Model-assisted basis weight control of a board machine

The incentive to increase production whilst minimising variance was the motivation for a joint project between Karlstad University and Stora Enso to develop a dynamic model of a 5-ply board machine using the pulp and paper dynamic simulation package Simons IDEAS. This rigorous model while suitable to quantify the possible benefits of suggested process improvements offline, quickly became too unwieldy for model-based control studies so a slightly simplified model was re-engineered in Simulink. Finally transfer function models, suitable for online control and optimisation tasks were identified from production data with a structure given by semi-rigorous reasoning. All three models were adjusted and subsequently validated against almost one year of plant operating data. This paper highlights the difficulties in validating a large industrial system with widely varying time constants, dubious transducers and the ubiquitous noise. We also demonstrate the use of simplified models to improve the performance of an adaptive IMC controller for the basis weight in the finished board and the improvement of MPC over the commonly employed IMC for grade changes. 1 Unique aspects of pulp and paper simulators The renewed interest in modelling in the pulp and paper industry can be explained by the now ubiquitous PC with adequate computing power in the control room and on the engi∗Author to whom correspondence should be addressed. Karlstad University, SE-651 88 Karlstad, Sweden. Email: david.wilson@kau.se neers desk coupled with the often underestimated attractive user-interface. Such a tool can tackle not only problems in design, but also analyse operational problems, assist in retrofitting and provide test possibilities for modelbased control studies. The result is that users, with automated modelling tools and little training, are encouraged to construction complex models and experiment with “what-if” scenarios in a wide variety of industries and applications. An industry-wide overview of modelling with particular attention to control is given in [18, 10]. When considering just paper and board machines, studies such as [6, 11] are typical of steady-state investigations while data driven models (or blackbox models) such as [1, 19, 13, 8], are typical of dynamic models. Models based on first principles such as mass, energy and momentum balances, [15, 14, 4, 22], while potentially more useful are less common due to difficulties in development, validation and maintenance. Despite the readily available computing power and a wide variety of simulation tools, the tackling of realistic industrial problems by the nonspecialist is still an open problem. This is particularly true for the pulp and paper industry which is still poorly served by the standard chemical process simulators, [9, p1141] as evidenced by the only cursory attention paid to dynamic simulation tools in [21, p2-5]. General purpose simulators (such as SpeedUp, [16]; gProms, [17]; Omola, [3]; Dymola/Modelica [7]) tend towards strong underlying numerical routines and often innovative thinking fall short on industry specific requirements such as physical property data on components such as pulp fiber types, fines, clay additives, non-Newtonian flow models, and correlations of paper properties such as bending stiffness, tear strength and brightness. These latter issues are addressed by simulation products with a specific pulp and