Machine learning enhanced metal 3D printing: high throughput optimization and material transfer extensibility

Abstract Metal 3D printing holds great promise for future digitalized manufacturing. However, the intricate interplay between laser and metal powders poses a significant challenge for conventional trial-and-error optimization. Meanwhile, the “optimized” yet fixed parameters largely limit possible extensions to new designs and materials. Herein, we report a high throughput design coupled with machine learning (ML) guidance to eliminate the notorious cracks and porosities in metal 3D printing for improved corrosion resistance and overall performance. The high throughput methodologies are mostly on obtaining the printed samples and their structural and physical properties, while ML is used for data analysis by model building for prediction (optimization), and understanding. For 316L stainless steel, we concurrently printed 54 samples with different parameters and subjected them to parallel tests to generate an extensive dataset for ML analysis. An ensemble learning model outperformed the other five single learners while Bayesian active learning recommended optimal parameters that could reduce porosity from 0.57% to below 0.1%. Accordingly, the ML-recommended samples showed higher tensile strength (609.28 MPa) and elongation (50.67%), superior anti-corrosion (I _corr = 4.17 × 10⁻⁸ A·cm⁻²), and stable alkaline oxygen evolution for >100 hours (at 500 mA·cm⁻²). Remarkably, through the correlation analysis of printing parameters and targeted properties, we find that the influence of hardness on corrosion resistance is second only to porosity. We then expedited optimization in AlSi7Mg using the learned knowledge and feed hardness and relative density, thus demonstrating the method’s general extensibility and efficiency. Our strategy can significantly accelerate the optimization of metal 3D printing and facilitate adaptable design to accommodate diverse materials and requirements.