Machine Learning-Based Fault Detection in Refrigeration Chambers Using Multi-Sensor Vibration Monitoring

Predictive maintenance has become increasingly relevant in modern engineering due to the demand for reliability and cost reduction in industrial systems. Vibration analysis plays a crucial role in detecting mechanical faults in rotating equipment, and its application to cold storage chambers ensures operational efficiency and food safety. This paper presents an experimental and computational study on vibration monitoring of refrigeration systems, integrating traditional signal processing techniques with artificial intelligence for fault detection. Similar approaches have been applied to industrial refrigeration systems using machine learning-based fault diagnosis [1]. Vibration and temperature data were collected from a cold chamber at IFPE, using accelerometers and a National Instruments CompactDAQ platform. A custom data-acquisition board based on an Arduino was also developed for thermal monitoring and door-cycle control. Several fault conditions were induced, including unbalance, condenser obstruction, lubrication deficiency, refrigerant leakage, and evaporator icing. A machine learning pipeline in Python was implemented for preprocessing, feature extraction (time and frequency domains), feature selection, and model training. Results show that treebased models, particularly XGBoost, achieved outstanding performance, with 98% accuracy and an ROC-AUC of 0.999 in fault classification. These findings confirm the feasibility of combining vibration analysis with machine learning for predictive maintenance in refrigeration systems, supporting the transition towards Industry 4.0.