Classificação de defeitos de soldagem em imagens radiográficas PDVD de tubulações de petróleo: uma abordagem com ensemble de Extreme Learning Machines
The inspection of radiographic images of welded joints is very subjective and is subject to errors of interpretation by the inspector. In this context, a great effort has been made in the last years to develop automatic and semiautomatic methods for detecting defects in welded joints. This research...
| Autor principal: | Boaretto, Neury |
|---|---|
| Formato: | Tese |
| Idioma: | Português |
| Publicado em: |
Universidade Tecnológica Federal do Paraná
2018
|
| Assuntos: | |
| Acesso em linha: |
http://repositorio.utfpr.edu.br/jspui/handle/1/2890 |
| Tags: |
Adicionar Tag
Sem tags, seja o primeiro a adicionar uma tag!
|
| Resumo: |
The inspection of radiographic images of welded joints is very subjective and is subject to errors of interpretation by the inspector. In this context, a great effort has been made in the last years to develop automatic and semiautomatic methods for detecting defects in welded joints. This research work presents an automated method for the detection and classification of defects in radiographic images of welded joints of pipes obtained by the double wall double image (DWDI) exposure technique obtained in real field situations and which generally have a lower quality than the images used in other studies. The proposed methos identifies the region of the weld bead, detects the discontinuities and classifies them as defects and non-defects, highlighting in the image the result. Classifiers are evalueted using methods of classification by multilayer perceptron (MLP) neural networks, extreme learning machines (ELM) neural networks, and Support Vector Machines (SVM). The proposed method for identifying the region of interest reached 100% precision in the segmentation od the weld bead. The SVM classifier performed better than the MLP and ELM classifiers in all scenarios tested. Using ELM ensembles, an F_score of 85,7% was obtained for a test patterns database, satisfactoryresults when compared to similar works. The use of ensembles of ELMs represents a gain of only 0,5% in the F-score compared to the best result of the individually trained network, however, with the use of ensemble decision threshold ranges, the presented method allows to show the discontinuities about which the ensemble is not sure, highlighting in the image these discontinuities as a region of uncertainty, leaving to the specialist the final evaluation of these discontinuities. The image resulting from the application of the method serves as an aid to the expert in the elaboration of reports. |
|---|