Machine Learning Algorithms for Liquid Crystal-Based Sensors.

Area of Science:

• Materials Science
• Chemical Sensing
• Machine Learning Applications

Background:

• Liquid crystal (LC) based sensors offer potential for chemical detection.
• Optimizing specificity and speed in LC sensors remains a challenge.
• Traditional analysis methods for LC responses have limitations in accuracy.

Purpose of the Study:

• To develop a machine learning (ML) framework for optimizing LC chemical sensor performance.
• To improve the specificity and speed of chemical detection using LC sensors.
• To demonstrate the capability of ML in extracting valuable feature information from LC responses.

Main Methods:

• Utilized ML techniques to analyze optical responses from surface-driven LC orientational transitions.
• Trained ML classifiers using feature information from thousands of optical micrograph images.
• Designed an experimental LC system to differentiate between DMMP and RH analytes.

Main Results:

• Achieved classification accuracies exceeding 99% using the ML framework.
• Demonstrated that traditional feature analysis yielded only 60% accuracy.
• Showcased the ability to achieve high accuracy with early time snapshots for fast sensing.
• Identified linear support vector machines as preferred classifiers.

Conclusions:

• ML frameworks can significantly enhance the accuracy and speed of LC-based chemical sensors.
• ML effectively extracts complex feature information from LC responses, outperforming traditional methods.
• The developed framework allows for systematic analysis of sensor data quality and noise filtering.
• Simultaneous exploitation of multiple feature sources is crucial for high-accuracy sensing.