Fiber-Optic Liquid Level Sensor Based on a Peanut-Shaped Mach-Zehnder Interferometer and Random Forest Network

Abstract Liquid level monitoring ensures precise control of liquid level changes, which is crucial for preventing accidents, ensuring production safety, and maintaining stable system operation. In this work, we theoretically proposed and experimentally demonstrated the integration of a Mach-Zehnder Interferometer (MZI) based on cascaded peanut-shaped fiber structures with a Random Forest network for liquid level monitoring. Because of the phase difference arising during the transmission process between core mode and cladding modes, changes in the liquid level sensed by the sensor result in corresponding changes in the phase difference, subsequently causing a shift in the transmission spectrum. The experimental results show that within the range of 0 mm to 25 mm, the sensitivity of the sensor is 0.084 nm/mm, with a surprisingly high Mean Squared Error (MSE) of 3.826. The significant error limits its practical application potential. To improve upon the high MSE, a Random Forest network was introduced for data optimization and training. By constructing multiple decision trees and using their predictions to obtain a final liquid level prediction through voting or averaging, the accuracy and robustness of the model were enhanced, improving the MSE to 0.000355, which represents a reduction in error by over 10,000 times. The designed sensing system holds promise for potential applications in soil moisture systems and liquid level detection in tank trucks. Additionally, the Random Forest network employed in this system demonstrates universal applicability within point-type optical fiber sensing systems which is possible to be widely utilized.