• Random Forest accurately predicts deposit geometry in air-based CSAM. • Height, feed rate, and scan speed control deposit shape and uniformity. • RF model effectively captures complex non-linear process-geometry links. • Enables data-driven optimization for scalable CSAM process development. This study develops and validates a Random Forest (RF) model for predicting deposit geometry in air-based cold spray additive manufacturing (CSAM). The model captures non-linear relationships among key process parameters—gas pressure, temperature, powder feed rate, spray angle, scan speed, and target height—and the resulting deposit profiles. Compared with a Neural Network (NN), the RF model demonstrates higher accuracy, achieving RMSE = 0.28 mm and R 2 = 0.98, whereas the NN attains RMSE 0.59 mm and R 2 = 0.91. The model reproduces experimental cross-sections across multiple targets (2 × 6, 4 × 6, 10 × 6 mm) and defines process windows yielding balanced height-to-width ratios. The optimized RF models reveal that geometric fidelity is strongly influenced by height, feed rate, and scan speed, providing crucial insight for process optimization and reproducible deposit formation. These results highlight the potential of interpretable machine-learning frameworks to enhance predictive accuracy and enable scalable process design in industrial CSAM applications.
Hutasoit et al. (Sun,) studied this question.