SpinCastML

SpinCastML is an open-source, executable framework for distribution-aware and chemically constrained inverse design of electrospinning processes. It replaces trial-and-error optimization with a reproducible, data-driven, and probabilistic decision-making approach aligned with the intrinsic variability of electrospinning. The framework is built on a rigorously curated dataset of 68,480 fiber-diameter measurements from 1,778 peer-reviewed studies across 16 polymers, making it one of the largest electrospinning datasets reported to date. SpinCastML integrates polymer-balanced Sobol and D-optimal sampling to address data imbalance and ensure robust generalization across chemically distinct systems. Machine-learning models are benchmarked under strict validation, and predictions retain the full fiber-diameter distributions rather than single averaged values. Inverse electrospinning design is formulated as a probabilistic inference problem using a chemically constrained Inverse Monte Carlo (IMC) engine. For a user-defined target diameter and tolerance, SpinCastML returns families of feasible electrospinning conditions with quantified success probabilities, explicitly accounting for non-uniqueness, uncertainty, and regime-dependent behavior. Chemical realism is enforced through explicit polymer–solvent solubility and solvent–solvent compatibility constraints, ensuring that predicted formulations are experimentally actionable. The software is distributed as a standalone executable, enabling laboratories to analyze their own datasets, reproduce results, and perform inverse design without prior machine-learning expertise. By reframing electrospinning as a distributional, inverse, and chemistry-aware design problem, SpinCastML establishes a new standard for data-driven nanofiber manufacturing and provides a generalizable blueprint for inverse design in complex, variable nanomanufacturing systems.