Comprehensive unraveling of CT-derived defect characteristics in fused deposition modeling via machine learning: clustering, classification, regression and process correlation

Additive manufacturing, enabling the integrated fabrication of lightweight and multifunctional components, is increasingly used in aerospace and advanced manufacturing. However, fused deposition modeling layer-by-layer deposition and parameter sensitivity of microstructural evolution inevitably introduce defects with diverse characteristics. To overcome challenges of existing works related to insufficient defect populations coverage, low-dimensional defect descriptors, and fragmented analysis frameworks, this work investigated fused deposition modeling fabricated PA6/CF cubic components under 27 process conditions using full-volume computed tomography. By integrating multi-dimensional defect extraction with statistical, a machine learning framework combining clustering, classification, and regression was developed, enabling systematic mappings from defect characteristics to process correlation. The results demonstrated that 9 defect characteristics spanning 4 categories varied with layer thickness, printing speed, printing density, and printing temperature; K-Means clustering identified 2 representative cluster centers driven by process combinations; among 4 classification models, Decision Tree reliably predicted defect categories from process parameters, while among 4 regression models, AdaBoost enabled quantitative prediction of key defect characteristics. Overall, the proposed machine learning-driven framework provides a viable approach to clarify fused deposition modeling defect characteristics, optimize process, and facilitate robust process-defect relationship modeling.