Advanced search
Start date

Modeling and optimization of descontinuous kraft digesters using neural networks and hybrid model-integration of processes in real time

Natascha Vigdis Polowski
Total Authors: 1
Document type: Doctoral Thesis
Institution: Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Química
Defense date:
Examining board members:
Song Won Park; Carlos Eduardo Rossell; Daniel Ibraim Atala; Reginaldo Guirardello
Advisor: Rubens Maciel Filho

In this work three models were developed for prediction of Kappa number (Deterministic Model, Neural and Hybrid Model). The Deterministic Model includes mass transfer and reaction kinetics based on intrinsic parallel reactions of lignin, cellulose and hemicellulose. This divides the process of Delignification in 3 stages or phases, with the 3 phases correspond to 3 different types of lignin (initial, bulk and residual). The model is specific to the eucalyptus pulp (short fiber) and batch digester. One of the contributions to this proposed model were the inclusion of reactions for lignin, cellulose and hemicellulose total, carbohydrates, than the Kappa number. Deterministic model was used for experimental data collected in the RAIZ - Instituto de Investigação da Floresta e Papel and data generated by simulations. The operating variables used as input data were: thickness, initial temperature, effective alkali, liquor ratio wood, the cooking time, density and porosity. The Deterministic Model was developed for continuous digester, and was validated with industrial data for short fiber eucalyptus. The modeling done for a batch digester is the same as for a continuous digester, which are continuous in 3 to 4 steps inside the equipment. Wherefore the continuous digester is modeled as a single batch that fractionation (temperatures and times for each stage of the equipment and sequential manner). This study employed a method of optimization (Successive Quadratic Programming) to define the procedure for operation in continuous digester allowed to obtain the product (pulp) with levels below 1.5% of residual lignin. Industrially this value is around 3% of residual lignin. The Neural Model is proposed as "feedforward" and training by backpropagation. For this model the input variables were temperature, effective alkali and H-Factor. The variable output is the Kappa number. The number of hidden neurons was defined by the neural model that showed the smallest error for the set of validation and training. The number of interactions was also determined from the smallest error generated by simulations. This model was validated with experimental and industrial data. The hybrid model used as input variables and Kappa Neural theoretical, temperature, and factor H and the output variable is the number Kappa Hybrid. This was validated with industrial data. The models presented (deterministic, Neural and Hybrid) are useful tools for the manufacture of pulp and paper, since there is the possibility to be applied to simulation of processes, optimization and control. The models can be tested for different operating conditions without changing the output. Besides, allowing better control of some variables in the manufacturing process, ie without loss of quality of product. In this study, the neural network and the kinetic models showed similar results. Keywords: Kappa number, Deterministic Model, Neural Model and Hybrid Model (AU)