(546g) Application of Chemometric Methods to Generate Reaction Pathway Hypotheses for the Thermal Cracking of Athabasca Bitumen

The use of advanced hyphenated analytical techniques in characterizing complex mixtures like bitumen is becoming increasingly prevalent in recent times, ^1,2 since they provide information on physical and chemical properties that would be otherwise difficult to determine. However, the challenge in dealing with the data is that it is multi-dimensional and contains a large amount of overlap between the signals of the constituent components, which hinders efficient chemical interpretation. Further processing of this raw information is required for analyzing changes based on the chemical reactions occurring in the complex mixtures. This work focuses on applying multivariate curve resolution (MCR) methods and Bayesian clustering and learning methods on Fourier transform infrared (FTIR) spectra of the liquid products obtained from thermal cracking of Athabasca bitumen in the temperature range of 300 – 420 ºC. The reaction times vary for each temperature and range from 15 min to 27 h.

The curve resolution methods applied in this study consist of three parts: (i) determining the chemical rank of the system at each temperature using singular value decomposition (SVD) to identify the number of chemical species undergoing a change; (ii) obtaining initial estimates for the concentration profiles using evolving factor analysis (EFA) ³ that would be fed to the final optimization; (iii) final resolution using alternating least squares (ALS) ⁴ and particle swarm optimization (PSO) ⁵ algorithms embedded into ALS to obtain the spectra and concentration of the rank-identified pseudo-components. The Bayesian methods are also applied on the FTIR spectra and consists of two steps: (i) Bayesian hierarchical clustering (BHC) ⁶ that incorporates some prior chemical knowledge of the bituminous system to split the wavenumbers into 5 groups based on maximized marginal posterior probability; (ii) Bayesian structure learning, which is a probabilistic graphical model uses score-based algorithms like hill climbing and Tabu search to produce directed acyclic graphs (DAGs), which represent causal relationships between the identified groups in BHC. Here, the wavenumbers are designated as random variables with a multinomial distribution, the hyperparameters of which have a Dirichlet distribution. The MCR-ALS-PSO methods are used to identify the nature of reactions occurring during different stages of thermal cracking while this knowledge is utilized to develop a plausible reaction pathway from the DAGs obtained from the Bayesian network.

Previously, Tefera et al. ⁷ conducted a similar study on thermal conversion of Cold Lake bitumen in the temperature range of 150 – 400 ºC with the help of quantitative parameters derived from the resolved spectra, and suggested that hydrogen transfer and methyl transfer led to a rise in the methylene content that was responsible for an increase in naphthenic content at lower temperatures and increase in the length of side-chains attached to aromatics at higher temperatures. Ring closure reactions were also indicated to happen due to an increase in the intensities corresponding to aromatic C=C stretching and decrease in aromatic C-H bending. These indicated the formation of heavier products which corroborated with an observed increase in viscosity even at 400 ºC. Incorporating chemical sense into the Bayesian network for Cold Lake bitumen conversion between 150 – 300 ºC indicated the formation of aliphatic lighter molecules along with condensed aromatics. The objective of our work was to identify the credible reaction chemistry involved in the thermal conversion of Athabasca bitumen and propose a reaction network based on the chemometric results and compare with that of Cold Lake bitumen.

PSO was used as an improvement over ALS for optimization of the concentration profiles in MCR in this work. However, the chemical changes as interpreted from PSO-optimized profiles were not much different to the ALS-optimized profiles but PSO embedded inside the ALS loop was found to converge faster than ALS alone in arriving at the final solution for curve resolution. Three pseudo-components were determined to be representative of bitumen transformation at all temperatures in this work. A reaction pathway for the pseudo-components was shown to move from the first pseudo-component representing the feed to the third component representing the final product. In some cases, the second component was at a lesser concentration than the third component throughout the reaction times after an initial rise. Hence, the trends in the spectra-derived quantitative parameters indicating chain length, aromatic condensation extent was important for the second pseudo-component as well. One quantitative parameter that was used in addition to the ones used by Tefera et al. ⁷ is the extent of aromatic substitution (EOS). This was calculated as the ratio of intensities of aromatic C-H bending from meta, para and ortho disubstituted aromatics to that of mono-substituted aromatics. The chain length parameter, calculated as the intensity ratio between methylene and methyl groups was constant at lower temperatures but eventually showed a steep decrease at higher temperatures. This combined with sufficient gas production indicated that cracking was indeed taking place. In another work by Sivaramakrishnan et al. (book chapter accepted in ACS books but not yet online) they showed that lighter products were formed (through distillation profiles) when Athabasca bitumen was thermally cracked at 400 ºC for up to 1440 min. EOS was also seen to decrease at all temperatures except at 380 ºC, where sufficient energy was not available to break the aromatic C-aliphatic C bond, which was anticipated to occur at 400 ºC and 420 ºC. For Cold Lake bitumen, EOS increased for most temperatures indicating the dominance of side-reactions like ring closure that might have also been responsible for an increase in viscosity along with free-radical recombination.

