Propeller Fault Classification for Quadrotor Aerial Vehicle using Spatial Displacement Statistical Features

Unmanned Aerial Vehicles (UAVs) have become increasingly vital in various applications, from surveillance to logistics. However, ensuring operational reliability is crucial, mainly in detecting potential faults that could compromise flight safety. This study explores spatial displacement data captured through optical sensors as valuable information to translate a quadrotor UAV’s flight behaviour. Incorporating spatial displacement statistical features into a fault classification model can enhance the detection of spatial variations, leading to the identification of abnormalities and potential malfunctions. This work integrates spatial displacement statistical analysis with a machine learning (ML) classification approach to propose a malfunction detection model for vertical take-off and landing quadrotor UAVs. Data from hovering quadrotors in healthy and faulty states were collected using an optical camera system. Essential spatial parameters were derived through statistical analysis, providing informative features for the classification model. These features revealed subtle flying characteristics concealed by raw flight data. The extracted features are used to train and classify different propeller faults using the MATLAB Classification Learner App. The ML classification model evaluations are employed for assessing the performance of three classification algorithms: Quadratic Support Vector Machine (Q-SVM), Kernel Naïve Bayes and Fine Tree algorithms. The result shows that Q-SVM achieves the highest training accuracy and is suitable for developing a reliable UAV fault diagnosis system.