PARTICLE TECHNOLOGY AND FLUIDIZATION Dynamic Multiscale Method for Gas-Solid Flow via Spatiotemporal Coupling of Two-Fluid Model and Discrete Particle Model Xizhong Chen and Junwu Wang State Key Laboratory of Multiphase Complex Systems, Institute of Process Engineering, Chinese Academy of Sciences, P.O. Box 353, Beijing 100190, P.R. China DOI 10.1002/aic.15723 Published online April 12, 2017 in Wiley Online Library (wileyonlinelibrary.com) Various computational fluid dynamics methods have been developed to study the hydrodynamics of gas-solid flows, however, none of those methods is suitable for all the problems encountered due to the inherent multiscale characteristics of gas-solid flows. Both discrete particle model (DPM) and two-fluid model (TFM) have been widely used to study gassolid flows, DPM is accurate but computationally expensive, whereas TFM is computationally efficient but its deficiency is the lack of reliable constitutive relationships in many situations. Here, we propose a hybrid multiscale method (HMM) or dynamic multiscale method to make full use of the advantages of both DPM and TFM, which is an extension of our previous publication from rapid granular flow (Chen et al., Powder Technol. 2016;304:177–185) to gas-solid two-phase flow. TFM is used in the regions where it is valid and DPM is used in the regions where continuum description fails, they are coupled via dynamical exchange of parameters in the overlap regions. Simulations of gas-solid channel flow and fluidized bed demonstrate the feasibility of the proposed HMM. The Knudsen number distributions are also C 2017 American Institute of Chemical Engineers AIChE J, 63: 3681– reported and analyzed to explain the differences. V 3691, 2017 Keywords: two-fluid model, discrete particle method, hybrid model, spatial coupling, dynamic coupling, gas-solid flow Introduction Gas-solid flows occur in many industries and thus have attracted a lot of attention from engineers and scientists over the past decades. Recently, computational fluid dynamics (CFD) has already emerged as a popular tool in aiding the design and scale-up of relevant processes. Various CFD methods, such as direct numerical simulation (DNS), discrete particle model (DPM) and two-fluid model (TFM), have been developed to study the hydrodynamics of gas-solid flow.1,2 However, none of those methods is suitable for all the problems encountered due to the inherent multiscale characteristics of gas-solid flows.3 Both TFM4–6 and DPM7–9 are very popular in fluidization community. TFM treats both gas and solid as interpenetrating continua, the model is commonly used in the study of largescale problems as it is computationally efficient, but the drawback lies at its continuum treatment of solid phase which results in significant uncertainties of this model. The constitutive equations for particulate phase stresses is generally based on the assumption that there is a clear scale separation in granular systems, or from thermodynamic viewpoint the local thermodynamic equilibrium postulate is valid. Unfortunately, this is usually not the case.10–12 For example, Continuum method Correspondence concerning this article should be addressed to J. Wang at jwwang@home.ipe.ac.cn. C 2017 American Institute of Chemical Engineers V AIChE Journal is inadequate due to the existence of Knudsen layer.13,14 Conversely, DPM treats the fluid as a continuum phase and treats the solid phase as a collection of discrete particles that obeys the Newton’s second law. The DPM is much more accurate compared with TFM, however, the computational costs are often very demanding, especially in case of dense suspensions encountered in industrial reactors. Please note that highly resolved TFM may offer insight of the physics of gas-solid two-phase flow, but it meets difficult when there are big velocity, density and/or temperature gradients (or in general, the magnitude of thermodynamic forces is large). More detailed discussion on this topic could be referred to a recent publication of Fullmer and Hrenya.15 In present study, a hybrid multiscale method (HMM) or dynamic multiscale method is developed for modeling the complex gas-solid flow. The HMM aims to have both the efficiency of TFM and the accuracy of