Page 78 - Reliability and Maintainability of In service Pipelines

Pipeline Inspection and Maintenance 67

Machine learning algorithms are used in the analysis of the MFL images.
These algorithms look at two primary issues:
Binary classification of image segments into injurious or metal loss defect,
and noninjurious safe or nonharmful objects.
Estimating the severity of defects.

2.3.5 RADIOGRAPHIC TESTING

Radiographic testing (RT) is carried out by recording degree of absorption of pen-
etrating radiation throughout the pipe wall. This produces a latent image of the
object being examined on a film which is then chemically processed to transform
latent image into a permanent shadow image of internal and external conditions
of the object. A greater amount of radiation passes through a defected area com-
pared to a region without defective issues. These radiographic images can be
evaluated by either human operators or the automated computer vision system
currently being tested for accuracy.
Radiography film has been the foundation of NDT for the last 50 years. As
discussed previously, manual inspection methods and similar procedures are
extremely time-consuming and often cannot be 100% accurate or reliable.
A large number of radiographic images are required over the long distances of
pipeline networks and this makes the identification process of welding defects
very time-consuming. Additionally, it is even more challenging as human opera-
tors have to assess and evaluate the large number of images and often people will
have different opinions regarding defects. Developments in digital image proces-
sing and computer vision have allowed the extensive use of automated visual
inspection to be tested and applied more frequently since it is much more consis-
tent and effective in terms of operating compared to human operations. It also
allows the inspection of pipelines in unsafe conditions to be analyzed in more
detail. The automated computer vision system utilizes the use of radiographic
films of welded pipelines which are produced by a radiographic testing.

2.3.6 ACOUSTIC DETECTION

Lohr and Rose (2003) spotted the potential of using ultrasonic guided circumfer-
ential and longitudinal waves propagating in a pipe wall to detect the pipe and
load properties. For guided waves in a circumferential experiment, it was shown
that the amplitude of the received signal decreases as the properties of the load
inside the pipe (air, oil, water, vegetable shortening) or wall thickness changes.
Alterations of up to 1 mm could be recorded. In longitudinal experiments, it was