melt-pool
lpbf-melt-pool-quality-dataset
LPBF 熔池监测与成形质量合成数据集
面向激光粉末床熔融（Laser Powder Bed Fusion, LPBF）金属增材制造过程质量控制的教学/算法验证数据集。
数据集将熔池图像、工艺参数与传感器时序三路数据以同一样本编号对齐，可用于成形质量分类、
致密度回归与过程异常检测等任务。
⚠️ 数据性质声明：本数据集为程序合成的模拟数据（synthetic），非真实实验或企业产线数据。
生成脚本为 generate_sample_data.py（配套课程设计项目），随机种子固定为 42，可完全复现。
如需真实工业数据，请使用 NIST AM-Bench 等公开基准数据集。
数据集背景
LPBF 过程以高功率激光逐层熔化金属粉末成形零件，熔池的温度、形貌与稳定性直接决定成形件的致密度与缺陷水平。
实际产线中，熔池图像由同轴高速相机采集，温度与氧含量由过程传感器逐层记录，而激光功率、扫描速度、
层厚与扫描间距是可调的工艺参数。本数据集按上述物理关系构造三路相互关联的数据，用于在没有产线条件… See the full description on the dataset page: https://huggingface.co/datasets/zoumingwu/lpbf-melt-pool-quality-dataset.vqa-meltpool-dataset
NIR Meltpool Thermal-State VQA
A visual-question-answering dataset for training vision-language models to classify
the thermal state of a laser powder-bed-fusion (LPBF) meltpool from a single
near-infrared (NIR) frame. Derived from the hyperspectral_nir_meltpool_dataset
(sample 588).
Task
Given one 250x250 NIR meltpool image, answer the fixed query by choosing one of
six thermal-state classes.
id
label
id
label
0
baseline
3
strong underheat
1
edge
4… See the full description on the dataset page: https://huggingface.co/datasets/MoreGeometrico/vqa-meltpool-dataset.AM12_MeltPoolKinetics
Roles
Roles: canon repo — annot is the source label (normal / underheat / overheat / edge for the 4-class question, normal / anomaly for the binary one), kept machine-parseable as the gold for verification and reward parsing; the model reads query + image, and metadata.output_type says which of the two questions a record asks. The reasoning column is empty, so this repo is not itself a training view for chain-of-thought. metadata is provenance and must never be fed to a model.… See the full description on the dataset page: https://huggingface.co/datasets/AI4Manufacturing/AM12_MeltPoolKinetics.AM12_MeltPoolKinetics-perception
Roles
Roles: canon repo — annot is the source label (normal / underheat / overheat / edge for the 4-class question, normal / anomaly for the binary one), kept machine-parseable as the gold for verification and reward parsing; the model reads query + image, and metadata.output_type says which of the two questions a record asks. The reasoning column is empty, so this repo is not itself a training view for chain-of-thought. metadata is provenance and must never be fed to a model.… See the full description on the dataset page: https://huggingface.co/datasets/AI4Manufacturing/AM12_MeltPoolKinetics-perception.melt-pool-classificationAM13_MeltpoolNet-annotated
AM13_MeltpoolNet — CoT-annotated (T-AM2)
Chain-of-thought SFT annotation of the T-AM2 melt-pool-mode task from AM13_MeltpoolNet.
Each row: an LPBF process record (query), the gold melt-pool mode (annot), and a worked
physical reasoning chain ending in FINAL ANSWER: <mode>.
Source (un-annotated, 181-style queries): AI4Manufacturing/AM13_MeltpoolNet.
Teacher: claude-sonnet-5 (gold-conditioned rationalization; each reasoning derived from the
record's raw values, with error-prone… See the full description on the dataset page: https://huggingface.co/datasets/AI4Manufacturing/AM13_MeltpoolNet-annotated.