Development of a Remote Refrigerant Leakage Detection System Using Machine Learning Technique

Abstract

Reducing refrigerant leakage from refrigeration and air-conditioning equipment is one of the essential issues to solve the global warming problem. Many countries are enacting laws requiring owners of large refrigeration and air-conditioning equipment to carry out regular inspections for refrigerant leaks and to repair any leaks that are detected. There are two inspection methods: direct inspections using visual checks or a gas sensor leak detector, and indirect inspections using equipment operating data to detect leakages. However, large equipment has many inspection points, and manual inspection using the direct method is very time-consuming and labor-intensive, placing a heavy burden on both the equipment owner and inspector. Furthermore, when the leakage rate is small, it is difficult to detect leakage by direct method due to the limitation of sensor sensitivity, etc. On the other hand, many countries offer incentives such as exemption from inspections or halving the number of inspections by installing a leak detection system with continuous monitoring. The authors are developing a highly accurate and continuous refrigerant leakage detection system using indirect method based on machine learning techniques. In this paper, the developed leak detection method is applied to VRFs and chillers, and the evaluation shows that the detection performance can be significantly improved compared to the conventional method.