Data set for "Generalisable 3D printing error detection and correction via multi-head neural networks"

The dataset contains 1,272,273 labelled images of the the extrusion 3D printing process. A camera mounted next to the nozzle of the printer was used to capture images of material deposition for 192 different printed parts covering a range of geometries, material colours, and lighting conditions. Each image is labelled with: flow rate, lateral speed, Z offset, hotend temperature, hotend target temperature, bed temperature, timestamp, and nozzle tip x and y coordinates. To collect the data an automated pipeline was created to acquire and automatically label images from a fleet of 8 extrusion printers and to sample different combinations of printing parameters. The dataset provides a CSV of 948,396 pre-filtered images where complete failures, parameter outliers, dark images, and images just after parameter changes are removed. A raw CSV is also included labelling all images in the dataset. This dataset can be used for numerous applications such as real-time error detection, closed-loop control, and parameter prediction.

本数据集包含1,272,273张带标注的挤出式3D打印过程图像。我们在打印机喷嘴旁安装摄像头，用于捕获192个涵盖不同几何形状、耗材颜色与光照条件的打印件的耗材沉积过程图像。
每张图像均附带以下标注信息：挤出流量、横向移动速度、Z轴偏移量、热端实时温度、热端目标温度、热床温度、时间戳，以及喷嘴尖端的X、Y坐标。
为采集该数据集，我们搭建了自动化流水线，可从由8台挤出式3D打印机组成的机群中采集图像并自动完成标注，同时采样不同的打印参数组合。
本数据集附带一份包含948,396张预过滤图像的CSV文件，该文件已剔除完全失败的打印图像、参数异常图像、过暗图像以及参数变更后即刻拍摄的图像。
此外还附带一份原始CSV文件，用于标注数据集中的全部图像。
本数据集可应用于多项场景，例如实时错误检测、闭环控制以及打印参数预测等。