Open Access

Niima Es-sakali, Zineb Zoubir, Samir Idrissi Kaitouni, Mohamed Oualid Mghazli, Moha Cherkaoui, Jens Pfafferott

• Variable refrigerant flow (VRF) systems are constantly prone to failures during their lifespan, causing breakdowns, high energy bills, and indoor discomfort. In addition to correctly identifying these defects, fault detection, and diagnostic studies should be able to anticipate and predict the anomalies before they occur for efficient maintenance. Therefore, this study introduces an efficientVariable refrigerant flow (VRF) systems are constantly prone to failures during their lifespan, causing breakdowns, high energy bills, and indoor discomfort. In addition to correctly identifying these defects, fault detection, and diagnostic studies should be able to anticipate and predict the anomalies before they occur for efficient maintenance. Therefore, this study introduces an efficient self-learning predictive maintenance system, CACMMS (Cloud Air Conditioning Monitoring & Management System), designed to anticipate refrigerant leaks in VRF systems. Unlike previous efforts, this system leverages advanced fault detection and diagnosis strategies in a real existing building to enhance prediction accuracy. The study employed three noise filtering models (Kalman filter, moving average, S-G smoothing) in the preprocessing phase. Ten features were selected for assessment, and four machine learning models (decision tree, random forest, K-nearest neighbor, support vector machine) were compared. The accuracy, precision, sensitivity, computation time as well and confusion matrix were used as performance indicators and metrics to evaluate and choose the best performant model. Results indicated that decision tree and random forest models achieved over 95 % accuracy with execution times between 0.70 s and 3.32 s, outperforming K-nearest neighbor and support vector machine models. These findings highlight the system’s potential to reduce downtime and energy costs through effective predictive maintenance.…