Modelling of Kraft Cooking Kinetics Using Near Infrared Spectroscopy

Sammanfattning: A kinetic model describing the kraft cooking degradation reactions for softwood lignin and carbohydrates was developed in this thesis. The model satisfactorily predicted the data for modified as well as conventional cooking schemes, something which eluded earlier published models. The model considers four wood components (lignin, cellulose, glucomannan and xylan), each divided into three species. The dependence of hydroxide and hydrogen sulfide ion concentrations and temperature was considered. The rate expressions for the wood components were accompanied by an algebraic expression, denoted the distribution model, developed for describing how the equilibrium distribution of the medium and low reactive species depends upon the current cooking conditions. The proposed model, alongside previously published models was explored in terms of cooking selectivity in numerical experiments. It was seen that the earlier models gave diverse results, while the proposed model was in accordance with the accepted rules and industrial practice. The quantitative dependence of pulp yield and viscosity was determined, and the cooking variables could be given the following order of importance in terms of governing cooking selectivity: increased concentration of hydrogen sulfide ions has a larger beneficial effect than decreased concentration of hydroxide ions, which in turn has a larger beneficial effect than decreased temperature.

A procedure for separation of the black liquor after an autoclave cook was developed, enabling analysis of the liquor bound inside the chips. The concentration trends for e.g. hydroxide ions and organic matter showed substantial differences between the free and the bound liquor. Near infrared spectroscopy (NIR) and multivariate calibration techniques were used for black liquor characterisation, and the extracted data was subsequently used for kinetic modelling. Predictive calibration models were built for e.g. concentrations of hydroxide ions, dissolved lignin and dissolved organic matter. A scheme for improving the prediction errors was proposed. The effect of performing spectral pre-processing before, or integrated inside, cross-validation was investigated, using a Monte Carlo simulation approach which showed that pre-processing should be applied inside the cross-validation rounds in order to achieve a reliable estimate of the model performance.