CASCADED SECURE STATE ESTIMATIONS AS CYBER RESILIENCE AGAINST SENSOR AND ACTUATOR ATTACKS ON AUTONOMOUS SYSTEMS

• Main Author: Hilmi, Muhammad
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84189
• Institution: Institut Teknologi Bandung

The emergence of autonomous systems in various industrial sectors has resulted in increased efficiency and safety in operations. The Industrial Revolution 4.0 provides a foundation for applying the digitization of various autonomous systems, optimizing processes, and creating opportunities for implementing smart industries. However, the increasing reliance on digital technology in industrial activities makes autonomous systems vulnerable to cyber threats. In particular, threats or attacks on sensors and actuators pose a significant risk of direct harmful consequences. This research focuses on addressing this challenge by developing cascaded secure state estimation methods designed to improve cyber defense against sensor and actuator attacks on autonomous systems. The proposed cascaded secure state estimations, in the form of two nonlinear observers, is developed to estimate the true state of the system and the magnitude of the attack on the sensors and actuators by each observer. In this development process, the necessary conditions to ensure the convergence and stability of the state estimation and attack estimation are formulated as a linear matrix inequality. Validation of this secure state estimation method for its efficacy is conducted through implementation on autonomous vehicles, such as numerical simulations on an autonomous container truck and experimental testing on an autonomous container truck under sensor and actuator attacks. The implementation results show high true state estimation accuracy and similarly for estimating the magnitude of sensor and actuator attacks. This method yields an RMSE of 2.77 m and 0.0983 rad for position and orientation sensor attack estimation, and errors of 0.1079 m and 0.0143 rad for actuator attack impacts estimation in a simulated autonomous container truck. In experiments, it achieved a trajectory lateral error of 0.0245 m and an orientation error of 0.0054 rad. These results demonstrate the potential for significantly improving the security and reliability of autonomous system implementations against sensor and actuator attacks.