No abstract

This work is a study to estimate the accuracy of classification of the main classes of weld defects detected by radiography test, such as: undercut, lack of penetration, porosity, slag inclusion, crack or lack of fusion. To carry out this work non-linear pattern classifiers were developed, using neural networks, and the largest number of radiographic patterns as possible was used as well as statistical inference techniques of random selection of samples with and without repositioning (bootstrap) in order to estimate the accuracy of the classification. The results pointed to an estimated accuracy of around 80% for the classes of defects analyzed. Introduction: The non-destructive radiographic method of inspection has been widely used over the decades to evaluate the integrity of material and equipment in a wide range of industries. In the specific case of radiographs of welded materials, the research for the development of an automatic or semiautomatic system of analysis of radiographs of welded joints has grown considerably in the last years and especially in the last 10 to 15 years [1-10]. The latest publications are mainly concerned with this last stage of defect classification where the authors normally use techniques of neural networks, Fuzzy logic and hybrid systems to implement classification patterns. As in all cases the number of samples used to estimate the parameters of the classifiers (in the case of neural networks: their synapse vectors and bias) is small, making it extremely difficult to divide the training and test sets with a number of statistically significant samples to estimate the accuracy of the classification adequately with data not used in the training of the classifiers. The question arises: What is the true accuracy of weld defect classification? There are few results published relating to this matter, but notably among them is one of the last publications of Liao [7]. The aim of this present work is to present the methodologies used and the results obtained in this study to estimate the classification accuracy of the main classes of weld defects, such as: undercut (UC), lack of penetration (LP), porosity (PO), slag inclusion (SI), crack (CR) and lack of fusion (LF). The non-linear classifiers were implemented using artificial neural networks. The largest possible number of radiographic patterns and statistical interference techniques of random selection of samples with and without repositioning (bootstrap) was used to estimate the accuracy of the classifiers. The results are presented in tables with estimated accuracies for each classification defect studied. It should be pointed out that this work is the continuation of previous works already published, which will be commented on briefly [11-14].

No abstract

Corrosion under insulation (CUI) is one of the major concerns of oil and petrochemical installations as damage evolves invisibly under insulation layers and usually revealed on the occurrence of leaking or more catastrophic failure. Methods to early detect CUI and its causes is an urgent necessity to assure safety and performance of insulated process piping. Oil and petrochemical plants are often considered explosive environments in which thermal excitation devices are forbidden. The present work aims then on the consolidation of a passive thermographic methodology to reliably detect moisture trapped under insulation layers that will cause corrosion. The proposed methodology focuses on the thermal behaviour of the piping structure during process variations and interactions with the external ambient. The partial least-squares analysis showed promising performance on separating different physical phenomena and creating cleaner images for defect detection.

No abstract