Although selective laser melting (SLM) has the advancements of fabricating complex geometric, multi-material and multi-functional structures, several defects still affect the process stability, in particular spatter. The formation of spatter in SLM depends on the process parameters, and these can potentially be used to tune the process to obtain better product quality. However, there is still a lack of efficient methods for processing spatter images to allow in situ detection of the onset of spatter. In this paper, an in situ monitoring method for acquiring spatter images in SLM was presented. A maximum-entropy double-threshold image processing algorithm based on genetic algorithm (MEDTIA-GA) was proposed to recognize spatter from images, and its results were compared with three conventional threshold segmentation methods: Otsu’s method, Triangle threshold segmentation algorithm, and K-means clustering algorithm. Results show that MEDTIA-GA method was able to eliminate three types of errors: noise sensitivity, spatter conglutination, and spatter omission. In addition, the average processing time of 37 ms for MEDTIA-GA method was far shorter than those for other three conventional threshold segmentation methods. Finally, the relationship between the spatter area as well as the number and the laser energy density were analyzed

尽管选择性激光熔化（SLM）具有制造复杂的几何，多材料和多功能结构的进步，但仍有一些缺陷会影响工艺稳定性，特别是飞溅。SLM中飞溅物的形成取决于工艺参数，这些参数可潜在地用于调整工艺以获得更好的产品质量。然而，仍然缺乏用于处理飞溅图像以允许原位检测飞溅发生的有效方法。本文提出了一种在SLM中获取飞溅图像的原位监测方法。提出了一种基于遗传算法的最大熵双阈值图像处理算法（MEDTIA-GA）来识别图像飞溅，并将其结果与三种常规阈值分割方法进行了比较：Otsu方法，K均值聚类算法。结果表明，MEDTIA-GA方法能够消除三种类型的误差：噪声灵敏度，飞溅物粘连和飞溅物遗漏。此外，MEDTIA-GA方法的平均处理时间为37 ms，远远短于其他三种常规阈值分割方法的处理时间。最后，分析了飞溅面积，数量与激光能量密度之间的关系。