Abstract: The whole mission of fixed-wing UAV can be divided into multiple working conditions, and the analysis of UAV working conditions is the precondition of fault diagnosis. According to the control principle of UAV, 9 variables (deflection angle of elevator, rudder deflection angle of left and right ailerons, course angle, elevation angle, angle of inclination, height, airospeed, and cylinder temperature) of the horizontal, vertical and speed control loops are selected to represent the real-time working conditons of UAV. For the characteristic of UAV data set, a modified density clustering algorithm coupled with shared nearest neighbor is proposed, which is notated as SNN-DBSCAN^* to classify UAV working conditions. An inflexion heuristic parameter optimization approach combined with data shear rate and satisfied curves is specifically proposed for SNN-DBSCAN^*. An independent component analysis and support vector machine fusion algorithm which is notated as ICA-SVM is proposed for condition matching of UAV. ICA is intentionally employed for feature extraction and reconstruction of the UAV variables so as to improve the disturbance resisting capacity of the condition matching algorithm. The test result of the real UAV flight data set shows that the SNN-DBSCAN^* algorithm can reasonably classify UAV working conditions without increasing prior knowledge, and a satisfied matching accuracy can be achieved with ICA-SVM model which shows good disturbance resisting capacity to the deviations of the UAV variables.

Key words: fixed-wingunmannedairvehicle, multi-workingconditionclustering, workingconditionmatching, sharednearestneighbor, densityclustering, independentcomponentanalysis, patternmatching