Extraction and expression of deformation features for wear images from four-ball friction test: For classification purpose

Abstract

The shape characteristics of wear spot in the four-ball wear test are often qualitatively identified by human testers based on experience, which are subjective, unquantifiable and biased. This paper proposes an automated method for extracting and expressing the shape features, which is effective for the deformed classification of the wear images. First, the wear spot is segmented from the background, and the geometric parameters of the wear spot are extracted. Based on this, the optimal standard wear spot which match the segmented (or actual) wear spot best is constructed. The established feature variables (i.e. abnormality rate and eccentricity rate) are used to quantitatively assess the deformation degree of segmented wear spot, by comparing the segmented wear spot with the constructed standard wear spot. Finally, Kmeans clustering algorithm is used to verify the validity of deformation features in the classification application. Simulation test results show that feature variables are valuable for identifying the deformation of wear images, and three effect evaluation indicators (CH/SI/DB) shows that the classification effect of the wear images is better.