Title: SHARP TH Simulation Effort
1SHARP TH Simulation Effort
- Paul Fischer
- Mathematics and Computer Science Division
- Argonne National Laboratory
- J. Lottes, A. Siegel, S. Thomas, C. Verma
Work sponsored by U.S. Department of Energy
Office of Nuclear Energy, Science Technology
2Outline
- Long term objectives / Overview
- 2007 Accomplishments
- Code Development
- Nek5000
- Low-Dimensional Code
- Simulations
- DNS
- LES
- RANS
- Low-Dimensional Models
3Long Term Objectives
- Exploit DOEs Petascale computing facilities ( P
gt 100,000 processors) and state of the art
simulation tools to improve TH predictive
capabilities at the design level - temperature distributions, under a broad range of
loading conditions - pressure drops and flow resistance through the
system - Provide validated predictive capabilities based
on a fidelity hierarchy - DNS ? LES ? RANS ? low-dimensional modeling
- enable investigation of new designs (e.g.,
outside validated range of current codes) - Coupled simulation capability
- spanning a range of scales,
- integrated with other physics (e.g., neutronics,
structural mechanics, ) - integrated with other codes
- Allow simultaneous coupling of say, LES in some
areas low-dimensional models elsewhere
neutronics - Ultimately, simulate full reactor
4Petascale Computing at DOE
- Argonne
- 100 Tflops IBM BG/P Nov. 07
- 32,000 processors, 850 MHz
- 500 Tflops IBM BG/P Aug. 08
- 140,000 processors, 850 MHz
- Oak Ridge
- 100 Tflops Cray XT4 Now
- 23,000 processors, 2.6 GHz
- 1 Petaflops Cray XT4 Late 08
- 200,000 processors, 2.6 GHz
- Its time to be thinknig about Exaflops
5Overview, SHARP Thermal-Hydraulics Plan
- Develop design analysis capabilities that span
desktop ? Petaflop - Design rapid turn-around reactor scale
- Analysis detailed simulations providing
information previously accessible only
through experiment. - Input to design codes
- Understanding of basic phenomena (e.g., thermal
striping) - Design validation
- Large scale multiphysics simulations at reactor
scale (out years, PFLOPS) - Reduce of experiments, not replace.
6Targeted Range of Simulation Capabilities
- Target Platform Model
- Desktop Subchannel Modeling
- Conservative
- low-resolution
- DG codes
- RANS
- LES
- Petaflops DNS
7Targeted Range of Simulation Capabilities
- Target Platform Model Current Capabilities /
Efforts - Desktop Subchannel SAS
(T. Fanning) - Modeling
- Conservative Starting w/ Nek (S.
Thomas) - low-resolution
- DG codes
- RANS Star CD (D.
Pointer) - LES Nek (F., D. Sheeler,A. Siegel)
- Petaflops DNS Prism (C. Pantano-UIUC)
8Approaches to TH analysis of subassemblies
- DNS direct numerical simulation of all scales
parameter-free - LES large eddy simulation dissipation
parameter-free -
- RANS Reynolds-averaged Navier-Stokes
tuning required - Subchannel modeling empirical input
- 400 x 200 subchannels in the core
- Subchannel analysis will continue to be used for
reactor design. - RANS will inform design process.
- LES can help to validate / inform RANS and
subchannel analysis.
impractical 107 p. per channel 105 p. per
channel steady state 100 p. per
channel steady state
9Current TH Capabilities within ANL SHARP team
- Nek5000 ANL code for fluids / heat
transfer (Fischer, Lottes, Thomas) - High-order accuracy
- Scales to P gt 10,000 processors
- State of the art multilevel solvers
- 2 decades of development / verification /
validation - Supports conjugate heat transfer, variable
properties, MHD, ALE, URANS - Extensive reactor TH experience (Fanning,
Pointer, Yang) - RANS modeling Star CD
- Subchannel codes (SAS)
10Validation Nek5000 ComputationsRod bundle
flow at Re30,000 w/ C. Tzanos (ANL)
- Low-speed streaks in a rod bundle
-
-
- Log-law profiles
11 Rod Bundle Validation Nek5000 Comparison w/
Experimental Data (F. Tzanos,
05)
12Outline
- Long term objectives / Overview
- 2007 Accomplishments
- Code Development
- Nek5000
- Low-Dimensional Code
- Simulations
- DNS
- LES
- RANS
- Low-Dimensional Models
13Code Development Efforts 07
- Nek5000
- Improved parallel coarse-grid solver for
multigrid solution of pressure - work in progress low-memory but not scaling
as expected - Working with European collaborators on low-Mach
number formulation for non-Boussinesq thermal
expansion effects - New mesh reading capabilities for large element
counts and non-native mesh generators - Coupled to VisIt (D. Bremer, LLNL)
- Low-Dimensional Modeling
- Surrogate mass-conserving velocity fields derived
from LES/RANS used for thermal transport in
larger systems (i.e., full-length fuel
assemblies) - Developing a conservative super-parametric
formulation that will be volume preserving
(non-faceted geometries) with few
degrees-of-freedom
14Simulations 07
- First Simulation Study wire-wrapped fuel pins
- DNS
- LES
- RANS
- Low-Dimensional Models
15First TH Study analysis of wire wrapped pins in
subassembly
- Starting point for TH simulation development and
deployment - Uniformity of temperature controls peak power
output - A better understanding of flow distribution
(interior, edge, corner) can lead to improved
subchannel models. - Wire wrap geometry is relatively complex
16Objectives for LES / RANS
From Bogoslovskaya et al.
- Potential surrogate for benchtop experiments
- Provide geometry-specific input to subchannel
codes - Consider sequence of 7, 19, , 217 pins to
provide a detailed picture of the hydrodynamics
and heat transfer in a single assembly.
17Approaches to TH analysis of subassemblies
- DNS direct numerical simulation of all scales
parameter-free - LES large eddy simulation dissipation
parameter-free -
- RANS Reynolds-averaged Navier-Stokes
tuning required - Subchannel modeling empirical input
- 400 x 200 subchannels in the core
- Subchannel analysis will continue to be used for
reactor design. - RANS will inform design process.
- LES can help to validate / inform RANS and
subchannel analysis.
impractical 107 p. per channel 105 p. per
channel steady state 100 p. per
channel steady state
18Direct Simulation of Wire in Turbulent Channel
with Crossflow
Carlos Pantano UIUC
- Channel-wire flow model
- Model turbulent flow around wires in reactor core
- Target large DNS with accurate spatio-temporal
resolution - Derive turbulence statistics for validation of
RANS/LES models - Preliminary results (spectral element code)
- Domain size Lx4 ?, Ly 2, Lz2 ?
