Bayesian Solutions to Multi-model Inferential Sensing Problems

  • Author / Creator
    Khatibisepehr, Shima
  • In many industrial plants, development and implementation of advanced monitoring and control techniques require real-time measurement of process quality variables. However, on-line acquisition of such data may involve difficulties due to inadequacy of measurement techniques or low reliability of measuring devices. These concerns motivate the design of inferential sensors to infer process quality indicators from real-time measurable process variables. Development and implementation of inferential sensors entail many challenges that are often addressed in a rather ad hoc manner. Although many of the encountered challenging issues are interconnected, most of the existing solutions are disjoint. The main contribution of this dissertation is development of an integrative and holistic Bayesian inferencing paradigm to provide general and integrated solutions to certain outstanding inferential sensing problems. The core component of an inferential sensor is the process model which is identified through first-principles and process data analysis. The problem of model identification from contaminated data is formulated under a hierarchical Bayesian framework to simultaneously consider different aspects of data analysis and inferential modeling. A Bayesian approach is developed for identification of multi-modal systems switching among non-linear continuous-state dynamics. The proposed procedure provides a framework to accommodate the overlapping operating regions, facilitate the inclusion of prior knowledge about the operating conditions, and include a global adaptation mechanism within the envelope of previously identified operating conditions. Real-time identification of inferential models can be viewed as a special modeling technique for design of multi-model inferential sensors with infinite number of local models. A Bayesian framework is developed to provide a systematic and computationally feasible method for real-time similarity function parametrization and model structure selection in just-in-time/space modeling methods. One of the practical challenges faced in implementation of inferential sensors is to assess the accuracy of their real-time predictions. A data-driven Bayesian approach is proposed to capture conditional dependence of the reliability of inferential sensor predictions on characteristics of the input space and reliability of the empirical process model. The practicality and validity of the proposed Bayesian frameworks are verified using data from various simulation configurations, experimental set-ups, and industrial processes.

  • Subjects / Keywords
  • Graduation date
  • Type of Item
  • Degree
    Doctor of Philosophy
  • DOI
  • License
    This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. 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.
  • Language
  • Institution
    University of Alberta
  • Degree level
  • Department
  • Specialization
    • Process Control
  • Supervisor / co-supervisor and their department(s)
  • Examining committee members and their departments
    • Huang, Biao (Chemical & Materials Engineering)
    • Lee, Jay H. (Chemical & Biomolecular Engineering)
    • Shah, Sirish L. (Chemical & Materials Engineering)
    • Prasad, Vinay (Chemical & Materials Engineering)
    • Zhang, Hong (Computing Science)