Anomaly Detection in Cold Storage Systems: A Machine Learning Approach for Fault Diagnosis

Cold storage systems play a crucial role in preserving temperature-sensitive goods. However, they are susceptible to various faults that can compromise operational efficiency and product safety. Traditional rule-based fault detection methods are limited by their rigidity and lack of adaptability. In contrast, this study introduces a machine learning (ML)-based framework for anomaly detection and fault diagnosis in cold storage environments. The proposed framework combines autoencoders for unsupervised anomaly detection with gradient boosting classifiers for supervised fault categorization. It addresses key challenges such as data imbalance, temporal drift, and sensor noise. Experimental results on an industrial cold storage dataset show that the framework achieves high fault detection accuracy, reduced false alarm rates, and strong generalization to unseen anomalies. These findings demonstrate the effectiveness of ML approaches in enabling proactive and scalable fault diagnosis in cold storage systems.