Robust continuous liquid-level monitoring for energy applications using a heater-assisted optical fiber and temporal convolutional networks

초록

Accurate liquid-level measurement is essential for the stability, efficiency, and safety of large-scale energy and process systems, including nuclear reactors, LNG storage tanks, and molten-salt tanks. However, rapid and reliable level sensing remains challenging in harsh or inaccessible environments. This study proposes a heater-assisted liquid-level measurement system based on distributed optical fiber temperature sensing, in which adjacent heater wires generate phase-dependent thermal responses by exploiting the distinct convective heat-transfer behavior of liquid and air. The small heater-sensor spacing enables a fast thermal response, allowing stable interface detection within about 5 s at a nominal power of 5 W and within about 25 s at a reduced power of 2.5 W. To ensure robust detection during both transient and steady heating, two complementary preprocessing features (temporal and spatial) are used as inputs to a Temporal Convolutional Network (TCN). The TCN classifies each spatial point as liquid or air, and the liquid level is then determined from the classified phase profile using a continuity-based post-processing rule. In static tests and an unseen dynamic draining test conducted in a room-temperature water–air system, the proposed approach achieved about 95 % point-wise phase classification accuracy; static level estimates showed mean absolute errors of 1.56–4.02 mm, while the draining test yielded MAE/RMSE values of 8.6/9.4 mm. The interface is sampled at 0.64 mm spatial intervals, and the flexible sensor structure supports installation in constrained geometries, making the proposed approach a scalable option for applications where conventional ultrasonic or visual measurements are impractical. © 2026

키워드