Open-source software for computational engineering
Transcription
Open-source software for computational engineering
OpenQBMM – An open-source implementation of quadrature-based moment methods. Application to population balances and turbulent mixing A L B E R TO PA S S A L A C Q U A 1 , 2 , R O D N E Y O . F OX 2 1Department 2Department of Mechanical Engineering, Iowa State University of Chemical and Biological Engineering, Iowa State University 1ère journée française des utilisateurs de OpenFOAM May 18th, 2016, Rouen, Normandie, France OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 1 Outline What is OpenQBMM? Example problems we aim at solving Population balance equations Models for turbulent mixing The structure of OpenQBMM Some example applications OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 2 What is OpenQBMM? An open-source implementation of quadrature-based moment methods into OpenFOAM® ◦ Population balance equations ◦ Turbulent reacting flows ◦ Polydisperse multiphase flows Where do I find OpenQBMM? ◦ Website: www.openqbmm.org ◦ On GitHub: https://github.com/OpenQBMM ◦ On Twitter: @OpenQBMM Who is behind OpenQBMM? ◦ Principal investigators: ◦ Alberto Passalacqua (lead) ◦ Rodney O. Fox ◦ Simanta Mitra ◦ Gas-liquid flows ◦ Gas-particle flows Project details ◦ Funded by the US National Science Foundation ◦ ACI – SI2 – SSE program ◦ Funding period ◦ Oct. 1st, 2014 – Sep. 30th 2017 ◦ Post-docs: Xiaofei Hu ◦ Students: Ehsan Madadi, Jeffrey Heylmun, David Williams ◦ External contributors: Frederique Laurent (UP Saclay), James Guthrie (U. of Strathclyde), Matteo Icardi (U. of Warwick), Daniele Marchisio and Dongyue Li (Politecnico di Torino) OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 3 Example problems Chemical reactors ◦ Mixing ◦ Reaction ◦ Particle formation and evolution ◦ Precipitation and nucleation ◦ Aggregation and breakup ◦ Growth Soot formation ◦ Flames ◦ Engines Gas-liquid flows ◦ Stirred tanks ◦ Bubble columns Mixing and reaction in multi-inlet vortex reactor (drug production) Gas-particle flows ◦ Fluidized beds ◦ Risers Gas-liquid-solid flow in stirred tank OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 4 Population balance equation Univariate population balance 𝜕𝑛(𝜉, 𝐱, 𝑡) + 𝛻𝐱 ⋅ 𝑛 𝜉, 𝐱, 𝑡 𝐔 − 𝛻𝐱 ⋅ Γ𝛻𝐱 𝑛 𝜉, 𝐱, 𝑡 𝜕𝑡 Advection in physical space Diffusion in physical space + 𝛻𝜉 ⋅ G 𝜉 𝑛 𝜉, 𝐱, 𝑡 Advection in space of internal coordinate ഥ 𝑎 𝜉, 𝐱, 𝑡 +𝐵ത 𝑏 𝜉, 𝐱, 𝑡 − 𝐷 ഥ 𝑏 𝜉, 𝐱, 𝑡 + 𝑁 𝐱, 𝑡 = 𝐵ത 𝑎 𝜉, 𝐱, 𝑡 − 𝐷 Birth and death due to aggregation Birth and death due to breakup Nucleation Number density function (NDF) - 𝑛(𝜉, 𝐱, 𝑡) ◦ 𝜉: internal coordinate (size, length, surface area, …) ◦ 𝜉 ∈ ℜ+ = [0, +∞[ ◦ 𝐱: position vector in physical space ◦ t: time OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 5 Turbulent mixing Turbulent mixing 𝜕𝑓(𝜉, 𝐱, 𝑡) + 𝛻𝐱 ⋅ 𝑓 𝜉, 𝐱, 𝑡 𝐔 − 𝛻𝐱 ⋅ Γ𝛻𝐱 𝑓 𝜉, 𝐱, 𝑡 𝜕𝑡 Diffusion in physical space Advection in physical space = 𝑆(𝜉, 𝐱, 𝑡) Turbulent mixing model Probability distribution function of the mixture fraction (NDF) 𝑓(𝜉, 𝐱, 𝑡) ◦ 𝜉: mixture fraction ◦ 𝜉 ∈ [0,1] ◦ 𝐱: position vector in physical space ◦ t: time Turbulent mixing models ◦ Interaction by exchange with the mean (Villermaux and Devillon, 1972) ◦ 𝑆 𝜉, 𝑥, 𝑡 = −𝛻𝜉 ⋅ 𝜀𝜉 𝜉 ′2 𝜉 −𝜉 𝑓 ◦ Fokker-Planck (Fox, 2003) ◦ 𝑆 𝜉, 𝑥, 𝑡 = −𝛻𝜉 ⋅ 𝜀𝜉 𝜉 ′2 1 𝜉 − 𝜉 𝑓 + 2 𝛻𝜉2 𝜀|𝜉 𝑓 ◦ Scale similarity model (Fox, 2003): 𝜀𝜉 𝜉 ′2 𝜀 𝜅 = 𝑐𝜑 ; 𝑐𝜑 = 2 OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 6 The method of moments 𝜕𝑛(𝜉, 𝐱, 𝑡) + 𝛻𝐱 ⋅ 𝑛 𝜉, 𝐱, 𝑡 𝐔 − 𝛻𝐱 ⋅ Γ𝛻𝐱 𝑛 𝜉, 𝐱, 𝑡 𝜕𝑡 + 𝛻𝜉 ⋅ G 𝜉 𝑛 𝜉, 𝐱, 𝑡 = 𝑆(𝜉, 𝐱, 𝑡) 𝑚𝑘 = Ω 𝜉 𝑘 𝑓 𝜉, 𝐱, 𝑡 d𝜉 𝜕𝑚𝑘 + 𝛻𝐱 ⋅ 𝑚𝑘 𝐔 − 𝛻𝐱 ⋅ Γ𝛻𝑚𝑘 = 𝐴ҧ𝜉,𝑘 (𝐱, 𝑡) + 𝑆𝑘ҧ (𝐱, 𝑡) 𝜕𝑡 Closure problem ◦ Compute the source terms of the moment transport equations ◦ Several approaches: we use Gaussian quadrature to provide closures OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 7 The quadrature method of moments Approximate the distribution with a Moments are expressed in terms of sum of Dirac delta functions the quadrature approximation: (McGraw, 1997): 𝑁−1 N−1 𝑛 𝜉, 𝑥, 𝑡 = 𝑤𝑖 𝛿(𝜉 − 𝜉𝑖 ) 𝑚𝑘 = 𝑤𝑖 𝜉 𝑘 𝑖=0 𝑖=0 Properties Moments of the distribution ◦ N quadrature nodes ◦ 2N conserved moments ◦ Discrete representation of the distribution Moment inversion Quadrature weights 𝑤𝑖 and abscissae 𝜉𝑖 Problem ◦ How do we evaluate the distribution at arbitrary values of 𝜉? OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 8 The extended quadrature method of moments In QMOM, the distribution is “discrete”: ◦ Represented by a sum of few Dirac delta functions 𝑛 𝜉 In some problems (evaporation, condensation, …), we would like a smooth representation 𝑛 𝜉 𝜉 𝜉 At similar computational cost! OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 9 The extended quadrature method of moments Approximate the NDF with a weighted sum of non-negative kernel density functions (Yuan et al., 2012) N 𝑛 𝜉, 𝑥, 𝑡 = 𝑤𝑖 𝛿𝜎 (𝜉, 𝜉𝑖 ) 𝑖=1 Choose kernel density function based on: The parameter 𝜎 is shared among the kernel density functions ◦ Only one additional moment needs to be transported with respect to standard QMOM ◦ Only one non-linear equation has to be solved to find 𝜎, instead of a system of non-linear equations ◦ A suitable value of 𝜎 may not exist! ◦ Support (range of 𝜉) ◦ ℜ Gaussian ◦ ℜ+ Gamma, lognormal ◦ 0, 1 Beta ◦ Known recurrence relation for orthogonal polynomials to the kernel density: ◦ Used to determine Gaussian quadrature Remember that the final objective is to accurately integrate the source terms of the PDF evolution equation to find their moments OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 10 Code structure of OpenQBMM We leverage the common structure of the equation for code re-use. Both population balance and mixing problems can be recast in the form: 𝜕𝑛(𝜉, 𝐱, 𝑡) + 𝛻𝐱 ⋅ 𝑛 𝜉, 𝐱, 𝑡 𝐔 − 𝛻𝐱 ⋅ Γ𝛻𝐱 𝑛 𝜉, 𝐱, 𝑡 𝜕𝑡 Advection in physical space Realizable advection scheme Diffusion in physical space + 𝛻𝜉 ⋅ G 𝜉 𝑛 𝜉, 𝐱, 𝑡 Advection in the space of internal coordinate Traditional FVM implementation OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM = 𝑆 𝜉, 𝐱, 𝑡 Source terms Realizable ODE solver 11 Requirements Smooth integration into OpenFOAM® ◦ Same user experience ◦ Same pre- and post-processing features ◦ Re-uses the existing infrastructure as much as possible Must interface to ◦ Incompressible and compressible solvers ◦ Reaction module ◦ Single/multi-phase solvers The solution algorithm ◦ Must maximize the number of preserved moments in the transported moment set ◦ We transport 2N + 1 ◦ NDF may degenerate: QMOM ◦ Must ensure, verify and preserve the realizability of the moment set ◦ Must be sufficiently accurate for the target applications ◦ Order of schemes ◦ Realizable schemes for advection Run-time selection for ◦ Models ◦ Quadrature setup (number of moments, accuracy, …) OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 12 Moment inversion and realizability univariateMomentSet object ◦ A single moment vector ◦ Set of moments to be inverted in each computational cell or face ◦ Verifies a set of moments is realizable based on the support of the distribution ◦ Determines the maximum number of invertible moments ◦ Computes the quadrature weights and abscissae for the moment set ◦ Stores the quadrature formula ◦ Can compute moments from the quadrature ◦ Manages the cases of degenerate moment sets ◦ Different for different quadrature formulae ◦ Even for Gauss ◦ Odd for Gauss-Radau The core of basic quadrature operations and moment realizability check OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 13 Realizability of a moment set We say that a moment vector 𝐦 = 𝑚 0 , … , 𝑚𝑀 is realizable when a measure 𝜇(𝜉) exists on a support Ω, so that 𝑚𝑘 = න 𝜉 𝑘 𝑑𝜇 Ω 𝑃𝑚 𝜉 are the orthogonal polynomials to 𝜇(𝜉) over Ω, defined by the recurrence relation Realizability depends on the support of 𝜇 𝜉 . If Ω is ◦ The real line ◦ Realizability condition: 𝛽𝑘 ≻ 0 ◦ The positive real line ◦ Realizability condition: ◦ 𝜁𝑘 = 𝐻𝑘 𝐻𝑘−3 𝐻𝑘−2 𝐻𝑘−1 > 0, 𝑘 = 0, 1 … ◦ 𝐻𝑘 are the Hankel determinants ◦ The compact interval 0,1 ◦ Canonical moments: 𝑝𝑘 ∈]0,1[ ◦ 𝜁𝑘 = 𝑝𝑘 1 − 𝑝𝑘 , 𝑘 > 0 𝑃𝑘+1 𝜉 = 𝑥 − 𝛼𝑘 𝑃𝑘 𝜉 − 𝛽𝑘 𝑃𝑘−1 𝜉 , ◦ 𝑝0 = 0 𝑃0 = 1 𝑃−1 = 0 Performed by univariateMomentSet before 𝛼𝑘 = 𝜁2𝑘 + 𝜁2𝑘+1 , 𝛽𝑘 = 𝜁2𝑘−1 𝜁2𝑘 each inversion, transparently to the user OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 14 Extended moment inversion extendedMomentInversion ◦ Computes the secondary quadrature ◦ Acts on an individual moment vector ◦ Maintains independence from fields, as univariateMomentSet Performs the EQMOM reconstruction of the NDF Principle of the procedure: 1. Start from a guess for the parameter 𝜎 ◦ From realizability conditions on the first few moments 2. Compute quadrature weights and abscissae 3. Enforce the conservation of the last moment 𝑚2𝑁 4. Solve the non-linear equation obtained at point 3 to find sigma 5. Repeat from 1 until convergence 6. If 𝜎 is not found, minimize moment error OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 15 An overview of other key classes univariateQuadratureApproximation PDFTransportModel univariatePDFTransportModel Stores moment and quadrature fields, and takes care of initialization and quadrature update populationBalanceModel mixingModel