MEMS fault-tolerant machine learning algorithm assisted attitude estimation for fixed-wing UAVs

Abstract

The utilization of Micro-electromechanical Systems (MEMS) sensors is widespread for directly detecting attitude angles, such as Accelerometer, Gyro, and Magnetometer readings. However, these MEMS sensors are prone to flaws, leading to inaccurate estimates of attitude angles and, consequently, causing UAVs to lose control. Given that UAVs are operated remotely over long distances, ensuring accurate attitude estimates becomes crucial. This study aims to address this issue by employing machine learning algorithms (MLA). These algorithms were trained and evaluated to overcome the problem by predicting missing data from a malfunctioning MEMS sensor using the available data from other MEMS sensors. To calculate the attitude angles, the study utilizes the Extended Kalman Filter (EKF) technique. Furthermore, a novel fault-tolerant machine learning-aided estimation algorithm has been proposed specifically for estimating the attitude angles (φˆ, θˆ, ψˆ) of fixed-wing UAVs. The significance of this research becomes even more prominent when considering the occurrence of MEMS sensor failure. In such cases, the machine learning algorithm plays a crucial role as it has been pre-trained specifically to handle these scenarios. The algorithm is equipped with the ability to effectively address and mitigate the challenges posed by MEMS sensor failures. By leveraging its pre-existing knowledge and learned patterns, the algorithm can accurately predict missing data caused by malfunctioning MEMS sensors. This capability proves invaluable in ensuring the reliable estimation of attitude angles, even in the face of sensor failures. Thus, the integration of machine learning into the estimation process enhances the resilience and robustness of the system, making it well-equipped to handle the uncertainties introduced by MEMS sensor failures.