Machine learning-based printability assessment and process control of aerosol jet 2D and 3D Printed PEDOT:PSS-based microstructures

Aerosol Jet Printing (AJP) is a high-resolution additive manufacturing technique applied for two-dimensional (2D) and three-dimensional (3D) microstructures. Typical inks refer to metal solutions, but also conductive polymers, such as poly(3,4-ethylenedioxythiophene):polystyrene sulfonate (PEDOT:PSS), have been widely printed by AJP. However, optimising the AJP process remains challenging due to the non-linear, geometry-dependent effects of parameters including carrier gas flow, sheath gas flow, and platen temperature. This study introduces a physics-informed machine learning (ML) framework for predicting print quality and selecting process parameters. To our knowledge, this is the first unified, data-driven comparison of 2D lines and 3D high-aspect-ratio (HAR) pillars with PEDOT:PSS inks, revealing how process physics shift from single-layer deposition to layer-by-layer growth. Datasets were expanded with virtual sample generation and physics-based feature engineering. Five ML models, namely Random Forest, Gaussian Process Regression, Support Vector Machines, Generalised Additive Models, and Neural Networks, were trained and tuned with nested cross-validation and Bayesian optimisation. The best-performing models achieved high predictive accuracy (R² > 0.9 for 2D; R² ≈0.7 for 3D). The framework shows that 3D AJP operates within a significantly narrower process window, with carrier gas flow requiring more than four times tighter control than in 2D AJP.