High Accuracy and Cost-Effective Fiber Optic Liquid Level Sensing System Based on Deep Neural Network.

Area of Science:

• Instrumentation and Measurement
• Artificial Intelligence
• Sensor Technology

Background:

• Traditional liquid level sensing systems face challenges in cost, accuracy, and monitoring multiple points simultaneously.
• Multiplexed sensor systems, while cost-effective, often suffer from signal overlap or crosstalk as liquid levels change.

Purpose of the Study:

• To propose a novel, cost-efficient liquid level sensing system with enhanced capacity and accuracy.
• To address the sensor crosstalk issue in multiplexed sensing systems.
• To develop a robust method for accurately predicting liquid levels at multiple points.

Main Methods:

• Development of a multiplexed liquid level sensing system utilizing a single sensor per measurement point.
• Implementation of a deep neural network (DNN) model to interpret sensor data and resolve signal overlap.
• Evaluation of the DNN model's performance against conventional machine learning algorithms like Random Forest (RF) and Support Vector Machines (SVM).

Main Results:

• The proposed DNN model accurately predicts liquid levels across different scenarios.
• The DNN approach effectively resolves the sensor crosstalk problem inherent in multiplexed systems.
• The DNN model demonstrated superior performance compared to RF and SVM for liquid level prediction.

Conclusions:

• The novel multiplexed liquid level sensing system, enhanced by a DNN, offers a cost-effective and accurate solution for multi-point monitoring.
• Deep neural networks provide a powerful tool for overcoming challenges in complex sensor data interpretation.
• This approach significantly improves the reliability and precision of liquid level sensing.