DPM for modeling the hydrodynamics of gas-solid flows in large-scale reactors, which is an extension of our previous works from rapid granular flow to gas-solid two-phase flow.16,17 Please note that only numerical comparisons between all three models are carried out to verify the developed HMM as its accuracy would not surpass the DPM method theoretically. When compared with the experimental results, some uncertainties about the particle properties (shape, size distribution, moisture) and geometries simplification may inevitably cause some ambiguities. The advantage of the numerical comparison is that the initial condition of all the three models could be controlled to make sure September 2017 Vol. 63, No. 9 3681 conditions for continuum description of solid phase in TFM. The solid volume fraction, solid velocity, and granular temperature are calculated as follows 1 X mi (1) eJ 5 VJ qs i J Method The TFM used in present study is a standard TFM with kinetic theory of granular flow for closing particulate phase stresses and Gidaspow’s model4 for closing the interphase drag force term. The Johnson and Jackson model18 is used to calculate the tangential velocity and granular temperature of solid phase at the wall boundary. The governing equations of TFM are listed in Appendix. The discrete particle model used in present study is a linear spring-dashpot discrete element method for particulate phase and Navier-Stokes equations for gas phase. Gidaspow’s model is also used for closing the interphase drag force term. Please note that to consist with the kinetic theory of granular flow used in TFM, the particles are smooth in DPM. The governing equations of discrete particle model are listed in Appendix. A general schematic of the geometry of the hybrid multiscale method proposed in this work is shown in Figure 1. DPM is used to solve the part of the channel near the wall, while TFM is used to solve the remaining part. TFM and DPM coincidently overlapped in overlap regions which are delicately designed to exchange solid phase boundary from each side of the individual model to enforce the mass and momentum conservation in the total system. The overlap region contains three subregions: a discrete boundary subregion, a buffer subregion and a continuum boundary. The buffer subregion x1 x2 is used to minimize the disturbance caused by boundary data exchange. Detailed description of the handling solid phase in overlap region could also refer to Chen, Wang, Li.17 The physical quantities of discrete particle in the subregion x2 x3 are sampled and averaged to generate the boundary 1 X mi vi VJ eJ qs i J (2) 1 X mi ðvi 2 uJ Þ2 3VJ eJ qs i J (3) uJ 5 HJ 5 The subregion x0 x1 is used to provide boundary for solid phase in discrete particle model from flow field of two-phase model. The mass conservation is achieved through a flux monitor placed at the termination of the region of discrete particle model (x0). The number of particles to be inserted or deleted is calculated by n 5 ðAJ uc Þ eJ qs 䉭tc = mi (4) where AJ uc is the flux of cell J calculated by TFM. The velocities of newly inserted particle are generated from Maxwell distribution with mean and standard deviation provided by the local information of TFM. A virtual wall is placed at the termination of discrete region to prevent particle from freely drifting to TFM region. After the compensation of mass conservation, a constraint dynamics of particle velocities in subregion x0 x1 is enforced to ensure the momentum conservation. The new velocity of particle i in this region is relaxed to the local solid velocity of TFM uc calculated by vi 0 5 vi 1 u c 2 NJ 1 X Vi NJ i51 (5) After the solid phase flow field is resolved, the whole solid volume fraction and drag force information are passed to the gas phase. A unified gas phase solver based on the modified SIMPLE algorithm is used to obtain the gas velocity and pressure of the whole simulated domain. Figure 1. Schematic of the coupling strategy of hybrid multiscale model. [Color figure can be viewed at wileyonlinelibrary.com] 3682 DOI 10.1002/aic Published on behalf of the AIChE September 2017 Vol. 63, No. 9 AIChE Journal 15475905, 2017, 9, Downloaded from https://aiche.onlinelibrary.wiley.com/doi/10.1002/aic.15723 