CLASSIFICATION AND CORRELATION OF SURFACE ROUGHNESS IN METALLIC PARTS USING TEXTURE DESCRIPTORS

In this paper we present a method to classify the surface roughness in metallic part after machining processes using an artificial vision system. Two texture analysis methods are used: Co-occurrence matrix (GLOW) and the energy of the texture obtained by Laws' method. These descriptors are classified with Lineal and Quadratic Discriminant Analysis (LDA and QDA) and Artificial Neural Networks (ANN.) The best results have been achieved using the Laws mask R5R5 (94.03%) and the combined correlation descriptor extracted from the GLCM (94.23%), both classified using Neural Networks. These results show the success of the method and the possibility to correlate these descriptors with the average roughness (Ra).