An integrated experimental–computational framework for optimizing SLA 3D printing parameters via machine learning and genetic algorithms

This study presents an integrated experimental and computational framework for identifying optimal stereolithography (SLA) printing process parameters. A full factorial experimental design with 81 printing configurations evaluated the effects of build direction, build orientation, layer thickness, and post-curing time. Results showed clear mechanical anisotropy in SLA printed parts. Flat-built specimens at 0° showed the highest tensile strength, while upright builds performed the worst. Thinner layers (0.1 mm) and longer post-curing (∼45 min) further improved ultimate tensile strength, yield strength, and Young’s modulus. Four machine learning models (XGBoost, GPR, SVR, and FFNN) were trained to predict ultimate tensile strength. XGBoost achieved the best performance (R 2 = 0.74, lowest RMSE, MAE). Based on this result, XGBoost was coupled with a genetic algorithm (GA) to optimize the printing conditions. The optimization reached stable convergence within 200 epochs. The optimal setting was flat build direction, 1° build orientation, 0.1 mm layer thickness and 33 minutes of post-curing. Under these conditions, the GA predicted an ultimate tensile strength of 40.62 MPa, a yield strength of 38.01 MPa, and a Young's modulus of 2.347 GPa. Experimental validation of the GA-optimized setting yielded an ultimate tensile strength of 45.9 MPa, a yield strength of 38.05 MPa, and a Young’s modulus of 2.795 GPa, which confirmed conservative yet reliable predictions. Thus, the XGBoost-GA framework provides a reliable route for data-driven optimizing of SLA process parameters.