Resilient Secure Control of Networked Industrial Control Systems under AI-Driven Feedback Poisoning via ACK Bundling

This paper investigates resilient secure control for networked industrial systems under AI-driven feedback poisoning attacks. Such attacks exploit feedback channels and timing regularities to degrade closed-loop performance while remaining difficult to detect. A data-driven poisoning model is constructed based on a real-world SWaT dataset, enabling the emulation of adversarial feedback manipulation beyond idealized assumptions. An acknowledgment (ACK) bundling mechanism is introduced to regulate feedback transmission by aggregating ACKs within a tunable time window. The bundling window Tb is adaptively adjusted via gradient descent (GD), facilitating a flexible trade off between control performance and communication efficiency in adversarial environments. In addition, a lightweight Lyapunov Krasovskii functional (LKF) is developed to establish closed loop stability in the presence of communication delays and poisoned feedback. Simulation results on a networked unmanned surface vehicle (USV) platform demonstrate improved tracking performance and robustness against feedback poisoning under constrained communication conditions. These results indicate that the proposed framework provides a practical and effective solution for secure networked industrial control.