(175d) Development and Implementation of Process Analytical Toolbox | AIChE

(175d) Development and Implementation of Process Analytical Toolbox


Xu, F. - Presenter, University of Alberta
Tamayo, E. - Presenter, Syncrude Canada Ltd.
Huang, B. - Presenter, University of Alberta

Advanced Process Control (APC) applications, such as Model Predictive Control (MPC) technology, have been widely accepted and applied in process industries. Monitoring of APC performance especially economic performance has been a great interest both in academia and in industry. This motivated us to develop a state-of-the-art software that can be used to monitor the performance of APC applications and facilitate the task of APC maintenance. The software package that we have developed during the last few years is called Process Analytical Toolbox (PAT).

The package of PAT was written in MATLAB, which is widely used in academia by researchers. MATLAB itself is a very powerful language in computation with large amount of speciality toolboxes in different research areas. It also provides a compiler that can compile most MATLAB codes directly into executable stand-alone applications. Moreover, the MATLAB OPC Toolbox makes it possible to retrieve real-time process data directly from DCS system via standard OPC interface, which is available in almost all different kind of DCS systems in process industries. We took full advantage of these features in our development of PAT package. On the other hand, it is possible for researchers to run their applications on line with direct real-time process data. This makes it a very convenient shortcut for the technology transfer from academia to process industries.

PAT is composed of application components and interface components. Two interfaces, PI Driver and Data Collector, are provided to retrieve process data, either historical process data off-line from the PI server or real-time process data on-line from the DCS system. They are designed to provide process data to the application components from difference data sources. The main application components include Cluster Analysis, Process Model Identification, Laguerre Network On-line System Identification, Univariate Time-Varying Controller Performance Assessment, Multivariate Controller Performance Assessment, Robust Minimum Energy Control with Output Covariance Constraints, MPC Economic Performance Analysis and Tuning Guidelines, and many others. The output results of these application components will tell process engineers useful information on the investigated process or controller directly based on the process data, most of which are routine operating data. The outcome will be process models, controller performance of time-varying processes, multivariate controller performance, optimal designed controller, MPC economic performance, MPC tuning guidelines, and so on. The most updated version of PAT includes some new algorithms developed very recently such as model-free subspace approach to multivariate control performance assessment. The synthesis of variety of results makes this toolbox a powerful package for process and control engineers. For example, the controller performance indices obtained from multivariate controller performance monitoring component can be imported into MPC economic performance analysis and tuning component. It can then give the possible MPC economic benefit potential as well as some corresponding MPC tuning guideline. These two application components together with Data Collector will make a plant oriented solution for APC performance monitoring.

In this presentation, we will discuss the software architecture, implementation platform, mathematical principles behind this package, and its application potentials. Case studies will be provided to demonstrate some of its important functionalities.