AutoML-Based Prediction of Process Outcomes in Expanded Polypropylene Autoclave Foaming

The foam injection molding process for expanded polypropylene (EPP) offers advantages in energy efficiency, material savings, and mechanical performance, making it suitable for automotive and packaging applications. However, its nonlinear and multivariable nature makes accurate prediction and process optimization difficult using traditional methods. This study presents an AutoML-based surrogate modelling framework for predicting and optimizing two key process outcomes: cycle time and warpage. Experimental data were obtained from 81 production trials conducted on a Teubert injection molding machine using five process parameters. The PyCaret library was used to automate model selection, hyperparameter tuning, and performance evaluation, and 18 regression algorithms were compared. The results showed that tree-based ensemble models clearly outperformed traditional linear models. Gradient Boosting Regressor achieved an R2 of 0.9671 for cycle time, while Extra Trees Regressor reached an R2 of 0.9874 for warpage. Beyond prediction, the finalized surrogate models were used for Monte Carlo-based design space exploration with 1000 synthetic parameter combinations. Non-dominated solutions were identified to construct the Pareto front, and the top-ranked settings were compared with experimental results. The findings show that AutoML-based surrogate modelling can support both accurate prediction and process window identification in EPP foam injection molding.