“Large Eddy Simulations of Ethanol Spray Combustion”. Flavio Galeazzo – EP/USP.

“Large Eddy Simulations of Ethanol Spray Combustion”. Flavio Galeazzo – EP/USP.

• 1.
  Laboratory of Thermaland Environmental Engineering Combustion Research Center Large Eddy Simulations of Ethanol Spray Combustion Dr.-Ing. Flavio Cesar Cunha Galeazzo, [email protected] Prof. Dr.-Ing. Guenther Carlos Krieger Filho 3. Workshop de High Perfomance Computing 17. April 2015
• 2.
  Laboratory of Thermaland Environmental Engineering Combustion Research Center Experimental setup  PhD of Newton K. Fukumasu (2010 - 2014)  Open burner derived from an automotive fuel injector and a swirler  Laser diagnostics (PDI, LIF e PIV) 3WHPC-LCCA Flavio Galeazzo 2
• 3.
  Laboratory of Thermaland Environmental Engineering Combustion Research Center 100 Hz 250 Hz 400 Hz Case Injection frequency Duty cycle Combustion pattern ESF100 100 Hz 29% Long anchored flame ESF250 250 Hz 43% Short anchored flame ESF400 400 Hz 59% Lifted flame Boundary conditions Air flow 20 m3 / h Ethanol flow 2 L / h 3WHPC-LCCA Flavio Galeazzo 3
• 4.
  Numerical setup  OpenFOAM2.1.1  LES (Large Eddy Simulation)  Smagorinsky subgrid scale model (SSG)  Spray  Lagrangian particle tracking  Combustion model  Simple “mixed is burned” approach Laboratory of Thermal and Environmental Engineering Combustion Research Center 3WHPC-LCCA Flavio Galeazzo 4
• 5.
  Laboratory of Thermaland Environmental Engineering Combustion Research Center Grid 1 Grid 2 Number of elements 600 000 1.6 million Grid sensitivity  Hexahedral elements  Two grids with different refinement  Cold flow  Better agreement with refined grid  Grid 2 was used for the LES simulations 3WHPC-LCCA Flavio Galeazzo 5
• 6.
  • Cold flow •Mean axial velocity UX at the centerline • Experiments • RANS simulation using Reynolds-stress turbulence model • LES using Smagorinky SSG • The negative velocities indicate the recirculation zone created by the swirl flow • Much better agreement with LES Laboratory of Thermal and Environmental Engineering Combustion Research Center 3WHPC-LCCA Flavio Galeazzo 6
• 7.
  • Combustion • Axialvelocity(m/s) • Temperature (K) Laboratory of Thermal and Environmental Engineering Combustion Research Center 7
• 8.
  100 Hz 250Hz 400 Hz • Experiment chemiluminescence • Mean temperature (K) Laboratory of Thermal and Environmental Engineering Combustion Research Center 8
• 9.
  Laboratory of Thermaland Environmental Engineering Combustion Research Center Computational resources  BlueGene/P – USP/Rice  42 racks, 128 CPUs, 512 computing cores each  4 GB per CPU  Proprietary interconnect  SGI Altix cluster – LETE  12 nodes, dual CPU, 8 computing cores each  24 GB per node / 12 GB per CPU  Infiniband interconnect Problems using BG/P  Compilation of OpenFOAM in the BG/P  Need GCC compiler – can’t use native IBM XL compiler  Need shared libraries – complicated to setup in the BG/P  Run the OpenFOAM simulation in the BG/P  Need aprox. 2.5 GB per grid partition  Can’t use all computing cores of each CPU 3WHPC-LCCA Flavio Galeazzo 9
• 10.
  Laboratory of Thermaland Environmental Engineering Combustion Research Center Performance comparison  BlueGene/P – USP/Rice  42 racks, 128 CPUs, 512 computing cores each  4 GB per CPU  Proprietary interconnect  SGI Altix cluster – LETE  12 nodes, dual CPU, 8 computing cores each  24 GB per node / 12 GB per CPU  Infiniband interconnect SwirlFlameLETE/1/H1/les_onlyFlow – cold flow Grid = 1.6 million elements Mode Number of partitions Simulation time (s) Computing time (s) Relative time Blue Gene/P SMP 128 0,002 517 1 LETE openmpi 24 0,002 450 0,87 SwirlFlameLETE/1/H1/EPS250/les_fireFoam - combustion Grid = 1.6 million elements Mode Number of partitions Simulation time (s) Computing time (s) Relative time Blue Gene/P SMP 128 0,002 8423 1 Blue Gene/P SMP 256 0,002 6077 0,72 Blue Gene/P SMP 384 0,002 5892 0,70 3WHPC-LCCA Flavio Galeazzo 10
• 11.
  Laboratory of Thermaland Environmental Engineering Combustion Research Center Conclusions  Better agreement of LES results with the experimental data  More refined computational grids and longer simulation time are required for LES  For more refined grids there is a need of more computational power  OpenFOAM 2.1 was successfully compiled in the BlueGene/P  The performance is not ideal  Old GCC compiler (version 4.3.2, from 2008)  Memory usage of OpenFOAM do not permit the use of all computing cores  Use of OpenMP is an alternative  Current development, up to now no speedup using Paralution OpenMP solver 11
• 12.
  Laboratory of Thermaland Environmental Engineering Combustion Research Center Thank you Dr.-Ing. Flavio Cesar Cunha Galeazzo, [email protected] Prof. Dr.-Ing. Guenther Carlos Krieger Filho 12