Machine learning analysis on nanomaterials literature data and knowledge exploration

Abstract

In this thesis study, regression- and classification-based machine learning analyzes were used to understand process-property and structure-property relationships of thin film materials. As an aspect of materials paradigm, two distinct efforts were comprehensively performed on poly(3-hexylthiophene-2,5-diyl) (P3HT) thin films, and aluminum doped zinc oxide (AZO) thin films. The extended automated machine learning workflows were used to find proper regression algorithm, and hyperparameters, that can demonstrate high prediction capability by coefficient of determination, 𝑅2. These models can be used as thin film design tools for screening promising design areas before real laboratory experiments to be performed.These models can improve the efficiency of laboratory studies, in terms of both reducing time and financials. In this way, the formed models has been named for two most critical challenges in scientific development. The study is also reducing the reproducibility issue for both film structures.In this endeavor, a tree-based method and a hierarchical agglomerative clustering algorithm were both applied on the dataset compiled from published literature. According to the performance of nanomaterials, the clustering method generated performance classes (for example, field effect charge carrier mobility for transistors, electrical resistivity, energy bang gap characteristics etc.). After creating performance classes, tree-based algorithms looked for the most important process variables, conditions, or structural elements that could improve the performance of nanomaterials. The relationship between the inputs to the fabrication process, structural characteristics, and material properties can be defined in quantitative manner by the relevance each other. Sol-gel deposited aluminum doped zinc oxide films have been shown to be a promising and cost-effective transparent conductive oxide for achieving high performance in solar cells, transistors, and diodes due to their wide energy band gap and low electrical resistivity. The reasons for variations in energy band gap and electrical resistivity of aluminum doped zinc oxide films have been discussed as doping concentration, crystalline size crystallinity, film thickness, and lattice constants.