Just-in-time and Adaptive Methods for Soft Sensor Design

  • Author / Creator
    Ma, Ming
  • In many industrial processes, critical variables cannot be easily measured on-line: they are either obtained from hardware analyzers which are often expensive and difficult to maintain, or carried out off-line through laboratory analysis which cannot be used in real time control. These considerations motivate the design of inferential sensors or so-called soft sensors to infer process quality variables in real time from on-line process measurements. Numerous modeling techniques have been proposed and successfully applied to soft sensors for many industrial processes. Despite the popularity of these techniques in industry, development and implementation of soft sensors are still challenging due to complexity of industrial processes. The main contribution of this thesis is the development of several soft sensing methods that can achieve and maintain satisfactory performance while handling multi-mode, nonlinear and time-varying problems.

    Real time identification of local process model, also known as Just-in-time (JIT) modeling, is a special modeling technique for design of infinite-mode soft sensors. It is widely used in dealing with nonlinear and multi-mode of industrial processes. The performance of JIT model depends on parameters of the similarity function as well as the structure and parameters of the local model. A Bayesian framework is proposed to provide a systematic method for real time parameterization of the similarity function, selection of the local model structure, and estimation of the corresponding model parameters in JIT modeling methods. Another challenging issue in JIT modeling is the selection of most relevant samples from database by considering input-output information. Thus, a new input-output similarity function is defined and integrated into a Bayesian framework for JIT modeling.

    To cope with time-varying behaviour of processes, on-line adaptation is usually integrated in the implementation procedure. Although there are a number of publications dealing with adaptation of soft sensors, few of them have considered the adaptation of nonlinear grey-box models which are popular in process industry. Thus, a new adaptation mechanism for nonlinear grey-box models is proposed based on recursive prediction error method (RPEM). Adaptive data preprocessing and cautious update strategy are integrated to ensure robustness and effectiveness of the adaptation.

    The effectiveness and practicality of the proposed methods are verified using data from industrial processes. Some of the proposed methods have also been implemented for industrial applications.

  • Subjects / Keywords
  • Graduation date
    Spring 2015
  • Type of Item
  • Degree
    Master of Science
  • 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.