Data-driven fault diagnosis of FW-UAVs with consideration of multiple operation conditions

Fixed-wing Unmanned Aerial Vehicles (FW-UAVs) are intelligent aircrafts. It is of significance to carry out fault diagnosis of FW-UAVs to improve reliability and safety. An entire mission of FW-UAVs contains couple of phases; correspondingly, this paper treats FW-UAVs as multiple operation condition processes. An innovative framework is then proposed for fault diagnosis of FW-UAVs, where the process dynamics, multiple operation conditions, variable data density, and process disturbance are considered. Firstly, augmented matrixes are constructed with the data samples to involve the dynamic characteristic of FW-UAVs. Secondly, a modified DBSCAN algorithm employing Shared Nearest Neighbor based Distance (SNND-DBSCAN) and a K Nearest Neighbor algorithm employing SNND (SNND-KNN) are proposed respectively. They cooperate with each other to realize offline operation condition classification and online recognition. Thirdly, Multiple condition oriented Dynamic KPCA (M-DKPCA) algorithms incorporated with Weighted sliding window denoising (WM-DKPCA) is proposed for fault diagnosis. Finally, the proposed algorithms are tested with real flight data sets in terms of linear and nonlinear faults; and the comparisons between KPCA, DKPCA, M-DKPCA and WM-DKPCA are presented. The results confirm that the multiple condition oriented M-DKPCA and WM-DKPA algorithms are more suitable for fault diagnosis of FW-UAVs; and WSW denoising can indeed improve the fault diagnosis performance.