Challenges and developments in fusion neutronics a CCFE perspective

Transcription

1 Challenges and developments in fusion neutronics a CCFE perspective A. Turner Applied Radiation Physics group SERPENT fusion neutronics workshop, Cambridge, June 2015 CCFE is the fusion research arm of the United Kingdom Atomic Energy Authority. This work was funded by the RCUK Energy Programme [grant number EP/I501045].

2 Nuclear Fusion The process that heats the sun and stars Easiest reaction is D-T; but fusion is hard! Need to heat D,T to over 100 million C. Researched since the 1950s. An ideal source of energy: No carbon emissions Abundant fuel sources (millions of years) High energy density: 1 kg fusion fuel = 10 7 kg of fossil fuel MeV neutron. This guy is the main issue for fusion shielding design. Significantly less high level radioactive waste compared to fission. Fusion power plants could provide stable baseload energy. Slide 1

3 Magnetic Confinement Fusion (MCF) DT plasma contained in vacuum chamber using magnetic fields (isolate from vessel walls to permit effective heating). Tokamak device the most successful device to date. Conceived in 1950 Soviet Union. Now many experiments around the world. CCFE hosts world-leading EFDA-JET tokamak and MAST. Fusion works! JET D-T campaign, 1997: 4 MW fus for 4 s; 16 MW fus peak. Peak Q value = 0.65 (ratio of input heating power). Slide 2

4 The goal commercial fusion power plants by Lots to do: plasma physics, tritium breeding, materials, remote handling, diagnostics, neutronics ITER: CCFE - Fusion goals Next step fusion device being build in Cadarache, France. Feasibility of fusion power (physics, engineering, technology). Sustained D-T plasmas, Q 10, develop power plant related technologies (e.g. tritium breeding) You are here DEMO construction DEMO operations IFMIF design IFMIF construct IFMIF materials testing Fusion PP ITER construction ITER operations Slide 3

5 Fusion Neutronics Approach to power-plant conditions: JET ITER DEMO Higher neutron emission rate, longer pulses, more availability Increasing levels of neutron production JET ITER DEMO/power plant Neutron production 1.4x10 18 n/s 1.8x10 20 n/s 8.5x10 21 n/s Fusion power 4 MW 500 MW 2.4 GW Peak Q value Pulse length 40 s 500 s Steady state? Slide 4 Annual neutron production (theoretical maximum) ~10 21 n/year ~10 26 n/year n/year

6 ARP neutronics work supports Applied Radiation Physics Slide 5

7 Fusion entering a nuclear phase, with significant emphasis on nuclear safety, shielding and activation. It will play a key role in the development of enabling technologies for fusion power plants. Neutronics support is required to ITER in the areas of: Blanket and divertor heat loadings and component lifetime. Remote maintenance of in-vessel components Design of port plug systems penetration and shielding. Assessment of decay gamma dose during maintenance. Development of tritium breeding modules. Assessing superconducting magnetic coil heating Radioactive waste assessment Radiation mapping studies. ITER neutronics analysis Slide 6

8 ITER neutronics analysis Neutronics support often relate to port plug design House heating / diagnostics and provide shielding. IC heating antenna 25 m 16 m Shutdown dose rate (SDDR) around ICH Slide 7

9 ITER neutronics analysis but sometimes extend into the port cells and neutral beam (NB) cells ITER NB ducts SDDR in NB cells Slide 8

10 ITER neutronics analysis or even the tokamak complex. Photon dose around activated divertor cassette in cask Photon dose around activated coolant loop 100 m Slide 9

11 ITER neutronics analysis ITER analysis is usually a shielding and activation study. Questions we are trying to answer for design engineers: What is the damage and dose rate in this material? Will it survive? What is the shutdown dose rate? Can we maintain this without remote handling? Can we improve the shielding? What is the radiological inventory? What is the dose when placed in the hot cell? Quantities needed: Neutron flux, nuclear heating, absorbed dose rate, gas production, radiological inventory, shutdown dose rate. 3-D mesh results and component (cell) results. Slide 10

12 Typical ITER neutronics workflow CAD simplification and conversion to MCNP ICRH port plug CAD model Integration with ITER reference models ITER reference model C-lite Slide 11

13 Typical ITER neutronics workflow Global variance reduction (GVR) NP transport (10 9 histories with GVR under MCNP) MCR2S FISPACT-II 3-D decay gamma source (10 6 FISPACT-II runs coordinated by MCR2S) WWITER Primary responses: High resolution neutron, photon flux maps. V/V gas production S/C coil heating Absorbed dose/dpa rates 6 order of magnitude neutron flux attenuation Secondary responses: Radionuclide inventory Dose rates during maintenance Shutdown dose rate field: (Decay gamma source run in MCNP) Slide 12

