Multivariable predictive control of a TMP plant

This thesis describes the development of a novel control strategy for a two-stage thermo-mechanical pulping (TMP) plant. Desired pulp quality is achieved by selecting the set-points of specific energy and refining intensity at each stage. The targets of specific energy and refining intensity are obtained through the control of motor load, production rate and refining consistency by manipulating closing pressure, chip flow rate and dilution flow rate at the inlet of each stage. A constrained predictive controller is developed based on the generalized predictive control (GPC) algorithm because of its simplicity, ease of use and ability to handle problems in one algorithm. Future control actions are determined by minimizing the predicted errors without violating input and output constraints. A multi-input multi-output (MIMO) CARIMA model identified through identification experiments is used to predict the future process outputs. Model parameters are estimated on-line to handle a time-varying nature of the process. An analytical solution of a constrained MIMO GPC subject to input and output constraints is derived by solving a quadratic programming (QP) problem. The computation required by the analytical solution is substantially lower than that required by an algorithmic solution. For general cases of constrained MIMO GPC, an optimal solution is derived by solving a mixed-weights least-squares (MWLS) problem. In the use of MWLS, a control performance index can be easily augmented and the weights can be modified in a manner that encompasses both the requirements for the future control movements to lie inside the feasible region and to minimize the control performance index. If the constrained optimization problem is unfeasible, the MWLS will converge to the point that minimizes the maximum constraint violation. The proposed control schemes were tested on the simulation model developed using the mechanistic and empirical methods to describe the behavior of a real process. Simulations demonstrated the proposed control schemes' efficiency and capability of handling problems in one algorithm. A linear model-based control strategy may lead to system oscillation or even instability if a refiner in the process is operated near maximum load point because at this point the nonlinearity between refiner motor load and plate gap becomes severe. To overcome the problem, a nonlinear Laguerre model - a type of orthonormal functions - is used to represent the nonlinear relationship. A MIMO Laguerre model-based predictive controller is then derived as an alternative for the control of a wood chip refiner. The Laguerre model can be arranged in linear form in model parameters so that the standard recursive least squares (RLS) can be used for on-line parameter estimation according to which controller parameters are adjusted. Simulations demonstrated that in the use of the Laguerre model-based control scheme, the nonlinearity of the process can be represented appropriately and the plate clashes resulting from the severe nonlinearity can be avoided. In addition, in the use of the Laguerre function representation the dynamics of an actual process can be described appropriately without the need for assumptions about the plant order and the time delay, i.e., accurate assumptions about their true values are not necessary.Applied Science, Faculty ofElectrical and Computer Engineering, Department ofGraduat