ERA

Download the full-sized PDF of Camera based Primary Separation Vessel Interface Level Detection and Estimation Utilizing Markov Random Field based Image ProcessingDownload the full-sized PDF

Analytics

Share

Permanent link (DOI): https://doi.org/10.7939/R3B56DH2M

Download

Export to: EndNote  |  Zotero  |  Mendeley

Communities

This file is in the following communities:

Graduate Studies and Research, Faculty of

Collections

This file is in the following collections:

Theses and Dissertations

Camera based Primary Separation Vessel Interface Level Detection and Estimation Utilizing Markov Random Field based Image Processing Open Access

Descriptions

Other title
Subject/Keyword
Image segmentation
Primary separation vessel interface level
Markov random field
Auto-logistic model
EM algorithm
Type of item
Thesis
Degree grantor
University of Alberta
Author or creator
Liu, Zheyuan
Supervisor and department
Huang, Biao (Chemical and Materials Engineering)
Afacan, Artin (Chemical and Materials Engineering)
Examining committee member and department
Prasad, Vinay (Chemical and Materials Engineering)
Yeung, Tony (Chemical and Materials Engineering)
Department
Department of Chemical and Materials Engineering
Specialization
Process control
Date accepted
2016-12-16T15:21:00Z
Graduation date
2017-06:Spring 2017
Degree
Master of Science
Degree level
Master's
Abstract
The level of the froth middling interface in primary separation vessel plays an important role in overall bitumen recovery in conventional oil sands bitumen extraction process. To maintain the interface within a certain range of level, the accurate measurement is always desired. Online camera detector usually has the best performance. However, it is not always reliable. The objective of this thesis is to develop a new approach to improve the reliability of camera sensor, which is Markov random field based image processing technique. An experimental setup was designed to simulate the liquid interface. Oil and water were used to form an interface because they are immiscible under normal condition. An online camera was installed to capture the image contain the interface level. Two differential pressure sensors were also installed, and the actual interface height could be determined based on those measurements. A Markov random field based supervised image segmentation technique was proposed to convert the raw image to a binary image. The interface level could be determined based on the vertical profile of averaged horizontal pixel values of the segmented image. The interface level estimations obtained using image processing technique were validated and compared with the estimations using traditional differential pressure sensors for various operating conditions. All results showed the agreement. An extended iterated conditional modes algorithm was proposed by introducing the k-means clustering as the initial estimation in the aim to reduce the computational cost. A customized neighborhood system was designed and aimed to reduce interface estimation variance. An approach of Gaussian mixture model and Markov random field based unsupervised image segmentation was also proposed. Expectation maximization algorithm was used to estimate the parameters in the model. The segmented images were compared with those obtained using Gaussian mixture model based unsupervised segmentation approach, and only the segmentation obtained using proposed Markov random field based approach agreed with those using the supervised technique. Interface level estimations were also compared and showed the agreement. To predict the segmented image and interface level, the spatial-temporal Markov random field based auto-logistic model was proposed to obtain the prediction. A two-step approach was proposed: the observed image at the future time was predicted using a modified random walk model, and then the image segmentation at the future time could be estimated using the proposed spatial-temporal Markov random field based auto-logistic model. Both predicted observed image and segmented image at the future time were validated by comparing with the real observation and segmentation. Though a number of pixels were mis-segmented, the interface level prediction was still agreed with real estimation.
Language
English
DOI
doi:10.7939/R3B56DH2M
Rights
This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for the purpose of private, scholarly or scientific research. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.
Citation for previous publication

File Details

Date Uploaded
Date Modified
2016-12-16T22:21:01.245+00:00
Audit Status
Audits have not yet been run on this file.
Characterization
File format: pdf (PDF/A)
Mime type: application/pdf
File size: 4002529
Last modified: 2017:06:13 12:15:45-06:00
Filename: Liu_Zheyuan_201611_MSc.pdf
Original checksum: c0e176634e7b12f9e80f3d212f1c3efd
Activity of users you follow
User Activity Date