IoT-Enabled Real-Time Condition Monitoring of Electrical Machines Using Predictive Analytics

Abstract

Functional performance of electrical machines is critical towards uninterrupted productivity in industrial, manufacturing, energy generation, and transportation systems whose unforeseen failure may result into huge economic loss and accidents. This paper presents an end-to-end IoT enabled real-time condition monitoring system with high-resolution sensor networks, edge computing, and predictive analytics to identify faults at early stage and estimate Remaining Useful Life (RUL) of electrical machines. The proposed system uses multi-modal sensing, i.e., combination of the measurements of vibration, temperature, acoustic, and electrical signals by employing embedded IoT nodes that have the robustness in signal measurement over the changing operation and environmental conditions. The noise filtering and feature extraction both over time and frequency domains along with local anomaly detection models are used in pre-processing the acquired data at the edge, where latency is minimized and bandwidth utilization optimized. The implementation of a hybrid edge-cloud architecture is also proposed, namely, initial diagnostics at an edge level and the use of cloud-hosted deep learning models performing high-level fault classification and RUL prediction (e.g., CNN-LSTM hybrids and LSTM regression networks). The validity of the framework was executed on a controlled testbed consisting of three- phase induction motor exposed to typical fault conditions, such as bearing fault, rotor imbalance, and stator winding insulation fault, at varying load and speed. The experiment shows fault detection classification accuracy surpassing 95 per cent, RUL mean absolute error of 5.2 and a decrease in mean fault detection latency, by 2.3 to 0.45 seconds between the edge assisted and cloud-only models. Also, the system allowed a 35 percent less unplanned downtime over the traditional reactive approaches in maintenance. This proposed solution provides a scalable, cost effective and interoperable solution that could be used on Industry 4.0 predictive maintenance ecosystems and could apply in smart factories, energy infrastructure, and transportation system. The research helps fill the gap between existing industrial IoT solutions that are meant to work in real-time and sophisticated predictive analytics solutions used to monitor the health of electrical machines.