A hybrid Taguchi – Machine learning framework in optimizing FDM printing parameters

Fused Deposition Modeling (FDM) is one of the most widely adopted additive manufacturing technologies; however, the mechanical performance of printed components is strongly influenced by printing parameters. This study proposes a hybrid Taguchi–Machine Learning framework to optimize the printing parameters of ABS PA-747H using the Taguchi method, analysis of variance (ANOVA), and Random Forest Regression (RFR). An L9 orthogonal array was employed to evaluate the effects of layer thickness, build orientation, infill density, and printing speed on tensile and flexural properties. The optimal parameter combination (A1B1C1D1: 0.2 mm layer thickness, 0° build orientation, 100% infill density, and 10 mm/s printing speed) increased tensile strength from 30 to 40 MPa (33.3%), flexural strength from 69 to 92 MPa (33.3%), tensile modulus to 1.90 GPa, and flexural modulus to 2.15 GPa. The RFR model, validated using 5-fold cross-validation, demonstrated excellent predictive performance with R² values exceeding 0.92 for all mechanical responses, together with low MAE, RMSE, and MAPE values, confirming its high prediction accuracy. Fractographic analysis revealed fewer voids, improved interlayer bonding, and shorter crack propagation paths under the optimized printing conditions, explaining the enhanced mechanical performance. The results demonstrate that the proposed hybrid Taguchi–Machine Learning framework provides an accurate and robust strategy for optimizing FDM process parameters and improving the structural integrity of additively manufactured components.