Data-driven machine learning optimisation of silver nanoparticle synthesis

Precise control of silver nanoparticle (Ag-NP) size is critical for application-specific performance, yet identification of optimal synthesis conditions remains reliant on trial-and-error experimentation. Here, we report a machine learning inverse design framework trained on a combined experimental dataset spanning batch and continuous flow reactor configurations. Five regression algorithms were evaluated as candidate surrogate models, with Gaussian process (GP) regression emerging as the best-performing model (coefficient of determination (R ²) = 0.91 and cross-validation score = 0.80). A random sampling strategy was employed to explore the parameter space and identify synthesis conditions capable of achieving user-defined target Ag-NP hydrodynamic diameters. To differentiate between the multiple feasible solutions, a composite candidate scoring function was implemented to rank synthesis conditions according to the error, estimated Ag-NP dispersity, and posterior uncertainty. The framework was validated across 21 inverse design predictions spanning a target size range of 20-70 nm, achieving average size errors of 3.4% and 7.4% for the batch and flow reactors, respectively. Collectively, this work establishes an experimentally validated inverse design framework capable of identifying synthesis conditions for target Ag-NP hydrodynamic diameters via citrate- and tannic acid-mediated reduction of silver nitrate.