Prediction and optimization of PETG flexural properties using Taguchi analysis and artificial neural networks for the FDM process

Abstract This study investigates the influence of various Fused Deposition Modeling (FDM) printing parameters on the flexural properties (flexural strength, modulus, and strain) of polyethylene terephthalate glycol (PETG) parts. Five printing parameters were selected—layer thickness, print speed, infill density, infill pattern, and nozzle temperature—each with five different levels. Using an L25 Taguchi orthogonal array, a total of 125 specimens were fabricated and subjected to three-point bending tests in accordance with the ISO 178 standard. Analysis of variance (ANOVA) identified infill density and nozzle temperature as significant parameters for the three flexural properties, while layer thickness exerted a significant influence on flexural strength and strain. The print speed impacted only modulus and infill pattern impacted the flexural strength. A multi-output Artificial Neural Network (ANN) model was developed to simultaneously predict all three flexural responses. The Bayesian Regularization algorithm (trainbr) with a {9-8-3} architecture outperformed all tested configurations, achieving a mean 5-fold cross-validation R 2 of 0.9571 and an MSE of 0.80792. To the authors’ knowledge, this is the first multi-output ANN framework simultaneously predicting flexural strength, modulus, and strain of PETG, extending beyond the single-response models and RSM approaches of prior work. A composite desirability-based multi-objective optimization identifies the globally optimal configuration (NT = 230 °C, LT = 0.20 mm, PS = 70 mm/s, ID = 100%, IP = Line) with a composite desirability of D = 0.907, explicitly quantifying trade-offs between the three flexural responses.