Robust estimation-based fault detection and isolation framework for autonomous aerial and navigation systems

Abstract

Autonomous aerial and navigation systems have become an integral part of modern engineering applications, including unmanned aerial vehicles (UAVs), integrated navigation platforms, and autonomous transportation systems. As these systems increasingly operate in complex and safety-critical environments, their reliability and fault tolerance have emerged as key requirements. Sensor and actuator faults, whether caused by hardware degradation, environmental disturbances, or operational wear, can significantly degrade system performance and may lead to instability or mission failure if not detected and isolated in a timely manner. Consequently, the development of reliable fault detection and isolation (FDI) methods remains a critical research topic for autonomous systems. Kalman filter (KF)-based estimation techniques and their nonlinear extensions are widely used in autonomous systems due to their computational efficiency and optimality properties under nominal conditions. However, traditional innovation-based FDI approaches relying on standard Kalman filtering structures exhibit inherent limitations when faults are present. In particular, slowly-growing sensor faults may be masked by the recursive nature of the filter through a phenomenon known as the error tracking effect, where biased measurements are gradually absorbed into the state estimates. This behavior reduces the sensitivity of traditional FDI methods and may result in delayed or missed fault detection. These limitations motivate the need for alternative estimation strategies that enhance fault detectability while maintaining reliable estimation performance under faulty conditions. This thesis aims to develop a unified robust estimation-based FDI framework that improves fault detection performance in autonomous aerial and navigation systems. The proposed framework integrates robust estimation principles into Kalman filtering structures to mitigate the error tracking effect and enhance sensitivity to faulty measurements, particularly for slowly-growing ramp faults. One of the methodological aspects addressed in this thesis is the examination of the error tracking effect in Kalman filtering and its influence on fault detection performance. While the error tracking effect has been discussed in the literature, this thesis revisits the phenomenon in the context of KF-based FDI and highlights its practical implications, particularly for the detection of slowly-growing faults. By describing how fault induced estimation errors propagate through the recursive filtering process, the thesis provides additional insight into the limitations of conventional FDI approaches and motivates the use of robust estimation principles within the proposed framework. The proposed robust FDI framework is designed in a general and system-independent manner, allowing it to be applied to both linear and nonlinear systems. To demonstrate its effectiveness and generality, two representative application cases are considered. The first application focuses on a linear Global Navigation Satellite System/Inertial Navigation System (GNSS/INS) integrated navigation system. In this case, the framework is implemented within a linear Kalman filter structure. Simulation studies are conducted under various sensor fault scenarios, including step and slowly-growing ramp faults. The results show that the proposed approach improves fault detection sensitivity compared to traditional innovation-based methods, particularly in scenarios where ramp faults are difficult to detect due to error tracking. The second application extends the framework to a nonlinear quadrotor unmanned aerial vehicle system. In this case, an extended Kalman filter (EKF)-based structure is employed, and the proposed robust FDI framework is implemented using the robust augmented state EKF (RASEKF) formulation. Actuator faults are dynamically modeled and incorporated into the estimation process through an augmented state representation, enabling simultaneous sensor and actuator FDI within a unified estimation architecture. Simulation results demonstrate that the RASEKF-based framework provides improved capability to detect gradually developing ramp sensor faults and and maintains robust estimation performance in the presence of faults when compared to traditional EKF-based and representative robust filtering approaches. Throughout both application cases, the proposed framework is evaluated through extensive simulation studies to assess its fault detection performance and robustness under faulty conditions. The results confirm that integrating robust estimation principles into KF-based FDI structures effectively mitigates the adverse effects of error tracking and enhances fault detectability without introducing excessive computational complexity. The framework remains suitable for real-time implementation and can be integrated into existing autonomous system architectures with minimal modification. While the proposed framework demonstrates effective performance under the considered scenarios, certain limitations are acknowledged. The current formulation primarily focuses on step and slowly-growing ramp fault profiles and assumes stationary fault sources. More complex fault behaviors, such as intermittent faults or nonlinear fault dynamics, are not explicitly addressed. In addition, the framework assumes that sensor and actuator faults do not occur simultaneously, which reflects a realistic but simplifying assumption for many practical systems. Extending the approach to handle simultaneous sensor and actuator failures represents a potential direction for future research. Furthermore, the selection of certain design parameters, such as sliding window length and weighting thresholds, remains dependent on system characteristics and noise properties, suggesting that adaptive or systematic parameter tuning strategies could further enhance the generality of the framework. In conclusion, this thesis presents a robust, unified, and practically applicable fault detection and isolation framework for autonomous aerial and navigation systems. By combining robust estimation principles with model-based Kalman filtering structures, the proposed approach improves fault detection sensitivity, mitigates the error tracking effect, and maintains reliable estimation performance in the presence of faults. The results obtained from both linear and nonlinear system applications demonstrate the potential of the proposed framework to enhance the reliability and fault tolerance of modern autonomous systems and provide a solid basis for future studies on robust estimation-based FDI in autonomous platforms.