Detecting Contextual Faults in Unmanned Aerial Vehicles Using Dynamic Linear Regression and K-Nearest Neighbour Classifier
A. Alos, Zouhair Dahrouj · Gyroscopy and Navigation · 2020
Abstract Unmanned aerial vehicle (UAV) is a complex system. Its design involves control, aerodynamics, and communication systems. We use the complex linear relationships among UAV attributes (sensor readings, and commands) to propose a new technique to detect contextual faults. The contextual faults mean that a defective sensor shows invalid values concerning the context of other attributes. The proposed approach depends on estimating the values of a focused attribute using dynamic linear regression. Next, it calculates the estimation error at each time step. The values of the estimation error are classified using K-NN (Nearest Neighbour) classifier into two classes (Normal, Abnormal). The abnormal points are flagged as potential faults. Moreover, comparison with other algorithms (K-Means and One-Class SVM) is made. The proposed approach showed better results in most of the cases.