Towards computationally-efficient modeling of transport phenomena in three-dimensional monolithic channels
Introduction
Monolithic reactors are widely used for reducing emissions through conversion of toxic by-products of combustion into less harmful substances in gasoline-powered vehicles. In addition to their well-established role as automotive converters, they represent potential alternatives in many other applications: e.g., catalytic combustion of methane, catalytic oxidation, hydrogenation or dehydrogenation of aromatic compounds, hydrogen generation for fuel cells, steam reforming of light hydrocarbons and methanol, water gas shift reactions, etc. [1], [2]. Typically, a monolithic reactor – ceramic or metallic – comprises several hundreds of slender, parallel, and straight channels as shown in Fig. 1. The channels are in turn coated with a porous material that serves as a support for the catalytic materials: usually a noble metal such as platinum (Pt), rhodium (Rh) or palladium (Pd). The resulting structure provides a high surface area and gives rise to a low pressure drop in comparison with, for example, packed bed reactors.
Mathematical modeling and simulation have found widespread use in research and development of monolithic reactors: numerous mathematical models have been reported with varying degrees of complexity in terms of model dimensionality, washcoat modeling, chemical kinetics and number of species, physics and dynamics [1], [3], [4], [5], [6], [7], [8]. These models, as a complement to experimental design and materials research, help to gain a fundamental understanding of the complex interactions between reactions, mass and heat transfer, and fluid-flow phenomena. They are also useful in elucidating the effect of design parameters, material properties, and operating conditions – such studies would otherwise be time-consuming and expensive to carry out experimentally. Furthermore, they can be employed in optimization, control and system studies.
Detailed mechanistic models consider conservation of mass, momentum, species and heat transfer and thus result in a computationally intensive system of highly-coupled partial differential equations (PDEs). Monolith channels are manufactured with various cross-sectional shapes, including circles, squares, hexagon, triangles and sinusoids. All these geometries require three-dimensional (3D) models in order to capture the physical phenomena – except for the circle-shaped cross-section. As such, simulating the mechanistic models for 3D geometries requires significant compute resources and computing time, where the latter can range from hours to days [1], [4], [9]. This limits their applicability, because they cannot easily be exploited in wide-ranging parametric studies, optimization, and in simulations of more than one channel in the monolith [10], [11], [12], [13], [14].
To lower the computational cost, simplifications are commonly invoked to reduce the number of spatial dimensions: numerous one- (1D), and two-dimensional (2D) models have been reported [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], and there have also been attempts to derive reduced three-dimensional models [27], [28], [29], [30], [31], [32], [33] that can predict the effect of geometry on the performance of a monolithic channel. However, the nature of the physical phenomena is usually simplified in the reduction process; the simplifications usually take the form of model assumptions: e.g., the channel is treated as isothermal; the flow is assumed to be fully developed; zero thickness washcoat; or the flow is incompressible, i.e., there is no change in density despite exothermic reactions. While these models require fewer computing resources as compared to the full 3D models, they do not preserve geometrical resolution and/or physics. Moreover, these model reductions have been limited to modeling of single channels; their scalability to reactor-scale simulations has not been established.
In this regard, the aim of this paper is twofold: first, to develop a reduced 3D model that lowers the computational cost and preserves the 3D nature of a monolithic channel and the essential physics; and second, to explore how well the reduced 3D model scales numerically for simulation of an entire monolith reactor comprising hundreds or thousands of channels. For this purpose, we employ an asymptotic reduction methodology, whereby the steady-state governing equations are systematically simplified by exploiting the slenderness of monolith channels and the impermeable nature of the washcoat as compared to the flow channel. Thus, the full set of conservation equations that consist of a system of 3D elliptical PDEs are reduced to a system of 3D parabolic PDEs and 2D elliptic PDEs, as illustrated schematically in Fig. 2(b). As a result, a 3D elliptic steady-state model can be reduced to a transient-like propagation counterpart by marching along the streamwise direction ; that is, the 3D steady-state model is reformulated as a 2D model for the cross-section and integrated in the streamwise direction . Here, we demonstrate the procedure on a monolith channel with a square-shaped cross-section. The resulting reduced model and numerical implementation are then verified with the full set of equations and validated with experiments. Further, we explore the scalability of the reduced formulation for full monolith reactors comprising multiple number of channels, and discuss the limitations of the proposed reactor model. We finish with conclusions and a discussion of extensions of the current methodology for transient simulations, complex surface chemistry, and other chemical engineering systems.
