| Details: |
The traditional trial-and-error experimental approach to materials
design is time consuming, expensive and uncertain. Over the past couple
of decades, rapid progress in computational hardware and software have
helped accelerate materials design and discovery. High throughput
computing strategy has been particularly helpful in these efforts. Yet,
a comprehensive search of the chemical space to identify materials with
a desired set of properties is highly computation-intensive, and is time
consuming.
In recent years, Machine Learning methods have emerged as an alternative
to expensive first principles calculations. We have adopted two
strategies to use ML for search and design of useful materials. In the
first strategy, available materials data is used to train a series of ML
models to predict material properties. These models are used to
hierarchically screen a set of new materials to identify the ones that
satisfy all the desired properties. The small set of materials
identified via this screening procedure is tested in DFT. Materials
passing the DFT test are proposed as new candidate materials with
specific functionalities. Success in designing new rare earth free
permanent magnets will be presented.
The second strategy is to ‘inverse’ design materials with a set of
target properties using generative artificial intelligence (AI). We have
used two different generative AI models: constrained variational
auto-encoder (cVAE), and a diffusion-based model, to design novel
magnetic materials. Success of the property-embedded cVAE over previous
models, and that of the diffusion-based model over the cVAE in
generating novel magnetic materials will be highlighted.
[1] A. Dutta and P. Sen, J Mater. Chem. C 10, 3404 (2020).
[2] S. Mal and P. Sen, Leveraging available data for efficient
exploration of materials space using Machine Learning: A case study for
identifying rare earth-free permanent magnets J Magn. Mag. Mater 589,
171590 (2024).
[3] S. Mal, G. Seal and P. Sen, MagGen: A Graph-aided deep generative
model for inverse design of permanent magnets, J Phys. Chem. Lett. 15,
3221 (2024).
[4] S. Mal, S. Mishra and P. Sen, Diffusion-based model for inverse
design of novel magnetic materials, (in preparation). |