Machine learning-assisted material development via Laser powder bed fusion process

Metal additive manufacturing (AM) revolutionized the fabrication of complex metal components, providing remarkable precision and flexibility in producing complex geometries. Integrating artificial intelligence (AI) can further revolutionize this field by highlighting complex relationships within manufacturing systems and enhancing quality control. Machine learning (ML) methods provide innovative solutions to optimize resource consumption, improve process efficiency, and address manufacturing challenges by correlating process parameters, material properties, part geometry, microstructural characteristics, and their resultant properties. In metal AM processes, ML applications extend beyond process optimization to include defect detection, in situ monitoring, and the enhancement of manufacturability and repeatability of components. This study optimizes key process parameters in laser powder bed fusion (L-PBF) to correlate the processing parameters and defect content in AISI 316L-2.5%Cu components. By applying ML algorithms, this research identifies optimal process parameter combinations to achieve specific objectives such as high production rate, low defect content, or superior surface quality. Seven ML algorithms (Bayesian Regression, Decision Tree Regression, Gradient Boosting Regression, Gaussian Process Regression, K-Nearest Neighbors Regression, Random Forest Regression, and Support Vector Regression) were systematically evaluated for their predictive accuracy across varying training and testing dataset sizes. Support Vector Regression (SVR) with a training size of 80% was chosen as the most accurate model for relative density prediction, with an average error of 0.62%. The optimized process parameters, derived from the best-performing ML model prediction, demonstrated a precise relationship between process parameters and defect content for achieving relative density values above 99.5% or high productivity. The optimized parameters obtained from this approach highlight the potential of ML-driven methodologies to balance productivity and defect content in AM processes. These findings demonstrate the importance of ML in advancing L-PBF technology and its broader applicability in metal AM.