• +605 249 0500
  • enquiries@qiup.edu.my

Seminar 15: Oil and Gas Pipeline Failure Prediction System using Long Range Ultrasonic Transducers and Euclidean-Support Vector Machines Classification Approach

“Oil and Gas Pipeline Failure Prediction System using Long Range Ultrasonic Transducers and Euclidean-Support Vector Machines Classification Approach”
by: Assoc. Prof. Dr. Lee Lam Hong
School of Engineering and Computer Technology
Faculty of Integrative Sciences and Technology
Quest International University Perak
DATE: 11 December 2013 (Wednesday)
TIME: 3 – 4 p.m
VENUE: Lecture Room 4, Applied Sciences Building, QIUP

ABSTRACT

This work proposes an enhanced SVM (Support Vector Machines) approach as such performance has low impact on the implementation of kernel function and parameters.  We named this approach “Euclidean-SVM”.  We introduce the use of Euclidean distance function to replace the optimal separating hyper-plane in the conventional SVM as the classification decision making function.  This is due to the fact that the construction of the optimal separating hyper-plane is based on kernel function and parameters.  Eliminating the hyper-plane as the classification decision surface will make the classifier less dependent on kernel function and parameters.  We utilize the SVM training algorithm to reduce training data points by identifying and retaining only the SVs, and eliminating the rest of the training data points.  In classification phase, Euclidean distance function is used to make the classification decision based on the average distance between the testing data point to each group of SVs from different categories.  The experimental results show that an Euclidean-SVM approach makes the classification accuracy to have a low impact on the implementation of kernel function and parameters. The Euclidean-SVM classification framework has been implemented into the prototype of an oil and gas pipeline failure prediction system.  This prototype is developed by using a network of long range ultrasonic transducers to examine the thickness and condition of the pipeline wall, and the sensor data is transmitted to the centre processing unit which is equipped with an Euclidean-SVM classifier to generate results in predicting the condition of the pipeline.  The feasibility of the Euclidean-SVM approach for implementation into both prototypes has been proven with promising experimental results.

Back to FIST Seminar List

Scroll to Top