Intelligent Classification of Tungsten Inert Gas Welding Defects: A Transfer Learning Approach

Automated and intelligent classification of defects can improve productivity, quality, safety various welded components used in industries. This study presents a transfer learning approach for accurate tungsten inert gas (TIG) welding while joining stainless steel parts. In this approach, eight pre-trained deep models (VGG16, VGG19, ResNet50, InceptionV3, InceptionResNetV2, Xception, MobileNetV2, DenseNet169) were explored to classify images into two-class (good weld/bad weld) multi-class weld/burn through/contamination/lack fusion/lack shielding gas/high travel speed) classifications. Moreover, four optimizers (SGD, Adam, Adagrad, Rmsprop) applied separately each the maximize prediction accuracies. All evaluated based on testing accuracy, precision, recall, F1 scores, training/validation losses, accuracies over successive training epochs. Primary results show that VGG19-SGD DenseNet169-SGD architectures attained best (99.69%) (97.28%) classifications, respectively. For “burn through,” “contamination,” “high speed” defects, most ensured productivity quality assurance TIG joints. On other hand, weld was promoted during “lack fusion” gas” defects. Thus, methodology help boost joints by good bad welds.