Machine-Learning Based Rotor Fault Diagnosis in a Multicopter with Strain Data

Abstract

An innovative machine-learning-based probabilistic framework for online rotor fault diagnosis in multicopters is presented. The proposed scheme employs in-flight out-of-plane strain measurements at each of the multicopter booms to detect, identify, and quantify rotor faults while distinguishing them from the aircraft response to random gusts. Its robust performance is demonstrated with application to a 2-foot-diam hexacopter flying under varying forward velocity and gross weight configurations, as well as atmospheric disturbances and uncertainty. The rotor fault diagnosis takes place in two steps. First, a simple perceptron classifies the aircraft’s health condition. If a rotor fault is detected it is simultaneously identified and the fault magnitude estimation step initiates. Here, linear regression models are used to predict the respective rotor degradation values with their 95% confidence intervals. The generalization capability of the method is established with several test data under unmodeled operating conditions (not used in the training phase). The proposed framework can accurately diagnose even minor rotor faults of 8% degradation while distinguishing them from aggressive gusts of up to [Formula: see text] magnitude. The maximum fault detection time is less than 0.3 s. The health state classification and the rotor fault magnitude quantification accuracy are over 99%.