Machine learning-assisted prediction modeling for anisotropic flexural strength variations in fused filament fabrication of graphene reinforced poly-lactic acid composites

The objective of this study is to conduct a comparative analysis of various machine learning algorithms on the flexural properties of graphene-reinforced poly-lactic acid fabricated through fused filament fabrication. The selected process parameters include raster orientation (0°, 45°, 90°), layer thickness (0.05 mm, 0.1 mm, 0.15 mm, 0.2 mm, 0.25 mm, 0.3 mm), and feed rate (20 mm/s, 40 mm/s, 60 mm/s). The fabricated specimens underwent flexural testing, and fractography was performed after the testing. The flexural strength variations depicted maximum and minimum values of 129.511 MPa and 59.959 MPa, respectively. Further, the flexural testing data is used for the evaluation of the effectiveness of various machine learning algorithms, namely linear regression, random forest regression, gradient boosting regression, extreme gradient boosting regression, voting regression algorithms, and artificial neural networks. The results show that linear regression outperforms all other alternatives, obtaining a coefficient of determination of 98.9%, as compared to 95.7%, 96.8%, 97.3%, and 97.8%, respectively, for random forest regression, gradient boosting regression, extreme gradient boosting regression, and artificial neural network. Consistently superior performance is observed for linear regression over other performance metrics as well, i.e., mean absolute error and root mean square error. By demonstrating the efficacy of machine learning in conjunction with optimization techniques inspired by nature, this study advances predictive methodology in material science, especially in the field of additive manufacturing.