DATA-DRIVEN QUALITY OPTIMIZATION FOR METAL ADDITIVE MANUFACTURING THROUGH MACHINE LEARNING

Abstract

Metal additive manufacturing has emerged as a transformative production technology for fabricating geometrically complex, lightweight, customized, and high-performance components across aerospace, automotive, biomedical, energy, tooling, and advanced engineering sectors. Despite these advantages, consistent quality remains a major challenge because final part characteristics are influenced by complex interactions among process parameters, feedstock properties, thermal history, melt-pool dynamics, machine condition, scanning strategy, layer geometry, environmental conditions, and postprocessing operations. Conventional quality control generally depends on fixed parameter windows, trial-and-error experimentation, destructive testing, offline inspection, and operator experience. Such approaches are expensive, time-consuming, and insufficient for identifying dynamic process variations during layer-by-layer fabrication. This paper proposes a data-driven quality optimization framework for metal additive manufacturing through machine learning. The proposed methodology integrates heterogeneous process sensing, edge preprocessing, multi-source data fusion, process parameter management, layer-wise quality monitoring, machine-learning-based defect prediction, Digital Twin-assisted process representation, adaptive parameter recommendation, explainable quality analytics, secure API interoperability, MLOps lifecycle governance, and sustainability-aware computational management. Real-time and historical information involving laser power, scanning speed, hatch spacing, layer thickness, build orientation, melt-pool temperature, optical intensity, acoustic emissions, vibration, chamber conditions, energy consumption, feedstock characteristics, surface quality, porosity, dimensional deviation, and defect observations are continuously collected and analyzed. Machine-learning models identify relationships between process conditions and quality outcomes and predict the probability of defects before completion of the entire build. The optimization layer recommends feasible process configurations according to quality, productivity, material utilization, energy consumption, and equipment constraints. A representative evaluation demonstrates improvements in defect prediction accuracy, porosity reduction, surface quality, dimensional consistency, first-pass yield, process stability, anomaly response time, material utilization, and energy efficiency compared with conventional parameter selection and isolated monitoring approaches. The findings indicate that machine-learning-driven quality optimization can transform metal additive manufacturing from a predominantly reactive inspection process into a predictive, adaptive, and continuously improving production system.