Machine Learning Assisted Laser-Powder Bed Fusion Process Optimization for AISI 316L-Cu Alloy

Metal Additive Manufacturing (AM) has revolutionized the production of complex metal components by enabling the fabrication of intricate geometries with high precision. This technology's potential can be significantly enhanced by integrating artificial intelligence (AI) methods, particularly Machine Learning (ML), which offers advanced capabilities in establishing complex interrelationships and improving system and product quality control. ML algorithms present a transformative opportunity to address manufacturing challenges, optimize resource consumption, and enhance process efficiency by exploring the intricate linkages between process parameters, material properties, part geometry, microstructural characteristics, and their resultant properties. In metal AM processes such as Directed Energy Deposition (DED) and Laser Powder Bed Fusion (L-PBF), ML applications extend beyond process optimization to include defect detection, in-situ monitoring, and the enhancement of manufacturability and repeatability of components. This thesis investigates optimizing critical process parameters in L-PBF using ML techniques to establish a relationship between process parameters and defect content of AISI 316L-2.5%Cu. Recognizing these relationships enables the optimization of process parameters to attain specific objectives, such as high productivity, minimizing defect content, or low surface roughness. This optimization method facilitates applications, balancing productivity and quality, allowing the selection of parameters that satisfy both criteria. The prediction accuracy of seven ML algorithms, Bayesian Regression (BR), Decision Tree Regressor (DTR), Gradient Boosting Regressor (GBR), Gaussian Process Regressor (GPR), K-Nearest Neighbors Regressor (KNN), Random Forest Regressor (RFR), and Support Vector Regressor (SVR) were analyzed. Following the assessment of multiple models with varying training and testing sizes for the density of samples, the Support Vector Regression (SVR) model has been identified as the most effective model. The optimized process parameters, derived from the best-performing ML model prediction, demonstrated an accurate relationship between process parameters and defect content for achieving relative density values above 99.5% or high productivity. The findings of this thesis validate the effectiveness of ML in enhancing AM processes and underscore the potential of data-driven methodologies to advance the field of laser-based AM.