14 ITER analysis challenges source routines Custom source routines are regularly needed in ITER analysis. Such routines permits user to read external data and specify the sampling of source histories. Parametric plasma source MCR2S decay gamma source essential to provide novel capability. Plasma neutron source, 14 MeV, centrally peaked Time-integrated dose from moving cask in ITER building Slide 13

15 ITER analysis challenges - HPC Large, shielded MCNP model Geometric complexity and streaming paths Results needed across large spatial regions Large history count needed ( ) Global variance reduction needed High performance computing needed X High resolution activation mesh tallies (> 10 GB) Slide 14

16 ITER analysis challenges activation 3-D neutron activation is a key capability for fusion neutronics. Radionuclide inventory. Shutdown dose rate: Equipment / people that go into activated regions. Components removed from activated regions. MCR2S: Mesh-Coupled Rigorous 2-Step. a parallel, 3-D mesh-based decay gamma source generator. Passes MCNP mesh tally data to FISPACT-II. Could be extended to Serpent. N N N Neutron flux/spectra on mesh Mesh-based decay gamma source Shutdown dose rate Slide 15

17 ITER analysis challenges activation ITER often require knowledge of the dose rate for activated components in other environments. E.g. activated divertor cassette in a shielded cask. MCNP6, SERPENT are developing coupled transportactivation capabilities (i.e. instead of R2S) however need to support this portable gamma source requirement. Slide 16

18 ITER analysis challenges model creation CAD models are provided by design engineers. Usually high level of detail, unnecessary for transport. Simplify the CAD, keep neutronically important details Requires experience and judgement Time-consuming process (this model, 8 weeks). Slide 17

19 ITER analysis challenges model creation Model simplified until components can be automatically translated to MCNP using conversion software Need to also create void cells (all space to be defined). Often time is spent repairing resulting MCNP by hand to remove small geometry errors. (this model, 4 weeks) Materials and densities adjusted to preserve presimplification mass. Slide 18

20 ITER analysis challenges model creation Desire by ITER organisation for geometric accuracy is pushing the limits of the traditional modelling approach. Model simplification is a significant portion of the analysis time (and can be frustrating!) The requirement to define a void and to make manual repair operations means that once the model is created it is not easily modified.? Models are also running much slower with increasing complexity (number of surfaces, nested universes ). Slide 19

21 ITER analysis challenges model creation Issues: Designs have to be approximated, requires experience. Unable to respond to changes in geometry e.g. design updates or proposals to study. In general, analysis is taking too long. IO is procuring components and faster analysis turn-around is needed. A simpler, faster way of producing detailed geometry is desperately needed for neutronics analysis whilst also being practical to use on current computing resources. Slide 20

22 Alternative transport geometry Mesh-based geometry tools well established in other engineering fields. Potential to deal with a major analysis bottleneck. ARP group performed (limited) investigations into Unstructured mesh capabilities of MCNP6v1.0 Faceted geometry tracking in DAG-MCNP5. Serpent2 volume and surface mesh geometry (just started). Unstructured mesh (3M elements) of ICH antenna, neutron flux Slide 21

23 Key issues: Alternative transport geometry Does it get the right answer. Ease of use, assigning material regions, tallies etc. Practical on current machines for realistic geometry resolutions (run time, memory). Ease of geometry production (free/commercial meshing tools). Not yet accepted by IO for analysis. New methods have to be benchmarked and accepted by IO (and nuclear regulator). Personal note, I would like to see these techniques go mainstream in the coming year or two. Slide 22

24 Alternative transport geometry DAG-MCNP5, key features: Surface tracking in faceted model, created from ACIS (SAT CAD file by CUBIT. Uses CUBIT CGM libraries for geometry tracking. Implicit void. MCNP6 unstructured mesh, key features: Volume element tracking Takes unstructured mesh (Abaqus format needed at present). 1 st or 2 nd order elements, tet, pent, hex. Implicit void. Overlaps resolved (e.g. non-conformal curved boundaries) Support for OMP threads for mesh geometry.

