A Study of Online Melt Pool, Plume, and Spatter Tracking in Laser Powder Bed Fusion using DBSCAN

Additive manufacturing offers unmatched design flexibility, but this advantage comes with inherent process uncertainties that can lead to defects affecting mechanical properties. Ensuring consistent quality requires reducing part-to-part variations, yet traditional ex-situ characterizations, such as X-ray computed tomography, are time-consuming and constrained by part complexity. To address this challenge, we propose a low-cost, in-situ monitoring framework utilizing unsupervised learning to detect and classify defects in Laser Powder Bed Fusion (LPBF). Our approach leverages optical imaging data captured during printing to identify anomalies. In this study, image processing techniques, including contrast enhancement, normalization, and noise filtering, are applied to extract critical features such as melt pool, plume, and spatter behavior, which are key indicators of print stability. Principal Component Analysis (PCA) is then used to reduce dimensionality while preserving key process variations. Subsequently, unsupervised clustering techniques, such as Density-Based Spatial Clustering of Applications with Noise (DBSCAN), are applied to classify printing conditions to detect outliers indicative of defect formation. This study provides valuable insights into the relationship between in-situ monitoring data and defect formation in LPBF, demonstrating the feasibility of machine learning-driven real-time quality assurance. By correlating detected outliers with actual porosity and defects, this approach has the potential to reduce reliance on post-processing inspections, improve defect prediction accuracy, and enhance the reliability of additive manufacturing processes.