Recognition of Crack Formation Modes in Laser Thermal Splitting Using Convolutional Neural Networks

Abstract

This paper considers the problem of predicting crack behavior during laser thermal splitting of silicate glass, an important material used in the production of microelectronics and optics components. Given the high requirements for edge quality and the need for early detection of deviations, it is proposed to use computer vision and deep learning methods to automate the control. The proposed approach is based on the ResNet-50 convolutional neural network adapted to the task of analyzing video data in real time. Fine-tuning of the last layers of the network made it possible to achieve high accuracy in classifying crack development. The results demonstrate the promise of using ResNet for problems of monitoring laser thermal splitting of brittle non-metallic materials.