PyGGi software has been featured by ACerS news!
PyGGi software has been featured by ACerS news!
Our latest work on district-level COVID spread modeling in India featured in News. Check it out at PRACRITI
Machine learning software launched from M3RG to predict glass properties, PyGGi, featured in news. Do check it out at pyggi.iitd.ac.in
Research on realistic atomic structure of geopolymer gels chosen as Editor's pick in JCP! (link)
Research on the fracture of nanoscale phase-separated glasses chosen as Editor's suggestion in Phys. Rev. Materials! (link)
Research on the effect radiation on quartz was featured in news article! (http://physicsworld.com/cws/article/news/2017/may/29/solid-becomes-liquid-like-when-irradiated; https://www.eurekalert.org/pub_releases/2017-05/aiop-aso051817.php; https://phys.org/news/2017-05-atomic-irradiated-materials-akin-liquid.html; http://www.azom.com/news.aspx?newsID=47734
Research in M3RG is at the intersection of computational modeling, scientific machine learning, materials, and mechanics. The main research areas we work on currently are as follows.
Machine-learning aided materials design: In materials, understanding and predicting the composition–structure–property relationship is the key to developing novel materials. Such predictions are typically hindered by the complex physics happening at different length and time scales, along with the large number of structural and compositional arrangements possible. As an alternative route, data-driven approaches such as machine learning can prove key to predict structure and composition of materials for tailored applications. The aim of this research is to rely on the large database available in the literature from previous experiments and simulations to design and test new compositions and structures of materials for targeted applications. We have launched a software package Python for Glass Genomics (PyGGi, http://pyggi.iitd.ac.in), which has the trained model for eight properties and growing!
Scientific natural language processing: Decades of research on materials is documented in the form of multiple modes of data, including text, tables, and figures in scientific articles. Extracting information from this literature to develop a large knowledge base of materials is the key towards consolidating the domain knowledge. This involves several challenges as the domain specific jargons and the usage is quite different from the general english literature. Our research aims to address these challenges by developing datasets, benchmarks, and models for materials domain that enable information extraction. As part of this research, we developed the first materials domain language model, namely, MatSciBERT and also the first general purpose model for information extraction from materials tables, namely, DiSCoMaT. More such models are in pipeline!
Graph neural networks for materials modeling: Materials are made of atoms. Learning the dynamics of atoms is at the heart of understanding materials response. This involves several challenges such as developing an efficient and transferable representation for the atomic systems, learning their dynamics happening at multiple length and time scales, to name a few. To this extent, we have developed several novel architectures such as Lagrangian and Hamiltonian graph neural networks, that can learn the dynamics directly from the trajectory. We also developed benchmarking dataset that analyzed equivariant graph architectures that can model atomic systems. Based on this approach, our research is aimed at learning the loss and energy landscapes of these systems. This, in turn, could be used to develop models that are generalizable to unseen temperatures, pressures, and compositions.
Machine learning for cement manufacturing: Cement is the second most manufactured material after water. Cement is also one of the largest emittor of green house gases. This process aims to employ artificial intelligence for automating industrial scale cement production. Specifically, we use historical data from cement plants including raw materials, fuels, process parameters, emissions, and the clinker quality to develop an end-to-end model that can mimic the cement plant functioning. Through this model, we aim to optimize the cement manufacturing to obtain target quality with controlled emission.
FUNDING
1. INSPIRE Faculty award for the project titled, "Atomic and mesoscale modeling of radiation-damage in concrete", Department of Science and Technology, India: 31/10/2017–30/10/2022 (INR 3,500,000).
2. SERB Early Career Researcher Award for project titled, "Elucidating Composition-Structure-Property Relationships in Silicate Glasses for Fracture- and Scratch-Resistant Applications", Department of Science and Technology, India: 15/03/2019–14/03/2022 (INR 5,000,000).
3. SPARC project titled, "Design and Manufacturing of Ultra-high Velocity Bullet and Fragments Resistant Armor" with Indian PI: Prof. Naresh Bhatnagar (ME, IIT Delhi), Indian co-PI: Prof. N. M. Anoop Krishnan, International PI: Prof. Subramaniam D. Rajan (Arizona State University), International co-PI: Dr. Ashok Bhatnagar (Honeywell International), 15/03/2019–14/03/2021 (INR 10,000,000).