univariatePopulationBalanceModel turbulentMixing OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 16 Verification and validation - PBE We validated the implementation against rigorous solution of population balance equations (Vanni, 2000) ◦ Zero-dimensional problem ◦ Aggregation and breakup with different kernels ◦ EQMOM with log-normal kernel density function Simplified form of the PBE 𝜕𝑛(𝜉,𝐱,𝑡) 𝜕𝑡 ഥ 𝑎 𝜉, 𝐱, 𝑡 = 𝐵ത 𝑎 𝜉, 𝐱, 𝑡 − 𝐷 ഥ 𝑏 𝜉, 𝐱, 𝑡 +𝐵ത 𝑏 𝜉, 𝐱, 𝑡 − 𝐷 Five cases ◦ Different aggregation and breakup kernels ◦ Different daughter distribution functions Comparison to ◦ Rigorous solution of Vanni (2000). ◦ EQMOM solution in MATLAB® OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 17 PBE kernels Aggregation kernels ◦ Constant ◦ Constant ◦ 𝛽 𝜉, 𝜉 ′ = 1 ◦ 𝑎 𝜉 =1 ◦ Sum ◦ Power law ◦ 𝑎 𝜉 = 𝜉𝛼 ◦ 𝛽 𝜉, 𝜉 ′ = 𝜉 3 + 𝜉 ′3 ◦ Hydrodynamic: ◦ 𝛽 𝜉, 𝜉 ′ = 𝜉 + 𝜉 ′ Breakup kernels ◦ Exponential 3 ◦ 𝑎 𝜉 = 𝑒 𝛿𝜉 ◦ Differential force ◦ 𝛽 𝜉, 𝜉 ′ = 𝜉 + 𝜉 ′ 2 |𝜉 + 𝜉 ′ | 3 Daughter distribution functions ◦ Symmetric fragmentation ◦ Uniform ◦ Mass ratio 1 to 4 OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 18 Results – Case 1 Submodels Initial moments: ◦ Constant aggregation and breakup ◦ Symmetric fragmentation ◦ 𝑚𝑘 = 1, 𝑘 = 0, … , 6 Zero-order moment OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 𝑑43 = 𝑚4 /𝑚3 19 Results – Case 2 Submodels ◦ Sum aggregation ◦ Power law breakup ◦ Symmetric fragmentation Initial moments: ◦ 𝑚𝑘 = 1, 𝑘 = 0, … , 6 Zero-order moment OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 𝑑43 = 𝑚4 /𝑚3 20 Results – Case 3 Submodels ◦ Hydrodynamic aggregation ◦ Exponential breakup ◦ Symmetric fragmentation Initial moments: ◦ 𝑚𝑘 = 1, 𝑘 = 0, … , 6 Zero-order moment OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 𝑑43 = 𝑚4 /𝑚3 21 Results – Case 4 Submodels Initial moments: ◦ Differential force aggregation ◦ Power law breakup ◦ Uniform fragmentation ◦ 𝑚1 = 1, 𝑚2 = 1.13, 𝑚2 = 1.294 ◦ 𝑚3 = 1.5, 𝑚4 = 1.76, 𝑚5 = 2.09237 ◦ 𝑚6 = 2.51269 Zero-order moment OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 𝑑43 = 𝑚4 /𝑚3 22 Results – Case 5 Submodels ◦ Hydrodynamic aggregation ◦ Exponential breakup ◦ 1 to 4 mass ratio fragmentation Initial moments: ◦ 𝑚𝑘 = 1, 𝑘 = 0, … , 6 Zero-order moment OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 𝑑43 = 𝑚4 /𝑚3 23 Multi inlet vortex reactor (MIVR) The multi-inlet vortex reactor (MIVR) is used for flash nanoprecipitation to manufacture functional nanoparticles: ◦ Electronics: provided a powerful path to developing small and powerful electronic components. ◦ Drug Delivery: nano-sized micelles accumulate in tumors via the enhanced permeability and retention effect. ◦ Nanoparticles doped on hydrogels for cancer treatment The MIVR achieves fast mixing by inducing turbulent swirling flow from four inlet streams. We study mixing inside the MIVR: ◦ Turbulent (RANS) flow field solution ◦ Mixing model with OpenQBMM J. C. Cheng and R. O. Fox, "Kinetic Modeling of Nanoprecipitation using CFD Coupled with a Population Balance," Industrial & Engineering Chemistry