Page 105 - IJEEE-2023-Vol19-ISSUE-1
P. 105
Abed, Wali, & Alaziz | 101
force of soil stress [12]. Pipelines utilized in real-time Adebayo [23] points effect on pressure and velocity
operations are frequently located in these kinds of profiles were obtained
environments. The most serious issue is the difficulty in Sharma [24]
discovering leaks caused by physical corrosion and pipes age Shen and Cheng use data-driven intelligent models of
[13]. The situation deteriorates when used over an extended [25] machine learning to find the small
period without being discovered. Most leaks could be leakage depending upon the behavior
located using the trial-and-error method that involves Proposed of flow and heat of the gas pipeline
watching a change in the fluid flow pattern [14]. There may investigation
be a sudden change in pressure across the free stream due to use SVM for leakage detection in the
the leak, which would cause the flow to be distorted. The gas pipeline using image processing.
performance of systems that include the mobility of tiny
spherical particles about the fluids in which they could be use on-site data for leakage detection
immersed involves a wide range of phenomena that are utilizing machine learning models in
significant to researchers and engineers. The terminal Water Distribution Systems(WDS)
settling velocity of a single spherical particle in an infinite
fluid is of interest to various fields, as is the fluid flow drag - Fluid flow and temperature transfer
(pressure drop) that a sphere experiences during fluid flow. modeling of various fluids pipelines
Consequently, several empirical and theoretical in presence of leakage detection ball.
investigations have been concentrated on them. One may
reasonably presume that all significant problems were - Resultant data from numerical
addressed some decades ago. Significant physical and simulation using in training of
mathematical ramifications are associated with using an machine learning (SVM and DT)
analytical drag formula. The intricacy of the flow prevents algorithm for realistic leakage
an analytical description of the friction factor and viscous detection.
forces from being possible across a wide range of velocities
where the pressure drop of leakage detection cannot be II. MACHINE LEARNING IN PIPELINE LEAK DETECTION
easier to detect theoretically [15]–[17]. Research on the
pressure drop in the sphere area has been undertaken Support Vector Machine (SVM), which is selected since it
throughout a wide range of flow velocities, fluid viscosity, works well in a wide subspace, can efficiently manage vast
and density ever since the pioneering investigations of volumes of data, and is ideal for categorizing non-linear
Stokes, Osteen, and others for flow through a ball. The input, is the algorithm that will be used [26]. A support
characteristics of the fluid will change in general as a result vector machine is a type of supervised machine learning
of the change in the fluid [18], [19]. The recent studies are model that solves issues involving two groups of
performed using numerical simulation for leakage detection categorizations by using classification techniques. When an
modeling as seen in Table I in addition to processing the real SVM model is provided with sets of labeled training data for
or numerical data utilizing various machine learning each category, the model can classify newly encountered
Algorithms. text. When compared to the most recent algorithms, such as
neural networks, they offer two primary advantages:
TABLE I increased speed and improved performance with a
constrained quantity of data points (in the thousands).
LEAK DETECTION MODELING IN RECENT STUDIES AND Because of this, the approach is well suited for solving issues
PRESENT WORK SCOPE. involving the categorization of text, which often involves
having access to datasets containing no more than a few
Reference Scope thousand annotated examples at most [13].
Barbosa et al. [20] Modeling of leakage effect on Adebayo [23] uses data-driven intelligent models to find
pressure and velocity profiles of small leakage depending upon the behavior of the flow and
(oil-gas) two-phase flow in a gas heat of the gas pipeline. He found that SVM gives a good
pipeline using leak localization (LL) approximation for leakage detection but does not provide the
functionality most accurate measurement and DT is the most sensitive
algorithm. Sharma [24] uses SVM for leakage detection in
Sousa et al. [21] Modeling of leakage behavior on the pipeline using image processing. This approximation
pressure and velocity distribution gives a superior leakage indication based on practical reality.
under two-phase flow (water-oil) The image resolution promoted no superior accuracy (15 %
conditions in the presence using CFX tolerance). Shen and Cheng [25] use on-site data for leakage
(ANSYS) software where the Finite detection utilizing a machine learning algorithm. Adaboost,
Element Method and CFD modeling random forest, and Discussion tree are used in their
were applied investigation. DT has the most accurate approximation with
a minimum false rate.
a numerical investigation of two
Araújo et al., [22] leakage points on oil pipeline using This paper proposes techniques to estimate leak location
and leakage rate by creating a model to simulate leak
CFX simulation where the leakage detection in the oil transmission pipeline utilizing a movable