Dimensionless framework improves transferability in prediction of porosity of aluminum and steel welds

High-power laser beam welding (LBW) has gained wide use in advanced manufacturing for its ability to produce deep and narrow welds at high processing speeds. Porosity of the welds, however, remains a major quality issue. Prediction of pore formation relies on a complex set of interactions, assumptions of which cannot be transferred between materials.

Meng et al. developed a universal machine learning framework that leverages physical data to predict porosity levels in the LBW of aluminum and steel. Combining porosity ratios obtained from systematic laser welding experiments with a validated 3D multiphysics simulation model, the group’s approach characterized molten pool behavior and keyhole dynamics under corresponding conditions by using dimensionless features derived from physical properties.

“This work demonstrates a possible route from material-specific defect prediction toward more transferable and physically interpretable models for laser manufacturing,” said author Xiangmeng Meng. “In many welding and additive manufacturing studies, machine learning models are highly dependent on the specific material, machine, or parameter window used for training. Our results suggest that using physically meaningful, dimensionless features can help overcome part of this limitation.”

Conventional data-driven models train directly on process parameters and other inputs, which limits their operating range because they lack incorporation of underlying physical mechanisms. The dimensionless physical features in the group’s approach incorporated physical mechanisms such as keyhole stability and bubble transport, improving the model’s interpretability and transferability across different metallic materials.

The framework’s predicted porosity values achieved promising predictive performance. The authors found that Stokes number and the keyhole ratio were the two most dominant physical factors driving porosity formation.

The group next looks to test the framework with a wider range of materials, welding constructions, and defect types.

Source: “A universal physics-informed machine learning framework for the prediction of porosity defects in high-power laser beam welding with different metallic materials,” by Xiangmeng Meng, Marcel Bachmann, Fan Yang, and Michael Rethmeier, Journal of Laser Applications (2026). The article can be accessed at https://doi.org/10.2351/7.0002164 .

This paper is part of the Proceedings of the International Congress of Applications of Lasers and Electro-Optics (ICALEO 2026) Collection, learn more here .