Accurate ultimate tensile strength classification in friction stir welding of symmetric AA5052 weld seams using ensemble deep learning model

Abstract

This research introduces a comprehensive classification and analysis system tailored for accurately determining the ultimate tensile strength (UTS) of weld seams. Traditional UTS assessment methods typically depend on destructive testing, which tends to be both lengthy and expensive, limiting their continuous application in quality control of welding procedures. This study leverages deep learning techniques, segmenting the dataset into subsets for training and validation in both multi-class and binary classification scenarios. The model devised in this study incorporates cutting-edge methodologies including geometric enhancement, U-Net based image segmentation, an image augmentation of diverse deep learning models, and decision fusion tactics. A significant aspect of this research was the success of Experiment 33, which skillfully combined various methodologies, resulting in outstanding performance. This experiment demonstrated exceptional accuracy in multiclass classification, alongside impressive outcomes in binary classification, achieving a high accuracy rate of 97.4% and an F1 score of 96.5%. This level of accuracy is indicative of the average performance across all models that incorporated the He-UWA for decision fusion strategy. It encompasses the efficacy of all models using He-UWA, with or without image segmentation. These findings underscore the effectiveness of our proposed model in accurately classifying UTS in friction stir welding. This represents a crucial advancement in assessing the quality of welding processes and provides a solid foundation for future investigations in this area.