(476d) Automatic Reaction Network Generation for Pyrolysis of Biomass

This contribution discusses the followed methodology in the Laboratory of Chemical technology at Ghent University to develop a more fundamental understanding of the combustion and pyrolysis of biomass.

In Europe, Asia and the United States of America combustion and pyrolysis of complex hydrocarbon mixtures is the main process for the production of energy and valuable base chemicals such as ethene and propene. However the oil reserves are not infinite and sooner or later the world will run out of oil. In fact, the annual oil consumption is growing continuously while the proven oil reserves remain almost invariant because of the limited amount of new wells found during the last decades. Moreover, the remaining oil reserves are of a lesser quality and more difficult to extract because of the heavier nature of the oil and because these new oil reserves are located in deep sea and Polar Regions. Therefore many scientists claim that global production of oil is set to peak in the next decade before entering a steepening decline. According to this "peak oil" theory our consumption of oil will catch, then outstrip our discovery of new reserves and we will begin to deplete known reserves. The latter will have massive consequences for the world economy and the way that we live our lives. One of the possible alternatives is to produce energy and base chemicals partly or completely from renewable resources. The use of renewable feedstocks is also encouraged by the growing awareness about the greenhouse effect.1 The latter is now generally recognized as being caused by the use of fossil fuels. Renewable resources are central to all scenarios in which biomass and bio-fuels replace more than a few percent of currently used energy resources.

Large-scale detailed kinetic models find increasing use in the modeling of combustion processes, atmospheric chemistry, soot formation, and other areas of industrial and environmental interest. Because such reaction networks may contain thousands of reactions and species, constructing them by hand can be tedious and error-prone. Therefore many research groups have developed computer tools to automatically generate these mechanisms, e.g. Broadbelt2,3, Green4, Froment5, etc. A key difficulty with mechanism generation programs is that they are usually combinatorial, producing large numbers of kinetically unimportant reactions and species. Moreover, sometimes non-physicochemical criteria and/or expert user involvement are employed to limit mechanism growth, thereby risking the possibility that important reactions are not included. In contrast, rate-based termination of computer-generated reaction mechanisms provides a physicochemical criterion for including reactions and species 3, 4, 6; only those pathways whose flux exceeds some minimum flux criterion, Rmin, are included in the network. XMG (Exxon-mobil Mechanism Generator) was the first network generation software applying the rate based termination criteria3. XMG has been further elaborated, adding several new features. Matheu et al.6 created XMG-Pdep, the first mechanism generator to systematically include pressure-dependent reactions. Recently a new mechanism generator, RMG, belonging to the same family has been developed4. RMG includes XMG-PDep's capabilities but also features the implementation of advanced technologies, such as graph representation of reaction families and a hierarchical tree-structured database for retrieving thermodynamic and kinetics parameters, and the use of object-oriented technology. These features strongly facilitate the continuous improvement of the level of detail in the description of the chemistry as compared with prior network generation software.

The basic approach followed by RMG can be summarized as follows. RMG uses a set of ?reaction families? to generate all possible reactions that a given chemical species can undergo as such and in the presence of the other species in the mechanism. Each reaction family represents a particular type of elementary chemical reaction, such as bond-breaking, or radical addition to a double bond. Currently thirty-four primary reaction families are defined; this is the richest set of reaction families ever compiled. RMG represents the individual chemical species as a 2-dimensional connectivity graph and defines the possible reactions by considering the possible mutations of the graph. The newly formed species are then considered as candidates for further reactions and their reactions can be added to RMG's chemical kinetic model.

As most mechanism generation tools, RMG obtains the necessary thermochemical data from an electronic database of literature values, whenever possible. In most instances, however, it must resort to group contribution methods to estimate enthalpies of formation, heat capacities, entropies, and certain other data required for modelling pressure-dependent reactions. Generally, RMG can rely on accurate literature thermochemistry data from assembled databases.

