Classification of Tribological Surfaces Without Surface Parameters

Quantitative measures are obtained from images of tribological surfaces. Based on these data, decisions are made regarding manufacturing and maintenance processes, machine-condition monitoring and failure analysis of engineering components. These decisions are often guided by an automated pattern recognition system. Components of this system are: surface topography data acquisition, surface characterization, surface classification and performance evaluation. The characterization and classification of tribological surfaces are major challenges that make the development of a reliable pattern-recognition system difficult. The reasons are that: (i) tribological surfaces often exhibit a non-stationary and multiscale nature, while most surface characterization methods currently used work well with surface data exhibiting a stationary random process, (ii) changes in topography that might occur between the interacting surfaces usually need to be known in advance, and (iii) the selection of surface parameters that separate different classes of surfaces is usually time-consuming and cumbersome. Because of these difficulties, characterization and classification methods which do not use surface parameters have been developed. In the classification methods, a measure of dissimilarity (e.g., Euclidean distance) calculated between a surface to be classified and already classified surfaces was used, instead of surface parameters. The unclassified surface was assigned to the class (of classified surfaces) with the lowest value of dissimilarity measure. The suitability of different classifiers; such as k-nearest neighbour classifier, linear-discriminant-analysis based classifiers and different dissimilarity measures; for the classification of tribological surface topographies (without the need for surface parameters) is investigated in this paper. Recent developments in this area, i.e., a fractal measure and a hybrid fractal-wavelet measure, are also discussed. The most suitable method for the classification of tribological surfaces has been selected.