Data‐driven hypotheses of reaction networks for thermochemical conversion of a physical mixture of levoglucosan and 2‐phenoxyethyl benzene

Abstract

In this work, we analyze the hydrous pyrolysis of a physical mixture of the model components representing cellulose (levoglucosan) and lignin (2-phenoxyethyl benzene). Fourier transform infrared (FTIR) and proton nuclear magnetic resonance ( 1 H-NMR) spectroscopy was used to characterize the products of the reaction. The main objective of the work was to use datadriven methods to develop a reaction network for this system based on the spectroscopic data. This was achieved using Bayesian hierarchical clustering to identify pseudocomponents and Bayesian networks to develop a reaction network between these pseudocomponents. The data-driven reaction network was shown to be consistent with the known chemistry of the pyrolysis of cellulose and lignin, and the chemistry of the physical mixture incorporated/combined elements of the reaction mechanisms of cellulose and lignin.

Identification of Reaction Network Hypotheses for Complex Feedstocks from Spectroscopic Measurements with Minimal Human Intervention

“…These networks also seemed to match with the general trend noticed in the hydrolysis of biomass. 14,15,83 It is also interesting to note that the candidate molecules identified in each step closely resemble model compounds presented in the literature for the thermal pyrolysis of biomass. 6,84 4.2.…”

Data Fusion-Based Approach for the Investigation of Reaction Networks in Hydrous Pyrolysis of Biomass

“…If synergistic effects occur, then predicting a biomass feedstock’s behavior would be considerably more complex. In our previous work, , cellulose and lignin independently underwent HTL conversion using their model components: levoglucosan and 2-phenoxy-ethyl benzene, respectively, and their most probable BNs were developed. Figure shows those networks along with the BN for the HTL of biomass developed in this work.…”

“…Data fusion is a productive means of combining such independent measurements to reveal shared as well as distinct information. In this work, a signal-level data fusion approach was followed to concatenate measurements from the two spectroscopic devices. , The data was scaled and normalized with respect to minimum peak intensity prior to fusion, along with dimensional reduction (removal of uninformative variables/wavenumbers) using principal component analysis. The resultant signal had a better signal-to-noise ratio compared to the raw data from the individual signals.…”

“…The Bayesian information criterion was used as a scoring metric, and the graph that maximizes this score was chosen. The details about the implementation of Bayesian learning for HTL data have been described in detail in our previous work. , It is important to note here that the structure learning task identifies a non-loopy graph structure that is directed, thereby rejecting the reversibility of the transition (reaction) between species. Nonetheless, the approach was implemented as the graph structure learned depicts the more dominant transition between the components, i.e., the direction of the reaction with a larger rate constant.…”

“…Following glycosidic bond cleavage, hydrogen from the hydroxyl group of the carbon atom 6 C is transferred to 5 C. The SMCR framework reveals a similar reaction mechanism as represented by the network structure derived after decomposition, as depicted in Figure 11. The decomposition of cellulosic structures from previous works 26,34 is well described by the Bayesian network for biomass. The signatures of alcohols, ethers, and alkenes collected in SMCR analysis of levoglucosan are found in A1 of the biomass network.…”