Actuator fault detection in a hexacopter using machine learning

POSE CLAUDIO; ALESSANDRO GIUSTI; JUAN GIRIBET

Buenos Aires

Congreso; Congreso del la Asociación Argentina de Control Automático; 2018

We present an approach based on machine learning for detecting total motor failure in a tiled-rotor hexacopter. We generate large flight datasets using a Simulink model based on a 6-DoF block and position and attitude PID controllers; these datasets are used to train and validate a random forest classifier that is able to detect if a motor has failed, and identify which one. The algorithm uses as inputs the attitude of the vehicle and the signals commanded to the motors, and outputs a flag per motor that indicates its healthy or faulty state. The classifier achieves a good performance, resulting in a detection of the fault in less than 100mS after it has occurred, with a high rejection of false positives, less than one false positive per hour of flight.