为检测飞轮系统的微弱故障，提出一种基于符号动力学的故障检测方法。首先，利用符号动力学算法将飞轮的电流数据进行字符映射，实现信号降噪、数据压缩。其次，利用D阶马尔科夫过程定义电流字符序列的变异过程，且根据字符序列的熵值变化率确立算法的字符个数，并计算字符的概率状态转移矩阵。最后，根据概率状态转移矩阵求解字符概率特征向量，并利用特征向量之间的距离阈值检测飞轮的故障。仿真结果表明：该方法能够根据字符概率特征向量之间的距离区分不同严重程度的飞轮系统故障，实现微弱故障的检测；与扩展卡尔曼滤波算法相比，该方法不需复杂的动力学建模，且仅利用单变量即可实现飞轮的故障检测。此外，利用过程数据，该方法可以快速学习卫星其他部件的行为，并检测故障。%The fault detection method of symbolic dynamics is proposed to detect the tiny faults of flywheels. Firstly, the symbolic dynamics algorithm transforms the flywheel current into a series of symbols. The signal noise reduction and data compression can be accomplished by the symbol generation. Secondly, the D⁃Markov machine can define the abnormal transitions of symbol sequences and produce the probability state transition matrix. Then, the number of symbols is selected from the change in entropy of symbol sequences. Moreover, the symbol probability vector can be obtained according to the eigenvector of the probability state transition matrix. The threshold of distance among the symbol probability vectors can be used to detect the faults of the flywheel. Finally, The simulation results show this method can identify varying degrees of faults in the flywheel and achieve the tiny fault detection by the distance among the vectors. Compared with the extended Kalman filter method, the proposed method can detect the faults of flywheel by using the single variable without the complicated kinetic modeling. The proposed fault detection method in this paper can be also used in the other components of satellites, which can learn the system behavior from process data and have the merit of portability.
展开▼
