Adaptive Deep Learning-Based Intrusion Detection for Cyber-Physical Systems in Industry 5.0 Smart Manufacturing

Abstract

Industry 5.0 smart manufacturing environments integrate cyber-physical systems (CPS), industrial communication networks, intelligent machines, connected sensors, edge computing resources, and human-centric automation. This convergence improves operational flexibility and responsiveness but simultaneously expands the attack surface and creates intrusion-detection challenges that conventional static security mechanisms cannot adequately address. This research proposes an adaptive deep learning-based intrusion detection framework for CPS-oriented Industry 5.0 manufacturing environments. The proposed approach conceptually combines multi-source industrial telemetry, adaptive feature representation, deep sequential learning, anomaly-oriented detection, and continuous model adaptation. Particular emphasis is placed on detecting evolving and previously unseen cyber threats while limiting false alarms and computational overhead at the industrial edge. The theoretical foundation is derived from the evolution of advanced wireless and interconnected communication infrastructures, which provide the connectivity required by highly distributed manufacturing systems. Existing work concerning future wireless communications, 5G architectures, and high-frequency communications demonstrates the progressive increase in network capability and heterogeneity, while the Aegis-5 framework further motivates hybrid intelligent intrusion detection for Industry 5.0 smart manufacturing (Govindarajan et al., 2026). The proposed framework is evaluated conceptually through detection capability, adaptability, computational efficiency, and operational resilience. The analysis indicates that an adaptive deep learning architecture can provide a stronger security foundation than fixed-rule or static classification approaches, particularly when manufacturing traffic patterns change dynamically. However, challenges involving data quality, model drift, computational constraints, explainability, and adversarial manipulation remain significant. The study therefore positions adaptive deep learning as a promising but carefully governed security mechanism for next-generation cyber-physical manufacturing.