- 15th order polynomial, 52 elements in x-y plane,
64 Fourier modes (750K grid points) - Bulk Reynolds numbers Rex500 and Rez1200 (?
67o) - Friction Reynolds numbers 42 and 86 (core flow
region)
19Flow visualization
Presence of spiral recirculation bubbles
(isocontours of mean spanwise velocity and
streamlines of transverse velocity)
20Turbulence statistics
Mean Velocity Components
Normal Reynolds stresses
Kolmogorov scale in false color logarithmic scale
(dark regions denote smaller ???not fully
converged statistics)
21LES of Single and 7 Pin Wire Wrap Nek5000
- Single Pin
- Mimics infinite array (no assembly walls)
- Cheap, first case for exploratory convergence
studies, etc. - 7-Pin
- Geometry is current ARR design
- P/D 1.135
- H/D 17.74 (2/3 of current ARR design)
22Relationship to Inflow / Outflow Configuration
- Flow establishes a fully turbulent state within
1 flow-through time - ? spatial development length H/D
- To be checked by multi-pitch inflow / outflow
simulations
23Cross-Sectional Velocity Distributions
- Flow tends to follow in the wake of the wire
- Near the contact point, the flow separates and
forms a strong standing vortex in the assembly
cross section, as also reported in RANS
computations of Ahmad Kim
24Subchannel Interchange Velocities
- Interchange velocity distributions
- left instantaneous
- right time-averaged
25Subchannel Interchange Velocities
- Close fit to sinusoid, with amplitudes
- H / D 13.4 a 0.290 Uz
- H / D 20.1 a 0.225 Uz
- H / D 26.8 a 0.150 Uz
- Amplitude higher than predicted by geometric
factors alone
H/D 26.8
20.1
13.4
267 Pin Simulatons
- E132,000, N 7
- nv 44 M
- np 28 M
- niter 30 / step
277 Pin Visualization
- Time-averaged axial (top) and transverse
(bottom) velocity distributions.
Snapshot of axial velocity
28Subchannel Interchange Velocities 7-Pin, with
Sidewalls
- Inter-channel exchange is no longer a simple
sinusoid - Edge channels have non-zero mean ? swirling flow
C
D
C
D
B
B
A
A
29Subchannel Interchange Velocities 7-Pin, with
Sidewalls
- Inter-channel exchange is no longer a simple
sinusoid - Edge channels have non-zero mean ? swirling flow
Single- (Infinite-) Pin Distributions
H/D 17.7
307-Pin RANS Using Star CD D. Pointer (ANL)
31Fine Polyhedral Mesh
- 2.5 million cells
- Based on fine triangulated surface
- Surface extrusion layer not used in current cases
to allow use of high Re and two-layer k-epsilon
turbulence models. Will be used with low Re
models. - Generated from fine triangulated surface using
Star-CCM meshing tools
32Coarse Polyhedral Mesh
- 1 million cells
- Based on coarse triangulated surface
- Surface extrusion layer not used in current cases
to allow use of high Re and two-layer k-epsilon
turbulence models. Will be used with low Re
models. - Generated from coarse triangulated surface using
Star-CCM meshing tools
33Fine Polyhedral Mesh Results
- Re15000 (Vmean 1, Dpin1)
- H/D 26.6
34Coarse Polyhedral Mesh Results
- Re15000 (Vmean 1, Dpin1)
- H/D 26.6
35LES / RANS Comparison
- Same basic features
- Significant scaling discrepancies (1.5 x due to
different H/D, rest tbd)
Star CD RANS Model (note scale difference)
H/D 26.6
H/D 17.7
36Low-Dimensional Representations
- A step towards subchannel modeling
- allows full-core simulations
- less geometric detail (no wire)
- Wire-induced transport compensated by
interchannel exchange velocities - currently generated by helical forcing
- future projection onto LES/RANS results
- Intra-channel mixing enhanced diffusion
- Allows rapid turn-around of coupled multi-physics
simulations - Some issues
- How to smear wire-wrap volume into reduced
geometry? - Increased clad thickness?
- Maintain cross-sectional area?
- Other
37Low-Dimensional Models, Full Length Subassemblies
- Effects of interchannel mixing with
- no-wire vs. wire-wrap
- pin conductivity
- thermal loading
- large pin counts
- Sacrifices detailed intra-channel mixing
- Surrogate velocity field generated by spiral
forcing to match effect of wire-wrap - Desktop (or small cluster)
38Conclusions
- Software Development
- Advances to Nek5000 to incorporate additional
physics, low-resolution conservative formulations
underway - Pushing the envelope on problem size and
processor count - Continually comparing with commercial and other
codes as reality check - Simulations
- First 7-pin LES study is near completion
- RANS LES comparison underway
- 19-pin simulations within the next few weeks
(EDF) - Low-resolution TH w/ 7 pins ready to couple with
UNIC - Low-resolution 217-pin simulation nearly ready