Materials, 8, 153 (2022).

Interpretation of machine learning models

The treatment of machine learning models as a black-box poses serious limitations on their usefulness as an exploratory tool. At IDEAs Lab, we are challenging this idea by building interpretation frameworks that can provide deep insights into the decision-making process of machine learning models. We have developed a Compositional-Stimulus Model-Response (CoSMoR) framework approach that utilizes local partial dependencies extracted from the model to provide "exact" contribution of each feature in the decision-making process along continuous composition pathways. This gives critical material-specific insights into how the target property manifests in a material as a function of composition. We are also developing methodologies to replace complex ML models with much simpler mathematical models that are fundamentally interpretable and thus lead to physical insights.

Publications:

1. D Beniwal & PK Ray, CoSMoR: Decoding decision-making process along continuous composition pathways in machine learning models trained for material properties, Physical Review Materials, 7 (2023) 043802.

2. D Beniwal & PK Ray, FCC vs. BCC phase selection in high-entropy alloys via simplified and interpretable reduction of machine learning models, Materialia, 26 (2022) 101632.

Quantitative phase fraction estimation in High-Entropy Alloys

Efficient alloy design requires knowledge of phase selection, phase-fractions and microstructure. With this in mind, we have developed a machine learning model at IDEAs Lab to predict the formation of FCC, BCC and Intermetallics (or a combination of these phases) in as-cast MPEAs. It goes a step further by also estimating the relative phase fractions of FCC, BCC and Intermetallic phases, and shows a good match with experimental and CALPHAD data. We explored three diverse MPEA systems using this model: Fex-Aly-(CoCr0.5Ni2.5)1-x-y, Alx-Tiy-(CrFeNi)1-x-y and Crx-Moy-(VNbTi)1-x-y, wherein the model correctly predicts both the phase boundaries as well as the non-linear variations in phase fractions as a function of composition.

Publications: 

1. D Beniwal and PK Ray, Learning phase selection and assemblages in High-Entropy Alloys through a stochastic ensemble-averaging model, Computational Materials Science, 197 (2021) 110647.