Surface defect detection and prediction in carbide cutting tools treated by lasers

Abstract

Laser surface engineering of cutting tools is used to improve the performance of cutting processes via altering the material interaction between the tool surface and workpiece. Laser processing applied to cemented carbide cutting tools can induce various thermal and mechanical surface defects including porosity, splatter, cracks, balling, spherical pores, voids, and dissociation. Those defects could be detrimental to the integrity of the tool, therefore parametric optimization is crucial to limit and control possible post-processing defects. This study aimed to identify and classify surface defects in post laser processed carbides to better understand the relationship between parameters and resultant surface integrity. A region convolutional neural network (R-CNN) was trained for identification and classification of these surface defects using scanning electron microscopy images (SEM) as inputs. The R-CNN provided a quantitative analysis of each defect with an average accuracy of 91%. Using the data from the R-CNN matched with the laser parameters, a back propagation neural network (BPNN) was trained to act as a predictive network. The network predicts the number and proportion of defects when the tool grain size, roughness and laser parameters are entered. The accuracy of this predictive network was 96.6%. The effect of individual laser parameters on the surface integrity is estimated by this method, enabling the optimization of laser processing in cutting tools. For the first time this can be used to predict tool performance based on tool’s surface integrity.