The dataset is organized into three roots:
good_weld: labeled non-defect reference runs (750 runs in 43 configuration folders).
defect_data_weld: labeled defect runs (1580 runs in 80 configuration folders).
test_data: anonymized evaluation set (90 samples named sample_0001 ... sample_0090).
Training-available labeled pool (good_weld + defect_data_weld) contains 2330 runs.
Each run/sample is multimodal and typically includes:
one sensor CSV… See the full description on the dataset page: https://huggingface.co/datasets/H4R/therness.
