Real-Time Anomaly Detection in Industrial IoT Systems Using Hybrid Deep Learning and Edge Computing

Abstract

Industrial Internet of Things (IIoT) systems generate vast volumes of sensor data in real-time, demanding immediate analysis to detect anomalies that could indicate faults or cybersecurity threats. Traditional cloud-centric models struggle with latency, bandwidth, and privacy issues. This paper proposes a hybrid deep learning framework deployed on edge computing devices for real-time anomaly detection. The system uses Convolutional Neural Networks (CNN) for feature extraction and Long Short-Term Memory (LSTM) networks for temporal pattern recognition. Experimental evaluation on a benchmark sensor dataset demonstrates a 97.4% detection accuracy with latency under 50ms per inference. The model significantly reduces network traffic and enhances on-site decision-making in critical industrial environments.