Research, vol. 49, pp. 10651-10662, Nov 3 2010 PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 24 Reference Multi Inlet Vortex Reactor All dimensions are in mm. Y. Shi, J. C. Cheng, R. O. Fox and M. G. Olsen, "Measurements of turbulence in a microscale multi-inlet vortex nanoprecipitation reactor" Journal of Micromechanics and Microengineering, vol. 23, 075005(10pp), 2013. PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 25 Simulation workflow A sequential workflow is used to investigate the MIVR: Flow field Fields of U, k, epsilon Mixing Mixture fraction field Population balance Particle size distribution Assumptions ◦ Particles have small Stokes number and do not affect the fluid motion ◦ The flash nano-precipitation does not significantly affect mixing PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 26 Flow field ◦ Incompressible steady-state RANS simulation (simpleFoam solver in OpenFOAM®) ◦ MIVR generates swirling flow, with higher turbulent intensity in the center of the mixing chamber ◦ The turbulent dissipation rate 𝜀 is higher in the center of the chamber: more intense mixing Steady state solution of the velocity (left), turbulent kinetic (middle) and turbulent dissipation (right) of the flow. PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 27 Reactive mixing modeling We consider the evolution equation for the joint composition PDF 𝑓(𝝓, 𝐱, 𝑡): rate of change ฐ 𝜕𝑓 𝜙 𝜕𝑡 𝜕 =− 𝜕𝜓𝑖 macromixing + 𝜀𝜙 𝜙 ′2 mesomixing 𝜕𝑓𝜙 𝜕𝑓𝜙 𝜕 𝑈𝑖 − Γ 𝜕𝑥𝑖 𝜕𝑥𝑖 𝑇 𝜕𝑥𝑖 𝜙𝑖 − 𝜓𝑖 + 𝑆𝑖 𝜓 𝑓𝜙 Closed by gradient-diffusion model where ◦ ◦ ◦ ◦ ◦ ◦ 𝜙 is the passive scalar 𝜙 ′2 is the scalar variance 𝜀𝜙 is the dissipation factor 𝑈𝑖 is the mean velocity Γ𝑇 is the turbulent diffusivity 𝑡 is time Interaction by Exchange with the Mean mixing model (Fox, 2003) PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY Source term due to reaction 29 Chemical kinetics We consider a simple case of two competitive consecutive reactions to study mixing efficiency in the MIVR Fast reaction 𝑘1 𝐴 + 𝐵՜𝑅 Slower reaction 𝑘2 𝐵 + 𝑅՜𝑆 where ◦ 𝑅 is the desired product ◦ 𝐵 is the byproduct ◦ 𝑘1 𝑘2 = 𝑂(103 ) Rewrite PDF in terms of ◦ Mixture fraction 𝜉 ◦ Reaction progress variables 𝑌1 , 𝑌2 𝑓(𝛟, 𝐱, 𝑡) ՜ 𝑓(𝜉, 𝑌1 , 𝑌2 , 𝐱, 𝑡) 𝑐𝐴0 𝜉𝑠𝑡 = 𝑐𝐴0 + 𝑐𝐵0 𝑐𝐴 = 𝑐𝐴0 1 − 𝜉 − 1 − 𝜉st 𝑌1 𝑐𝐵 = 𝑐𝐵0 𝜉 − 𝜉st 𝑌1 + 𝑌2 𝑐𝑅 = 𝑐𝐵0 𝜉st 𝑌1 − 𝑌2 𝑐𝑆 = 𝑐𝐵0 𝜉st 𝑌2 PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 30 Conditional quadrature method of moments We rewrite the joint PDF in terms of conditional PDFs: 𝑓 𝜉, 𝑌1 , 𝑌2 = 𝑓 𝑌2 𝜉, 𝑌1 𝑓 𝜉, 𝑌1 = 𝑓 𝑌2 𝜉, 𝑌1 𝑓 𝑌1 𝜉 𝑓 𝜉 we then consider the conditional moments 𝑗 𝑌1 𝑗 𝜉 ≝ න 𝑌1 𝑓 𝑌1 |𝜉 d𝑌1 𝑌2𝑘 (𝜉, 𝑌1 ) ≝ න 𝑌2𝑘 𝑓 𝑌2 |𝜉, 𝑌1 d𝑌2 and we represent the moments of PDF as 𝑗 𝑀𝑛𝑗𝑘 𝐱, 𝑡 = ම 𝜉 𝑛 𝑌1 𝑌2𝑘 𝑓 𝜉, 𝑌1 , 𝑌2 ; 𝐱, 𝑡 d𝜉 d𝑌1 d𝑌2 PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 31 Conditional quadrature method of moments We represent the PDF as (assuming one node for 𝑌1 and 𝑌2 ): N 𝑓 𝜉, 𝑌1 , 𝑌2 ; 𝐱, 𝑡 = 𝑤𝛼 𝛿 𝜉 − 𝜉𝛼 𝛿 𝑌1 − 𝑌1𝛼 𝛿 𝑌2 − 𝑌2𝛼 𝛼=1 and we represent the moments of the joint PDF as: N 𝑗 𝑀𝑛𝑗𝑘 𝐱, 𝑡 = 𝑤𝛼 𝜉𝛼𝑛 𝑌1𝛼 𝑌2𝑘𝛼 𝛼=1 We consider: ◦ Two