A Method for Identifying Girth Weld Defect Based on Fusion of Domain Features in Transformer Model
The identification of types of girth weld defects is of great significance for the quality evaluation of welding seams in long-distance pipeline and the evaluation of pipeline service life.Due to the small feature differ-ences and low contrast in radiographic inspection images of different types of defects,the features extracted by the existing defect type identification methods have insufficient characterization capability,and their accuracy and credibility are difficult to meet industry requirements.Therefore,a weld defect type identification method based on defect domain feature extraction and Transformer fusion was proposed.First,based on the knowledge of technical personnel for girth weld defect evaluation and the defect domain features,14 features were defined from 3 aspects such as defect geometric features,defect location features and defect background areal features to achieve feature characterization of different types of defects.Second,based on the Transformer model,the above 14 features were fused to present a depth adjustable defect type identification model and method.Finally,the actual girth weld de-fect data from the enterprise were used to verify the model.The results show that the model has improved the classi-fication accuracy by 18.2% and 14.3% respectively compared to the ResNet50 model for lack of fusion and lack of penetration defects,while for the undercut and concave defects,the classification accuracy reaches over 90%,Proving that this method can effectively improve the defect type identification accuracy,and can be extended to the type identification field such as radiographic inspection of casting defects and TOFD inspection of weld defects.
万方数据
维普
