INSTANT/PHISICS – RELAP5 coupling A. Epiney, C. Rabiti, Y. Wang, J. Cogliati, T. Grimmett, P. Palmiotti

Overview Adding the PHISICS tool suite to RELAP5-3D –A new transport/diffusion solver: INSTANT –A New cross section model: XS-MIXER Examples –Typical PWR –Takeda 4 benchmark –NGNP MHTGR Summary and outlook

Vision Increase of neutronics modeling accuracy Modeling flexibility Uncertainty assessment PHISICS RELAP5- 3D Lean software inter-dependency Low impact for user

RELAP5-INSTANT vs RELAP5-NESTLE FeaturesRELAP5 - NESTLERELAP5 - PHISICS Energy group2-4Not bounded DiffusionYes TransportNoYes Triangular MeshNoYes Unstructured MeshNoYes AdjointNoYes Multi-Dimensional Cross Section TablesNoYes SpeedWinLose (Future ?) Discontinuity FactorsYesFuture ? Cylindrical GeometryNoFuture Perturbation TheoryNoFuture DepletionNoFuture Localized refinementNoFuture

New transport core solver: INSTANT Accessible through new keyword “INSTANT” $ $ REACTOR KINETICS INPUT $ instant gen no-gamma e … Compatible with RELAP5: cross section models control rod model Existing RELAP5 inputs will run just by changing “nodal” to “instant” INSTANT control parameters (if different from default) could be provided through separate input file

New cross section model PHISICS PHISICS gives additional access to: –Unlimited number of energy groups (memory limit) –Transport XS vs. Diffusion Coeffs. –Square root vs. linear structure temp. FB –Simple HTML input –Different FB tabulations for different materials Table allows to account for cross terms Multiple points address non linearity Functional representation in the future

New cross section model PHISICS Accessible through new keyword “PHISICS” $ $ REACTOR KINETICS INPUT $ instant phisics no-gamma e … –Compatible with RELAP5 CR model –Kinetic nodes to TH mapping: as in “Gen” FB Zones and Regions for: –Structure temperature –Fluid temperature –Fluid density –Poison concentration

The “Gen” feedback structure Every kinetic node is assigned to a FB Zone –(Zone figures are assigned to axial meshes like for compositions) Every FB Zone has: –a number of heat structure FB regions Every HSFB region feeds back one (weighted) structure temperature –a number of volume FB regions Every VFB region feeds back: (weighted) density, temperature and poison concentration Total feedback variables: HSFB + 3*VFB

Implementation (using RELAP5 XS models) RELAP 5: Plant and TH INSTANT XS, Geometry Power Compatibility with existing RELAP5 XS and control rod models

Implementation (PHISICS XS) RELAP 5: Plant and TH XS-MIXER INSTANT T f, T c, ρ c,… CR positions XS Power Steady state search using several energy group (>4) has been already implemented Geometry Controls …

Software Structure RELAP Input Reader Feed input Branch to a special input file RELAP /PHISICS driver Data Type PHISICS input file reader Construction- destruction Data Type INSTANT DRIVER (pointing local data type interface) Data Type Feeding RELAP5INSTANT Neutronics TH coupling

Cartesian Geometry: Typical PWR Rod in/out cases Full core model 17x17 nodes 13 axial levels 11 Materials 36 Feedback zones 2 Energy group CR out CR in

PWR Rod Out Easy visualization with INSTANT(VTK file) *surface order 1, volume 4 Convergence evolution Keff InitialConverged INSTANT* RELAP Delta

PWR Rod Out: K eff Evolution

PWR without control rods Power distribution difference (%) NESTLE/INSTANT First iteration Converged Feedbacks tend to reduce the difference in power distribution

PWR Rod Out Comparison Comments –Difference in k eff reasonable –Difference in assembly power higher than expected. Possible reasons: Higher spatial resolution in INSTANT (NESTLE mesh refinement study could confirm this) Different implementation of vacuum BC

PWR Rod Out Spatial convergence and computational time –INSTANT P0 Initial conditions Converged Computational time [%] Surface orderVolume order S. orderVolume order S. orderVolume order INSTANT converges spatially Best computational cost – accuracy ratio for Surface order 1, Volume order 4

