A ROBUST ALGORITHM FOR NONDESTRUCTIVE TESTING OF WELD SEAMS

The detection of welding flaws by means of nondestructive inspection methods remains open to the development of new algorithms and methods of inspection. One of the most widely used techniques is radiographic analysis, which requires interpretation by trained inspectors. Unfortunately, manual inspection is subject to various factors that can alter performance in the detection of the faults. An automated welding fault segmentation algorithm is presented using a set of digitized radiographic images. The result of the study has allowed the development of the following scheme: first, use the median filter to reduce noise; second, apply the bottom-hat filter to separate the hypothetical faults from the background; third, determine the segmented regions by binary thresholding; fourth, use the filters provided by morphological mathematics to eliminate over segmentation; and fifth, use the watershed transform to separate the internal regions. The results of the study have generated a general ROC curve on a set of 10 images with an area A_z = 93.6%.