Spatial Attention-based Segmentation of Spatter Particle Defects from X-ray Computed Tomography for Metal Additive Manufacturing

Metal additively manufactured surfaces differ notably from machined surfaces due to the printing process and material characteristics.Unlike machined surfaces, Additive Manufacturing (AM) surfaces often exhibit defects such as spatter particles, cracks, and surface-breaking pores.Consistently and efficiently detecting these defects is challenging, particularly for internal surfaces.X-ray Computed Tomography (XCT) offers an effective solution for capturing both external and internal surface data.However, its efficiency is limited, and the process remains subjective due to the reliance on manual inspection.In this work, a spatter particle defect dataset is established and we develop a spatial attention-based segmentation algorithm to accelerate defect detection and eliminate the subjectivity of manual segmentation.The model is trained to segment spatter particles from XCT measurements of metal AM samples.Comprehensive comparisons during experiments show that our segmentation approach outperforms existing methods.