Real-time IoT data analysis for HVAC system maintenance

Abstract IoT-enabled heating, ventilation, and air-conditioning (HVAC) control systems are widely used in buildings. Consequently, the need for predictive maintenance processes has increased to enhance system efficiency and prevent failures. This study was conducted to improve the efficiency and reliability of HVAC systems by transitioning from traditional time-based maintenance approaches to AI-based predictive maintenance. The primary research question focuses on the effectiveness of machine learning techniques in detecting anomalies and predicting failures in HVAC systems using real-time IoT sensor data. The aim was to develop a proactive maintenance framework capable of identifying and addressing potential faults before they impact system performance. The experiment is conducted in an indoor environment at Tezpur University, utilizing a Raspberry Pi 4B, NodeMCU, DHT11, and BMP280 sensors. Real-time data were transmitted using the MQTT protocol and processed using Node-RED. The Isolation Forest and Random Forest models were trained on the collected sensor data to detect anomalies and predict failures. The results confirmed that both models effectively identified anomalies and forecasted HVAC failures, with the Random Forest model achieving high accuracy, precision, and recall. These findings validated the hypothesis that machine learning can enable proactive HVAC maintenance. Temperature and pressure were identified as the most significant indicators of system health. The results align with the existing literature and demonstrate the value of integrating AI with IoT-based building management. Future work may explore the use of additional sensor inputs, deep learning models, and explainable AI techniques to further enhance the prediction capabilities and system transparency.