Fractal analysis of weld defect patterns obtained by radiographic tests

TL;DR

This study applies fractal analysis techniques to radiographic images of weld defects, demonstrating that fractal features can effectively aid in classifying different defect types in radiographic testing.

Contribution

The paper introduces a novel application of fractal analysis and classification methods to improve pattern recognition of weld defects in radiographic images.

Findings

Fractal features successfully differentiate defect types.

Karhunen-Loeve transformation improves classification accuracy.

Fractal analysis enhances radiographic defect pattern recognition.

Abstract

This paper presents a fractal analysis of radiographic patterns obtained from specimens with three types of inserted welding defects: lack of fusion, lack of penetration, and porosity. The study focused on patterns of carbon steel beads from radiographs of the International Institute of Welding (IIW). The radiographs were scanned using a greyscale with 256 levels, and the fractal features of the surfaces constructed from the radiographic images were characterized by means of Hurst, detrended-fluctuation, and minimal-cover analyses. A Karhunen-Loeve transformation was then used to classify the curves obtained from the fractal analyses of the various images, and a study of the classification errors was performed. The obtained results indicate that fractal analyses can be an effective additional tool for pattern recognition of weld defects in radiographic tests.