Explainable Multi-Modal Deep Learning Framework for Enhancing Cyber-Physical Systems Security

Cyber-physical systems (CPS) combine sensing, control, and communication for automated operation across industrial domains. The increasing integration of these systems has introduced vulnerabilities that demand secure and explainable protection mechanisms. This work presents an explainable multi-modal deep learning framework for CPS intrusion detection. The model integrates temporal and contextual signals from HAI, ICS Process, and MQTT-IoT-IDS2020 datasets, achieving 99.96 % accuracy and 99.92% F1-score. LIME-based interpretation provides transparent reasoning for detected anomalies, while optimized fusion and encoding modules enhance computational efficiency. Experimental outcomes show that the framework delivers reliable, interpretable, and scalable security suitable for real-time CPS environments.