This work is an initial study in a project where the influence of different transport phenomena and chemical kinetics on the overall pulping kinetics is investigated.
The current work has been divided into three main parts: 1) Investigation of the kinetics of earlywood and latewood respectively. 2) Determination of the lignin concentration profile in wood chips. 3) Modelling of the transport phenomena and reaction kinetics inside wood chips.
In the first part reaction kinetics of milled earlywood and latewood were determined. In the experimental part wood mill samples were impregnated with cooking liquor and then cooked at high liquor to wood ratio (200:1) in a 1.3 l autoclave. Different soda and sulfide concentrations were investigated. Measurements showed that it took approximately 23 minutes to raise the liquor temperature from a initial temperature of 20 degrees Celsius to the final temperature of 166 degrees Celsius. This heating period was taken into consideration when the mathematical modelling was performed. The experimental results from this part of the study showed that the kinetics of the delignification is different for these two morphologically diverse regions. The experimental data was fitted to a kinetic expression where the influence of temperature on the rate of delignification and yield has been considered.
In the second part of this study the concentration profile of lignin inside a 3 x 3 x 6.5 cm sapwood chips was experimentally investigated. The sapwood chips used were carefully cut so that the annual rings became perpendicular to the radial direction. This was done in order to make it easy to specify transport properties of different chemical species in the tangential radial and longitudinal directions inside the wood structure when modelling was made. The experimental procedure was the same as in the first part with one exception: the impregnation was repeated twice. Wood samples were then taken at different positions in the height and thickness as well as in the longitudinal direction inside the cooked sapwood chip. The experiments were performed at two different cooking times and for two different wood species (oak and pine). Results from these experiments show that there is a concentration gradient of lignin inside the sapwood chip in the longitudinal as well as in the thickness direction.
Finally a two-dimensional simulation was performed where transport phenomenon and reaction kinetics were considered at the same time. The result indicates that it takes a long time, approximately 55 minutes, to reach the final reaction temperature of 166 degrees Celsius through out a chip with the dimensions mentioned above. Before 55 minutes there is a multidimensional temperature distribution that must be considered when modelling the overall pulping kinetics of wood chips. The differences in heat conductivity of the fibre- and cross fibre direction respectively together with the chip geometry is one parameters that must be considered when explaining the concentration gradient of lignin inside the sapwood chip in the longitudinal and thickness direction.
The results from this study indicate that the delignification inside a wood chips may during the initial heating period be controlled by an existing temperature gradient. It is also shown that it is important to consider different reaction rates inside morphologically different regions when making calculations on a smaller scale.
Paper 316b: The Development and Optimization Steps of a Predictive Neural Network Model for a Continuous Digester with a Small Training Set
The pulping process starts in the digester, where wood chips and cook liquor are submitted to high temperature and pressure. Initially, the cooking liquor has to penetrate into the wood in order to react and dissolve the lignin, undesirable substance in the paper making process. However, the cooking liquor can also react with other wood components, the extractives, which are dissolved very fast, and the carbohydrates, cellulose and hemicellulose, which must be preserved, for they are the desired final product. The process variables must be adjusted in a way that the lignin reactions are favored and the carbohydrate reactions avoided. The retention time in the digester can be rather long and therefore it is interesting for a mill to be able to predict the degree of delignification, measured in the pulp industry as Kappa Number, in an early stage.
This process involves penetration and diffusion of chemicals into the wood and dissolved components out, simultaneous exothermic reactions of wood components, heat transfer and the flow of wood chips and liquor in co-current and counter-current zones. Although it has been extensively researched, the occurrence of so many simultaneous phenomena and the fact that the lignin molecule is very complex and not fully determined qualify this as a complex process, with many questions still to be answered.
On the other hand, neural networks have been proclaimed as the solution for processes like this one, where knowledge is incomplete or complexity makes it harder to obtain good results with first principle models. This work presents a neural network model to predict Kappa number in a pulp mill, based on a small industrial training set.
In addition to the neural model, a deterministic and a hybrid model were also developed in order to compare the different modeling techniques, not only the results, but also development aspects, limitations and advantages of each one.
Aracruz Celulose SA, the largest pulp mill in Brazil, provided the data for the model development. The evaluation of published models set the criteria for the selection of the deterministic model to be used. The neural model is comprised of a feedforward network trained with backpropagation algorithm. After an evaluation of the industrial data, only the variables that showed to carry relevant information were used in the model.
The mill could not provide the expected amount of training data. However, the quality of the data and the rigorous data collection procedure, as well as the careful data evaluation made it possible to search for solutions for a successful model. The alternative was the production of a new data set formed with points of the correlation curve obtained from industrial data, which is the emphasis of this work.
The network structure and its parameters, such as number of hidden layers and neurons, training algorithm, number of inputs, etc. were optimized in order to improve model performance.
The combination of the result obtained from the deterministic model with a neural network composes the hybrid model.
Despite the small data set, the neural network produced satisfactory results, where the difference between the expected and predicted values was lower than the experimental error inherent to the lab test for determination of remaining lignin. The first principle model was able to reproduce delignification rate, which determines the pulping degree, and therefore was considered adequate to be used in the hybrid model. The hybrid network results were slightly better than the ones obtained with the pure net and its training was appreciably faster.
A model which is able to reproduce expected results in a timely fashion can be implemented in a control or supervisory system that works on-line. The results showed that when well trained and optimized, both, the pure nets and the hybrid models, fulfill these requirements. They also show that it is possible to use normal process variables, making it more feasible to develop customized models.
