The Abnormal Detection Strategy for Spacecraft Components with Multi-dimension Parameters

Aiming at realizing the abnormal detection for spacecraft components based on the monitoring during environmental testing, this paper proposes a novel strategy containing principal component analysis (PCA), one class support vector machine (OCSVM), and integrated learning. Firstly, product features are extracted from the raw data. Then, PCA is utilized to reduce the feature dimension and standardize the data. After that, sub-datasets are generated through resampling and utilized to train the individual OCSVM models. Finally, the decision results of these models are averaged to obtain the final classification results. A case study based on a thruster simulation dataset shows that the proposed strategy can obtain accurate detection results.