(460b) Molecule-Based Kinetic Modeling of Steam Cracking

For higher performance, increased selectivity and faster development of new processes, accurate models with a wide range of applicability are required. An adequate kinetic description of the reaction chemistry in combination with a sufficiently detailed reactor model is essential but not sufficient for that purpose. Reaction networks can be generated based on a limited number of rules corresponding to elementary reaction families defining the chemistry of the process. The kinetic parameters can be obtained by regression of experimental data but more recently also from ab initio calculations. The latter allows to obtain a more consistent kinetic data base as well as to obtain more insight.

The present contribution focuses on the kinetics of steam cracking and in particular on the development and application of a Single-Event MicroKinetic ( SEMK) model for the latter.

Steam cracking can be described by considering a limited number of elementary reaction families: (i) radical addition to an unsaturated bond and the reverse â scission, (ii) hydrogen abstraction, and (iii) C-C and C-H bond scission and radical recombination reactions. This allows to generate a full reaction network in a systematic way (1).

An ab initio based group contribution method ( 2,3) was applied to calculate the kinetic parameters. This method is a consistent extension of Benson's group additivity concept to transition states. Indeed, despite the increasing accuracy and computational performance, ab initio methods are still too demanding to determine accurate kinetic data for the larger species that occur in heavier feedstocks. Group additive values were derived from rate coefficients determined for elementary reactions involving small species only and, hence, allowing the use of the high level CBS-QB3 ab initio method. Tunneling and hindered internal rotation around the transitional bond was accounted for. A reasonable agreement with pilot and industrial data was obtained.

The SEMK methodology requires the characterization of a feedstock in terms of types of molecules. Both fossil or renewable feedstocks typically consist of a considerable amount of types of molecules. The available information is limited to commercial indices corresponding to macroscopic properties such as density and boiling point traject. So-called molecular reconstruction methods use this information to obtain a characterization of the feedstock in the terms required by e.g. a SEMK model (4). Three possible approaches were compared: multiple linear regression, artifical neural networks ( ANN) and Shannon entropy maximization. The variance of the training set, i.e. of the set of feedstocks from which both the commercial and the molecular characteristics are known, is an important issue. A principal component analysis of the commercial indices of a naphtha training set was performed. Three principal components allowed to capture most of the variance. Simulation of an industrial naphtha steam cracker with the SEMK model indicated that both multiple linear regression and ANN can provide adequate product yields as long as the so-called Mahalanobis distance of the cracked naphtha is smaller than two. The effectiveness of Shannon entropy maximization was found to depend very strongly on the choice of the molecules constituting the reconstructed naphtha. Provided that an appropriate choice is made this method is very versatile.

1. Automatic Reaction Network Generation using RMG for Steam Cracking of n-Hexane

K.M. Van Geem, M.-F. Reyniers, G.B. Marin, J. Song, D.M. Matheu, W.H. Green

AIChE Journal, 52 (2), 718-730, 2006

2. Ab initio group contribution method for activation energies for radical additions

M. Saeys, M.-F. Reyniers, G.B. Marin, V. Van Speybroeck, M. Waroquier

American Institute of Chemical Engineers Journal, 50, no. 2, 426-444, 2004

3. Carbon-centered radical addition and â scission reactions: modeling of activation energies and pre-exponential factors

M. K. Sabbe, M.-F. Reyniers, V. Van Speybroeck, M. Waroquier, G.B. Marin

ChemPhysChem

4. Molecular reconstruction of naphtha steam cracking feedstocks based on commercial indices

K. M. Van Geem, D. Hudebine, M-F. Reyniers, F. Wahl, J.J. Verstraete, G.B. Marin

Computers and Chemical Engineering 31, 1020-1034, 2007