A Visual Classification Method for Milling Surface Roughness Based on Convolutional Neural Network

Abstract

∗At present, most milling surface roughness detection is still completed by the traditional contact roughness measuring instrument. Aiming at the problem that the traditional contact roughness measuring instrument has a strong dependence and the measurement speed is slow, this paper will design a visual detection method that is consistent with the working environment, which belongs to the non-contact measurement. This method, firstly design special light source, illumination mode, complete the fill light to highlight image roughness related characteristics, and then the key hardware such as CCD camera is used to collect image data sets, finally based on convolutional neural network detection technology, namely through the end-to-end image analysis, then using convolution operation and comprehensive processing roughness classification model. Finally, the surface roughness can be classified accurately and rapidly predicted. The results show that the accuracy of surface roughness classification is 97%, and the time is much higher than that of the traditional contact roughness measuring instrument.