Identifying suitable, non-standard process parameters for any additive manufacturing process is oftentimes accompanied with extensive, resource-intensive parameter optimization, that does not generalize well across different processing conditions. Calculations and simulations as alternatives run into limits when it comes to highly complex multi-physics problems.

Artificial intelligence (AI) can be used together with large-scale data collection, to enhance process control, providing a possible solution to this challenge. To improve the performance of such models, a hybrid methodology is developed, based on the leading example of laser beam powder bed fusion (PBF-LB). This methodology combines the mathematical concept of dimensional analysis (DA) with AI. As a first step, DA is used to reduce the complexity of the high-dimensional PBF-LB process by identifying physically meaningful groups of key dimensionless numbers representing the underlying physics. The resulting dimensionless numbers are then used to train neural networks to assess process suitability and predict outcomes with reasonably high accuracy. The results potentially enable the derivation of process adaptations and design guidelines for PBF-LB.