25 Alternative transport geometry DAG-MCNP5, key findings: For high resolution (F tol < 10-3 ), results identical to MCNP5-CSG. Surface representation uses relatively little memory. Run speed is comparable with MCNP-CSG. Pseudo-cell tallies easy to set up. Material to region definitions very intuitive. CUBIT not the nicest tool, plus DAG compilation difficult. Needs very clean CAD to produce a watertight facet model. Not threaded, MPI only (currently). Slide 24

26 Alternative transport geometry MCNP6-UM, key findings: Unable to use high resolution model* (10M elements), hence results found to significantly deviate from MCNP-CSG. Volume mesh uses relatively high memory. But is threaded and memory does not increase (much) with threads (can parallelise high memory jobs but effectiveness of OMP parallelisation not yet tested). Volume meshes easy to generate (with suitable tools). Run speed is somewhat slower than MCNP-CSG (factor of ~3 for coarse model, should be improved in MCNP6 v1.1). Material to region definitions slightly counter-intuitive (cell cards > matcell index > material) *Geometry pre-processing time to create global tracking model grows quickly when number of elements in a part is large (> 100,000). E.g. days, weeks! Need to split up regions with large element counts during meshing. Slide 25

27 Alternative transport geometry - needs Something that combines the best features of these approaches would be ideal Efficient memory usage Threaded execution Limited thread-private storage Maximise use of cores and RAM Intuitive assignment of regions, materials, tallies But still parallelise well over threads Avoid issues with long model preprocessing times. Surface mesh in other regions Reduced memory use High resolution Volume mesh in required regions Interior element tallies, activation, burn-up Fast transport calculations Even for 100 s of millions of elements. Implied background material / void. Automatically resolve overlaps Support separately meshed regions in mesh file e.g. shared curved surfaces. Slide 26

28 Serpent-2 General requirement is to take large and high resolution geometry models, and remain practical on current computing! Serpent-2 is looking promising on many of these issues. Preliminary investigations into Serpent-2: Surface faceted geometry (STL) Automatically resolves overlaps Volume element geometry (OpenFOAM) Supports mixture: some cells filled with universe = STL and others with universe = OpenFOAM. (I m told) the tracking methods used in Serpent lend themselves to large models. To be tested. Looks promising! Slide 27

29 Serpent-2 Some observations/questions OpenFOAM appears to have a very steep learning curve, potentially not suitable for mainstream users Consider a more widely used / commercial format? E.g. MCNP6 uses ABAQUS format. Support for regions, materials, separately meshed components. Does OpenFOAM-Serpent support other mesh types e.g. (e.g. hex, tet-hex mixtures), 2 nd order elements? Unstructured mesh tallies supported? E.g. for unstructured mesh activation in MCR2S. Slide 28

30 Serpent-2 Successful import of (partial) DEMO tokamak model via STL into Serpent-2. DEMO CAD model DEMO model in SERPENT2 (STL file via SpaceClaim) STL from SpaceClaim, with Netfabb to heal. but no region hierarchy in STL files produced. Can manually splice multiple STL files together (solid, endsolid). Serpent cell-material assignment relies on the order of listing in the file. Any software recommendations? Solid facet endfacet facet endfacet Endsolid Solid facet endfacet facet endfacet Endsolid Slide 29

31 ARP group has students investigating various aspects of Serpent-2 and suitability for fusion neutronics. Benchmarking of results (neutron only at this stage). Application of mesh-based geometry for fusion, practical considerations, ease of use etc. Computational speed and memory for large mesh models. Effectiveness of MPI and threading for large models. Others? Serpent-2, to investigate It is worth discussing potential areas to investigate. Testing and benchmarking is essential prior to use in ITER analysis Code to code comparisons on simple ITER-relevant models. Comparisons with fusion-relevant experimental benchmarks. Slide 30

32 Summary Slide 31

33 Fusion neutronics workflow bottlenecks ITER entering final design Bottleneck Increasing geometrical complexity of models Time to simplify CAD and create transport models increasing Analysis computation increasing Analysis is longer and more expensive Higher likelihood of delays Inability to respond to changing requirements Geometry requirements are at the limits of conventional techniques. Investigating new capabilities such as unstructured mesh geometry in MCNP, Serpent and DAG-MCNP, as well as suitable mesh generation tools. Slide 32

34 Fusion neutronics needs for Serpent 2 VR, weight windows Activation capabilities Portable decay gamma sources Coupled N,P transport Custom sources Mesh and cell tallies Unstructured geometry Implicit background void Volume and surface geometry in same model Tolerant of overlaps Unstructured mesh element tallies Efficient tracking with large mesh models (100M elements) Effective parallelisation with efficient memory use (MPI, threads) Slide 33

35 Thank you for your attention. Slide 34