Early Bearing Fault Detection of Industrial Fans based on Vibration Analysis using Interpretable Deep Learning Framework

The global market of machine condition monitoring is projected to grow at a rate of 8.3% in next five years. The recent technological advancement in IoTs and AI has driven predictive maintenance to be one of the most effective approach in this domain. Vibration analysis is the most efficient technique used to carry out predictive maintenance, using advanced data- driven intelligent approaches. The vibration data carries most significant information regarding the health state of machine components especially bearings. The proposed hybrid framework utilizes CNN and transformer utilizing the complex weightsharing capabilities of CNNs, combined with ability of transformer to capture the broader context of spatial relationships in large-scale patterns, making it suitable for datasets of varying sizes. A fault detection accuracy of 98.86% is achieved through experimentation on a run-to-failure real-industrial environment dataset composed of vibration data of large-scale coaxial fans.