Automatic Detection of Root Pass Weld Defects of Gas Pipelines Using Expert Nonlinear Classifier

Automatic Detection of Root Pass Weld Defects of Gas Pipelines Using Expert Nonlinear Classifier

… Download …


Part of #Automatic Detection of Root Pass Weld Defects of Gas Pipelines Using Expert Nonlinear Classifier# :

Publishing year : 2009

Conference : Second Tube and Related Industries Conference

Number of pages : 10

Abstract: The welded joints in radiogram often contain defects which the interpreter must identify and quantify, before he decides on their acceptability, referring to nondestructive testing standards and codes. Once the radiographic segmentation is accomplished, providing a description in the term of regions (defect and background), then the problem is to interpret their contents. It is therefore a question of identifying effective attributes that allow to characterize these defect regions and even recognize them like class elements easily identifiable. In industrial radiography, we can obtain radiograms on which weld defects, if they exist, can have different sizes and orientations. In recent years, there has been a significant advance in the research for the development of an automated system for analyzing the weld defects detected by radiographs. In a normal welded gas pipeline there are four passes of welding including rootpass, hotpass, fillerpass and coverpass. This work describes the study of nonlinear pattern classifiers, implemented by artificial neural networks, to classify the weld defects existing in rootpasses which have seen in radiographic images of welds. Mainly the most important defects related to root passes are the lack of penetration (LOP), internal concavity or suck back, internal root subcorut and burn through. Using a novel approach for this area of research, a criterion of neural relevance was applied to assess the disability of the classes studied by the used features, with the aim of proving that the feature quality is more important than the used features. The results prove the efficiency of the techniques for the data used.