Edge-Based Real-Time Condition Monitoring System for Electrical Machines Using Machine Learning

Abstract

In industrial automation systems, electrical machines are essential resources, failure of which results in expensive downtime, safety and poor efficiency. Predictive maintenance plans have consequently become a necessary tool to detect fault in advance and reduce reliability. Yet, the traditional centralized and cloud-based systems have a critical factor of limitation, such as communication lag, reliance on bandwidth, and cybersecurity, as well as slow decision-making, which limits their use in real-time industrial settings. To solve these issues, the current paper will suggest a condition monitoring framework of electrical machines based on edges, which facilitates fault diagnosis on-device at a significantly low cost in terms of computation. The suggested system combines multi-sensor signal capture, time-frequency systematic preprocessing, and powerful feature extraction, then involving the classical machine learning classifiers, such as, Support Vector Machine (SVM) and Random Forest (RF), to classify the faults. High diagnostic performance is experimentally proven under varied operating conditions with high accuracy and F1-score and low inference latency to make it be applicable to real-time deployment. Further implementation of edges has confirmed that it will save power consumption and small memory usage, which support the potential of low-cost embedded platforms to implement industrial monitoring. The paper confirms a practical scalable and efficiently energy consuming energy solution to smart condition monitoring in current electrical automation systems using a combination of classical machine learning and effective feature engineering and edge computing has been developed.