An accurate prediction of the conversions, yields, energy consumption and production would greatly enhance not only the design but would also provide a practical tool to determine the optimal operation conditions and/or to evaluate the process economy as a function of feedstock composition. Historically, it was infeasible to accurately determine all of the parameters in large chemical kinetic models a priori, so developing accurate models involved fitting to a vast number of experimental data and required extensive and time-consuming experimental work to gather the necessary data. At present, estimation methods, particularly those based on quantum chemistry, have significantly improved, allowing a reliable a priori determination of kinetic and thermodynamic data. Biomass consists mostly of lignin, hemicellulose and cellulose. As model compound for lignin phenethyl phenyl ether (chroman) is considered. Xylose is used as model compound for hemicelluloses, while glucose is considered for cellulose. In the present contribution only the results for phenethyl phenyl ether will be discussed. The proposed methodology is similar for all the considered model compounds. First RMG is used for generating the detailed reaction network. The latter consist of thousands of reactions between different molecules and radicals. Next the model is validated using experimental reference data obtained at Ghent University, MIT or from literature.

The accuracy of the reaction rate coefficients used to construct the models is of the utmost importance, especially when using a rate based termination criterion, as the rate estimates determine whether or not an elementary step will be included in the final mechanism. Certain reaction pathways can indeed remain unidentified because the reaction rate coefficients are inaccurate or because the model generator omitted to include some reaction families, leading to incomplete reaction networks. Adjustment of the reaction rate coefficients of an incomplete reaction network to impose agreement with experimental data, then forces some of the reactions in the model to take over the roles of the unidentified reactions thereby reducing the physicochemical basis of the model and hence restricting its valid range of extrapolation.

Once the reaction network is generated and the thermodynamic properties and reaction rate coefficients are determined, reactor modelling is started. First an appropriate reactor model should be chosen because radial non uniformities can make the implementation of more dimensional reactor models necessary, e.g. Van Geem et al.8 Next, an appropriate solver is selected which is able to solve the stiff set of differential equations. Finally, the simulation model (reaction network + reactor model) is validated using the experimental data set obtained on the bench scale set-up and the pilot plant at the LCT. Optimisation of certain reaction rate coefficients is unavoidable to obtain a good agreement between simulated and experimental data. In the proposed methodology the reaction rate coefficients of the reference reactions seem are most suited for optimisation.

References:

1. Houghton J.T., Ding Y., Griggs D.J., Noguer M., van der Linden P.J., Xiaosu D., 2001: Climate Change 2001: The Scientific Basis: Contributions of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, 2001.

2. Broadbelt L.J., Stark S.M., Klein M.T. Computer-generated Pyrolysis Modeling ? on-the-fly Generation of Species, Reactions, and Rates. Ind. Eng. Chem. Res. 1994; 33 (4): 790-799.

3. Susnow R.G., Dean A.M., Green W.H., Peczak P., Broadbelt L.J. Rate-Based Construction of Kinetic Models for Complex Systems. J. Phys. Chem. A. 1997; 101: 3731.

4. Hillewaert L.P., Diericks J.L., Froment G.F. Computer-Generation of Reaction Schemes and Rate-Equations for Thermal-Cracking. AIChE Journal. 1988; 34: 17.

5. Green W.H. Predictive Kinetics: A New Approach for the 21st Century. Adv. Chem. Eng. 2007, 32, 1-50.

6. Matheu D.M., Green W.H., Grenda J.M. Capturing Pressure-Dependence in Automated Mechanism Generation: Reactions through Cycloalkyl Intermediates. Int. J. Chem. Kin., 2003, 35, 95-119.

7. Van Geem K.M., M.-F. Reyniers, G.B. Marin, J. Song, D.M. Matheu, W.H. Green Automatic Reaction Network Generation using RMG for Steam Cracking of n-Hexane, AIChE Journal

52 (2), 2006, 718-730.

8. Van Geem K.M., Heynderickx G. J., Marin G.B. Effect of radial temperature profiles on yields in steam cracking, AIChE Journal (2004) ; 50 (1) p 173 ? 183