AI-driven optimization of infill patterns and orientations for enhanced mechanical properties in 3D printing

Material-extrusion additive manufacturing offers substantial design freedom, but the tensile response of printed polymers remains strongly dependent on internal architecture and build orientation, which complicates simultaneous optimization of stiffness and strength. This study experimentally investigates the coupled effect of four infill patterns (Linear, Grid, Honeycomb, and Octagonal) and five build orientations (0°, 30°, 45°, 60°, and 90°) on Z-ULTRAT specimens and integrates the measurements into a data-driven multi-objective optimization framework. A full-factorial campaign comprising 20 printing configurations was conducted at fixed settings, including a nozzle temperature of 240 °C and an infill density of 50%, and quasi-static tensile tests were used to determine elastic modulus (E), yield stress (YS), tensile strength (TS), and breaking stress (BS). TS varies from 18.60 to 38.41 MPa, with the maximum obtained for Honeycomb at 90°, whereas E varies from 1.89 to 2.97 GPa and reaches its maximum for Honeycomb at 45°. The highest YS and BS are obtained for Grid at 60° (37.36 and 34.13 MPa, respectively). Four response-specific DNN models optimized by an improved grey wolf optimizer (DNN-IGWO) were used as surrogates for E, YS, TS, and BS. The model metadata report correlation coefficients R of 0.9805, 0.9820, 0.9815, and 0.9838, respectively; these values are interpreted as internal fit indicators because the original random split seed was not preserved. A fixed 14/4/2 train/validation/evaluation split with seed 2026 was used for supplementary sensitivity analyses with GridSearch, Optuna-TPE, BayesOpt, and low-data baseline models. MOGWO then generated Pareto solutions describing trade-offs among the four mechanical responses within the tested discrete design domain. The originality of the work lies in combining experimentally anchored response-specific surrogate modelling with Pareto-based optimization for the joint selection of infill architecture and build orientation in Z-ULTRAT.