Perspectives: Automation Strategies

Project Aims to Bolster Model-Based Control

Goal is to automate and reduce maintenance costs while sustaining benefits

By Valentijn de Leeuw, vice president, ARC Advisory Group

Last November, the European Union (EU) Autoprofit project held a workshop near Amsterdam for academia, users and suppliers.  The workshop organizers presented the results of the soon-to-be-completed model-predictive-control (MPC) /recovery time objective (RTO)-related project and generated discussions about what further research would be necessary and how this follow-up research could be organized and financed.  Prof. Ton Backx, dean of the faculty of electrical engineering at the Eindhoven University of Technology, provided a summary of the workshop at ARC Advisory Group’s European Industry Forum last week in Amsterdam.

The results of the project include several prototypes of autonomous performance monitoring and model identification, update and tuning for linear and nonlinear MPC.  These are intended to enable large- and small-scale industrial process plants to sustain MPC benefits over time.  The project benchmarked these prototypes via simulations and test applications on industrial production units.

Project Goal and Philosophy
There is general agreement that model-based control and optimization applications are costly to implement and maintain because this requires significant time and effort by experts.  As a result, even though the technology offers great potential benefits in other applications, currently most model-based applications have been installed on large-scale refining and petrochemical units.  Even on such units, due to cost and manpower considerations, the applications are not adequately maintained and thus often lag behind hardware, process, feedstock or other plant changes.  As a result, the performance of and benefits from the applications degrade over time.

The Autoprofit project, partially funded by the EU and executed by ABB, Boliden and SASOL together with the universities of Aachen, Delft  Eindhoven and Stockholm, aimed to initiate the development of autonomous performance monitoring, diagnosis and maintenance of model-based controllers.  The project objective is to largely automate and reduce the maintenance cost of these applications.  Because improving the economic performance of industry and applications was a project goal, the approach uses economic cost functions. The project researched algorithms and software to close the autonomous maintenance loop with validation by industrial case studies.  Here, the primary focus was on linear MPC, with extensions into nonlinear dynamics on part of the cycle.  A brief discussion of the project work packages follows.

Performance Monitoring and Diagnosis
An inexpensive, automated experiment generates an identified model of the plant, while minimizing the economic cost of the excitation of the process. A parallel loop of the identified model and the original controller is used to detect the difference between a disturbance change and a plant change. The data obtained are reusable for controller retuning (disturbance change) or system re-identification (plant change).

Experiment for Model Re-indentification
Several approaches were developed to excite the process only where and as much as necessary: open loop by constraining the excitation signal, and closed loop using experimental design or hiding the excitation from the controller.

Robust Techniques for Automatic Tuning
Two approaches were developed for robust automatic tuning of model-based systems after a model update or temporary detuning to improve performance without re-identification.  One is based on the impact of model uncertainty on closed-loop performance, and the other is optimization-based.  The first has been successfully tested on industrial installations.

Extension to Nonlinear Models  
A method for detecting nonlinearities was developed that first tests for the need for a nonlinear model. If such a model is required, the method then designs and executes experiments for identification, updates model structures and parameters, updates the controller, and assesses the performance gains of the nonlinear MPC generated compared to original linear MPC.  If benefits are found, the program tunes the NLMPC controller on-line.

Benchmarking and Industrial Applications
Simulation benchmarks have been performed using Matlab on a distillation column (a depropanizer at a Sasol plant in Secunda, South Africa) and are now being analyzed, and on a pulp digester at Boliden in Sweden.. Tests with hidden (stealth) excitation of a zinc flotation plant at Boliden in Sweden have shown that the method did not cause performance losses.

ARC’s Take
ARC Advisory Group believes that the project results are significant and —once further tested, and validated — should ultimately enable industrial process plants to both implement MPC applications more effectively and sustain the economic benefits over time to maintain the applications and keep them up to date as the underlying plant operations and process conditions inevitably change.


VdeLeeuw 300Valentijn de Leeuw, vice president at ARC Advisory Group’s European organization, also acts as independent expert-evaluator of research projects for the European Commission in the Information and Communication domains. Valentijn has a PhD in technical sciences from Delft University of Technology in cooperation with Ecole Nationale Superieure des Mines de Paris and IFP and also has a masters in chemistry from Utrecht State University, The Netherlands.
 

More from this perspective...

Title

Understand The Difference Between IoT and M2M

What you want to accomplish determines which approach is better to use.

04/24/2014

Project Aims to Bolster Model-Based Control

Goal is to automate and reduce maintenance costs while sustaining benefits

07/09/2014

Connected Plant: Get Ready for The Internet of Things

Industry has an opportunity to lead the revolution.

03/17/2014

Companies Address Challenges of Batch Manufacturing

Standards, IT technology, cloud-based solutions and advanced analytics work in concert to improve operations.

06/03/2014