Machine learning-based surrogate modeling for efficient prediction of Moldex3D injection molding

¹ DELTA Laboratory, ENSAM, Hassan II University, 20670 Casablanca, Morocco
² Process Engineering and Environment Laboratory, FST, Hassan II University, 28806, Morocco
³ Systems Engineering Laboratory, Hassania School of Public Works, Casablanca 20230, Morocco
⁴ Marie and Louis Pasteur University, UTBM, CNRS, FEMTO-ST Institute, F-90010 Belfort, France

^* e-mail: zineb.achor-etu@etu.univh2c.ma
^** e-mail: souad.tayane@univh2c.ma
^*** e-mail: zahraoui.yassine@ehtp.ac.ma
^**** e-mail: jaafar.gaber@utbm.fr

Abstract

Injection molding simulations using Moldex3D are essential for optimizing polymer processing, but they are often time-consuming and computationally expensive, especially when evaluating multiple materials and process conditions. This study examines the utilization of machine learning (ML) to accelerate simulation workflows by predicting key Moldex3D outputs without running full simulations. Two ensemble learning models–random forest (RF) and gradient boosting (GB)–were trained on simulation data from two polymer materials and used to predict three critical parameters for a third material, PP Domolen: filling time, packing sprue pressure, and warpage total displacement. The models’ performance was assessed with mean absolute error (MAE) and root mean squared error (RMSE). Quantitative results indicate that the RF model attained an MAE of 0.151 seconds for filling time and 3.409 MPa for packing pressure, closely matching the actual Moldex3D values. GB performed better in predicting warpage, with an MAE of 0.762 mm, compared to 1.094 mm for RF. These results demonstrate that ML models can provide accurate predictions with significantly reduced computational time. This approach offers a promising step toward real-time optimization in polymer engineering and supports the broader adoption of AI-assisted manufacturing.