Main contents


Symbiosis of mill modelling and computational methods

Tuning of industrial controllers is an age-old research topic in control engineering which has been tackled with numerous approaches ever since the dawn of the automatic control. In that regard, several guidelines and methods exist for tuning single loop controllers optimally. However, modern industrial control systems consist of dozens of controllers that are interacting and interfering with each others. Therefore holistic approaches are needed to manage the overall performance of the large and complex systems.

Iterative Regression Tuning (IRT) is a novel method that has been developed for tuning of multiparameter systems, such as industrial control systems and dynamic simulation models. The IRT method is a data based approach that makes use of statistical multivariate regression methods and iterative optimization algorithms. Statistical dependencies between system parameters and quality measures (defining the quality of the system performance) can be captured from data using methods like Partial Least Squares and Canonical Correlation Analysis. Since the optimization cost function is a nonlinear but smooth function of the parameters, using iterative gradient based optimization methods is justified. IRT method is independent from the application domain and up to some point also from the scale of the application.

Two feasibility studies of IRT method have been conducted on realistic simulation models. The first case study was made with a model of a combustion power plant, and the second one on a continuous pulp digester model. Both of them were constructed with Apros simulation software. The cases showed that IRT method is capable of tuning simultaneously several controllers with respect to multiple performance targets. The method is computationally heavy due to the slow simulation based data acquisition. Streamlining the applied iterative optimization method and by using statistical testing methods the time required by the IRT routine can be shortened to a practical level.

The Simbiot project focuses on combining different computational and information technological methods in order to establish solid groundings for a mill model based design environment for process industry applications. Such design environment would open the possibility to access the plant design data with a desired level of accuracy. This would facilitate the use of dynamic simulation and different optimization methods as design tools. Full-scale utilization of simulation based design principles awakes also the need for advanced optimization methods of dynamic simulation models. Therefore, the research also concentrates on the possibility to optimize the parameters of a reduced complexity simulation model with the IRT method.

Our research partners are TKK Laboratory of physics, VTT Industrial systems, and VTT Processes.

Simbiot is a part of the Tekes MASI program.

  • Keywords: Process control, Dynamic simulation, Controller tuning, Multivariate regression, Complex systems
  • Duration: 2005-2006
  • Research area: Process Control, Cybernetics
  • Related projects: TeMa
  • Publications (total: 8)



Doctoral dissertations

Master's Theses