Predictive quality assurance of material extrusion process using artificial neural networks and ensemble models

Abstract In this research, artificial neural networks and ensemble models were developed to predict multi-input process parameters to refine the quality of material extrusion 3D-printed parts. The input process parameters include layer thickness, build orientation, build temperature, extrusion temperature, and print speed. The predicted output parameters included dimensional accuracy, tensile strength, and porosity. The results of the study illustrate that, for the neural networks, both BPNN and RBFNN outperformed GRNN on multiple occasions, especially for both porosity and tensile strength predictions, their performance stays consistent across each output prediction at different k -fold cross-validation values (3, 5, and 8). In the ensemble models, XGB outperformed RF across all output parameter predictions. Aside from this, XGB mean square error values were relatively consistent for both k -fold cross-validation values of 3, 5, and 8 for all output parameters predictions. Overall, the XGB performed excellently in predicting all the output process parameters with consistent mean square error and coefficient of determination values for all k -fold cross-validation values. This research serves as a predictive quality assurance tool for researchers and practitioners to estimate geometric and mechanical properties for a variety of practical applications.