Section snippets
Mathematical formulation
As a first approximation, we first assume that there is no heat loss from the monolith to the surroundings, and all the channels in the monolith are perfectly built and are subjected to identical inlet conditions. This allows us to invoke symmetry whence modeling of one single channel to predict the behavior of the monolith reactor is sufficient [9], [20]. We will then address variations between monolith channels for reactor-scale simulations.
Let us consider a model for a slender 3D monolith
Analysis
In the coming section we shall employ scaling arguments to characterize the fluid compressibility, and justify the model reductions: i.e., to show how the system of 3D elliptical PDEs can be reduced to a system of parabolic PDEs and 2D elliptic PDEs for a slender monolith channel.
Numerics
A commercial finite element solver, Comsol 4.3a [37], was used to implement both the full and reduced models. For the full model in Comsol 4.3a, the following inbuilt application modes were used: the weakly-compressible Navier–Stokes equation for conservation of mass and momentum, and convective and conductive equations for conservation of species and energy. For the reduced counterpart, user-defined PDEs were employed for all governing equations. The dependent variables for the full model are u
Validation, verification and computational savings
So far, we have employed scaling arguments to reduce the full set of equations for a non-isothermal, multicomponent model of a 3D monolith channel. Now, in order to verify that the reduced model does indeed capture the behavior of the full model, we proceed by considering a three-way reaction mechanism for automotive monolithic converters [38]. The reaction scheme with corresponding rate expressions and reaction enthalpies is summarized in Table 1. In essence, the reaction mechanism comprises
Reactor-scale simulations
We have so far presented a computationally-efficient reduced model for a single monolith channel. Now, let us turn our attention to its scalability for a full monolith reactor comprising thousands of channels (see Fig. 1(a)), all of which might behave differently due to imperfections during manufacturing, a non-uniform inlet flow distribution, and external boundary conditions, such as heat transfer with the surroundings. The computational cost in terms of memory usage and execution time would
Conclusions
We have presented a validated and verified reduced model for a square-shaped monolith channel. This was obtained by reducing the full set of equations, a system of 3D elliptic PDEs, to a set of 3D parabolic PDEs and one ODE in the flow channel and 2D elliptic PDEs in the washcoat layer. The model reduction has been shown to lower the computational requirements, in terms of memory usage and convergence time, by around 2–3 orders of magnitude whilst preserving geometrical resolution and the
Acknowledgments
The financial support of the National University of Singapore is gratefully acknowledged. The third author would like to acknowledge the support of the Mathematics Applications Consortium for Science and Industry (www.macsi.ul.ie) funded by the Science Foundation Ireland Mathematics Initiative Grant 06/MI/005, and the award of a visiting researcher grant within the Government of Brazil’s “Science without Borders” programme.
References (45)
- et al.
Mathematical modeling of monolith catalysts and reactors for gas phase reactions
Appl. Catal. A: Gen.
(2008) - et al.
The application of monoliths for gas phase catalytic reactions
Chem. Eng. J.
(2001) - et al.
Modeling of transport and chemistry in channel flows of automotive catalytic converters
Chem. Eng. Sci.
(2010) - et al.
A critical evaluation of Navier–Stokes, boundary-layer, and plug-flow models of the flow and chemistry in a catalytic-combustion monolith
Catal. Today
(2000) - et al.
Recent advances in numerical modeling of catalytic monolith reactors
Catal. Today
(2005) - et al.
Modeling the effect of flow mal-distribution on the performance of a catalytic converter
Chem. Eng. Sci.
(2012) - et al.
Towards the simulation of the catalytic monolith converter using discrete channel-scale models
Catal. Today
(2012) - et al.
CFD modelling of the automotive catalytic converter
Catal. Today
(2012) - et al.
Toward simulation of full-scale monolithic catalytic converters with complex heterogeneous chemistry
Comput. Chem. Eng.
(2010) - et al.
Kinetic and 2D reactor modeling for simulation of the catalytic reduction of NOx in the monolith honeycomb reactor
Process Saf. Environ. Prot.
