Inferential Model Predictive Control of Continuous Pulping under Grade Transition

Abstract

Even though continuous pulp processes have been studied for many years, the absence of a model that can accurately describe the evolution of fiber morphology has impeded the application of advanced control techniques. In this study, a multiscale model for continuous Kraft pulping processes, which can capture the spatiotemporal evolution of wood chips and cooking liquor, is developed by integrating a macroscopic model (i.e., Purdue model) with a microscopic model (i.e., kinetic Monte Carlo algorithm). Then, an approximate model is identified to circumvent the high computational requirement of the multiscale model and to handle the input time-delay, followed by designing a soft sensor to infer state variables and primary measurements. This allows the use of an inferential model predictive control strategy in a continuous pulp digester to regulate the blow-line pulp properties (i.e., Kappa number and cell wall thickness) and achieve optimal grade transitions.

Sensor location selection for continuous pulp digesters with delayed measurements

“…When it comes to the controller design for the continuous pulp digester, one of the main objectives is to produce pulps that achieve specific quality according to different wood species while respecting distinctive operating conditions. To this end, great efforts have been made to meet various objectives based on the various control methods, such as model predictive control, adaptive control, dynamic matrix control, reduced dimension control, and genetic algorithms et.al 14–20 …”

“…To this end, great efforts have been made to meet various objectives based on the various control methods, such as model predictive control, adaptive control, dynamic matrix control, reduced dimension control, and genetic algorithms et.al. [14][15][16][17][18][19][20] The state information or the process knowledge is essential for such controllers and/or regulators design. However, the full state information is not often available due to the special features of the continuous digester, physical constraints of sensor installation, and/or the prohibitive expense of implementing spatially-distributed sensors.…”

Dynamic modeling and model predictive control of a continuous pulp digester

“…Alexandridis and Sarimveis 19 employed an adaptive MPC based on a radial basis function ANN model for Kappa number control of continuous pulp digesters. Choi and Kwon 20–23 developed a class of MPC for a continuous pulp digester and a batch pulp digester based on the proposed multi‐scale model. However, even though the aforementioned works have made a valuable contribution toward the modeling and controller design for the pulp digester, there still some aspects that did not receive much attention and/or accurate consideration.…”

“…Second, grade transitions have increased significance in pulping mills, which are necessitated by optimal production scheduling and planning policies. Hundreds of grade transitions are required every year for pulp digesters, which are even more frequent when the market demand increases 12,22 . The targets of operational variables are considerably different to meet various market demands, so shift operations need to be taken into account when the controller is developed and its realization.…”

“…Hundreds of grade transitions are required every year for pulp digesters, which are even more frequent when the market demand increases. 12,22 The targets of operational variables are considerably different to meet various…”