However, the development of the model shows that the statement that neural networks are a magic tool that transform a lot of data into a successful model is a myth, because in reality, it requires an extensive evaluation and optimization effort in order to thrive.
Paper 316c: Autocausticizing Sodium Carbonate with Borate
Recently there has been considerable interest in the use of borate as an autocausticizing agent. One reason for this renewed interest in borate as an autocausticizing agent is due to recent research (1), which shows that this process is much more effective then was originally believed. With borate based autocausticizing, sodium metaborate reacts with sodium carbonate in the bed of the recovery furnace, Eqn (1), to generate trisodium borate and to release carbon dioxide. On dissolving the smelt in water, the trisodium borate reacts with water to form NaOH and regenerate metaborate, Eqn (2).
NaBO2 + Na2CO3 = Na3BO3 + CO2 (1)
Na3BO3 + H2O = 2NaOH + NaBO2 (2)
This research presents a kinetic study of the reactions of sodium metaborate and sodium carbonate with and without carbon present. One of the major objectives of this work is to determine if this process can be used to generate sodium hydroxide from carbonate in a fluidized bed. Such a process could be used to generate hydroxide directly from carbonate for use in semichemical pulping. Many semichemical pulping processes use a combination of purchased sodium hydroxide and fluidized bed regenerated carbonate for pulping. Because of the high cost of sodium hydroxide, there is considerable interest in reducing or eliminating the purchased hydroxide.
The results from this research provide insight into the nature of the metaborate-carbonate reaction, demonstrate that this reaction can occur within the fluidized bed temperature range and to some extent before the melting borate of the borate and carbonate salts. A kinetic rate expression that shows the effect of the experimental parameters on the reaction rate is included.
1. H, Tran, J. Cameron, and C. Bair, "Autocausticizing of Kraft Smelt with Sodium Borates", Pulp and Paper Canada, (September 1999).
Paper 316d: Delignification Kinetics of Alkali Profiled Harwood
Cooks Using on-line Liquor Analysis
Alkali profiling techniques have proved to be promising methods to extend delignification so that the bleach demand can be reduced. Most of the research works focus on modifications to achieve low lignin content of softwood pulp. This paper focuses on the development of a kinetic model for a modified hardwood kraft pulping system, where the alkali concentration is maintained at the desired level during the cook. The laboratory digester used for the study is equipped with in-situ liquor analysis system, which enables measurement and control of alkali, and sulfide concentrations at desired level during the cook. In addition the system is also equipped to measure the concentrations of dissolved lignin and total dissolved solids in the liquor during the cook. The model describes the kinetics on the basis of real time chemical concentrations, measured by the liquor analysis system. This model can be used for controlling the delignification, to achieve lower kappa number pulp at comparable pulp strengths or enhance the strength at current kappa levels.
Paper 316e: A Kinetics-based Model for Predicting Kappa Number
during Batch Kraft Cooking
A Kappa number model was derived from the kinetics of delignification in batch kraft cooking as Ka=A-Bln[(H-Hb)EAb^R] where Ka is Kappa number, Hb and EAb are the H factor and effective alkaline concentration at the beginning of bulk delignification, and A, B, R are constant can be determined through data regression. The derived model avoids the alkali complex reaction in initial phase of delignification and the effect of unmeasurable moisture content disturbance. The parameters in the model can be easily determined. Compared with some models that require time-dependent EA concentration in the entire kraft cooking process, the present model is very simple and only requires to determent one point effective alkaline concentration. Therefore, the model has practical importance for predicting pulp kappa number during batch kraft cooking. The validity of the mathematical model was tested, and the model was found to be in good agreement with data obtained from laboratory pulping of masson pine. This model can be used for predicting the cooking endpoint.
Paper 316f: Effect of Lignin Diffusion on Kraft Delignification Kinetics as Determined by Liquor Analysis. Part II: Theoretical Analysis
This study provides the theoretical analysis of the effect of lignin diffusion on the kraft delignification kinetics of black spruce as determined by liquor analysis. The rate of lignin transport from wood chips to bulk liquor is modeled with the equation describing non-steady state and one-dimensional diffusion with an internal generation term from chemical reaction. An exact solution was derived for this equation. To calculate the numerical results, a program based on spreadsheet macro was developed. It is found that the measured lignin concentration in cooking liquor can be accurately described by the diffusion model. The theoretical calculation confirms the finding obtained experimentally that the effect of temperature on the lignin concentrations from liquor analysis could not be described by using the classical value of 134 kJ/mol for the activation energy for kraft cooking of black spruce. The lower activation energy measured experimentally is because diffusion has stronger effect on the measured lignin dissolution rate at higher temperatures. The model is also used to evaluate the lignin concentration profiles within wood chips and the bulk and intra-chip lignin concentrations as a function of chip thickness and temperatures.
Paper 316g: Generalized Delignification and Cellulose Degradation Curves for Ozone Bleaching
The kinetics of delignification during the initial phase of ozone bleaching of pulp are well described by the shrinking core model with difusion control of ozone in the reacted layer of the fibre wall. Based on this model, generalized delignification and cellulose degradation curves are constructed which apply to hardwoods and softwoods of different initial kappa numbers, and takes into account process conditions such as ozone partial pressure and pulp consistency. The only input required for the construction of these generalized curves are the initial kappa number and pulp viscosity, the fibre saturation point and the fibre wall width distribution of the pulp sample.