by Shanghai Jiaotong University, Wiley Online Library on [04/02/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License they are consistent. Simple geometry and flow condition are likewise preferred and chosen here while more complex and realistic industrial reactor simulation is the ultimate goal. The instantaneous and time averaged Knudsen number distributions predicted by discrete particle model are reported and analyzed to further clarify the differences between different models. The detailed flowchart of hybrid multiscale numerical algorithm for gas-solid flow is showed in Figure 2. In every time step, the algorithm begins with solving the individual particle movement in the regions of discrete particle model, the positions and velocities of the entire collection of particle are known and subsequently the boundary conditions for the solid phase in TFM could be determined. The governing equations of solid phase in TFM is solved in the regions belong to TFM given the boundary conditions, then the number of particles to be added or deleted in the discrete particle region are calculated based on the mass flux monitor of the solid phase Table 1. The Domain Decomposition Setting for HMM in Horizontal Direction Domain TFM DPM Buffer Grid No. 318 17 and 1420 46 and 1517 AIChE Journal September 2017 Vol. 63, No. 9 information of TFM. If a particle is to be added, its velocity will get from the Maxwell velocity distribution with the mean velocity and standard deviation provided by velocity and Table 2. Simulation Parameters of the Gas-Solid Channel Flow Parameter Value Gas viscosity (Pas) Gas temperature (K) Operation pressure (Pa) Particle density (kg/m3) Particle diameter (m) Spring stiffness coefficient (N/m) Particle–particle coefficient of restitution Particle–wall coefficient of restitution Height of channel (m) Width of channel (m) Time step (s) Published on behalf of the AIChE 1.8 3 1025 298 1,01325 2000 1.2 3 1023 3000 0.95 0.95 0.8 0.12 1.0 3 1025 DOI 10.1002/aic 3683 15475905, 2017, 9, Downloaded from https://aiche.onlinelibrary.wiley.com/doi/10.1002/aic.15723 by Shanghai Jiaotong University, Wiley Online Library on [04/02/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Figure 2. Flowchart of hybrid multiscale model algorithm for gas-solid flows. [Color figure can be viewed at wileyonlinelibrary.com] Figure 4. Comparisons of the radial gas velocity distribution results between different models for channel flow with uniform inlet velocity. [Color figure can be viewed at wileyonlinelibrary.com] Figure 6. The time series of Knudsen number in different monitor positions for channel flow with uniform inlet velocity. [Color figure can be viewed at wileyonlinelibrary.com] Figure 5. The probability density distribution of Knudsen number for channel flow with uniform inlet velocity. Figure 7. Radial distributions of solid volume fraction predicted by all three methods. [Color figure can be viewed at wileyonlinelibrary.com] 3684 DOI 10.1002/aic Published on behalf of the AIChE [Color figure can be viewed at wileyonlinelibrary.com] September 2017 Vol. 63, No. 9 AIChE Journal 15475905, 2017, 9, Downloaded from https://aiche.onlinelibrary.wiley.com/doi/10.1002/aic.15723 by Shanghai Jiaotong University, Wiley Online Library on [04/02/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Figure 3. Comparisons of the radial simulation results between different models for channel flow with uniform inlet velocity: (a) Solid volume faction (b) Solid velocity. Parameter Value Gas viscosity (Pas) Gas temperature (K) Operation pressure (Pa) Particle density (kg/m3) Particle diameter (m) Spring stiffness coefficient (N/m) Particle–particle coefficient of restitution Particle–wall coefficient of restitution Height of channel (m) Width of channel (m) Inlet gas velocity (m/s) Inlet solid flux (kg/(m2s)) Time step (s) Figure 8. The frequency density distribution of Knudsen number inside the channel. [Color figure can be viewed at wileyonlinelibrary.com] 1.8 3 1025 298 1,01325 2000 1.2 3 1023 24,000 0.95 0.95 1.6 0.12 7.0 60 5.0 3 1026 the momentum conservation. Once the solid phase volume fraction and velocity distributions on the entire region are known, the source term in gas phase governing equations