dc.description.abstract |
Determination of the size of a defect in a given material is important from industrial
usage point of view. In this work, a computational technique has been developed that
takes a humble step forward from just qualitative description of defect, such as “big” or
“small” to its area-wise quantification. Our program (by the name “DEFAREA”) accepts
a 2D grayscale image of an investigated specimen as input and sizes the irregular shaped
defects contained therein in terms of the area occupied by them. In case where a defect
feature is of regular shape being a projected image of a cylinder or a sphere the program
is also able to produce volumetric results. The program exploits the fact that defects offer
color contrasts that are different from the rest of the image (such as bone fracture in X-
ray radiograph). It is based on grayscale thresholding (GT) whereby it first iterates down
to compute a minimum value of graylevel that separates the first peak from the rest of the
distribution in the grayscale spectrum of the given input image. This threshold, which is
representative of a particular shade of gray color, is then used to identify, select and count
the number of pixels which have graylevel values below the computed threshold. The
number of segmented pixels within the whole image size then easily produces not only a
numeric fraction of the defective portion of inspected specimen but also the area
occupied by the defect if the physical sizes and dimensional measurements of the
specimen are known. The main part of the algorithm, however, revolves around devising
a reliable computational method to obtain a certainty range in the reported defect size.
Certainty range is needed as there physically exists a transition region (TR) between the
defective and the immaculate parts of the investigated object that can not be put in either
category. TR offers lesser contrast with the flawless part of the image than the pure defect
areas. So a given defect is doubly quantified with and without appending the transition
region around it with the aid of user-defined adjustability in the computed grayscale
threshold. Then finally an average value of defect size is calculated along with an
associated certainty.
The presented algorithm is validated against physical measurements of some
locally fabricated metallic plates having drilled holes of known sizes simulated as defects
in them in which the results indicate that it correctly selects and quantifies at least 94.7%
of the actual required regions of interest in a given image and it gives less than 8% false
alarm rate.
The algorithm is then applied to sizing of a wide range of defects commonly
encountered in nuclear industry regarding reactor fuels. The images of nuclear fuels used
as input in the program are collected from a reference standard source of neutron
radiographs. The present work confirms the ability to quantify various kinds of defects
such as chipping in nuclear fuel, cracks, voids, melting, deformation, inclusion of foreign
materials, heavy isotope accumulation and non-uniformity etc. The classes of fuel range
from those of research and power reactors to fast breeders, from fresh nuclear fuel to
post-irradiate, and from pellets to annular and vibro-compacted fuel. It is also
demonstrated that the program can handle a variety of image sizes, displays several
output modes of image segmentation and works well without the need of any
smoothening or eroding morphological operations. |
en_US |