TIG Al Welding Defect Classifier Computer Vision Project

LABELS:

good weld 0 burn through 1 contamination 2 lack of fusion 3 misalignment 4 lack of penetration 5

@article{BACIOIU2019603, title = "Automated defect classification of Aluminium 5083 TIG welding using HDR camera and neural networks", journal = "Journal of Manufacturing Processes", volume = "45", pages = "603 - 613", year = "2019", issn = "1526-6125", doi = "https://doi.org/10.1016/j.jmapro.2019.07.020", url = "http://www.sciencedirect.com/science/article/pii/S1526612519302245", author = "Daniel Bacioiu and Geoff Melton and Mayorkinos Papaelias and Rob Shaw", keywords = "Automation, Convolutional neural networks, HDR camera, Vision, Process monitoring, Quality assessment", abstract = "Weld defect identification represents one of the most desired goals in the field of non-destructive testing (NDT) of welds. The current study investigates a system for assessing tungsten inert gas (TIG) welding using a high dynamic range (HDR) camera with the help of artificial neural networks (ANN) for image processing. This study proposes a new dataset (https://www.kaggle.com/danielbacioiu/tig-aluminium-5083). of images of the TIG welding process in the visible spectrum with improved contrast, similar to what a welder would normally see, and a model for computing a label identifying the welding imperfection. The progress (accuracy) achieved with the new system over varying degrees of categorisation complexity is thoroughly presented." }