WOA-XGBoost-SHAP: An explainable CFD-driven surrogate framework for gas-assisted injection molding

Gas-assisted injection molding (GAIM) represents an advanced polymer processing technique for producing lightweight components, yet gas-flow behavior remains complex due to interfacial instability and fingering in branched domains. This study develops a computational fluid dynamics (CFD)-driven and eXplainable AI framework to quantitatively investigate and predict gas-flow behavior in a complex fishbone-shaped GAIM geometry. A numerical dataset focusing on gas penetration length and fingering behavior is generated through comprehensive CFD simulations and automated image-based post-processing. Given that these responses are highly sensitive to processing parameters and exhibit complex non-linear relationships, a robust surrogate modeling approach is required to map these interactions accurately. To ensure the foundational reliability of the dataset, the CFD model is first validated against existing literature, confirming its ability to capture physical gas-flow characteristics. To predict these responses and ensure model transparency, the integrated WOA-XGBoost-SHAP framework is proposed. Within this framework, the optimized WOA-XGBoost model significantly outperforms the inherently powerful eXtreme Gradient Boosting (XGBoost) in terms of predictive accuracy. Model performance assessed via leave-one-out cross-validation yields R CV 2 values above 0.85 for both responses. In addition, external CFD test cases are used to further examine the generalization capability of the trained surrogate models under unseen conditions. Furthermore, SHapley Additive exPlanations (SHAP) are utilized to interpret parameter contributions and elucidate deeper insights into coupled response behavior, effectively supporting more informed and intelligent process design in GAIM.