Data-driven optimization of droplet stability in binder jetting

Binder jetting additive manufacturing relies on stable droplet formation to ensure part quality and structural integrity. However, droplet behavior is governed by nonlinear interactions among operating parameters, making systematic optimization challenging. This study develops data-driven surrogate models to predict and optimize inkjet droplet stability using voltage and pulse width as input variables. Five stability indicators were extracted from imaging, and an aggregated stability score was defined based on 2,400 samples. Baseline regression models showed limited generalization capability, while four machine learning approaches were evaluated for comparison. Among them, Gaussian process regression (GPR) achieved the most reliable predictive performance, with a training R2 of 0.9182 and a leave-one-out cross-validation R2 of 0.8982. Optimization using the GPR surrogate identified an operating condition 81 V and 205 µs that yielded the lowest experimentally measured stability score, improving performance by more than 10% compared to the previous minimum. Although discrepancies in absolute predictions were observed, the model successfully captured the underlying stability trend. These results demonstrate that droplet stability in binder jetting should be interpreted as a multi-indicator surrogate-based optimization problem and highlight the importance of model selection for achieving robust predictive performance under limited experimental datasets. The proposed framework provides a practical strategy for surrogate-based process optimization in the binder jetting system.