Surface finish control in machining processes using textural descriptors based on moments

This paper presents a method to perform a surface finish control using a computer vision system. The goal pursued was to design an acceptance criterion for the control strategy. Class I would contain those parts with low roughness-acceptable- and class 2 those with high roughness -defective. We have used 140 images obtained from AISI 303 stainless steel machining. Images were described using five different methods - Hu, Flusser, Taubin, Zernike and Legendre moments. Classification was done using k-nn and neural networks. With k-nn the best error rate - 4.7% - was achieved using Hu and Flusser descriptors. With the neural network, a ten node hidden layer network with 300 cycles using Legendre descriptors leads to the optimal configuration - 4.7% error rate.