Identifying suitable process parameters in laser welding is challenging, as even small variations in process settings and energy distribution can significantly affect weld quality, while the number of feasible experiments is typically limited by cost and effort. The introduction of laser beam shaping further increases the dimensionality of the parameter space, making systematic exploration particularly difficult with conventional experimental approaches.

In this work, Bayesian optimization is employed to efficiently explore the expanded design space and jointly optimize process parameters and laser beam shape. Beam shaping is realized using a coherent beam combining (CBC) system, enabling dynamic control of the spatial intensity distribution. To facilitate efficient optimization, the beam shape is described by a low-dimensional parameterization of the intensity distribution, allowing systematic variation of spatial energy deposition while maintaining a compact representation of the beam profile.

Bead-on-plate laser welding experiments were conducted on AISI 316L stainless steel, varying laser power, welding speed, and beam shape parameters. The optimization objectives were defined based on weld quality criteria, including weld depth and seam width, as well as the avoidance of typical process defects such as pore formation, undercuts, and seam elevation.

The results indicate that the low-dimensional beam representation results in a structured optimization landscape that can be effectively explored using Bayesian optimization. Suitable parameter combinations meeting the defined weld quality targets were identified in fewer than 50 experiments, demonstrating the potential of the proposed approach for data-efficient process development in laser beam welding.