Machine learning-based prediction and experimental validation of electrospun PVDF fibers: unraveling the dynamics and control of the β-phase

The piezoelectric characteristics of polyvinylidene fluoride (PVDF) result from the all-trans configuration of polymer chains within the β-phase, predominantly achieved through the electrospinning technique. In this paper, the prediction of the β-phase in PVDF films produced via electrospinning is investigated using ensemble learning methods such as AdaBoost and random forest. This study additionally explores the effect of applied voltage on both the β-fraction and morphology. Enhanced material properties are observed with an increase in applied voltage, leading to the peak crystallinity at 20 kV (69.67%) and the maximum dielectric value at the same voltage (9.57). As the applied voltage increases from 14 to 22 kV, a reduction in nanofiber diameter from 70.62 to 61.03 nm is documented. A comprehensive dataset, incorporating parameters like applied voltage, molecular weight, tip-to-collector distance, and solvent selection, is curated from diverse literature sources for model training. This dataset is employed to predict the β-phase, and the model’s predictions are experimentally validated against our own results. Moreover, the outcomes predicted by the machine-learning model undergo analysis through the explainable artificial intelligence tool like LIME, offering insights into feature contributions at specific instances. The findings presented in this paper provide valuable insights for selecting the appropriate electrospinning parameter to achieve a specific percentage of β-phase, contributing to advancements in PVDF film fabrication.