Explainable Ensemble Machine Learning for Predicting Deposition Characteristics in Advanced Additive Manufacturing

Sandeep Jain, Pradyumn Kumar Arya

In advanced manufacturing processes, precise deposition behavior prediction is crucial for process parameter optimization. In order to forecast significant deposition responses such as bead width (w), bead height (h), energy input (EI), and volumetric input (VI) based on process parameters like laser power (P), travel speed (v), and wire feed rate (fw), seven different machine learning (ML) models were developed in this study, including Random Forest (RF), Gradient Boosting (GB), Extreme Gradient Boosting (XGB), LightGBM (LGBM), Extra Trees (ET), Support Vector Regression (SVR), and Elastic Net (EN). The predictive power of all models was assessed using the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R2). The discoveries showed that ensemble models performed better than traditional ML techniques. The GB model performed the best overall, followed by the XGB model, which showed strong generalization and high prediction accuracy across training, validation, and testing datasets. Additionally, computational efficiency research discovered that the GB model holds a moderate model size and practically quick training time, making it suitable for real-world application. The robustness of the chosen model was supported by a paired t-test, which verified that the performance differences between the GB model and other models are statistically significant (p < 0.05). Furthermore, the impact of input parameters on the anticipated responses was interpreted using SHAP (SHapley Additive exPlanations) analysis. The interpretation results exhibited that while wire feed rate mostly affects volumetric deposition behavior, laser power and travel speed are the key parameters monitoring energy input and bead geometry. The GB and XGB models were used to forecast deposition reactions using specific process parameters, and the predictions were compared with experimental findings in order to further confirm the predictive power of the developed models with better performance. Overall, the results display that this study, in combination with ensemble boosting models, offers a reliable framework for understanding complicated relationships between deposition features and processing parameters, providing insightful information for process optimization in advanced manufacturing applications.