Deep learning for automated defect recognition in tungsten inert gas welds of stainless steel 304

Identifying weld defects is crucial for ensuring the integrity of welded structures. Traditional visual inspections are subjective and prone to errors, while the integration of machine learning (ML) and artificial intelligence (AI) offers significant improvements in automated, accurate weld defect detection. For better accuracy, speed, and consistency in quality control, this study aims at how deep learning algorithms can be used to automatically find defects in SS304 TIG welds. Using visible spectrum camera images, AI-driven models can accurately identify defects in welding in real time, reducing the need for manual checking and reducing the chance of mistakes made by humans. This deep learning method improves weld quality assurance, making it easier to check the integrity of SS304 TIG welds in a more accurate and cost-effective way. The finding demonstrates significant improvements. A model with fewer epochs achieved an F1-score of 86% and an accuracy of 95%, while a model with more epochs attained an F1-score of 88% and an accuracy of 96%. The models demonstrated exceptional efficacy in predicting errors like inadequate shielding gas, with an F1-score of 65%. A fully connected layer model improved the F1-score by 39.2% and test accuracy by 37.7%, exhibiting remarkable efficacy in detecting the burn-through defect, attaining an F1-score of 0.89. The results of this study show that AI and ML models using advanced regularisation and deep learning techniques are a better and more reliable way to find weld defects. This makes TIG welding better and more reliable.