Multi-track Multi-layer Laser Additive Manufacturing Quality Monitoring Dataset

To construct the dataset, 48 multi-track and multi-layer specimens were fabricated using 30CrNi2MoVA steel powder under varied process parameters. The folder "pic_all" contains all the CCD image data collected during the printing process of these 48 sample sets, with each experimental file named according to the format "Power-Speed".A local-to-local spatiotemporal mapping strategy was adopted to generate a labeled dataset, leveraging multi-track, multi-layer melt pool image sequences to characterize the localized porosity states within the part.

Specifically, a 0.4-second sampling time window was employed to continuously extract molten pool image data along the temporal axis of the laser AM process, while simultaneously sliding along the specimen’s length to annotate the corresponding input data with the total volume of local defects within each spatial window. Each sample corresponds to the data from a single 0.4-second sampling window. This approach preserves the local dynamic variations of the molten pool and establishes a spatiotemporal correlation between melt pool evolution and defect formation.