Additive Manufacturing Powder Bed Four Types of Defects Dataset

This dataset is from the powder bed image acquisition experiment in the SLM250 selective laser melting (SLM) device. During the experiment, two high-resolution cameras (model VISION3D PowerScan-5M330) were used to perform multi angle imaging of the powder bed. The powder bed substrate is a square with a side length of 250 mm × 250 mm, and the image resolution is 2448 × 2048 pixels. Two monochrome cameras capture images from different angles to capture defect features on the powder bed. A total of 2477 powder bed images were collected in the experiment, including various artificially simulated defects, to simulate the types of defects that may occur during the actual printing process. Defect categories include: ditches (420 sheets), stacking (600 sheets), insufficient spreading (800 sheets), and craters (657 sheets). In addition, the dataset also includes five times the amount of data generated through random data augmentation and corresponding annotation files to improve the training performance of the model. The annotation file adopts two formats, XML * and TXT *, to select the defect areas in each image and provide defect category and coordinate information. The annotation of each defect includes category labels and position coordinates in the image, making the dataset suitable for research on object detection and defect classification.本数据集采自SLM250型选择性激光熔融（Selective Laser Melting，SLM）设备的粉床图像采集实验。实验中采用两台型号为VISION3D PowerScan-5M330的高分辨率相机，对粉床进行多视角成像。粉床基板为边长250 mm × 250 mm的正方形，图像分辨率为2448 × 2048像素。两台单色相机从不同角度采集图像，以捕捉粉床表面的缺陷特征。本次实验共采集2477张粉床图像，涵盖多种人工模拟缺陷，用以复现实际打印过程中可能出现的缺陷类型。缺陷类别包括：沟槽（420张）、堆粉（600张）、铺粉不足（800张）以及凹坑（657张）。此外，本数据集还包含通过随机数据增强生成的五倍体量数据及对应标注文件，以提升模型的训练性能。标注文件采用XML与TXT两种格式，可对每张图像中的缺陷区域进行框选，并提供缺陷类别与坐标信息。每一处缺陷的标注均包含类别标签与图像内的位置坐标，使得本数据集适用于目标检测与缺陷分类相关研究。