Fault Detection in Rotors of Quadcopter UAV Using Convolutional Neural Network

Abstract

Mostly fault in quadcopter occurs in its rotors due to numerous reasons which leads towards flight failure. And due to that speedy fault detection is of great significance in practical applications. This work proposes a new fault detection method based on Convolutional Neural Networks (CNNs) on IMU sensors data to detect and classify the faulty and nominal flight conditions. Induced failure are introduced in quadcopter rotor section by varying the accelerometer such that yaw, pitch and roll data in Simulink/MATLAB. The data of the faulty model later on is used as input for the CNN learning model on WEKA platform. The proposed architecture is capable of automatically learning features and parameters which lead towards failure of the quadcopter from the supplied data as well as analyzing spatial and temporal fluctuations. On simulated data, empirical results demonstrate that our solution can classify various rotor fault types with an accuracy of 94%.