Assessment of glass-to-glass welding by USP lasers with machine learning approaches

Abstract

Glass welding using ultra-short pulsed (USP) lasers has become one of the promising technologies in the past decades. With appropriate settings of parameters, USP lasers can provide many advantages for glass welding. However, there is still a lack of studies focused on predictions and relationships between successful welding and its correlates by machine learning models from glass welding experimental data, causing no guidance when implementing such experiments in the laboratory. In this study, we report the results of glass welding using a femtosecond laser system. The welding conditions (i.e., success or failure) under different process parameters such as focal position, power, and the other four parameters are analyzed by Neural Network (NN), Logistic Regression (LR), and Classification and Regression Tree (CART). The prediction accuracies of the models are from 84.3% to 97.3%. In other words, the process parameters can be applied to similar experiments to enhance the success rate of glass welding using USP lasers. Therefore, this study can fill the gap of lacking analytical results on predictions and relationships between successful welding and its process parameters in glass welding using USP lasers.