Machine Learning of Optimal Processing Conditions in Ultrafast Laser Welding of Glass Based on Laser Emission State

In this study, we developed a machine-learning framework to predict the success of ultrafast laser microwelding of glass substrates based on plasma-emission spectra and laser-processing parameters.Ten-millimetre linear welds were produced under varying laser power, repetition rate, and scanning speed.The emission spectra collected during welding, combined with the corresponding processing conditions, were used to train support vector machine (SVM) and neural network (NN) classifiers.Both models demonstrated strong predictive performance, achieving over 80% accuracy on test data.Although a slight decrease in accuracy was observed on newly acquired data, this result highlights the potential for further improvement through data diversification.By expanding the dataset and incorporating environmental factors such as temperature and humidity, we aim to enhance the robustness and generalisation capability of the models.This approach offers a promising path toward improving the reliability and efficiency of ultrafast laser microwelding processes.