反作用轮不完全数据故障诊断新算法

Novel Fault Diagnosis Algorithm for Reaction Wheel Incomplete Data

针对卫星反作用轮遥测数据存在不完备情况，提出一种基于核模糊均值聚类（Kernel Fuzzy C—Means，KFCM）的数据修复诊断算法KFCM—Imputation（KFCMI）。该算法通过KFCM聚类实现已知故障样本的聚类中心和聚类半径，通过相似度计算查找与不完备数据最相似的数据点，将该数据点填充于不完备数据点位置，保证数据的完备性，并通过数据的相似度进行故障诊断。考虑所有故障特征同步缺失数据和各故障特征随机缺失数据两种工况，对比工程上直接删除缺失数据的方法，在数据缺失量小于总数据量的13％时，KFCM-I诊断精度能达90％以上；当数据缺失量占总数据量13％～20％时，诊断精度仍能达到80％。KFCM-I算法故障诊断精度高、计算简单，对工程应用有较好的参考价值。

A novel algorithm called Kernel fuzzy C means imputation （KFCM I） was proposed for the incomplete data of satellite reaction wheel. The centers and radius of the known fault samples were clustered by the KFCM, and the incomplete data points were replaced by the chosen data points for keeping the sample data completely, which were chosen from the known fault data by the similarity calculation and were the most similar to the incomplete data point. Then the similarity was adopted to diagnose the test sample data. Two scenarios were considered, one was that all the feature data missed at the same time, and the other was that each data missed at the random time, but the locations of all the missing data points were random. Compared with directly excluding the outlier or missing data points in engineering, when the missing data is less than 13% of the total test data, the accuracy of KFCM-I is higher than 90%, while the missing data is between 13%-20%, the accuracy also achieves at 80%. The KFCM-I algorithm is more accuracy and has low computation, which is an useful reference to the engineering.