Monitoring of Industrial Processes via Non-stationary Probabilistic Slow Feature Analysis Machine Learning Algorithm

• Author / Creator

  Scott, David

• This research develops a first of its kind machine learning (ML) algorithm, called probabilistic slow feature analysis (PSFA), that monitors and detects faults for non-stationary industrial processes. The novelty of this ML algorithm is that it can monitor and detect faults for non-stationary processes. Non-stationary processes are intrinsically hard to monitor and control, as the process moves in unpredictable patterns that typical ML algorithms cannot handle. The intent with the ML algorithm is to first apply it on electric submersible pumps (ESP) in oil & gas operations in order to improve efficiency, increase the longevity of operations, and reduce failures. This research enables an inspiration to challenge industrial process standards and create an application that can decrease environmental impacts, improve process efficiency, reduce process water usage, and reduce costs for Canadian companies.

  Existing research has been performed on systems that do not have transferability to oil sands processes due to vast differences in operations. Thus, a robust framework to explore the implications of optimization, control, and monitoring of an ESP for application in Canadian oil sands processes is required to capitalize on the significant benefits it can provide. The objective of this research is to optimize ESP operation by utilizing the designed ML algorithm that combines process understanding, control knowledge, and deep learning. By utilizing the optimized control framework, ESP efficiency will increase by determining when the process is under poor operations. After that, it lowers the environmental impact of oil sands processes by reducing electricity and water consumption. The increased efficiency will generate a more extended run life for ESPs and lower operating and maintenance costs for Canadian oil sands producers.

  This past year, in collaboration with the Department of Automation at Tsinghua University in Beijing, China, the non-stationary-PSFA (NS-PSFA) ML algorithm was designed. This algorithm is capable of monitoring non-stationary process data. NS-PSFA utilizes deep learning algorithms that recognize behavior between time series data and detects approaching issues from historical data. The method utilizes data-driven modeling, where if the pattern of failures or inefficiency is detected, process or mechanical upsets will be inferred.

  The advantages of this research in ESP optimization utilizing ML and deep learning is significant, given the relatively unexplored opportunities for these methods in application with the Canadian oil sands process. This research includes being able to reduce the environmental impact of oil sands processes in addition to excess water savings that will have an immense impact on surrounding communities and populations for years to come. The horizons of this research are vast in terms of the application of the research, as this type of framework can be adapted to other processes across various industries. As the optimization of ESPs is a critical threshold to an economically viable and environmentally conscious process, this research is of utmost importance to ensure Canadian oil sands producers thrive in an ever-growing competitive global oil and gas market.

• Subjects / Keywords

  • Fault Detection
  • Machine Learning
  • Process Control
  • Slow Feature Analysis
  • Process Monitoring

• Graduation date

  Spring 2020

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/r3-v8yd-xw60

• License

  Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.