Skip to main content

Materials Discovery Through Supervised Machine Learning on Optical Properties

While computation has made it much easier to find characteristic properties of given systems, reversing this workflow to predict feasible materials and material structures with specific desired properties (i.e., "inverse design") remains difficult. However, being able to do so will increase the currently dragging velocity of material discovery, narrowing areas of exploration to more potent structural arrangements. In this project, we aim to utilize supervised machine learning techniques to generate materials corresponding to desired optical properties. We will use COMSOL Multiphysics software to first generate layered structures by varying material composition, layer thickness, and layer order. We then will use built-in Maxwell solvers to determine system absorption, reflection, and transmission spectra. From this, we will work to produce an algorithm using a quantitative structure-property relationship (QSPR) machine learning algorithms in Python to predict materials from desired material properties. Successful completion of this work will result in a workflow that can utilize a picture of desired absorption, transmission, or reflection spectra to generate a material, which will advance material discovery and design.

Aditi Munshi
Grinnell College
Computer Science
Research Advisor: 
Dr. Andrew Ferguson
Department of Research Advisor: 
Materials Science & Engineering
Year of Publication: