Background: The tensile performance of fused deposition modelling components depends on thermal bonding, internal geometry, and deposition conditions. This study evaluated whether printing speed, infill density, and layer thickness could be used to improve the tensile strength of polylactic acid specimens and model the resulting response. Methods: ASTM D638-type tensile coupons were designed in SolidWorks, sliced in SuperSlicer, and printed on a Creality Ender 3 with 1.75 mm PLA and a 0.4 mm nozzle. A three-factor, three-level full factorial design generated 27 parameter combinations: printing speed of 40, 60, or 80 mm/s; infill density of 5, 10, or 15%; and layer thickness of 0.12, 0.20, or 0.28 mm. Tests were conducted on a 100 kN universal testing machine at 5 mm/min. A 3-4-1 feed-forward artificial neural network trained by the Levenberg-Marquardt algorithm was used to estimate tensile strength. Results: Measured strength ranged from 8.76 to 17.46 MPa, with an overall mean of 13.58 MPa. Mean strength increased from 11.44 to 15.75 MPa across the investigated layer-thickness range and from 12.55 to 14.99 MPa across the infill-density range, whereas printing speed produced a smaller main-effect trend. For the 27 reported observations, model predictions achieved RMSE = 0.121 MPa, MAE = 0.081 MPa, R² = 0.9974, and r = 0.9987. The highest observed response occurred at 40 mm/s, 5% infill, and 0.28 mm layer thickness. Conclusions: Within the tested domain, layer thickness and infill density governed the principal response trends, and the neural network closely reproduced the available observations. Independent validation, raw replicate analysis, and a confirmation experiment are required before the reported setting can be generalized as an optimum.

Keywords: additive manufacturing; artificial neural network; fused deposition modelling; full factorial design; PLA; tensile strength