could be calculated and followed by a solving procedure to the flow field of gas phase in the entire domain. Results and Discussion Figure 9. The 3D distribution of Knudsen number inside the channel. [Color figure can be viewed at wileyonlinelibrary.com] granular temperature of local TFM correspondingly. A final adjustment of particle velocities in the near boundary interface is enforced according to constraint dynamics (Eq. 5) to ensure To verify the implementation of the hybrid multiscale numerical algorithm and further demonstrate the benefits of hybrid multiscale model, three gas-solid channel flow cases are designed and simulated. The boundary conditions at the wall are said to play an important role in the accuracy of the results of TFM for the gas-solid channel, due to the effect of Knudsen Layer which cannot be captured by TFM. Therefore, DPM is utilized in the near wall regions to simulate the gassolid flow inside the Knudsen Layer and the remaining part of the channel is simulated by TFM. Meanwhile, each case is also simulated using all the three methods, that is, pure TFM, pure discrete particle model and hybrid multiscale model. The domain decomposition setting in horizontal direction for hybrid multiscale model is listed Table 1. All the cases have a consistent grid setting, that is, the domain is discretized in 20 structural cells in the horizontal direction and 64 cells in the vertical direction. In addition, the third direction in discrete particle model is set as periodic boundary for solid phase and has a thickness of 6.1 times particle diameter. Figure 10. Radial distributions of gas and solid velocity predicted by all three methods. [Color figure can be viewed at wileyonlinelibrary.com] AIChE Journal September 2017 Vol. 63, No. 9 Published on behalf of the AIChE DOI 10.1002/aic 3685 15475905, 2017, 9, Downloaded from https://aiche.onlinelibrary.wiley.com/doi/10.1002/aic.15723 by Shanghai Jiaotong University, Wiley Online Library on [04/02/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Table 3. Simulation Parameters of the Gas-Solid Riser with Coarse Particles dp Kn 5 pffiffiffi 6 2g0 es L Figure 11. The snapshots of flow field predicted by all three methods at t 5 15 s (Ug 5 7.0 m/s, Gs 5 60 kg/(m2s)). [Color figure can be viewed at wileyonlinelibrary.com] (6) The probability density distribution of Knudsen number inside the channel predicted by DPM is showed in Figure 5. It can be seen that the majority of Knudsen number inside the channel are very small and the time averaged value is 0.00548, which indicates that most of regions may be suitable for continuum description of solid phase. Figure 6 presents the time series of Knudsen number in different monitor positions. The fluctuation and average value of Knudsen number in the near wall position are larger than other positions, which could be further used to explain the reason why the simulation results of TFM deviate from DPM. Gas-solid channel flow with uniform inlet velocity A uniform velocity-driven gas-solid channel flow without considering the gravity is first studied. Both gas and solid are flowed into the system from the bottom of the channel with uniform velocity (2.0 m/s) and the inlet solid volume fraction is 0.1. The gas density is calculated using idea gas law. Noslip boundary is imposed to gas phase at walls and pressure outlet boundary is prescribed at the top of the channel. Johnson and Jackson boundary condition for solid phase is applied at the wall in pure TFM, where the particle wall restitution coefficient is the same as particle–particle restitution coefficient and the specularity coefficient is set to zero. Total simulation time of 3 s is completed for each simulation and mean flow field information is extracted from the last 2 s numerical results. Detailed parameters used in the numerical simulation are listed in Table 2. Figure 3 shows the comparisons of the radial distribution of solid volume fraction and solid velocity simulations results predicted by all the three models. It is found that although both the inlet gas and solid velocities are uniform, DPM predicts that there are a small oscillation of solid volume fraction and a decline of solid velocity in the near wall region. The decline