Optimizing electrospinning parameters for nylon 6 nanofibers: A statistical and machine learning approach

It is a key challenge in the electrospinning of polymeric nanofibers to accurately control the fiber diameter due to the nonlinear and complex relationships among process parameters. When experimental datasets are limited, it is a resource-intensive and time-consuming practice to perform optimization using the conventional trial-and-error technique. In this data-driven approach, the electrospinning parameters for the fabrication of Nylon 6 nanofibers were optimized using machine learning techniques. The Taguchi L9 orthogonal array was used to generate experimental data by systematically varying parameters such as the polymer concentration, applied voltage, flow rate, and needle-to-collector distance, with the average fiber diameters measured using scanning electron microscopy. Various regression and ensemble machine learning models were trained and evaluated using systematic hyperparametric tuning. Among all the models used, the Extra Trees regressor provided the most reliable predictive performance, achieving a coefficient of determination R² score of 0.99 and Mean Average Error (MAE) of 0.2003 with an overall efficiency of 99.89%. The SEM-derived fiber diameters were used to validate the model predictions, confirming the experimental consistency. The feature-importance plot proved the dominance of the applied voltage and flow rate. The proposed approach significantly reduces the experimental effort required for Nylon 6 nanofiber synthesis.