Accelerated Composites Design through Machine Learning

In recent years, we have witnessed several pioneering advancements in the fields of machine learning (ML) and neural network (NN). Typically, a neural network can be characterized as a series of algorithms that can recognize the underlying relationships between a set of data accurately through a process that mimics the way the human brain operates. The accuracy of the NN depends especially on the accuracy of the available training data. Training data are extracted from physically realistic models of a system or process with different degrees of complexity. The idea of coupling computational physics and NN is fairly new and bears great promise.

Since my university does not offer a course in machine learning, I learnt the basics of machine learning through Coursera in 2019. Also Satyajit Mojumder of Northwestern University, USA provided me reading materials on deep neural network. Under his supervision, I undertook a project that uses deep multi-fidelity physics informed neural network to accelerate molecular dynamics simulations' predictive capability. The work has been accepted in Computational Material Science (Elsevier) and is in production. Feel free to check the paper and the codes.