PARTICLES 2013 Fully Ced Multiphase Simulation of a Bottom-spray Wurster-Coater usg a hybrid CPU/GPU CFD/DEM Approach D. Jajcevic, A.C. Radeke, E. Siegmann, G. Neubauer, J.G. Khast K1 Competence Center - Initiated by the Federal Mistry of Economy, Family and Yth (BMWFJ) and the Federal Mistry of Transport, Innovation and Technology (BMVIT). Funded by FFG, Land Steiermark and Steirische Wirtschaftsförderung (SFG).
Outle Motivation & Introduction Hybrid approach Validation problem Simulations at larger scale Apication to a Wurster coatg device Conclusions & tlook RCPE Präsentationstitel 2
DEM CFD Cg liquids / air flow : CFD runs on multi-co CPUs solids / particles : DEM runs on sgle GPU XPS L Cg 3
Ced simulation CFD:DEM XPS L Cg 4
DEM CFD Cg Methodology (CFD@CPUs + DEM@GPU) @ PC CFD L- ACCI L-Code-Cg-Interface DEM XPS Data Mappg To DEM-grid Data Mappg To CFD-grid XPS L Cg 5
DEM CFD Cg Validation Validation Setup: Pseudo 2D spt fluidized bed Sgle, dble and trie possibilities Particle image velocimetry (PIV) Number of particles: S: 12.000 / D: 47.000 / T: 107.000 Sgle Dble Trie Srce: Numerical and experimental study on multie-spt fluidized beds, Buijtenen et al. XPS L Cg 6
DEM CFD Cg Validation Pseudo 2D Spt Fluidized Bed XPS L Cg 7
DEM CFD Cg Validation Pseudo 2D Sgle Spt Fluidized Bed XPS L Cg 8
DEM CFD Cg Validation Sgle Spt Fluidized Bed XPS L Cg 9
DEM CFD Cg Validation Sgle Spt Fluidized Bed XPS L Cg 10
DEM CFD Cg Validation Pseudo 2D Dble Spt Fluidized Bed XPS L Cg 11
DEM CFD Cg Dble Spt Fluidized Bed XPS L Cg 12
DEM CFD Cg Validation Trie 2D Dble Spt Fluidized Bed XPS L Cg 13
DEM CFD Cg Trie Spt Fluidized Bed XPS L Cg 14
DEM CFD Cg Trie Spt Fluidized Bed XPS L Cg 15
Large-Scale CFD-DEM Simulation The Trie bed time = 1s XPS L Cg 16
Large-scale attempts N = 6.800.000 = 0.38mm dt (CFD)= 1e-3 dt(dem)= 1e-5 XPS L Cg 17
XPS L Cg 18 Large-Scale CFD-DEM Simulation The Trie bed time = 1s XPS L C l g
Large-Scale CFD-DEM Simulation Intel Co i7 960 4x3.20GHz CPU Computer A sgle NVIDIA GeForce GTX 580 graphical card with 3GB of memory XPS L Cg 19
Wurster coater setup CFD mesh >200.000 cells 180 72 275 440 Porosity: 0.15 Porosity: 0.1 Porosity: 0.5 15 Bndary conditons Mass flow ramp: 0.04 0.05 kg/s XPS L Cg 20
Wurster coater: bottom spray Spray nozzle Geometric, spatial volume Spray rate: 20 ml/m Spray angle: 20deg Particles get coatg accordg to RTD Particles grow & case mass high efficiency (20ms@ 1e6 particles/time step) => -processg apication! v XPS L Cg 25.09.2013 - Seite 21
Wurster coater: bottom spray N = 1.000.000 = 1.106 mm ~1kg solids 0.02 kg coatg total Process time 70s. (last 10s: flow ramp down) dt (CFD)= 1e-3 dt(dem)= 1e-5 Wall clock time: 3s/day 8CPUs,1GPU (Fermi class) XPS L Cg 25.09.2013 - Seite 22
Wurster coater: bottom spray N = 1.000.000 = 1.106 mm ~1kg solids 0.02 kg coatg total Process time 70s. (last 10s: flow ramp down) dt (CFD)= 1e-3 dt(dem)= 1e-5 Wall clock time: 3s/day 8CPUs,1GPU (Fermi class) XPS L Cg 25.09.2013 - Seite 23
Wurster coater: sults Ced hybrid CFD/DEM simulation + spray N = 1.000.000 = 1.106 mm ~1kg solids 0.02 kg coatg total Process time 70s. (last 10s: flow ramp down) dt (CFD)= 1e-3 dt(dem)= 1e-5 Wall clock time: 3s/day 8CPUs,1GPU (Fermi class) XPS L Cg 25.09.2013 - Seite 24
Wurster coater: sults Ced hybrid CFD/DEM simulation + spray N = 1.000.000 = 1.106 mm ~1kg solids 0.02 kg coatg total Process time 70s. (last 10s: flow ramp down) dt (CFD)= 1e-3 dt(dem)= 1e-5 Wall clock time: 3s/day 8CPUs,1GPU (Fermi class) XPS L Cg 25.09.2013 - Seite 25
Wurster coater: sults Ced hybrid CFD/DEM simulation + spray N = 1.000.000 = 1.106 mm ~1kg solids 0.02 kg coatg total Process time 70s. (last 10s: flow ramp down) dt (CFD)= 1e-3 dt(dem)= 1e-5 Wall clock time: 3s/day 8CPUs,1GPU (Fermi class) XPS L Cg 25.09.2013 - Seite 26
Conclusions and Outlook Xtended Partice System (XPS) Achievments / conclusions Validated CFD/DEM cg Modelg of rotation/rollg can be neglegted for highly dynamic systems? (sphes) Solid/fluid teraction for comex geometries Spray modelg (coatg, impgnation) System solution N: 10.000 => 100.000 => 1.000.000 => 10.000.000 => 100.000.00 Efficiency: deskside simulations: workstation, multi-cpu + sgle GPU Outlook Heat transfer liquid phase solid phase <= spray modelg Agglomeration Validated polydisperse particles Non-spherical particles, true shape: tablets, crystals Multi-GPU: N > 100.000.000 particles XPS L Cg 25.09.2013 - Seite 27