Real-Time Cell Gap Estimation in LC-Filled Devices Using Lightweight Neural Networks for Edge Deployment

Area of Science:

• Optoelectronics
• Materials Science
• Machine Learning

Background:

• Accurate liquid crystal (LC) cell gap measurement is crucial for optical device performance.
• Birefringent materials in LC cells distort transmission spectra, hindering traditional analysis.

Purpose of the Study:

• To develop a lightweight machine learning framework for estimating LC cell gap from transmission spectra.
• To enable accurate, non-destructive cell gap determination in filled LC cells.

Main Methods:

• A shallow multilayer perceptron (MLP) model was trained on experimentally acquired transmission spectra.
• Peak-to-peak interferometry provided ground truth cell gap values for training.
• Optimization algorithms, activation functions, and neuron configurations were systematically evaluated.

Main Results:

• The optimal MLP model achieved a correlation coefficient near 1 and root-mean-square error (RMSE) below 0.1 μm.
• The model demonstrated real-time inference on a Raspberry Pi 4 with low latency and resource consumption.
• Successful deployment validated the model for portable, edge-based applications.

Conclusions:

• A lightweight MLP framework provides accurate and efficient LC cell gap estimation from transmission spectra.
• The developed system is suitable for in situ diagnostics and quality control in LC-based optical applications.
• Edge-based deployment enables portable and real-time inspection capabilities.