Machine learning for fault detection in refrigeration systems using difluoromethane refrigerant

Abstract

Refrigeration systems are vital to residential and industrial cooling applications; however, faults such as refrigerant leakage and air filter clogging can severely compromise energy efficiency and reduce equipment lifespan. This study explores the potential of four machine learning (ML) models—Naïve Bayes, Generalized Linear Model (GLM), Decision Tree and Random Forest—for detecting two prevalent fault types in systems using difluoromethane (R32) as the working fluid. A synthetic dataset comprising 1998 samples was developed in Python, simulating normal operation alongside refrigerant leakage and filter clogging scenarios, based on the technical characteristics of R32 air-conditioning systems. Feature engineering and statistical visualization techniques were employed to enhance classification accuracy. All models demonstrated high predictive performance (accuracy >96 %), with Naïve Bayes achieving 100 % accuracy, indicating potential overfitting. Decision Tree and Random Forest models maintained strong generalization capabilities, with accuracies of 97.9 % and 97.4 %, respectively, suggesting practical applicability in real-time fault diagnosis. The proposed approach enables early detection of operational issues, thereby reducing energy losses and extending system service life. This research aligns with the United Nations Sustainable Development Goals (SDGs), particularly SDG 7 (Affordable and Clean Energy), SDG 9 (Industry, Innovation and Infrastructure) and SDG 12 (Responsible Consumption and Production), by fostering intelligent and sustainable solutions for residential cooling technologies.