Predicting Liquid Crystal Behavior with Artificial Neural Networks.

Area of Science:

• Materials Science
• Computational Physics
• Optoelectronics

Background:

• Liquid crystals (LCs) exhibit fluid and solid properties, crucial for displays and sensors.
• Accurate prediction of LC properties, such as polar angle and refractive index, is vital for device optimization.
• Traditional modeling methods are computationally intensive, necessitating efficient alternatives.

Purpose of the Study:

• To develop artificial neural networks (ANNs) for predicting LC mean steady-state polar angle and refractive index.
• To evaluate ANN performance against conventional simulation methods for LC modeling.
• To explore ANNs as a low-latency tool for optimizing LC-based technologies.

Main Methods:

• Artificial neural networks (ANNs) were trained to predict LC properties from surface viscosity and anchoring energy.
• The train, validation, test method was used to assess predictive accuracy.
• K-Fold cross-validation was employed to further validate ANN model performance.

Main Results:

• ANN models demonstrated high predictive accuracy, with ANN_A4 (R² = 0.9995) and ANN_B2 (R² = 0.9969) showing excellent results via the train, validation, test method.
• K-Fold cross-validation revealed different optimal models, with ANN_A5* (R² = 0.40767) and ANN_B4* (R² = 0.93799) performing best.
• ANNs offer significantly lower latency compared to traditional simulation techniques.

Conclusions:

• ANNs show substantial potential for modeling liquid crystals, providing accurate predictions of key optical properties.
• The low latency of ANNs makes them suitable for computationally intensive optimization tasks in LC technology.
• This study highlights ANNs as a powerful and efficient tool for advancing liquid crystal applications.