特征提取, 边界特征, 聚类特征, 相关度, 多特征分析

To advance efficiency and pertinence in monitoring engine wear condition applying oil analysis, oil spectrometric and direct-reading ferrography data denoised by wavelet packet was used to extract three boundary features. In combination with centroids feature obtained through K-means clustering, centroid-boundary multi-feature analysis was established to monitor wear state of main friction pairs of certain engine, and the decision table about wear condition was formed. The result shows that compared to the decision table obtained by traditional three-line method, the partition of wear state based on multi-feature analysis is more detailed, accurate and instructionally significant. The identification efficiency for wear state increases by one time. Moreover, the

study of the relativity between oil spectrometric signals indicates the abnormal wear of piston ring friction pairs, which further confirms the pertinence and availability of multi-feature analysis. 

feature extraction, boundary feature, clustering