Comparative study of evolutionary machine learning approaches to simulate the rheological characteristics of polybutylene succinate (PBS) utilized for fused deposition modeling (FDM)

Polymer filament fabrication and its printability are influenced significantly by rheological behavior. This influence can pose a significant obstacle when attempting to transition fused deposition modeling (FDM) from the laboratory to industrial or clinical settings. The aim of this study is to demonstrate how machine learning (ML) approaches can speed up the development of polymer filaments for FDM. Four types of ML methods: artificial neural network, support vector regression, polynomial chaos expansion (PCE), and response surface model, were used to predict the rheological behaivior of polybutylene succinate. In general, all four approaches presented significantly high correlation values with respect to the training and testing data stages. Remarkably, the PCE algorithm repeatedly provided the highest correlation for each response variable in both the training and testing stages. Noteworthy, variation differs between response variables rather than between algorithms. Taken together, these modeling approaches could be used to optimize filament extrusion processes.