Predictive Quality Modeling for Ultra-Short-Pulse Laser Structuring utilizing Machine Learning

Laser structuring offers precision and versatility for material processing but holds potential for optimization due to high-energy consumption and long production-times. Based on a process parameter study, we utilize Machine Learning and multi-modal data fusion of process parameters, high-frequency monitoring data and workpiece properties. We perform a benchmarking to analyze how data and algorithm characteristics impact modeling accuracy. We show that given enough data, ensembles achieve high accuracy and robustness and observe that accuracy strongly depends on initial workpiece properties. The higher the influence of laser structuring, the more superior is the inclusion of time-series features extracted from monitoring data.