PWR Rod Out Computational times INSTANT vs. NESTLE Spatial approximation used 4 th order  37 degree of freedom by node, by energy group  Flexibility comes at less computational efficiency We started at*: NESTLE 175sINSTANT 13760s (S1,V4)  INSTANT factor ~80 slower Ways of reducing the computational time Cross section threshold  gained factor of 10 Parallelization of INSTANT  how many cores you have? *Processor time in kinetics subroutines 1500 iterations

PWR rod Out Neutronics inactive if  XS<tolerance  Gained a factor of 10

PWR Rod Out Using parallelization for INSTANT  Scaling is almost perfect on shared memory otherwise dependent on node to node communication speed  factor of 10 or more possible by average users  Using both, TH skipping and parallelization the initial factor of 80 can be compensated The real conclusion is… Now you can choose your trade off between accuracy and computational cost

PWR Rod Out: Multi Group Test Cross section model test with PHISICS –2 groups (Diffusion coefficient / Total cross sect.) –2 group cross sections expanded to 8 and 20 groups Energy group i is expanded into j groups One can show that k eff does not change for X i,j =  i,j X i X=D,  fis,  abs,   s gi,j->gi’j’ =  i,j  i’,j’  gi->gi’ With some constraints:  = 1,  not  i,j  abs +sum  i,j sum  i’,j’  gi->gi’ < 1/ (3D i,j )

PWR Rod Out (PHISCS XS) Test confirms functioning of PHISICS XS model and multi-dimensional interpolation

PWR Rod In 1 control rod inserted XS by RELAP CR model Steady state calculation  Convergence evolution 2 nd group INSTANTNESTLE Keff converged

Takeda 4 benchmark 3D Hexagonal test 4 energy groups No TH feedback Small fast sodium cooled reactor NESTLE vs. INSTANT CR In CR Out

Takeda 4 benchmark (CR Out) Volume Order Surface Order INSTANT Diffusion NESTLE coarse mesh diffusion methodnodal expansion solution technique INSTANT: PN1 Solution converges towards PN2ND Solution Vol 6/ Surf 1 gives best computational time/accuracy ratio RELAP/NESTLE: coarse mesh method not appropriate, nodal expansion within 90pcm (between INSTANT Surf 0 and 1) UNIC Diffusion k eff =

Takeda 4 benchmark (CR) Volume Order Surface Order INSTANT PN1 NESTLE coarse mesh diffusion methodnodal expansion solution technique INSTANT: PN1 Solution converges towards PN2ND Solution Vol. 6/ Surf 0 gives best computational time/accuracy ratio RELAP/NESTLE: Coarse mesh diffusion not appropriate; nodal expansion still off by ~800 pcm UNIC Diffusion k eff =

MHTGR (NGNP) for the MHTGR benchmark Features needed by the benchmark not supported by RELAP3D-NESTLE –Linear independent feedbacks not good enough –26 energy groups –Need of multiple feedback regions (T fuel, T graph ) in each zone –Triangular mesh for CR location NGNP is supporting the RELAP/INSTANT coupling

NGNP MHTGR Reactor vessel Core barrel Coolant channels Central reflector Fuel blocks Side reflector Control rod channels Hexagonal geometry Inner and outer reflector 3 Fuel rings RELAP5 representation for feedback zones

NGNP MHTGR INSTANTNESTLE Keff converged XS by RELAP (2G) Steady state calculation  Convergence evolution 2 nd group INSTANT: Surf. 1/Vol. 6 NESTLE: coarse mesh diffusion method

Conclusion RELAP5-3D – PHISICS coupling add the following feature –Spatial/angular mesh refinement –Unlimited number of energy group –Cross section tabulation –In the future: depletion, time dependent, decay heat, adjoint sensitivity analysis We can match computational time with higher accuracy We preserve compatibility with past input deck

Extra Slides

Simple HTML input Different tabulations for different materials Different tabulations with Different number of dimensions and points One dimension for each FB variable considered

Simple HTML input 2.62E E … … Dimension 1 Dimension 2 => Full table input allows consideration of cross terms