Machine Learning-Based Taylor Cone Analysis for Electrospinning Process-Window Identification and Cell-Laden...

Electrospinning is a powerful biofabrication technique for constructing fibrous scaffolds that mimic extracellular matrix architectures. However, the process is highly nonlinear and involves strongly coupled parameters, making fiber architecture and biological performance difficult to control empirically and limits reproducibility. Most machine learning (ML) studies still focus on static correlations between processing parameters and outcomes rather than on dynamic modeling of electrospinning evolution and regulation, thereby hindering real-time, precise process control. Here, we establish a data-driven framework that links process parameters, Taylor cone geometry, and nanofiber/cell outcomes through high-speed imaging and ML. A custom image-processing pipeline is developed to automatically extract cone angle, height, and width from in situ images with improved accuracy via a rotating-ray-based geometric algorithm. Using datasets collected from three representative polymer systems (PVDF, PVB, and PEO), random forest models outperform support vector machine and minimum description length, achieving moderate-to-good prediction of Taylor cone descriptors (R²≈0.57-0.83) across varied voltages, flow rate, and needle-collector distances. We further extend this framework to PEO-based cell electrospinning. By correlating predicted cone geometries with experimentally measured viability of 293T cells and umbilical cord mesenchymal stem cells, we identify a Taylor cone "safe window" that simultaneously stabilizes jet formation and preserves cell survival (>95%). Guided by this window, a bilayer cell-laden scaffold comprising a PVB support layer and a PEO-cell layer is fabricated with improved process robustness. This enables stable delivery of high-density, highly viable cells and facilitates the fabrication of three-dimensional cell-laden nanofiber scaffolds. Overall, this Taylor cone analysis, ML-guided strategy provides a quantitative framework for process-window identification in polymer and cell electrospinning, and may support future integration with real-time monitoring, online inference, and closed-loop control in intelligent biofabrication.