Attention-enhanced neural network-based metaheuristic optimization frameworks for enhancing tensile strength in 3D printing

Abstract

A comprehensive framework to predict and optimize the tensile strength for fused deposition modeling (FDM) 3D printed polylactic acid specimens is presented in the study. The normalized printing speed, layer thickness, and nozzle temperature are used here in the proposed attention-enhanced neural network (AENN) prediction model, as input variables. A trained attention layer to assign weights per parameter and Monte Carlo dropout to quantify prediction uncertainty are utilized. Incorporating 27 full-factorial experiments, a mean absolute error (MAE) of 0.33 MPa and a maximum relative error (MRE) of 3.43% was achieved for AENN, outperforming the traditional Ridge Regression method (MAE of 2.61 MPa and MRE of 10%). Two metaheuristic optimization algorithms—Firefly algorithm and JAYA algorithm, are used to optimize print speed (39 mm/s), layer thickness (0.30 mm) and nozzle temperature (218°C) for maximum tensile strength. Both algorithms converge on nearly identical settings, with JA demonstrating slightly smoother convergence. Experimental validation with the average tensile strength of 43.52 MPa, confirms the AENN's predictive capability and optimization framework's robustness toward enhancing performance of FDM printed parts. Such frameworks can be adopted for other materials, additional parameters, or multiple objectives.