of solid velocity could be contributed to the decline of gas velocity caused by the no-slip boundary, which is presented in Figure 4. It is clear that the simulation results of HMM are consistent with DPM while TFM failed to capture the solid volume fraction oscillation near the wall. 3686 DOI 10.1002/aic Figure 12. The predicted time averaged axial voidage profiles. Published on behalf of the AIChE [Color figure can be viewed at wileyonlinelibrary.com] September 2017 Vol. 63, No. 9 AIChE Journal 15475905, 2017, 9, Downloaded from https://aiche.onlinelibrary.wiley.com/doi/10.1002/aic.15723 by Shanghai Jiaotong University, Wiley Online Library on [04/02/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Noting that the gas velocity predicted by HMM is also corrected near the wall boundary, which indicated that both gas and solid phase can be benefit from the design of hybrid multiscale method. The Knudsen number (Kn) defined as the ratio of mean free path (k) to the characteristic length scale (L) is often used in microscale gas flows to characterize the validation of the continuum assumption. TFM treats solid phase as a continuum media thus its application may be limited to the situation of small Knudsen number. Conversely, DPM tracks individual particle directly and thus do not rely on the magnitude of the Knudsen number. Therefore, we could use the simulation results of DPM to investigate of the distribution of Knudsen number inside the channel. Here, we choose the width of channel as the characteristic length and calculate the mean free path of particle using kinetic theory of granular flow.4 The formulation of local Knudsen number used in this case is given as follows [Color figure can be viewed at wileyonlinelibrary.com] Gas-solid channel flow with parabolic inlet velocity In this case, all simulation parameters are the same as the previous one except that the inlet gas and solid velocities are changed to be parabolic Uz 5Um ð12r 2 =R2 Þ, where Um is the maximum velocity (2.0 m/s) in the center of the channel. Due to the difference of the solid velocity across the channel, the radial solid volume also became nonuniform across the channel. Figure 7 shows that all the three models predict a similar flow pattern. Quantitatively, TFM has a relative large deviation from DPM in the center and the near wall region while the simulation results of HMM is more consistent with DPM. To disclose the reason, the Knudsen number (Eq. 6) based on DPM simulation results is then calculated. Figure 8 shows the frequency density distribution of Knudsen number inside the channel. It can be seen that the average Knudsen number is larger than the uniform inlet velocity case. Most of the large AIChE Journal September 2017 Vol. 63, No. 9 value of Knudsen number are observed in the near wall region as shown in Figure 9. As a result, it is expected that the simulation results of solid velocity in the near wall region between TFM and DPM will have some differences. Figure 10 also shows that the gas velocity predicted by HMM has a better agreement with DPM simulation results. Gas-solid fluidized bed Different from the first two cases, the gravity is finally added to the channel in this case. Detailed parameters used in the numerical simulation are listed in Table 3. Due to the formation of heterogeneous structures, a longer simulation time is needed. The flow field is found to arrive at the steady state after 5 s. Total simulation time of 16 s is completed for each simulation and mean flow field information is collected after the initial 5 s. Published on behalf of the AIChE DOI 10.1002/aic 3687 15475905, 2017, 9, Downloaded from https://aiche.onlinelibrary.wiley.com/doi/10.1002/aic.15723 by Shanghai Jiaotong University, Wiley Online Library on [04/02/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Figure 13. The predicted time averaged radial flow profiles. [Color figure can be viewed at wileyonlinelibrary.com] Figure 11 shows the snapshots of flow field predicted by all the three models. The characteristic of a typical gas-solid riser flow is the dynamic formation of heterogeneous structure like clusters. As it can be seen from Figure 11, all the models predict the nonuniform flow field inside the