Intelligent Optimization of Gas-Assisted Electrospinning via LLM-Guided Bayesian Inference.

Nanofiber-based structures have shown considerable potential in semiconductor-related applications, including ultra-thin dielectric layers and flexible electronic devices, owing to their tunable micro-/nanoscale morphology. However, the manufacturing of these structures is often hindered by the complex multiparameter coupling and poor reproducibility inherent in conventional electrospinning processes. To address these challenges, this study develops an intelligent optimization framework for gas-assisted electrospinning by integrating Large Language Models (LLMs) with Bayesian Optimization (BO). A Gaussian Process Regression (GPR) surrogate model was established to navigate the high-dimensional parameter space efficiently. Comparative studies demonstrate that the proposed BO+LLM strategy not only outperforms pure data-driven BO and pure knowledge-driven LLM approaches but also surpasses the conventional Response Surface Methodology (RSM) baseline, successfully locating a verified minimum fiber diameter of 239 nm. Furthermore, through response-surface analysis, this work identifies a specific multiphysics collaborative window where electrostatic stretching and aerodynamic assistance are balanced. These findings provide a robust pathway for the reproducible fabrication of nanofiber-based electronic devices.