SlideShare ist ein Scribd-Unternehmen logo
1 von 14
Profiling fluid flow past an
immersed object with CUDA
An Independent Study By
Aritra Kumar Nath
Lattice Boltzmann Method
• For simulating fluid flow
• Based on the microscopic kinetic equation for the particle distribution
function.
• Fictive particles propagating and colliding over lattice mesh
• Per node Density & Velocity used to determine macroscopic properties of
fluid (pressure, fluid velocity, viscosity)
• Linear so less computation intensive
• Straightforward parallel implementation
Lattice Boltzmann Method
• Simulation domain divided into 2D lattice (D2Q9)
• Each node with packet distribution
• Packet distributions have associated velocity vectors
Packet Distribution
0 v 0 -v 0 v -v -v v
0 0 v 0 -v v v -v -v
{ vqi , i = 0 . . . 8} =
Lattice Boltzmann Method
• Finite differencing used to compute new packet distributions
• Values computed iteratively
• Packet distributions based updated based on values in the previous
time step
• The Lattice Boltzmann equation:
Simulation Driver Rules
• Collision
• Propagation
ρ= ∑pqi u= 1/ρ ∑pqi vqi
Boundary Conditions
• Full-Bounce Back
• Half-Bounce Back
Implementation on GPU
• Symmetric velocity vector orientation in every node
• Scope for Parallelism inherent
• All velocity vectors with same direction grouped together for
computation
• Three primary kernels used are for:
– Collision
– Propagation
– Calculating boundary conditions
• NVIDIA GEForce 310M used
Velocity Vectors Resolved
Division of the D2Q9 model, according to its velocity directions. (Image
courtesy: GPU-Based Flow Simulation with Complex Boundaries)
Flow Chart - LBM Computation
(Image Courtesy: GPU-Based Flow Simulation with Complex
Boundaries)
LBM Fluid Simulation - Graphic
Visualization
Studying Flow Profiles of Fluids
• The following simulation parameters are variable:
– Lattice dimensions
– Fluid Viscosity
– Maximum Velocity
• The following values for the microscopic fictive particles are computed:
– Nodal velocity (microscopic fictive particles)
– Density (rho)
• The values obtained will further influence macroscopic fluid properties
Studies Results & Observation
• Study 1:
Compare different iteration sequences same simulation parameter values:
Parameter Value Set:
– Simulation Parameter:
– Fluid Viscosity: 0.05
– Maximum Fluid Velocity: 0.01
• Study 2:
Compare the same iteration sequence with different simulation parameter values
without causing turbulence:
Parameter Value Set 1:
-- Fluid Viscosity: 0.8
-- Maximum Fluid Velocity: 0.1
Parameter Value Set 2:
-- Fluid Viscosity: 0.01
-- Maximum Fluid Velocity: 0.1
Studies Results & Observation
• Study 3:
Compare the same iteration sequence with different simulation parameter
values random turbulence:
Parameter Value Set:
– Simulation Parameter:
– Fluid Viscosity: 0.05
– Maximum Fluid Velocity: 0.01

Weitere ähnliche Inhalte

Ähnlich wie Profiling fluid flow past an immersed object with

Nonlinear hydrodynamic forces - manufacture and transducer selection
Nonlinear hydrodynamic forces - manufacture and transducer selectionNonlinear hydrodynamic forces - manufacture and transducer selection
Nonlinear hydrodynamic forces - manufacture and transducer selection
Sachindra Max
 
Graybill-Hardison Final Poster
Graybill-Hardison Final PosterGraybill-Hardison Final Poster
Graybill-Hardison Final Poster
ANDREW HARDISON
 

Ähnlich wie Profiling fluid flow past an immersed object with (20)

Fluid Simulation methods in Video Games
Fluid Simulation methods in Video GamesFluid Simulation methods in Video Games
Fluid Simulation methods in Video Games
 
Hedera - Dynamic Flow Scheduling for Data Center Networks, an Application of ...
Hedera - Dynamic Flow Scheduling for Data Center Networks, an Application of ...Hedera - Dynamic Flow Scheduling for Data Center Networks, an Application of ...
Hedera - Dynamic Flow Scheduling for Data Center Networks, an Application of ...
 
CFD evaluation of lime addition in AMD
CFD evaluation of lime addition in AMD CFD evaluation of lime addition in AMD
CFD evaluation of lime addition in AMD
 
DevoFlow - Scaling Flow Management for High-Performance Networks
DevoFlow - Scaling Flow Management for High-Performance NetworksDevoFlow - Scaling Flow Management for High-Performance Networks
DevoFlow - Scaling Flow Management for High-Performance Networks
 
General presentation EPANET tutoriasls one
General presentation EPANET tutoriasls oneGeneral presentation EPANET tutoriasls one
General presentation EPANET tutoriasls one
 
PREDIcT
PREDIcTPREDIcT
PREDIcT
 
Applications of Lattice Boltzmann Method in Dynamic Modelling of Fluid Flows
Applications of Lattice Boltzmann Method in Dynamic Modelling of Fluid FlowsApplications of Lattice Boltzmann Method in Dynamic Modelling of Fluid Flows
Applications of Lattice Boltzmann Method in Dynamic Modelling of Fluid Flows
 
