Deep Learning in Materials Science

Our research leverages data-driven modeling and deep learning to accelerate the understanding and prediction of complex material behaviors across multiple length scales. Using molecular dynamics simulations combined with multi-fidelity neural networks, we extract material properties with enhanced speed and accuracy, significantly reducing computational cost. Machine learning models are further trained to predict the mechanical responses of 2D transition-metal dichalcogenides, supporting their use in next-generation flexible electronics. Additionally, MD-driven analysis of Leidenfrost phenomena provides insights into liquid thin-film phase transformations. Together, these efforts demonstrate how deep learning can transform materials science by enabling rapid, reliable predictions and guiding the design of advanced functional materials. 

Related Publications:

1. M. D. Rony, M. A. Islam, M. S. H. Thakur, M. Islam, M. N. Hasan, Molecular dynamics data-driven study of Leiden frost phenomena in context to liquid thin film phase transformation. International Journal of Heat and Mass Transfer 209, 124107 (2023). 

2. P. Malakar, M. S. H. Thakur, S. M. Nahid, M. M. Islam, Data-Driven Machine Learning to Predict Mechanical Properties of Monolayer Transition-Metal Dichalcogenides for Applications in Flexible Electronics. ACS Applied Nano Materials 5, 16489–16499 (2022). 

3. M. Islam, M. S. H. Thakur, S. Mojumder, M. N. Hasan, Extraction of material properties through multi-fidelity deep learning from molecular dynamics simulation. Computational Materials Science 188, 110187 (2021).