Neural network-based propeller damage detection for multirotors

This work presents a method for detecting and identificating possible damages to propeller blades in multirotor vehicles, for a particular case study of a quadrotor. The detection method is based on a neural network, which takes as input the energy of several spectral bands of the inertial measurements and control variables, and outputs a measure of how damaged a propeller is. The ability of the network to correctly generalize from a limited dataset will be shown by training it using data gathered from an indoor, controlled environment, and testing it using data from outdoor flights.