Multirotor fault detection based on supervised learning

POSE CLAUDIO; JUAN GIRIBET; JOSÉ I. GONZÁLEZ ETCHEMAITE

Buenos Aires

Congreso; Congreso Argentino de Control Automático; 2020

Asociación Argentina de Control Automático

Active fault tolerant systems require to precisely identify the cause of a failure to properly reconfigure the system in order to obtain the best performance under faulty conditions. The fault detection and identification is usually achieved by means of an observer-based system, however, the advances and popularity of learning-based methods provide an interesting approach to solve this problem.In this work, experimental results of a supervised learning-based classifier to detect total rotor failures in a hexarotor unmanned aerial vehicle are presented. Based on previous work, where a simulation environment was built to generate realistic flight data to train and validate the proposed approach, here, that environment is further improved by taking into account several perturbations and modeling errors that were not previously considered. The performance of the fault detection and identification classifier is then evaluated with recorded real flight data, for several different training methods.