Charles Darwin University

CDU eSpace
Institutional Repository

 
CDU Staff and Student only
 

Investigation of neural network modelling for the prediction of CO2 corrosion in the oil and gas industry

Kelly, Jacinta (2015). Investigation of neural network modelling for the prediction of CO2 corrosion in the oil and gas industry. Bachelor of Engineering (4th Year Project) Thesis, Charles Darwin University.

Document type: Thesis
Citation counts: Google Scholar Search Google Scholar
Attached Files (Some files may be inaccessible until you login with your CDU eSpace credentials)
Name Description MIMEType Size Downloads
Download this reading Thesis_CDU_48865_Kelly_J.pdf PDF version generated by student application/pdf 6.14MB 664
Reading the attached file works best in Firefox, Chrome and IE 9 or later.

Author Kelly, Jacinta
Title Investigation of neural network modelling for the prediction of CO2 corrosion in the oil and gas industry
Institution Charles Darwin University
Publication Date 2015
Thesis Type Bachelor of Engineering (4th Year Project)
Subjects 0913 - Mechanical Engineering
Abstract Carbon dioxide corrosion is responsible for a third of all corrosion related failures in the oil and gas industry. Due to the fact that there are too many environmental and metallurgical factors with complex interactions, it has been difficult to effectively predict the rate of carbon dioxide corrosion in any given situation.

Neural network modelling mimics the human brain by using pattern recognition and classification algorithms with given input and output data, learning from them and making decisions based on the learning. It is an empirical modelling tool that has been successfully applied in many diverse applications, including in the area of corrosion.

This work seeks to expand on previous studies and investigate neural network modelling and how well it can effectively predict corrosion rates in any given situation. The design process for this neural network model included the compilation of a comprehensive empirical database from a range of sources, a principal component analysis of selected input parameters, outlier detection and an iterative approach to creating the most appropriate network architecture.

Using a unique set of data for testing, the resulting neural network model has a correlation coefficient of 0.964, a root mean square error of 1.6mm/year and a relative mean square error of 100%. The final optimised ANN model was tested and compared against current industryused empirical and mechanistic models. Simulation results showed the proposed ANN model exhibits a more accurate and reliable method in predicting CO2 corrosion. The ANN model achieved a correlation coefficient of 0.964 with an accuracy of +/- 1.6mm and when simulated against other models, it achieved a mean relative error of 36%, which was superior against other industry-used models. It was also able to achieve 84% correct classification of the predicted corrosion rates against the Institute of Energy Technology (IFE) Severity Level ranking system. The other industry models only achieved a maximum of 55% correct classification.

Recommendations on the applicability and use of neural network modelling for industry use have been given. As a part of this work, the model has been subsequently developed and deployed as a MATLAB application with a graphical user interface as an example of how this model might be used in industry.
Keyword corrosion
carbon dioxide corrosion
modelling
neural network
probabilistic modelling


© copyright

Every reasonable effort has been made to ensure that permission has been obtained for items included in CDU eSpace. If you believe that your rights have been infringed by this repository, please contact digitisation@cdu.edu.au.

 
Versions
Version Filter Type
Access Statistics: 104 Abstract Views, 664 File Downloads  -  Detailed Statistics
Created: Tue, 18 Aug 2015, 09:32:28 CST by Jessie Ng