CDI Project: Physics-Informed Deep Learning Model for Data-Driven Prediction and Discovery
PI: Malachi Schram
Project Team: Jenna Pope, Jan Strube, Robert Rallo
Project Term: October to September 2021
Project Description: This project aims to develop and apply machine learning (ML) techniques with domain aware models to improve performance, robustness, and reduce the need for large datasets. The project has worked on three efforts:
- use of recurrent neural network (RNN) to predict the evolution of multiple species resulting from a series of interconnected iodine reactions
- use of data-driven deep-learning model to classify and model the additive manufacturing build process
- use of methods to optimize the inter-atomic mixture to maximize the heat conductivity for materials at interfaces
Under normal operation of a nuclear power plant, little iodine is released into the environment. However, if an incident occurs, such as the incident at the Fukushima Daiichi Nuclear Power Plant in 2011, radioiodine can be released into the environment. We study the use of RNNs to address the chemical kinetics problem of predicting the chemical evolution of species as a reaction progresses. The project team combines kinetic models from two systems previously published in the literature to examine the ability of RNNs to predict the chemical evolution of iodine species in a supposed nuclear reactor incident.
The advent of additive manufacturing processes brought with it intense research into various materials and manufacturing processes. At the same time, the need for validation of material properties and study and forecasting of aging has arisen. Modern imaging techniques like X-ray computed tomography are a convenient vehicle for such studies. Still, the large data sets they produce a call for novel analysis techniques that can extract the required information efficiently. Deep Learning techniques promise to deliver the necessary throughput. We have developed a 3D extension of the ResNet architecture for image classification and studied its performance on a simple classification task, using X-ray computed tomography data as input.
In material science, we are frequently interested in understanding the properties and design implications of material at interfaces. These interfaces can be manipulated to improve the desired characteristics of the bulk material. In this study, we are interested in understanding and optimize the impact of interfacial atomic defects on the thermal transport across a Cu/Si junction. To that end, we developed a learning-based framework to optimize over a potentially large parameter search. Using this technique allows us to accumulate knowledge of the system when applying actions to the current state of the system. In this study, we are focusing on optimizing the thermal transport by varying the fraction and length of the interfacial atomic defects using molecular dynamics (MD) simulations.