Bayesian networks relating the 5 groups of chemically assigned components indicated the gradual transformation from heavy feed to lighter products with not much increase in aromatic content. The formation of lighter material was possible through cleavage of aromatic C-alkyl C bonds and transfer of hydrogen to replace substituent in the aromatic ring by a radical hydrogen transfer (RHT) mechanism that was shown to exist in bitumen free-radical reactions by Blanchard and Gray. ⁸ Breaking of C-H and C-C bonds attached to benzene was also shown to occur in the temperature range of 398 – 450 ºC in the work by Khorasheh and Grey ⁹ so there is enough evidence to support the proposed mechanism in this work. The key finding of this work was the formation of lighter products with not much increase in aromatic content in the liquid phase and simultaneous release of hydrogen rich gases, prominently consisting of methane. Also, at 300 ºC, Athabasca bitumen did not react much while Cold Lake showed significant viscosity decrease. The slight difference in the reactivities of Athabasca and Cold Lake bitumen was highlighted in this work and this could be due to the increased saturate content in Cold Lake compared to Athabasca bitumen. ¹⁰

The primary contribution of this work is the development of a modeling framework to identify reaction pathways and the associated (pseudo-)kinetics for complex reacting mixtures from spectroscopic data, which can be used for real-time process monitoring, and to apply it to the investigation of the mild thermal vis-breaking of bitumen.

References

(1) Yoon, S.; Bhatt, S. D.; Lee, W.; Lee, H. Y.; Jeong, S. Y.; Baeg, J. O.; Lee, C. W. Separation and Characterization of Bitumen from Athabasca Oil Sand. Korean J. Chem. Eng. 2009, 26 (1), 64–71.

(2) Varanda, C.; Portugal, I.; Ribeiro, J.; Silva, C. M.; Silva, A. M. S. NMR Spectroscopy in Bitumen Characterization. In Analytical Characterization Methods for Crude Oiland Related Products; John Wiley & Sons Ltd: Chichester, UK, 2017; pp 141–161.

(3) Keller, H. R.; Massart, D. L. Evolving Factor Analysis. Chemom. Intell. Lab. Syst. 1992, 12, 209–224.

(4) Tauler, R.; Marques, I.; Casassas, E. Multivariate Curve Resolution Applied to Three‐way Trilinear Data: Study of a Spectrofluorimetric Acid–base Titration of Salicylic Acid at Three Excitation Wavelengths. J. Chemom. 1998, 12 (1), 55–75.

(5) Shinzawa, H.; Jiang, J.-H.; Iwahashi, M.; Noda, I.; Ozaki, Y. Self-Modeling Curve Resolution (SMCR) by Particle Swarm Optimization (PSO). Anal. Chim. Acta 2007, 595 (1–2), 275–281.

(6) Tefera, D. T.; Yañez Jaramillo, L. M.; Ranjan, R.; Li, C.; De Klerk, A.; Prasad, V. A Bayesian Learning Approach to Modeling Pseudoreaction Networks for Complex Reacting Systems: Application to the Mild Visbreaking of Bitumen. Ind. Eng. Chem. Res. 2017, 56, 1961–1970.

(7) Tefera, D. T.; Agrawal, A.; Yañez Jaramillo, L. M.; De Klerk, A.; Prasad, V. Self-Modeling Multivariate Curve Resolution Model for Online Monitoring of Bitumen Conversion Using Infrared Spectroscopy. Ind. Eng. Chem. Res. 2017, 56, 10756–10769.

(8) Blanchard, C. M.; Gray, M. R. Free Radical Chain Reactions of Bitumen Residue. ACS Div. Fuel Chem. Prepr. 1997, 42 (1), 137–141.

(9) Khorasheh, F.; Gray, M. R. High-Pressure Thermal Cracking of n-Hexadecane in Aromatic Solvents. Ind. Eng. Chem. Res. 1993, 32 (9), 1864–1876.

(10) Selucky, M. L.; Chu, Y.; Ruo, T.; Strausz, O. P. Chemical Composition of Athabasca Bitumen. Fuel 1977, 56, 369–381.