Combined non linear geometric approach and learning RBF-NN for quadcopter Fault Tolerant Control

The rapid utilization of Unmanned Aerial Vehicles in urban areas underlines the pressing need for mitigating ground risks associated with their operations. Therefore, ensuring the safety and reliability of these operations has become top priority. Consequently, there is a necessity to develop advanced fault detection and diagnosis systems for unmanned aerial vehicles. This thesis proposes an active fault tolerant control system based on a non linear geometric approach combined with neural networks for addressing quadrotor actuator faults. The non linear geometric approach enables the decoupling and isolation of faults associated with each actuator through the implementation of residue filters that solely affects the corresponding faulty actuators. The magnitude of the isolated fault is then estimated using a radial basis neural network integrated into the controller’s feedback loop to adjust its actions. The performance of the proposed methodology is comprehensively analyzed using a quadrotor simulator validated with real-world data. The robustness of the system is further confirmed through extensive experimentation involving variations in the characteristic parameters of the proposed simulator.