Real-Time Fan Anomaly Detection Using Embedded Machine Learning

Currently, detecting anomalies in electrical devices, particularly electric fans, is crucial. Ensuring the stable operation of industrial electric fans helps early prevention of electronic system failures, thereby saving repair costs before the issues become more severe. This study has developed an early fault detection method for industrial electric fans using machine learning techniques, which can be deployed on resource-limited embedded devices. The proposed system monitors vibration and current consumption through accelerometers and sensors, employing a machine-learning model to evaluate anomalies. Unlike conventional machine learning models, the results show that applying Convolutional Neural Networks (CNN) and Deep Neural Networks (DNN) significantly increases accuracy without requiring extensive feature extraction. Additionally, due to the lack of data on abnormal and normal states, we created a fan model in the laboratory to collect data for up to 7 hours. The trained model achieved an accuracy of 98.3% with six labels, Fan On, Fan Off, Fan Tapping, Fan Broken Wings, Fan Stuck, and Fan Unstable, on the self-collected dataset.