Computational modeling of the upgrading of fast pyrolysis oil
- Yields of hydrocarbons
- Improve utilization of hydrogen
The CCPC has targeted the modeling of two aspects of liquid phase upgrading (LPU) of bio-oil: the engineering of the reactors and the engineering of the reactions. The performance (yield) of a chemical reactor depends on three factors:
- Intrinsic kinetics of the reactions
- Residence time distribution of the reactants
- Micromixing of the contents, for reactions that are not 0 or 1st order,
Yield = kinetics ⊗ RTD ⊗ micromixing
Intrinsic reaction kinetics can be measured in reactors that afford rates of mass transport and residence times sufficient to span the relevant ranges of rates and conversions. The residence time distribution can, at least in principle, be measured using a bolus of an inert but detectable tracer. While micromixing can be difficult to characterize experimentally, modern simulations of the fluid dynamics can help define it.
Because measurements of detailed kinetics of the reactions involved in LPU are still pending, for the sake of exercising the differences occasioned by nonidealized residence time distributions, we set the kinetics represent two parallel reactions, a bimolecular deactivation and a first order hydrogenation, both as macroscopic expressions that included Langmuir-type adsorption of the reactant, A (a highly lumped representation of the bio-oil) and the product, B (the precursor of “gunk”, which fouls the catalyst) and the product, C (the desired upgraded product):
The rate constants, ki and equilibrium constants, Ki, (Table 1; note that all concentrations were normalized by the concentration of the feed, A) were chosen to reflect that, at the typical operating temperature of the stabilization reactor of the LPU process, the deactivation reaction is about 10-times faster than the hydrogenation.
The resulting concentration profiles (Figure 1) were calculated using the Matlab differential equation solver, ODE45.
Residence time distributions
Figure 2 compares the normalized Residence Time Distributions for the two scales of reactor with those of three idealized RTDs. The characteristics of the two scaled reactors were chosen to approximate those employed in a bench scale reactor (2.5 cm diameter) with the demonstration scale reactor (10.5 cm diameter) in operation at PNNL. The fluid dynamics were modeled the continuum multi-fluid flow and transport simulator STOMP (White and Oostrom, 2006).
In lieu of detailing the geometry of the beds, we employed an empirical correlation devised by Cohen and Metzner (1981) that describes the radial changes in void fraction in the bed (Figure 3), given the diameter of the packing and the diameter of the cylindrical reactor.
The bimodal residence time distributions for the packed bed reactors arise from channeling at the wall, which is evident from the flow fields calculated by the fluid dynamics code (Figure 4).
The residence time distributions were then used to estimate the conversion of bio-oil to gunk (product B), according to the simplified kinetics scheme described above. The model results show marked differences between the large and small reactors (Table 3). Happily, it appears that the larger reactor, in which less of the volume is channeled, could perform better. Yet, all the reactors appear to be susceptible to the formation of gunk because they all will maintain the polymerizable reaction mixture at temperatures sufficient to promote gunking for times long enough to complete the desired reactions. Evidently, all the results need to be validated to assure that we have employed faithful representations of the reactor geometry and, most importantly, of the reaction kinetics.
We are currently approaching the reaction engineering of the catalysts by calculating the effects of the solvophilicity of the support on the approach of the reactants to the vicinity of the surfaces of supported, small metal particles that catalyze the hydrodeoxygenation reactions. See the following articles for more information http://dx.doi.org/10.1016/j.cattod.2016.08.025
Staff at PNNL investigating liquid phase catalysts:
Cohen Y and AB Metzner. 1981. Wall effects in laminar flow of fluids through packed beds. AIChE J. 27(5):705-715.
White MD and M Oostrom. 2006. STOMP – Subsurface Transport Over Multiple Phases, Version 4.0, User’s Guide. PNNL-15782, Pacific Northwest National Laboratory, Richland, Washington.