Investigation of the influences of the process parameters on the weld depth in laser beam welding of AA6082 using machine learning methods
Maximilian Schmoeller, Christian Stadter, Markus Wagner, Michael F. Zaeh
Abstract
The high-strength aluminum alloys of the AA6xxx group are characterized by their high thermal conductivity and dynamic process behavior during laser beam welding. Thus, the development of models as the basis for a robust control of the weld depth is a challenge. Optical Coherence Tomography (OCT) has been available for several years as a sophisticated measurement method for determining the keyhole depth. The information about the process behavior measured with OCT in combination with a process model of the welding depth as a function of the process parameters is the enabler for a precise and real-time capable control of the process. Machine learning methods can be used to describe the transient process behavior of the weld depth. With the help of a Beta-Variational-Autoencoder (β-VAE), a novel, data-based method for the development of a generative model for the prediction of the welding depth based on the process parameters was implemented. As dominant process parameters the laser beam power and the feed rate were determined.