(2008)
Finite-element model for a catalytic monolith reactor
Comput. Chem. Eng.
A study of Nusselt and Sherwood numbers in a monolith reactor
Catal. Today
A discussion on transport phenomena and three-way kinetics of monolithic converters
Chem. Eng. Sci.
Comparison among monolithic and randomly packed reactors for the methanol-to-propylene process
Chem. Eng. J.
Conceptual design and CFD simulation of a novel metal-based monolith reactor with enhanced mass transfer
Catal. Today
A fully distributed model for the simulation of a catalytic combustor
Chem. Eng. Sci.
The effect of washcoat geometry on mass transfer in monolith reactors
Chem. Eng. Sci.
3-D modeling of monolith reactors
Catal. Today
Simplified method for effectiveness factor calculations in irregular geometries of washcoats
Chem. Eng. Sci.
A finite element method for the weakly compressible parabolized steady 3D Navier–Stokes equations in a channel with a permeable wall
J. Comput. Fluids
Modelling transport phenomena and chemical reactions in automotive three-way catalytic converters
Chem. Eng. J.
Computationally-efficient hybrid strategy for mechanistic modeling of fuel cell stacks
J. Power Sources
Cited by (8)
Diffusion–reaction phenomenon for negative-order reactions in flow reactors
2019, Chemical Engineering Research and DesignCitation Excerpt :Significant advances have been made in the development of reactor models for packed-bed reactors, fluidized bed reactors, structured reactors, and other reactors in which heterogeneous catalytic reactions take place (Deen et al., 2007; Jurtz et al., 2019; Mahecha-botero et al., 2009; Mladenov et al., 2010; Raimondeau and Vlachos, 2002; Rudniak et al., 2004; Sharma et al., 2015; Stepanek et al., 2012; Tischer and Deutschmann, 2005; Wang et al., 2013).
Modeling of a square channel monolith reactor for methane steam reforming
2018, EnergyCitation Excerpt :The structure consists of many straight parallel channels in micron- and millimeter-scale (0.5–6.0 mm). These parallel channels can be manufactured with square, triangular, and sinusoidal cross-sectional areas to provide flow passages for the reactants and products to and from the reactions, respectively [25–27]. For an application for diesel particulate filter, a hexagonal shape cross-sectional area can provide higher filtration efficiency [28].
Investigating the effect of channel geometry on selective catalytic reduction of NO<inf>x</inf> in monolith reactors
2017, Chemical Engineering Research and DesignCitation Excerpt :Detailed modeling and simulation is required for in depth understanding the complexity of interactions between various physical and chemical phenomena occurring inside the monolith channels and on the channel walls (Chen et al., 2008; Tilland et al., 2015). In the recent years, many researchers have used different mathematical models for simulation and modeling of these types of reactors (Chen et al., 2008; Shrestha et al., 2015; Dhanushkodi et al., 2008; Mei et al., 2007; Ciardelli et al., 2007; Schildhauer et al., 2005; Ciardelli et al., 2004; Han et al., 2003; Stitt et al., 2001; Veser and Frauhammer, 2000; Goralski et al., 2000; Mozaffari et al., 2015; Sharma et al., 2015). Depending on the objectives, applications and numerical computational limitations, these models can be categorized as one-, two-, or three-dimensional models (1-D, 2-D or 3-D, respectively).
Comparison of random and monolithic fixed-bed reactors for the oxidative dehydrogenation of butene to butadiene
2016, Journal of the Taiwan Institute of Chemical EngineersValidity and scalability of an asymptotically reduced single-channel model for full-size catalytic monolith converters
2016, Applied Mathematics and ComputationCitation Excerpt :The reduced model thus comprises 3D parabolic PDEs for mass, momentum, species and energy in the flow channels, and 2D elliptic PDEs for energy in the other layers (see Fig. 2b). For the sake of brevity, we summarize the full set of the governing equations and reduced counterpart in Appendix A and Appendix B, and refer to Sharma et al. [29] for more details including the derivation of the reduced model and model validation with experiments. The methodology of nondimensionalization, scaling analysis, and simple asymptotics to obtain leading-order solutions is well illustrated with examples in our earlier work [31–37].