channel, which is qualitatively the same as the experimental observation.19 Figure 12 shows the quantitatively comparisons of time averaged axial voidage profiles along the channel height predicted by all the three models. Compared to TFM, the overall simulation result of HMM is closer to DPM, especially in the middle and top parts of the channel. However, both HMM and TFM show some differences in the near bottom parts of the channel. Figure 13 presents the time averaged radial solid fraction, gas velocity, and solid velocity predict by all the three models with two different heights. Qualitatively, all the three models predict dense solid volume fraction in the near wall region and dilute solid volume fraction in the centre of the channel. Quantitatively, if we choose the simulation results of DPM as benchmark, the overall performance of HMM is better than TFM. Due to clusters are very common in the predicted flow field, we choose the local gradient of the solid volume fraction (L5es =jres j) as the characteristic length to calculate the Knudsen number inside the channel in this case. The solid volume fraction inside and outside the cluster is significantly different, thus this Knudsen number calculation method takes into account for the structure of flow field. Figure 14 shows the frequency distribution of Knudsen number inside the channel. The majority of both the transient and time averaged value of the Knudsen number are not large. However, there do exist some large values as could be observed in Figure 15b. It is found that in the near bottom and top region of the channel, the time averaged Knudsen number is significant larger than the middle part. This can qualitatively explain the reason why the simulation results of TFM deviate more from DPM in those regions. Conversely, the transient value of Knudsen number does not have an evident regularity as shown in Figure 15a. Large value of transient Knudsen number could appear in many regions of the channel as the clusters are dynamically evolution inside the channel. However, the domain decomposition of DPM and TFM region in HMM simulation is always fixed at present, which causes the simulation results of HMM still have some difference with pure DPM simulation. This indicates that the domain decomposition should be automatically adjusted with a specific criteria such as given Knudsen number in future research. Figure 15. The 3D distributions of Knudsen number inside the channel: (a) Transient value (b) Time averaged value. [Color figure can be viewed at wileyonlinelibrary.com] 3688 DOI 10.1002/aic Published on behalf of the AIChE September 2017 Vol. 63, No. 9 AIChE Journal 15475905, 2017, 9, Downloaded from https://aiche.onlinelibrary.wiley.com/doi/10.1002/aic.15723 by Shanghai Jiaotong University, Wiley Online Library on [04/02/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Figure 14. The distributions of Knudsen number inside the channel: (a) Transient value (b) Time averaged value. Conclusions This work presents a hybrid multiscale model with spatiotemporally coupled TFM and discrete particle model. In the framework of hybrid multiscale model, the gas phase is solved using a unified solver throughout the whole domain. An overlap region is constructed to facilitate the boundary communication of solid phase between TFM and discrete particle model. Gas-solid channel flows without gravity were simulated by pure TFM, pure discrete particle model and hybrid multiscale model. The simulation results indicated that the hybrid multiscale model was consistent with pure discrete particle model while pure TFM had some differences with pure discrete particle model in the near wall region. The distribution of Knudsen number based on the simulation results of pure discrete particle model was analyzed to explain the reason of deviations. Further simulations of the gas-solid fluidized bed also showed the feasibility of the proposed hybrid multiscale model. Large number of Knudsen number appeared in many regions of the fluidized bed due to the dynamically changed heterogeneous structures. Future works will involve improving the model with the development of a criterion of the validity of TFM and the implementation of adaptive domain decomposition method