Hybrid Edge–Cloud Deep Learning Framework for Real-Time Anomaly Detection in Industrial IoT Networks

Abstract

Industrial Internet of Things (IIoT) networks produce huge amounts of diverse data from different sensors, actuators, and smart objects distributed in mission-critical industrial environments. Maintaining the safety and trustworthiness of these systems requires real-time anomaly detection mechanisms that can identify malicious operations, malfunctioning devices, or abnormal traffic patterns with high accuracy and low latency. Cloud-based deep learning solutions provide computational scalability but may introduce significant communication delays. Although edge solutions deliver proximity and fast inference, they lack sufficient computational capability to handle complex deep models. Motivated by this gap in the existing literature, in this study, we introduce a novel hybrid edge–cloud deep learning framework for real-time anomaly detection in IIoT networks that capitalizes on the strengths of both paradigms. The edge layer conducts lightweight preprocessing, feature extraction, and initial detection with the help of compressed neural modules, while the cloud layer runs advanced deep models (e.g., CNN–LSTM hybrids and transformer-based ensembles) to carry out high-level analysis. A joint inference pipeline minimizes the end-to-end delay with adaptive offloading strategies modelled as an optimization problem that trades off detection accuracy for network delay. Extensive experiments on the Edge-IIoTset dataset show that the proposed hybrid architecture achieves a detection accuracy of 98.7% with a 37% average inference latency reduction compared to purely cloud-based solutions. The comparison also demonstrates improved robustness across diverse attack categories. These results also demonstrate the feasibility of hybrid edge–cloud intelligence for secure, robust, and real-time anomaly detection in IIoT.