Improved UAV blade unbalance prediction based on machine learning and ReliefF supreme feature ranking method

As unmanned aerial vehicles (UAVs) are witnessing a rapid increase in usage in regard to many different applications, it has become paramount to classify blade faults and unbalances in preflight processes. This paper aims to introduce and show the effectiveness of new unbalance classification models using two different machine learning techniques; support vector machine (SVM) and k nearest neighbor (kNN). Screw loosening, motor base shifting, and other faults were simulated as weight-adding unbalances on the blades of a quadcopter UAV. The vibration-based signal processing features in the time domain were extracted with the aid of a 3-axis accelerometer sensor and a data acquisition system while in hover mode. ReliefF supreme feature scoring and ranking method were used to improve prediction accuracy to acquire effective features and suppress insufficient ones. The performance of the models was validated using several methods by comparing them with the basic models. Accuracy increased from 92.52 to 98.85% and from 95.1 to 96.41% in SVM and kNN machine learning classifiers, respectively. The enhanced models proved reliance on locating unbalanced blades as the processing time decreased noticeably. The results stipulate that the proposed system transcended current developments in predicting blade faults of UAVs and showed good promise for future development of embedded systems-based quadcopter fault diagnosis.