Additive manufacturing method of lattice structure based on material manufacturing performance driven: Using machine learning to optimize manufacturing process

The process parameters for the additive manufacturing (AM) of lattice structures are typically derived from standard component testing. However, size effects can influence the mechanical properties of structures with small-scale features, resulting in the degradation of the mechanical performance of lattice structures. In this study, we developed a machine-learning approach based on the manufacturing performance of materials to predict, optimize, and expand the process window of laser powder bed fusion, establishing a mapping relationship between geometric dimensions and mechanical properties. Using AlSi10Mg as an example, we employed a genetic algorithm (GA)-optimized neural network as the predictive model, incorporating multiple input parameters (laser mode, laser power, scanning speed, hatch spacing, strut diameter, and building angle) to predict the mechanical properties of basic strut units (ultimate tensile strength and elongation after fracture). In addition, a graphical user interface was developed to guide the production process of lattice structures. The experimental results demonstrate that the GA-optimized backpropagation neural network can accurately characterize the mapping relationship between sample build parameters and mechanical properties. Pulsed-laser processing was found to effectively enhance the mechanical performance of submillimeter struts. Compared with lattice structures produced with ideal material process parameters, the AM method that considers size effects and is driven by material manufacturing performance, achieved a maximum mechanical property enhancement of 14% for a lattice structure with a unit cell size of 5 mm. The proposed method enables the effective control of size effects and guides the optimization of AM processes involving structures with small-scale features.