Thermoacoustics in random fibrous materials

1
Thermoacoustics in random fibrous materials

• Seminar
• Carl Jensen
• Tuesday, March 25 2008

2
Outline

• Thermoacoustics
• Computational fluid dynamics
• High performance computing

3
Thermoacoustics

• Discovery and early designs such as Sondhauss
  tube (right) and Rijke tube
• Developed into more efficient designs
• Stacks
• Gas mixtures
• High pressure
• Traveling wave devices

4
Engine Cycle
Stack temperature gradient

• A conceptual parcel of gas in the stack moves
  back and forth in the acoustic wave
• The changing pressure causes the temperature of
  the parcel to vary with position in the acoustic
  cycle
• The parcel is warmer on the left, but cooler than
  the stack so it absorbs heat
• The parcel is cooler on the right, but warmer
  than the stack so it rejects heat

Gas parcel temperature
Temperature
Sound
Position
QH
QC
QH
QC
TPltTS
TPTS
TPgtTS
5
Stack types

• Parallel pore
• Ceramics
• Stainless steel plates
• Irregular materials
• Wools (Steel, glass, etc.)
• Foams
• RVC
• Aluminum

6
Porous media theory

• Material approximated as rigid framework of tubes
• Roh and Raspet extended thermoacoustic solution
  for propagation in a tube to capillary framework
  of porous media to create a thermoacoustic theory
  for porous media
• Empirical model based on measured parameters
• Tortuosity, q
• Thermal and viscous shape factors, nµ and n?
• Porosity, O

?
7
Computational fluid dynamics

• Based on kinetic theory
• Solves for particle distributions in discretized
  phase space
• Simple dynamics particles move across lattice
  links and collide

8
Collision models

• In reality, the collisions represented by O are
  very complicated
• Conservation laws and assumption of velocity
  independent collision time gives the BGK
  collision operator
• Same dynamics as Navier-Stokes equations for low
  Mach number with sound speed , and
  viscosity
• Single relaxation time means Pr1

9
Collision models

• Multiple relaxation time
• Same principle but different moments of the
  distribution are relaxed differently
• Sound speed, bulk/kinematic viscosity, and Pr are
  all adjustable parameters
• Enhanced stability

10
Hybrid thermal model

• Energy conserving LB hampered by spurious mode
  coupling
• Dodge by using athermal LB and finite difference
  for temperature
• Breaks kinetic nature of simulation but enhances
  stability

11
Validation

• First test is sound propagation in 2 dimensional
  pore
• Infinite parallel plates

2R
12
Analytical solution
13
Computational setup

• Temperature set to ambient at each wall
• No slip on top/bottom walls
• Driving wave at left
• Non-reflecting at right

T1, u0
p(t) T1
T1
T1, u0
14
ResultsF(?)
15
ResultsF(?T)
16
High Performance Computing

• CPU (Athlon X2 4800)
• 2 cores
• 9.6 Gflops
• 6.4 GB/s memory bandwidth
• 2 GB RAM
• GPU (GeForce 8800 GTX)
• 128 stream processors
• 345.6 Gflops
• 86.4 GB/s
• 768 MB RAM

Control
Arithmetic
Cache
17
GPU Programming

• Massive threading
• Up to 12,288 threads in flight at once
• Threads batched into blocks
• Each multiprocessor block runs one block of
  threads
• Many threads per block
• Many blocks per process

Block 0
Block 1
Reg.
Reg.
Reg.
Reg.


Thread 0
Thread 1
Thread 0
Thread 1

Shared Mem.
Shared Mem.
Main Memory
18
Results

• Compute time
• Matlab 5 hours
• CUDA 25 seconds
• Other GPGPU issues
• Constrained memory
• Single precision
• Complex programming

19
Supercomputer
Nodes
Host
Image from http//www.olympusmicro.com/micd/gall
eries/oblique/glasswool.html
20
Supercomputer

• Much larger memory
• Less strict synchronization
• More flexible programming
• Double precision
• Non-local job queues, remote debugging, etc.
• Lower overall throughput without using a lot of
  processors

21
Current Work

• Sound impulse over 3D sphere

22
Conclusions

• Hybrid thermal lattice Boltzmann method contains
  proper physics to simulate thermoacoustic
  phenomena
• A lot of increasingly accessible options for high
  performance computing