Modular, Energy-Efficient, Hybrid DL Framework for Real-Time Anomaly Detection in Industrial IoT

Industrial IoT networks are increasingly used to monitor machinery, safety, and environmental conditions, but face challenges of anomaly detection under resource and privacy constraints. Existing deep learning models often demand high computation, are domain-specific, or unsuitable for energy-limited edge devices. This chapter introduces a modular monitoring framework built on a hybrid CNN with Global Feature Concatenation (CNNwGFC), designed to capture spatial irregularities across diverse sensor inputs while remaining lightweight for edge deployment. To preserve privacy and support scalability, federated learning is integrated, enabling decentralized training without sharing raw data. Explainability methods such as SHAP and LIME are employed to improve transparency and trust. The framework will be evaluated on benchmark datasets (TON_IoT, Edge-IIoTset) to demonstrate accuracy, efficiency, and adaptability, highlighting its potential for sustainable, privacy-preserving anomaly detection in smart mining and industrial operations.