Multiclass Classification Fault Diagnosis of Multirotor UAVs Utilizing a Deep Neural Network

A fault diagnosis algorithm using a deep neural network for an octocopter Unmanned Aerial Vehicle (UAV) is proposed. All eight rotors are considered in the multiclass classification fault diagnosis problem. The latest angle time history is fed to the proposed algorithm to determine rotor failure in real time. The normal case and fault case of each rotor are considered with appropriate output pairs to form a dataset. The proposed classifier can distinguish a failed rotor from the others with the help of different patterns of Euler angles during the training process. Two hidden layers are constructed using sigmoid and softmax activation functions. A generalized delta rule is adopted, and a stochastic gradient descent scheme is used to calculate the weight update of the neural network. The proposed fault diagnosis algorithm can be augmented to a fault-tolerant controller to construct an integrated system that involves solving a convex optimization problem. Numerical simulations are conducted to validate the performance of the proposed diagnostic algorithm. It is demonstrated that the performance can be adjusted by controlling the design parameters.