A hybrid search-tree discriminant technique for multivariate wear debris classification

Wear debris analysis (WDA) is an effective machine health monitoring technique in which wear debris attributes can reveal wear mechanisms and particle colours can be used to detect oxidation. However, most of the existing approaches to WDA are based on analytical ferrography. Online wear debris typing is still difficult to achieve because of the low resolution of online particle images. In this work, a hybrid search-tree discriminant technique is described. It permits identification of online wear particles by combining their colour attributes and multi-view particle features. A multi-class support vector data description multi-SVDD, the K-means and a support vector machine SVM are integrated to establish a three-level search-tree model to discriminate multivariate wear debris including red and black oxides, cutting, spherical, fatigue and sliding particles. First, the oxides are identified based on their special colour information using the multi-SVDD. Second, the K-means is used to look for the clustering centres of cutting and spherical particles by utilizing their distinct features of aspect ratio and sphericity. The third level built by SVM is adopted to distinguish fatigue and sliding particles based on their height aspect ratio and height-to-width aspect ratio. Experiments have demonstrated that the search-tree discrimination model is effective for multivariate wear debris classification. The proposed method provides a solution to the existing problems in online wear particle identification for wear mechanism analysis.