Neural network based fault identification for quadrotor thrust loss and propeller misalignment

This paper proposes a neural network (NN) based fault identification technique for thrust loss and propeller misalignment. The proposed identification method is demonstrated using a quadrotor, and the stabilization is achieved using standard proportional, derivative, and integral controller. First, the fault is introduced into the quadrotor mathematical model, and then a database is created to train the NN. In the first case, the NN model to identify the thrust loss and location of rotor failure is developed. The idea is to assign the thrust loss to each rotor and then train the NN using thrust loss and rotor speed. In the second case, the propeller misalignment NN model also has been developed. The idea here is to assign the misalignment to each rotor that is responsible for the additional yaw. This results in a difference in diagonally placed pair rotor speed. The contribution of this paper is the fault identification method that is suggested, where fault classification and type are established after the NN is trained using misalignment angle and rotor speed. Simulation results reveal that, by taking the rotational speed of all four rotors as input, the trained NN determines the percentage of thrust loss and propeller misalignment angle.