An Adaptive Support Vector Machine-Based Workpiece Surface Classification System Using High-Definition Metrology

The shape of a machined surface significantly impacts its functional performance and exhibits different spatial variation patterns that reflect process conditions. Classification of these surface patterns into interpretable classes can greatly facilitate manufacturing process fault detection and diagnosis. High-definition metrology (HDM) can generate high density data and detect small differences of workpiece surfaces, which exhibits better performance than traditional measurement methods in process diagnosis. In this paper, a novel adaptive support vector machine (SVM)-based workpiece surface classification system is developed based on HDM. A nonsubsampled contourlet transform is used to extract features before classification with its characteristics of multiscale, multidirection, and less dimension of feature vectors. An adaptive particle swam optimization (APSO) algorithm is developed to search the optimal parameters of penalty coefficient and kernel function of SVM and is helpful to escape from the local minimum by its strong ability of global search. A varied step-length pattern search algorithm is explored to optimize the global point in every iteration of the APSO algorithm by its good performance in local search. These two algorithms are combined with their relative merits to find the optimal parameters for building an adaptive SVM classifier. The results of case studies show that the proposed adaptive SVM-based classification system can achieve a relatively high classification accuracy in the field of workpiece surface classification.