A Deployable Edge Computing Solution for Machine Condition Monitoring

Machine condition monitoring and predictive maintenance are critical components of smart manufacturing. The realization of machine predictive maintenance relies on synthetic integration of advanced sensing, Internet of Things (IoT), cloud computing, and Artificial Intelligence (AI)-enabled data analytics for machine condition diagnosis and prognosis. One challenge that hurdles the manufacturing shop floor from implementing predictive maintenance is its outdated hardware and limited internet bandwidth, which prevents the floor from transmitting all collected data to the cloud for centralized data analytics. One promising solution is edge-cloud computing and decision-making architecture. In such an infrastructure, collected sensing data will be first processed on edge to directly obtain machine diagnosis results (e.g., normal vs. abnormal), and only extracted features or selected raw data will be transmitted to the cloud for further analytics (e.g., remaining life prognosis). To improve the affordability and applicability of edge computing, this study proposed a novel and deployable edge device that adopts a feature-based Tiny Machine Learning (Tiny ML) model to achieve vibration data processing and machine condition monitoring while maintaining the modeling generalizability and computing efficiency. It integrates a Microelectromechanical Systems (MEMS) accelerometer for vibration data sampling (up to 20 kHz) and a microcontroller for edge computing and communication to the cloud. The integral IoT system, which involves the edge device, cloud services, and wireless communication between the edge and cloud is presented, emphasizing its plug-and-play capability. In this framework, MQTT protocol is adopted for communication among edge, cloud, and end users, allowing processed signal information and machine diagnosis results to be published to end users in real-time. Extensive experiments have been conducted in the context of industrial motor condition monitoring and fault diagnosis. High classification accuracy has been demonstrated by the device-deployable Tiny ML, comparable or even better than the integration of high-cost commercial sensors with more advanced ML analytics. Additionally, the edge device's durability and power efficiency are also evaluated.