Abstract:
Pipelines are considered to be the most widely used source for transportation of oil/gas
worldwide. For the past few decades the incidents of oil and gas pipeline failures are becoming
quite frequent, causing financial costs, environmental damages and health hazards. This situation
is due to the lack of accurate methods of inspection for oil and gas pipelines. Wireless sensor
network facilitates efficient supervising and checking of environment from distant locations
reliably. The negative pressure wave (NPW) technique is commonly used for fast leakage
detection. The difficulty with the NPW method is the complexity of the analysis of the pressure
signatures under high noise scenarios and presence of slow leakages. Hence, a full-proof
inspection method is required to provide monitoring of the condition of oil and gas pipeline. In
light of the issues of low efficiency and high false alarm rates in traditional pipeline condition
monitoring, we propose a wireless sensor monitoring framework which provides a real time
parametric view and sufficient information for critical oil and gas pipeline fault/leakage
detection. Data is acquired in real-time and processed in order to make decision to predict
presence of leakage in pipelines. The real-time nature of the data requires automated decision
making than personal inspection. For this purpose an intelligent machine learning based
algorithm is designed which can provide considerable accuracy for detection of slow /small
leakages in natural gas/oil pipeline monitoring wireless sensor networks (WSN). The information
processing algorithms in WSNs are mainly modified techniques from the field of
multidimensional data series analysis. We apply the methods of support vector machine (SVM)
using optimal kernel functions parameter and Gaussian mixture model (GMM) in multidimensional feature space. The proposed technique is validated using a series of experiments on
a field deployed test bed, with regard to performance of detection of leakages in pipelines.
