Experimental and <scp>ANN</scp> Modeling of Mechanical Properties and Surface Quality in Multi‐Material <scp>FDM</scp> Printing: Effects of Retraction Parameters

ABSTRACT In fused deposition modeling (FDM), retraction related parameters are usually adjusted to reduce stringing and oozing; however, their combined effects on mechanical performance and surface quality, particularly in multi‐material printing, remain insufficiently quantified. This study investigates how retraction control influences tensile strength and surface roughness in multi‐material FDM printing and evaluates the predictive capability of an artificial neural network (ANN) model in comparison with gaussian process regression (GPR), k‐nearest neighbors (K‐NN), and Taguchi based prediction. Five input factors were considered: material type (MT), retraction distance (RD), retraction speed (RS), retraction minimum travel (RMT), and extra prime amount (EPA), where the material types consisted of acrylonitrile butadiene styrene (ABS), polycarbonate (PC), and polyamide (nylon). A Taguchi experimental design, supported by analysis of variance (ANOVA) and machine learning (ML) based modeling, was employed to establish both statistical and predictive relationships between process parameters and the selected responses. Experimental results showed that tensile strength ranged from 18.21 to 64.57 MPa, with the maximum value of 64.57 MPa obtained for Nylon (RD = 4 mm, RS = 20 mm/s, RMT = 2.5 mm, EPA = 0.00 mm 3 ) and the minimum value of 18.21 MPa obtained for ABS (RD = 2 mm, RS = 20 mm/s, RMT = 0.9 mm, EPA = 0.08 mm 3 ). Surface roughness varied between 11.57 and 16.03 μm, where the lowest roughness (11.57 μm) was observed for ABS (RD = 4 mm, RS = 40 mm/s, RMT = 1.6 mm, EPA = 0.00 mm 3 ) and the highest roughness (16.03 μm) occurred for ABS (RD = 6 mm, RS = 60 mm/s, RMT = 2.5 mm, EPA = 0.08 mm 3 ). ANOVA results showed that all investigated factors had a statistically significant effect on both strength and roughness (). MT was identified as the dominant factor for strength, whereas roughness was mainly influenced by MT and EPA. In predictive modeling, ANN achieved the best performance with MSE = 0.9981, MAE = 2.4794, RMSE = 0.9908, MAPE = 1.6253, , and for tensile strength, and MSE = 0.4523, MAE = 1.4864, RMSE = 0.6724, MAPE = 1.0532, , and for surface roughness. The comparison models, GPR and K‐NN, yielded lower prediction accuracy, while Taguchi based estimates showed the weakest overall performance. These results show that retraction and priming parameters significantly affect strength and surface quality, and that the proposed framework effectively models and improves multi‐material FDM performance.