Online Fault Detection of Fixed-Wing UAV Based on DKPCA Algorithm with Multiple Operation Conditions Considered

基于DKPCA的固定翼无人机多工况在线故障诊断

¹ School of Ordnance Sergeant, Army Engineering University, Wuhan 430075, China
² School of Electrical Engineering and Automation, Wuhan University, Wuhan 430072, China

Received: 2 August 2019

Abstract

The mission execution process of a fixed-wing UAV has multiple phases and multiple operation conditions. Its parameters are nonlinear and dynamic. These characteristics make its online fault detection rather complicated. To carry out the fault detection, this paper selects nine key parameters of the transverse, longitudinal and velocity control loops of the UAV to characterize its real-time conditions. The core parameters are dynamically preprocessed to construct an augmented matrix so as to describe the dynamic characteristics of the UAV. Then, the improved k-mediods* algorithm is used to cluster the operation conditions of the UAVs by means of augmented dimensions. Neural networks are used to achieve the online matching of operation conditions. To overcome the nonlinearity of the UAV, the fault detection is performed by using the DKPCA algorithm; the fault monitoring is conducted through constructing the compound indexes of SPE and T², notated as FAI. Furthermore, the fault separation algorithm is proposed to specify the variables of fault from the augmented high-dimensional data set. In order to deal with the erroneous reporting of faults due to measurement errors, the paper conducts the wavelet denoising of FAI, the compound indexes of the DKPCA algorithm. Finally, the data set collected from a real UAV flight is used to verify the effectiveness of the DKPCA algorithm for operation condition clustering and matching, fault detection and wavelet denoising.

摘要

固定翼无人机（UAV）执行任务过程包含多个阶段，体现出多工况特征，且UAV参数具有动态性和非线性，导致UAV在线故障诊断复杂化。本文选取UAV横向、纵向和速度控制回路的9个核心参数来表征无人机实时状态，通过动态预处理构建增广矩阵以描述UAV的动态特征；采用改进k-mediods^*算法对UAV扩维数据进行工况聚类，并采用神经网络完成在线工况匹配。针对UAV非线性特征，采用了DKPCA算法进行故障诊断，基于SPE和T²构建合成指标FAI用于故障监测，并提出了分离算法从高维数据定位故障变量；为了应对测量误差带来的故障误报，还对DKPCA合成指标FAI进行了小波去噪处理。最后，以某UAV的实飞数据为对象，呈现以上算法的结果，验证了工况聚类、工况匹配、故障诊断及小波去噪等算法的有效性。