Detecção de defeitos em juntas soldadas de tubulações de petróleo em radiografias computadorizadas parede dupla vista dupla (PDVD) por redes neurais

Abstract

Detection of weld defects in radiographic images aims to ensure the safety of analyzed structures in order to avoiding financial losses and prevent against environmental damage. Nowadays, the inspection of welded joints is essentially a human activity and, therefore, it is subject to errors related to the inspector visual acuity, experience, fatigue and distractions, affecting the repeatability and reproducibility of this process. In this sense, this work presents a method to assist the detection of weld defects in welded joints of petroleum pipelines in computed radiography acquired by Double Wall Double Image (DWDI) technique. The developed method involved the application of contrast enhancement of treated images, segmentation of discontinuities and, the search space reduction by eliminating the central region of the DWDI weld. Thus, these procedures contributed to that segmented discontinuities which correspond to potential weld defects regions were classified by Multilayer Perceptron Neural Networks, performing the detection of weld defects.