Application of linear discriminant analysis tocharacterize effect of polishing cloths
                                                    ZHAO Dengchao, LIU Ming, ZHOU Chao, HUANG Yuxing
                       （School of Mechanical Engineering and Automation, Fuzhou University, Fuzhou 350116, China）

Abstract: Aiming at the problems about the method for evaluating quality of polished surfacesis limited to onedimensional roughness parameters and selected not uniform.The linear discriminant analysis, a method in machine learning, was used to investigate effects of polishing cloths. Experiments were carried out on copper with alumina polishing slurry and four types of polishing cloths including velvet, silk velvet, plain velvet, and woolen. Micromeasure was used to measure polished surfaces morphology. Threedimensional roughness parameters (e.g. the arithmetic mean height, the standard deviation of the surface height, the skewness, the kurtosis, etc.) were used for clustering and dimensionality reduction analysis. Logistic regression and oneway analysis variance was used to crossvalidation verify the accuracy of the model. The results indicate that four types of polishing cloths have different effects on copper surface and the model can provide its difference visualization, the arithmetic mean height is the best parameter to distinguish the polishing quality of the four cloths used, and the standard deviation of the surface heights is second best.

Key words: linear discriminant analysis; polishing cloths; logistic regression; the arithmetic mean height