FATTREE: A scalable Commodity Data Center Network Architecture
FATTREE: A scalable Commodity Data Center Network ArchitectureFATTREE: A scalable Commodity Data Center Network Architecture
FATTREE: A scalable Commodity Data Center Network Architecture
 
Development of computer program for fluid flow in pipes and open channel
Development of computer program for fluid flow in pipes and open channel  Development of computer program for fluid flow in pipes and open channel
Development of computer program for fluid flow in pipes and open channel
 
Engineering project non newtonian flow back step
Engineering project   non newtonian flow back stepEngineering project   non newtonian flow back step
Engineering project non newtonian flow back step
 
Nonlinear hydrodynamic forces - manufacture and transducer selection
Nonlinear hydrodynamic forces - manufacture and transducer selectionNonlinear hydrodynamic forces - manufacture and transducer selection
Nonlinear hydrodynamic forces - manufacture and transducer selection
 
Benchmarking Elastic Cloud Big Data Services under SLA Constraints
Benchmarking Elastic Cloud Big Data Services under SLA ConstraintsBenchmarking Elastic Cloud Big Data Services under SLA Constraints
Benchmarking Elastic Cloud Big Data Services under SLA Constraints
 
Graybill-Hardison Final Poster
Graybill-Hardison Final PosterGraybill-Hardison Final Poster
Graybill-Hardison Final Poster
 
Multipath Load Balancing for SDN Data Plane
Multipath Load Balancing for SDN Data Plane Multipath Load Balancing for SDN Data Plane
Multipath Load Balancing for SDN Data Plane
 
Imac10 Component
Imac10 ComponentImac10 Component
Imac10 Component
 
CFD_Lecture_1.pdf
CFD_Lecture_1.pdfCFD_Lecture_1.pdf
CFD_Lecture_1.pdf
 
Reservoir Simulation
Reservoir SimulationReservoir Simulation
Reservoir Simulation
 
10 sdn-vir-6up
10 sdn-vir-6up10 sdn-vir-6up
10 sdn-vir-6up
 
4 lockheed
4 lockheed4 lockheed
4 lockheed
 
Valiant Load Balancing and Traffic Oblivious Routing
Valiant Load Balancing and Traffic Oblivious RoutingValiant Load Balancing and Traffic Oblivious Routing
Valiant Load Balancing and Traffic Oblivious Routing
 

Profiling fluid flow past an immersed object with

  • 1. Profiling fluid flow past an immersed object with CUDA An Independent Study By Aritra Kumar Nath
  • 2. Lattice Boltzmann Method • For simulating fluid flow • Based on the microscopic kinetic equation for the particle distribution function. • Fictive particles propagating and colliding over lattice mesh • Per node Density & Velocity used to determine macroscopic properties of fluid (pressure, fluid velocity, viscosity) • Linear so less computation intensive • Straightforward parallel implementation
  • 3. Lattice Boltzmann Method • Simulation domain divided into 2D lattice (D2Q9) • Each node with packet distribution • Packet distributions have associated velocity vectors
  • 4. Packet Distribution 0 v 0 -v 0 v -v -v v 0 0 v 0 -v v v -v -v { vqi , i = 0 . . . 8} =
  • 5. Lattice Boltzmann Method • Finite differencing used to compute new packet distributions • Values computed iteratively • Packet distributions based updated based on values in the previous time step • The Lattice Boltzmann equation:
  • 6. Simulation Driver Rules • Collision • Propagation ρ= ∑pqi u= 1/ρ ∑pqi vqi
  • 7. Boundary Conditions • Full-Bounce Back • Half-Bounce Back
  • 8. Implementation on GPU • Symmetric velocity vector orientation in every node • Scope for Parallelism inherent • All velocity vectors with same direction grouped together for computation • Three primary kernels used are for: – Collision – Propagation – Calculating boundary conditions • NVIDIA GEForce 310M used
  • 9. Velocity Vectors Resolved Division of the D2Q9 model, according to its velocity directions. (Image courtesy: GPU-Based Flow Simulation with Complex Boundaries)
  • 10. Flow Chart - LBM Computation (Image Courtesy: GPU-Based Flow Simulation with Complex Boundaries)
  • 11. LBM Fluid Simulation - Graphic Visualization
  • 12. Studying Flow Profiles of Fluids • The following simulation parameters are variable: – Lattice dimensions – Fluid Viscosity – Maximum Velocity • The following values for the microscopic fictive particles are computed: – Nodal velocity (microscopic fictive particles) – Density (rho) • The values obtained will further influence macroscopic fluid properties
  • 13. Studies Results & Observation • Study 1: Compare different iteration sequences same simulation parameter values: Parameter Value Set: – Simulation Parameter: – Fluid Viscosity: 0.05 – Maximum Fluid Velocity: 0.01 • Study 2: Compare the same iteration sequence with different simulation parameter values without causing turbulence: Parameter Value Set 1: -- Fluid Viscosity: 0.8 -- Maximum Fluid Velocity: 0.1 Parameter Value Set 2: -- Fluid Viscosity: 0.01 -- Maximum Fluid Velocity: 0.1
  • 14. Studies Results & Observation • Study 3: Compare the same iteration sequence with different simulation parameter values random turbulence: Parameter Value Set: – Simulation Parameter: – Fluid Viscosity: 0.05 – Maximum Fluid Velocity: 0.01