quadrature nodes for the 𝜉 direction ◦ One quadrature node for the 𝑌1 direction ◦ One quadrature node for the 𝑌2 direction 𝑀000 , 𝑀100 , 𝑀200 , 𝑀300 𝑀010 , 𝑀110 𝑀001 , 𝑀011 PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 32 Moment transport equations in CQMOM 𝜕𝑀𝑛𝑗𝑘 + 𝛻 ⋅ 𝐔𝑀𝑛𝑗𝑘 − 𝛻 ⋅ Γ𝑇 𝛻𝑀𝑛𝑗𝑘 𝜕𝑡 𝑛𝜀𝜉 = ′2 𝑀𝑛−1𝑗𝑘 𝑀100 − 𝑀𝑛𝑗𝑘 𝜉 N Scale similarity 𝜀𝜉 𝜀 = 𝐶 𝜉 𝜉 ′2 𝑘 N 𝑗−1 𝑗 + 𝑤𝛼 𝑅1𝛼 𝑗𝜉𝛼𝑛 𝑌1𝛼 𝑌2𝑘𝛼 + 𝑤𝛼 𝑅2𝛼 𝑘𝜉𝛼𝑛 𝑌1𝛼 𝑌2𝑘−1 𝛼 𝛼=1 𝛼=1 1 − 𝜉𝛼 − 𝑌1𝛼 1 − 𝜉𝑠𝑡 𝜉𝛼 − 𝑌1𝛼 − 𝑌2𝛼 𝜉𝑠𝑡 𝑅2𝛼 𝜉𝛼 , 𝑌1𝛼 , 𝑌2𝛼 = 𝜉𝑠𝑡 𝑘2 𝑐𝐵0 𝑌1𝛼 − 𝑌2𝛼 𝜉𝛼 − 𝑌1𝛼 − 𝑌2𝛼 𝜉𝑠𝑡 𝑅1𝛼 𝜉𝛼 , 𝑌1𝛼 , 𝑌2𝛼 = 𝜉𝑠𝑡 𝑘1 𝑐𝐵0 PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 33 Mixing - Simulation setup Two opposing inlet streams for solvent 𝑀000 = 1 𝑀100 = 1 𝑀200 = 1 𝑀300 = 1 𝑀010 = 0 𝑀110 = 1 𝑀001 = 0 𝑀101 = 0 Two opposing inlet streams for nonsolvent Mixing and reaction fields are obtained using OpenQBMM (www.openqbmm.org) PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 𝑀000 = 1 𝑀100 = 0 𝑀200 = 0 𝑀300 = 0 𝑀010 = 0 𝑀110 = 1 𝑀001 = 0 𝑀101 = 0 34 Mixture fraction and concentrations ◦ ◦ ◦ ◦ Mixture fraction measures the mixing Two streams have 𝜉 = 0 and two other have 𝜉 = 1 Through micromixing, the mixture fraction relaxes to value 𝜉 = 0.5 In FNP processes 𝜉 is key, since the solvent and non-solvent should reach a specific ratio to precipitate The mixture fraction (left), concentration A (middle), and concentration B (right). PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 35 Mixture fraction and concentrations The mixture fraction (left), concentration A (middle), and concentration B (right). PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 36 Predicted mixture fraction ◦ The mean mixture fraction is defined as: 𝜉 = 𝑀1 Mean mixture fraction inside the mixing chamber shown in three planes, middle, top quarter (q2t) and bottom quarter (q2b). PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 37 Predicted variation of mixture fraction ◦ The variance of the mixture fraction an be written as: 𝜉 ′2 = 𝑀2 − 𝑀12 ◦ 𝜉 ′2 decays quickly after inlet streams enter the reactor ◦ Variance is higher in the quarter top and quarter bottom planes compare to middle plate inside the mixing chamber ◦ Higher variance is obtained at middle plate near to inlet streams ◦ Since the flow is turbulent, the variation of mixture fraction is not big Variance inside the mixing chamber shown at three planes, middle, top quarter (q2t) and bottom quarter (q2b). PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 38 Mixing with competitive consecutive reaction ◦ Simulations show that species B is consumed mostly inside the mixing chamber ◦ The main reaction happens between A and B since the rate of their reaction is 𝑂 1000 higher than the secondary reaction Concentration of 𝐶𝐴 , 𝐶𝐵 for competitive consecutive reaction (Middle plane displayed for the chamber). PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 39 Mixing with competitive consecutive reaction ◦ The main product R is mostly produced inside the chamber and moves toward the exit ◦ The byproduct S is produced due to the presence of R and