Edge-based predictive maintenance for industrial electrical equipment using IoT sensors

Abstract

Electrical industrial machinery's unexpected breakdown significantly influences modern manufacturing, threatening operational convenience and revenue. Although cloud-based predictive maintenance is generally the initial proposed solution, it often falls short in effective, real-time failure detection due to latency bottlenecks, bandwidth limitations, and data privacy issues. To address these problems, this study proposes an edge-based IoT predictive maintenance framework for industrial electrical equipment. Hence, a Random Forest classifier was trained using three modalities of sensor data (vibration, current, and voltage measurements) and its inference was tested on emulated edge environments. Results show that the proposed model can identify faulty motors with an accuracy of 95.33%. Furthermore, inference at the edge decreases latency to an average of 0.4658 ms. Finally, compared to traditional cloud infrastructure, the proposed solution reduces bandwidth consumption by a factor of 36.