Toward High Robustness and Resiliency using Fault/Attack Detection

Abstract

Modern Cyber-physical Systems (CPS) in the automobile industry are equipped with various sensors to provide the safety and the convenience such as the obstacle detection and the adaptive cruise control. However, those systems could be vulnerable to sensor attacks (e.g., GPS spoofing) and the malicious sensory data causes the system to be in the danger. Therefore, the robustness and resilience against the abnormal conditions is essential to improve the safety of the system. To achieve the robustness and resiliency, multiple sensors measuring the same physical (i.e., redundancy) can be used. Sensory data obtained from the multiple sensors is fused using various sensor fusion models to detect and identify the malicious sensor data. In this thesis, sensor fault/attack detection methods are elaborated, which can be classified into two types: Hardware redundant method and analytical redundant method. The hardware redundant method is based on the sensor data obtained from multiple sensors. It has the advantage that doesn’t require the process of modeling. However, using multiple sensors causes the high cost and the limitation of the space to be implemented. In contrast, the analytical redundant method is based on the model of the system. It doesn’t require the several sensors. However, modeling a system is complicated and if the modeling has the error, it could result in the false alarm during the fault evaluation. In this thesis, we first focus on the hardware redundant method using several sensors measuring the same physical value. Then, we also handle the analytical redundant method by modeling the unmanned ground vehicle. ⓒ 2017 DGIST