for the cases of more complex spatial heterogeneous flow structures. Acknowledgments We thank Professor Hans Kuipers at Eindhoven University of Technology for allowing the usage of his 3D-DPM code, on which the HMM is based. This study is financially supported by the National Natural Science Foundation of China (21422608) and the “Strategic Priority Research Program” of the Chinese Academy of Sciences under the grant No.XDA07080200. Notation TFM = DPM = HMM = dp = e= ew = F= g= g0 = I= kn = Kn = L= m= Np = pg = ps = ra = t= Re = ug = us = v= two-fluid model discrete particle model hybrid multiscale model particle diameter, m coefficient of restitution between particle–particle interaction coefficient of restitution between particle and wall interaction force, N gravitational acceleration, m/s2 radial distribution function unit tensor normal spring stiffness, N/m Knudsen number characteristic length scale, m particle mass, kg number of particles gas pressure, Pa particle pressure, Pa position of particle a time, s Reynold number gas velocity vectors, m/s solid velocity vectors, m/s particle velocity vectors, m/s AIChE Journal September 2017 Vol. 63, No. 9 Vc = volume of cell, m3 Vp = volume of particle, m3 Greek letters b= eg = es = es,max = lg, ls = ks = Hs = s= u= d= dn = cs = qg = qs = gn = drag coefficient for a control volume, kg/m3s voidage solid volume fraction solid volume fraction at packed condition fluid and solid viscosity, Pas solid bulk viscosity granular temperature, m2/s2 stress tensor specularity coefficient dirac function overlap between particles, m energy dissipation, J/(m3s) fluid density, kg/m3 solid density, kg/m3 normal damping coefficient Literature Cited 1. van der Hoef MA, van Sint Annaland M, Deen NG, Kuipers JAM. Numerical simulation of dense gas-solid fluidized beds: a multiscale modeling strategy. Annu Rev Fluid Mech. 2008;40(1):47–70. 2. Ge W, Wang W, Yang N, Li J, Kwauk M, Chen F, Chen J, Fang X, Guo L, He X, Liu X, Liu Y, Lu B, Wang J, Wang J, Wang L, Wang X, Xiong Q, Xu M, Deng L, Han Y, Hou C, Hua L, Huang W, Li B, Li C, Li F, Ren Y, Xu J, Zhang N, Zhang Y, Zhou G, Zhou G. Meso-scale oriented simulation towards virtual process engineering (VPE)–The EMMS paradigm. Chem Eng Sci. 2011; 66(19):4426–4458. 3. Li J, Ge W, Wang W, Yang N, Liu X, Wang L, He X, Wang X, Wang J, Kwauk M. From Multiscale Modeling to Meso-Science. Springer Berlin Heidelberg, 2013. 4. Gidaspow D. Multiphase Flow and Fluidization: Continuum and Kinetic Theory Description. Boston: Academic Press, 1994. 5. Enwald H, Peirano E, Almstedt AE. Eulerian two-phase flow theory applied to fluidization. Int J Multiphase Flow. 1996;22(Suppl):21– 66. 6. Syamlal M, Pannala S. Multiphase continuum formulation for gassolids reacting flows. In: Pannala S, Syamlal M, O’Brien TJ, editors. Computational Gas-Solids Flows and Reacting Systems. New York: IGI Global, 2011, 1–65. 7. Deen NG, Van Sint Annaland M, Van der Hoef MA, Kuipers JAM. Review of discrete particle modeling of fluidized beds. Chem Eng Sci. 2007;62(1–2):28–44. 8. Zhu HP, Zhou ZY, Yang RY, Yu AB. Discrete particle simulation of particulate systems: theoretical developments. Chem Eng Sci. 2007;62(13):3378–3396. 9. Zhu HP, Zhou ZY, Yang RY, Yu AB. Discrete particle simulation of particulate systems: a review of major applications and findings. Chem Eng Sci. 2008;63(23):5728–5770. 10. Goldhirsch I. Rapid granular flows. Annu Rev Fluid Mech. 2003; 35(1):267–293. 11. Fullmer WD, Hrenya CM. The clustering instability in rapid granular and gas-solid flows. Annu Rev Fluid Mech. 2017;49:485–510. 12. Goldhirsch I. Scales and kinetics of granular flows. Chaos. 1999; 9(3):659–672. 13. Campbell CS. Boundary interactions for two-dimensional granular flows. Part 2. Roughened boundaries. J Fluid Mech. 1993;247:137–156. 14. Galvin J, Hrenya C, Wildman R. On the role of the Knudsen layer in rapid granular flows. J Fluid Mech. 2007;585:73–92. 15. Fullmer WD, Hrenya CM. Quantitative assessment of fine-grid kinetic-theory-based predictions of mean-slip in unbounded fluidization. AIChE J. 2016;62(1):11–17. 16. Chen X, Wang J. Hybrid discrete-continuum model for granular flow. Procedia Eng. 2015;102:661–667. 