Indirect adaptive model predictive control of a mechanical pulp bleaching process using a Smart Delay Time Predictor

The classic way to control a process, in a model based framework, is to obtain a model of the system and then to use it for the design of a controller. A nonlinear time-varying process can be operated in real-time by an indirect adaptive controller. Part of this thesis is devoted to describing the particular structure of such a controller and applying it to a pulp bleaching process. We present and discuss all aspects of controlling a real-world delay time system application, the pulp bleaching process at Irving Paper Ltd. The bleaching process was thoroughly studied and models identified offline as a single-input single-output process then extended to a multivariable process. Then online identification methods were used, and the process was accurately modeled as a first order system plus a variable delay time. This is a difficult process to control, since the delay time varies with pulp flow into and out of the bleaching vessel.;Another major part of the thesis focuses on improving the controller performance by solving the variable delay time problem using a novel a Smart Delay Time Predictor approach and a recursive least squares (RLS) model identifier. This new approach is an extension of the variable delay time estimator technique based on time-variable flow processes. The present work has improved the approach proposed by Sayda and Taylor [6] in one important respect: the time delay prediction method presented here eliminates the adverse transients occurring in case of the uncertainty in the variable time delay, i.e., it removes transient spikes due to miscalculation of the forced response inside the controller.;The efficacy and robustness of this technique is demonstrated by controlling the pulp bleaching process using an indirect adaptive model predictive control (MPC) algorithm with an RLS identifier and a variable delay time predictor embedded in that controller. This algorithm produces control moves that account for good reference tracking in the presence of disturbances and actuator constraints. Further, a filter is added to the RLS parameter estimator to tackle the problem of small spikes occurring in the input and the output of that controller.;We extended the online identification methods to identify the pulp bleaching process when dealing with it as a multivariate system. Such a model would be used as the basis for multivariable control. However, the poor quality of the resulting model precluded that work.