Modeling of complex systems

The process industry offers considerable challenges from a control engineering perspective. Improving the control is vital link in the optimization of a production system. By controlling more exactly, it is possible to reduce the variations in quality and also reduce the consumption of resources, since the supply of these can be kept closer to optimum.

A process industry facility is characterized by many subprocesses that are connected through e.g. material flows, recycles, common resources, controllers, etc. The problems that arise due to this complex structure can, in principle, be divided into three categories

1. Many variables. The large number of process variables make it unreasonable to put a lot of labour into each individual variable for design of controllers and fault detection algorithms.
2. Coupled variables. The dynamic couplings between process variables make the entire facility an enormously large dynamic system which is not reasonable to handle as a whole.
3. Complex dynamics. The complex structure can give rise to complicated and unpredictable dynamics in the facility as a whole, which may result in e.g. oscillations.

A prerequisite for systematic design of a controller or fault detection algorithm is a model for the process as well as a quantitative description of  the uncertainty associated with this model and the disturbances to which it is subject. For process modelling there are many systematic methods described in the literature while methods for modelling of uncertainties and disturbances are largely undeveloped. Therefore, design of controllers and fault detection algorithms can not be automated to any large extend but requires manual effort of an experienced engineer. For point 1, our research will thus address systematic methods for modelling of uncertainties and disturbances.
 
A problem under point 2 is the question of how to divide a complex system into smaller subsystems without important properties, such as dynamics, controllability, and observability, are changed dramatically. This kind of partitioning is required for design of both controllers and fault detection algorithms.

Johansson, A 2008, 'Shift invariant signal norms for fault detection and control', Systems & Control Letters, vol. 57, nr. 2, s. 105-111.

Berndtsson, M & Johansson, A 2008, 'Design of a dynamic threshold generator for λ-tuned control loops', Control Engineering Practice.

Johansson, A 2007, 'Validation of state-space models with time-varying parameter uncertainty', I 46th IEEE Conference on Decision and Control Proceedings: Hilton New Orleans Riverside; December 12-14, 2007; New Orleans, LA, USA, IEEE s. 6406-6411.

Andreas Johansson