17. Chen X, Wang J, Li J. Multiscale modeling of rapid granular flow with a hybrid discrete-continuum method. Powder Technol. 2016; 304:177–185. 18. Johnson PC, Jackson R. Frictional-collisional equations of motion for particulate flows and their application to chutes. J Fluid Mech. 1987;176:67–93. 19. Cocco R, Shaffer F, Hays R, Reddy Karri SB, Knowlton T. Particle clusters in and above fluidized beds. Powder Technol. 2010;203(1): 3–11. Published on behalf of the AIChE DOI 10.1002/aic 3689 15475905, 2017, 9, Downloaded from https://aiche.onlinelibrary.wiley.com/doi/10.1002/aic.15723 by Shanghai Jiaotong University, Wiley Online Library on [04/02/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License In this case, TFM simulation of the whole domain is fastest, which takes 41 h of CPU clock time. Discrete particle model simulation of the whole domain takes 320 h of CPU clock time to complete the running, while hybrid multiscale model only takes 243 h to complete. Therefore, hybrid multiscale model is more efficient than pure discrete particle model even in this demo and it is promising for larger scale simulation as well. Table A1. Governing Equations of TFM Description Equation @ ðeg qg Þ 1r ðeg qg ug Þ50 @t @ ðes qs Þ @t 1r ðes qs us Þ50 Mass balance equations @ ðeg qg ug Þ 1r eg qg ug ug 52eg rpg 1r eg sg 1b us 2ug 1eg qg g @t @ ðes qs us Þ 1r ðes qs us us Þ52rps 2es rpg 1r ðes ss Þ1b ug 2us 1es qs g @t Momentum balance equations h i ðes qs us Hs Þ 5ð2ps I1es ss Þ : rus 2r ðks rHs Þ2c23bHs Granular temperature equation 3 @ðes qs Hs Þ 1r @t 2 Gas stress sg 5lg ðrug 1ruTg Þ1 23 lg ðr us ÞI ss 5ls ðrus 1ruTs Þ1 ks 2 23 ls ðr us ÞI Solid stress Ps 5es qs Hs 12qs ð11eÞe2s g0 Hs qffiffiffiffi pffiffiffiffiffiffi e qs dp pHs ls 5 45 qs dp e2s g0 ð11eÞ Hps 1 s 6ð32eÞ 11 25 ð11eÞð3e21Þes g0 qffiffiffiffi ks 5 43 es qs dp g0 ð11eÞ Hps qffiffiffiffi pffiffiffiffiffiffi 2 150qs dp Hs p Hs 6 2 11 e g ð11eÞ 12q e d ð11eÞg ks 5 384ð11eÞg s 0 p 0 s s 5 p 0 qffiffiffiffi Hs 2 2 4 c53ð12e Þes qs g0 Hs dp p 2ðr us Þ Solid phase pressure Solid shear viscosity Solid bulk viscosity Granular phase heat conductivity Inelastic collision energy dissipation 1=3 21 es g0 5 12 es;max 8 qg eg es jug 2us j 22:65 3 > > > eg > 4 CD < dp b5 > e2 l q es jug 2us j > > > 150 s 2g 11:75 g : eg dp dp ( ð24=Re Þð110:15Re 0:687 Þ; CD 5 0:44; Radial distribution function Interphase drag force coefficient Wall boundary condition of solid phase es 0:2 es > 0:2 Re < 1000 Re 1000 e q d ju 2u j Re 5 g g pl g s g 6ls es;max @us;w us;w 52 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffi 3puqs es g0 Hs @n pffiffi 3=2 3puqs es u2s;slip g0 Hs @H Hs;w 52 kcs Hs @ns;w 1 6es;max cs;w s;w Table A2. Governing Equations of Discrete Particle Model Description Equation Mass balance equation for gas phase @ ðeg qg Þ 1r @t Momentum balance equations for gas phase @ Momentum exchange term ðeg qg ug Þ50 ðeg qg ug Þ 1r ðeg qg ug ug Þ52eg rPg 1r @t ðX Np bVa Sp 5 V1c ðug 2va Þdðr2ra ÞdVc es a51 ðeg sg Þ2sp 1eg qg g 8 qg eg es jug 2us j 22:65 3 > > > eg es 0:2 > 4 CD < dp b5 > e2 l q es jug 2us j > > > 150 s 2g 11:75 g es > 0:2 : eg dp dp Particle movement equation ma dvdta 52Va rpg 1 X b2contactlist dra dt 3690 DOI 10.1002/aic Fab;n 1 Va b ðug 2va Þ1ma g es 5va Published on behalf of the AIChE September 2017 Vol. 63, No. 9 AIChE Journal 15475905, 2017, 9, Downloaded from https://aiche.onlinelibrary.wiley.com/doi/10.1002/aic.15723 by Shanghai Jiaotong University, Wiley Online Library on [04/02/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Appendix Description Equation Contact force Fab;n 52kn dn nab 2gn vab;n Overlap between particles dn 5ðRa 1Rb Þ2jrb 2ra j Normal unit vector a nab 5 jrrbb 2r 2ra j Normal component of the relative velocity vab;n 5ðvab nab Þnab pffiffiffiffiffiffiffiffiffiffiffiffi mab kn lnffi e gn 5 22pffiffiffiffiffiffiffiffiffiffiffiffi 2 2 Normal damping coefficient p 1ln e Manuscript received Nov. 3, 2016, and revision received Feb. 12, 2017. AIChE Journal September 2017 Vol. 63, No. 9 Published on behalf of the AIChE DOI 10.1002/aic 3691 15475905, 2017, 9, Downloaded from https://aiche.onlinelibrary.wiley.com/doi/10.1002/aic.15723 by Shanghai Jiaotong University, Wiley Online Library on [04/02/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License TABLE A2. Continued