EdgeRopeNet: Lightweight Neural Network for Real-Time Wire Rope Tension Monitoring Using FBG Sensors in Edge-Fog Mining Systems

Abstract

Wire rope tension monitoring in mining hoist systems demands real-time, high-accuracy detection to mitigate catastrophic failure risks, yet existing cloud-based solutions remain constrained by 300–800 ms latency and network dependence, and conventional FBG sensing lacks embedded intelligence at the edge. To address these limitations, EdgeRopeNet utilizes a compact GRU-based neural architecture with two dense layers (64 and 32 neurons) deployed on Raspberry Pi 4 edge devices (4 GB RAM), supported by fog-layer aggregation on Intel i7 hardware. Sensor data from FBG arrays undergo Savitzky–Golay filtering and Min–Max normalization prior to inference, enabling 19 ms real-time latency and 97.8% prediction accuracy on synthetic datasets emulating mining shaft dynamics. Performance was rigorously benchmarked against ten baselines: five traditional models (Linear Regression, SVM, Random Forest, k-NN, Naïve Bayes) and five deep learning methods (CNN, LSTM, GRU, CNN–LSTM hybrid, Transformer) sing an 80:20 train–test split across 100 epochs with Adam optimization. EdgeRopeNet delivered 97.8% accuracy, 97.4% precision, 98.1% recall, a 97.7% F1-score, and MAE of 0.012, surpassing CNN–LSTM (95.2% accuracy, MAE 0.029) and Transformer models (96.1% accuracy, MAE 0.023). Parameter-pruning reduced model size by 60% while preserving 97.4% precision and 98.1% recall, with edge inference sustained at 0.019 seconds per prediction. Overall, EdgeRopeNet achieves a 94% reduction in latency relative to cloud-based platforms while maintaining superior accuracy, providing a scalable, autonomous, and edge-resilient solution for safety-critical mining infrastructure. Keywords: Edge computing, wire rope tension monitoring, FBG sensors, lightweight neural networks, mining hoist systems, real-time calibration.