Machine Learning with Unconventional Liquid Crystals

Abstract

This study explores the integration of supervised machine learning techniques with liquid crystals, contributing to the growing body of work focused on identifying liquid crystals through analysis of their textures. Three different datasets containing unconventional nematic liquid crystal phases were used to test the ability of machine learning models to identify the ferronematic and twist-bend nematic phases through experimentally obtained texture images. Two different machine learning architectures – a sequential Convolutional Neural Network (CNN), and parallel CNN were evaluated on their ability to identify these textures at varying levels of complexity, augmentation, and regularisation. In all cases it was found that flip augmentation was the only augmentation trialled to yield positive results with an acceptable level of accuracy. The inclusion of dropout regularisation led to lower accuracies in almost all cases aside from its use in the inception model for the largest dataset. The first dataset comprising of texture images of four phases of the NT3.5 compound achieved accuracies of 0.97±0.01 with a sequential CNN. The second dataset comprising of texture images of four phases of the CB6O.7 compound achieved accuracies of 0.984± 0.003 with a sequential CNN. The third dataset, which comprised of select nematic phases from both compounds achieved accuracies of 0.971±0.003 with a sequential CNN, and 0.987±0.003 with the parallel CNN model: InceptionV3. The results of this thesis contributes to the growing body of research using machine learning to classify liquid crystal phases, building on existing best practices and applying a general methodology to previously untested liquid crystal phases, and serves as a proof of concept. The inclusion of modified pre-built models broadens the accessibility and repeatability of a proven methodology that can be applied to a wide range of research areas.

Date of Award	13 Jun 2025
Original language	English
Awarding Institution	The University of Manchester
Supervisor	T. R. Wyatt (Co Supervisor) & Ingo Dierking (Main Supervisor)