B, when A is absent (mixing limitation) The concentration of product and bi-product for competitive consecutive reaction (Middle plane displayed for the chamber) PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 40 Mixing with competitive consecutive reaction Concentration of 𝐶𝐴 , 𝐶𝐵 for competitive consecutive reaction (Middle plane displayed for the chamber). PASSALACQUA ET AL. - ICMF 2016 - IOWA STATE UNIVERSITY 41 Next steps Realizable ODE solvers with adaptive stepping ◦ Implemented ◦ Being validated Applications to ◦ Bubbly flows ◦ Poly-celerity (size-conditioned velocity) ◦ Velocity + size distribution High-order realizable advection schemes ◦ Gas-particle flows ◦ Anisotropic Gaussian ◦ Infrastructure implemented ◦ Schemes coming “soon” Multi-dimensional inversion algorithm ◦ Velocity distributions ◦ Joint distributions OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 44 What if I want to contribute? You can contact us: Pull requests on GitHub ◦ E-mail: [email protected] ◦ IRC: Freenode, #openqbmm ◦ Twitter: @openqbmm Types of contribution ◦ ◦ ◦ ◦ Code Documentation Test cases Bug reports ◦ Stable releases are made using GitHub tag/release mechanism ◦ Follow the OpenFOAM coding style guide ◦ Make sure things work ◦ Very simple approach ◦ No build = no merge ◦ Break master = no merge ◦ Regression in tests and validation = no merge (well… it happened ) Two ways of contributing code: ◦ A contribution repository ◦ ◦ ◦ ◦ ◦ Must be against master (which is our dev) Large code contributions Contributions of specific interest Example applications Prototypes ◦ Provide a test-case demonstrating the functionality ◦ Put references and document ◦ Be ready to maintain your contribution, if it’s large ◦ Unused to date ◦ Pull requests seem to be the favorite way ◦ Contributors don’t feel their work will be left to itself OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 45 Acknowledgments The support of the US National Science Foundation under the SI2 – SSE award NSF – ACI 1440443 is gratefully acknowledged. The precious support of Henry Weller (CFD Direct/OpenFOAM Foundation) in the design of the code structure and in the development of the implementation strategy is gratefully acknowledged. The following contributors are deeply thanked: ◦ Frederique Laurent (UP Saclay) ◦ Moment realizability, high-order schemes ◦ James Guthrie (University of Strathclyde) ◦ Patches to allow build vs. OpenFOAM 3.x ◦ Matteo Icardi (U. of Warwick) ◦ Growth models for PBE ◦ Daniele Marchisio and Dongyue Li (Politcnico di Torino) ◦ PBE kernels for liquid-liquid systems OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 46 Disclaimer This presentation and its content is not approved or endorsed by OpenCFD Limited (ESI Group), the producer of the OpenFOAM software and owner of the OPENFOAM® and OpenCFD® trade marks. CoMFRE, Iowa State University and the speakers are not associated to OpenCFD Limited and ESI. OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 47 Thank you! . OPENQBMM - JOURNÉE FRANÇAISE DES UTILISATEURS DE OPENFOAM 48