Refrigerant leakage detection in building heat pump systems based on dynamic modeling and sensitivity parameters

Abstract

Refrigerant leakage, a common problem in building heat pump systems, reduces operational efficiency, increases energy consumption, and raises greenhouse gas emissions, contributing to environmental degradation and energy loss. This research developed a dynamic modeling simulation platform for building heat pump systems, focusing on Water-to-Water Heat Pumps (WWHP). The WWHP system simulation investigated variations in temperature, pressure, and capacity under normal operation and refrigerant leakage scenarios. To assess the model’s accuracy, an experimental setup involving a WWHP system was used to observe changes related to refrigerant charge and leakage. A Python-based program for refrigerant charging and leakage detection was developed by integrating the sensitivity of key parameters across systems with specific evaluation metrics. This simulation was validated using experimental data from the WWHP system, showing average relative errors in pressure and capacity of 4.12 % and 4.66 %, respectively, with an average temperature deviation of 2.3 °C when altering the refrigerant charge. Under leakage conditions, the pressure and temperature values were 7.27 % and 2.7 °C, respectively. When replacing refrigerant R134a with R513A, the relative errors remained at the same level. The simulation identified 12.2 kg as the optimal refrigerant charge under standard test conditions, noting that undercharging reduces capacity while overcharging has minimal impact due to the accumulator’s presence. The high-temperature and high-pressure sections in the WWHP system exhibited significant sensitivity. Although the detection program successfully identified the charge state using generated datasets, it failed to detect microleakage due to the brief duration of these datasets. Detection efficacy was affected by small leakage sizes and low operating pressures and temperatures, with an average detection time of approximately 26 min.