Learning physical properties of liquid crystals with deep convolutional neural networks

Area of Science:

• Physical Sciences
• Materials Science
• Data Science

Background:

• Machine learning (ML) algorithms, developed in the 1990s, are increasingly applied in physical sciences.
• Their use in liquid crystal (LC) research remains limited despite optical imaging's prevalence in the field.
• ML, particularly deep learning with image analysis, has seen significant advancements.

Purpose of the Study:

• To investigate the application of convolutional neural networks (CNNs) for analyzing liquid crystal optical images.
• To predict physical properties of liquid crystals directly from images without manual feature engineering.
• To assess the accuracy of CNNs in identifying LC phases and predicting parameters.

Main Methods:

• Utilized convolutional neural networks (CNNs), a type of deep learning model.
• Applied CNNs to optical images of liquid crystals.
• Optimized simple CNN architectures for predictive tasks.
• No manual feature engineering was performed.

Main Results:

• CNNs accurately predicted physical properties of liquid crystals from optical images.
• Deep neural networks achieved near-perfect accuracy in identifying liquid crystal phases and predicting the order parameter for nematic LCs.
• CNNs precisely determined the pitch length of cholesteric liquid crystals and the temperature of an experimental liquid crystal sample.

Conclusions:

• Convolutional neural networks offer a powerful, data-driven approach for analyzing liquid crystal optical images.
• This method enables accurate prediction of various physical properties, simplifying research protocols.
• The findings suggest a promising future for ML in advancing liquid crystal science.