Robust methods for control structure selection in paper making processes

Abstract: Process industries have to operate in a very competitive and globalized environment, requiring efficient and sustainable production processes. As a result, production targets need to be translated into control objectives which are usually formulated as performance specifications of the process, i.e. tracking of references or rejection of process disturbances. This is often a hard and difficult task which involves assumptions and simplications because of the process complexity. Complexity arises often due to the large scale character of a process, i.e. a pulp and paper can host thousands of control loops. A critical step in the design of these loops is the choice of the structure of the control, which means that controllers need to be placed between sensors and actuators.Current methods for control structure selection include the Interaction Measures (IMs). The IMs help the designer to select a subset of the most significant input-output channels, which will form a reduced model on which the control design will be based. The IMs are traditionally evaluated using a nominal model of the process. However, all process models are affected by uncertainties as simplifications and approximations are unavoidable during modeling. Thus, the validity of the control structure suggested by the IMs cannot be assessed by only analyzing the nominal model. The first part of this thesis focuses in analyzing the sensitivity of the IMs to model uncertainties in order to determine a robust control structure which is feasible for all the uncertainty set.It also becomes clear that, control structure selection requires extensive knowledge about how the multiple process variables are interconnected. The second part of this thesis focuses on creating IMs which can help the control designers to understand the propagation of effects in the process, and express this propagation in directed graphs for an intuitive understanding of the process which will help to design a feasible control structure. These methods have been inspired by coherence analysis used in brain connectivity.Neurons and neural populations interact with each other in different brain processes related to events as perception, or cognition. Electroencephalography (EEG) is a measure of electrical activity in the brain which is acquired from sensors positioned on the surface of the head, each of the electrodes collects the aggregated voltage of a neuron population. Analyzing the flow of information between populations of neurons allows to understand the communication between different parts of the brain in different brain processes. In a very similar way, analyzing the flow of information between variables in an industrial process will provide designers with the required information to understand the behavior of the plant.