Presentation on theme: "Politecnico di Milano sede Bovisa Ottobre, 2007"— Presentation transcript:
1Politecnico di Milano sede Bovisa 15 - 16 Ottobre, 2007 MESOSCOPIC NUMERICAL MODELING OF REACTIVE MIXTURE FLOWS IN SOLID OXIDE FUEL CELLS BY LATTICE BOLTZMANN METHOD AND HIGH PERFORMANCE PARALLEL COMPUTINGPietro Asinari, Romano Borchiellini, Michele CalìPolitecnico di Milano sede Bovisa Ottobre, 2007
2OutlineMesoscopic modeling of SOFC electrodes by Lattice Boltzmann Method (LBM)Mixture modeling: MRT Gross & Krook modelNumerical scheme: semi-implicit linearized backward Euler formulation (SILBE)LABORA CodeCluster facilities and scaling performancesReconstruction techniquesGas permeation and diffusion: direct numerical simulation of tortuosity
3Lattice Boltzmann Method – LBM Microscopic TheoryKinetic TheoryMacroscopic Theory (Continuum) and Thermodynamics (Equilibrium)Deterministic Newton’s LawStatistical MechanicsEuler EquationsNavier – Stokes EquationsMolecular DynamicsLiouville EquationHilbert and Chapman – Enskog Analysis (Singular Perturbation Analysis)Finite Moments Multiple Relaxation TimesLattice Boltzmann EquationLBMLattice Gas AutomataBoltzmann EquationDiscretized Distribution Functions (DDFs)
4Why Mesoscopic Modeling and LBM No linear system of algebraic equations must be solved no need of iterative procedures.Explicit time numerical process transient simulations can be naturally performed.No need for staggered grids unphysical solutions are automatically avoided.Additional local information are available the computed variables of a single cell are enough to estimate higher order derivatives.Complex topologies can be efficiently included the models are stable for quite rough meshes.
5Application to SOFC Electrodes Mesoscopic Modeling is a very powerful tool for SOFC technology becauseit allows one to go deeply in the reaction core for investigating fuel cell portions, which are actually not accessible by direct measurement (spatial distribution of the concentration polarization, local fluid flow,…).However the reliability of numerical results strongly depends onthe reliability of the microscopic structure used in the simulations,the reliability of the input parameters, particularly the transport coefficients effecting the reaction rate.
6Mixture ModelingSelf collisions involve particles of the same type while cross collisions involve particles of different type
7Parallel Algorithm Calculations and communications at the same time ! Collision Step (Internal Layer)Collision Step (Core)Streaming Step (Core)Moment Calculation Step (Core)Streaming Step (Internal Layer)Moment Calculation Step (Internal Layer)Non – Blocking Send (Internal Layer)Calculations and communications at the same time !Non – Blocking Receive (External Layer)
8LABORA POLITOThe LABORA code (Lattice Boltzmann for Raster Applications) was developed from scratch at “Politecnico di Torino” (Italy), for solving mainly the fluid flow of reactive mixtures in porous media.The project started in 2003 (now 10,000 lines in C++).Main code features are:fully three dimensional formulation (D3Q19 lattice);optimized memory storage;parallelization based on automatic and arbitrary domain decomposition (open source MPI package);different tuning strategies.
9HPC Facility: System X @ Virginia Tech System X is a supercomputer assembled by Virginia Tech faculty members, staff, and students in the summer of 2003, comprising 1,100 Apple PowerMac G5 computers. System X is currently running at Teraflops, (20.24 peak), and was last ranked #47 (November, 2006) in the TOP500 list of the world's most powerful supercomputers. At that time, it was still the most powerful system categorized by TOP500 as "self made" at any university.
10HPC (?) Facility: ClusterLinux @ POLITO “Politecnico di Torino” (Italy): ClusterLinux, scalable grid computing facility, currently 64 Pentium single processor nodes (64 CPUs, 2.8 GHz, 512 MB RAM, 40 GB HD), LAN 100 Megabit Ethernet, up to 102 CPUs.This facility is based on PC from student laboratories which are under-used during night and/or vacations.The main goal is to fruitfully use computational resources which are already available in order to maximize the investment outcome.
14The (near) future: EnerGRID project EnerGRID: design and development of a grid infrastructure for high performance computing in modeling energy networks based on widespread sources of heat and power generationOn-going collaborations with research groups of Computer Science Department at Stuttgart (GE) in the framework of the program HPC – Europa.
15Reconstruction Techniques Non-destructive X-ray computed micro-tomography is not enough for SOFC application, this resolution is not sufficient reconstructions from reliable 2D techniques, such as standard and back scanning electron microscopy (SEM/BSEM), is the only viable alternative.(1) granulometry law grain shapes are assumed;(2) multiple–point statistics neighboring information are processed for more reliable reconstruction.
16Multiple-point Statistics Multiple-point statistics were used, based on two-dimensional (2D) thin sections as training images, to generate 3D pore space representations (Okabe & Blunt, Journal of Petroleum Science & Engineering, 2005).A 3D image can be generated that preserves typical patterns of the void space seen in the thin sections.The use of multiple-point statistics predicts long-range connectivity of the structures better than granulometry law.Essentially the algorithm is based on three steps:Borrowing multiple-point statistics from training images,Pattern reproduction,Image processing-noise reduction and smoothing.
18Fluid Flow at the Bottom Hexahedral mesh 2563=16.7 MCell MDof for binary mixture (H2O/H2) in 3D porous medium.100,000 collisions.Wall clock time 57 hours with a 64 CPU cluster.Parallelization efficiency 85 % with non-optimized domain decomposition.
19Surface Averaged Quantities Surface averaged quantities must be introduced for comparing the mesoscopic fluid flow with the macroscopic measurements and user-level expectations.<Concentration><Mass Flux>
20Optimal Refinement: Fluid Flow In order to recover the desired accuracy (<3%), the finest computational mesh, i.e (refinement X8) must be considered. Unfortunately, this means to simulate a portion too small of the anode, which is not representative of the whole electrode.
21Optimal Refinement: Tortuosity Fortunately the tortuosity has a small dependence on the mesh resolution (<5%). It depends on the path of the considered species flowing in the porous medium and even very coarse meshes allow one to at least estimate the path of the species with acceptable accuracy. This means that larger physical domains can be simulated.
23ConclusionsDirect numerical simulation of tortuosity for SOFC application is promising for comparing the performances of different materials and highlighting the possible ways to improve themThe required mesh resolution for solving the fluid flow with regards to the tortuosity calculation is not too demandingThe simulation of the local electro-chemical reaction must be improved the ion and electron flows in the solid phases must be accurately solved tooDifferent sintering technologies can be compared
24Further Documentation P. Asinari, M.R. von Spakovsky, M. Calì, B.V. Kasula, “Direct numerical calculation of the kinematic tortuosity of reactive mixture flow in the anode layer of solid oxide fuel cells by the Lattice Boltzmann Method”, Journal of Power Sources, 170, pp , 2007.P. Asinari, “Semi-implicit-linearized Multiple-relaxation-time formulation of Lattice Boltzmann Schemes for Mixture Modeling”, Physical Review E, 73, , 2006.P. Asinari, “Viscous coupling based Lattice Boltzmann model for binary mixtures”, Physics of Fluids, 17, , 2005.P. Asinari, “Asymptotic analysis of multiple-relaxation-time lattice Boltzmann schemes for mixture modeling”, Computers and Mathematics with Applications, 2007 (in press).