Current Non Destructing Testing (NDT) techniques for oil and gas pipeline inspection are accurate and reliable but there are limited numbers of continuous monitoring technique available that can automatically make real-time decisions on the status of the pipeline. Furthermore, most of the NDT methods are deployed at predetermined interval which can last for several months. Sudden onsets of defects are undetected and lead to pipeline failure and unscheduled shutdown. A reliable inspection method is required whereby the pipelines are monitored continuously and are able to provide the operators sufficient time to plan and organize shutdowns. In order to implement this, a continuous monitoring technique is needed which can detect defects automatically with minimal human intervention. Support Vector Machine (SVM) is a powerful machine learning technique for classification, however, the training phase requires batch data to find a model and this is not feasible for a continuous NDT system. This thesis proposes a novel method where the SVM training phase is able to find a model from the incremental dataset acquired from Long Range Ultrasonic Testing (LRUT) system. Results show that this method has comparable accuracy compared to the batch data method. Traditionally, SVM training data is labeled by an expert, however in a continuous monitoring NDT, it is not practical to assign an expert to label the continuously acquired data. Therefore, a novel unsupervised training technique is proposed. The technique is able to cluster the acquired data into a few clusters accurately. The performance of the proposed technique is compared to Self Organizing Map (SOM) method and shows better results. This thesis also proposes a novel method to implement a Genetic Algorithm (GA) as the Quadratic Programming (QP) solver in the SVM efficiently. Conventional SVM implement Sequential Minimal Optimization (SMO) which requires that the data be sparse for optimal operation. The performance of the method is evaluated and shows comparable result to traditional methods. As such, this thesis provides the framework to perform unsupervised continuous monitoring for oil and gas pipelines using LRUT in real time.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:734348 |
| Date | January 2016 |
| Creators | Nik Zulkepeli, Nik Ahmad Akram |
| Publisher | University of Nottingham |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://eprints.nottingham.ac.uk/31204/ |
Page generated in 0.002 seconds