Enhancing Predictive Maintenance Accuracy Through Edge Computing and Digital Twins for Industrial Motors

ABSTRACT

Machinery and equipment failures in the industrial sector often lead to significant losses in time and materials. To mitigate these losses, predictive maintenance processes are essential for identifying and preventing potential breakdowns. This paper proposes a model that integrates edge computing and digital twin technology to enable autonomous, efficient predictive maintenance of industrial electrical motors. The formulated model consists of three principal components: a custom motor testing kit equipped with an accelerometer for vibration data acquisition, a synthesis of edge computing and wireless communication for a digital twin prototype and the application of machine learning and federated learning to enhance the digital twin. Data from the testing environment is transmitted to a Raspberry Pi (an edge device) and to a central server. A Random Forest classification model, trained on sensor data, is implemented on the server. Through federated learning, updated model parameters are sent to the edge device, which adjusts its local model using live sensor data and returns the refined parameters to the server. This refined model is then deployed to a shared digital twin framework. To address sensor noise, a Simple Moving Average Filter was employed, which, when combined with the Random Forest Algorithm, achieved a notable success rate of 89 percent. This study successfully demonstrates the application of edge computing, federated learning and digital twins in predicting equipment failures using sensor data.