ARJAN KONING, DIMITRI ROCHMAN. Nuclear ... HANS HENRIKSSON ... The plan is to use a method proposed by Koning and Rochman [2] to investigate the ...
MASSIVE COMPUTATION METHODOLOGY FOR REACTOR OPERATION (MACRO) CECILIA GUSTAVSSON, STEPHAN POMP, HENRIK SJÖSTRAND, GUSTAV WALLIN, MICHAEL ÖSTERLUND
Division of applied nuclear physics, Department of physics and astronomy, Uppsala University, Lägerhyddsvägen 1, 751 20 Uppsala – Sweden
ARJAN KONING, DIMITRI ROCHMAN Nuclear Research and consultancy Group (NRG) Westerduinweg 3, Petten – The Netherlands
KLAESHÅKAN BEJMER Vattenfall Nuclear Fuel AB
Jämtlandsgatan 99, Vällingby – Sweden
HANS HENRIKSSON Vattenfall Research and Development AB Jämtlandsgatan 99, Vällingby – Sweden
ABSTRACT Today, nuclear data libraries do not handle uncertainties from nuclear data in a consistent manner and the reactor codes do not request uncertainties in nuclear data input. Thus, the output from these codes have unknown uncertainties. The plan is to use a method proposed by Koning and Rochman [2] to investigate the propagation of nuclear data uncertainties into reactor physics codes and macroscopic parameters. A project (acronym MACRO) has started at Uppsala University in collaboration with A. Koning and with financial support from Vattenfall AB and the Swedish Research Council within the GENIUS (Generation IV research in universities of Sweden) project. In the proposed method the uncertainties in nuclear model parameters will be derived from theoretical considerations and comparisons of nuclear model results with experimental crosssection data. Given the probability distribution in the model parameters a large set of random, complete ENDFformatted nuclear data libraries will be created using the TALYS code. The generated nuclear data libraries will then be used in neutron transport codes to obtain macroscopic reactor parameters. For this, models of reactor systems with proper geometry and elements will be used. This will be done for all data libraries and the variation of the final results will be regarded as a systematic uncertainty in the investigated reactor parameter.
The understanding of these systematic uncertainties is especially important for the design and intercomparison of new reactor concepts, i.e., Generation IV, and optimization applications for current generation reactors is envisaged.
1.
Introduction
While experimentally obtained nuclear data is published together with their uncertainties, which are not handled in a consistent manner in nuclear data libraries, and standard reactor core simulation codes do not make use of them. Thus, it is not possible to assign, in a rational way, uncertainties to the calculated macroscopic reactor parameters. For safety and economical reasons, the handling of uncertainties in the underlying nuclear physics data and their covariances is a key issue, especially for the design and comparison of new reactors concepts such as Gen IV. The questions of uncertainty propagation and which nuclear data measurements are of highest priority have been addressed by the CANDIDE [10] project and the NEA SG26 working group. The FP7 project ANDES (under preparation) will follow the recommendations of SG26 and CANDIDE. The goal of the MACRO project is to assign uncertainties to macroscopic reactor parameters, as well as calculating their sensitivity to uncertainties in specific crosssections and the model parameters resulting in those crosssections. This will be done using a brute force computational approach based on the TALYS [1] code developed at NRG. The project is based at Uppsala University, and will receive funding from Vattenfall AB and the Swedish Research Council due to being part of the GENIUS (GENeration IV research In Universities of Sweden) project.
2.
The total Monte Carlo approach
The proposed concept used within MACRO can be divided into several steps [2]: 1) Theoretical considerations and comparisons of nuclear model results with experimental reaction data yield a specific a priori uncertainty of each nuclear model parameter. In this way parameter ranges for all nuclear models can be determined and one is left with 20 to 30 parameters that cover nuclear reactions up to 20 MeV. The probability distributions for these parameters will be, at least as a starting point, assumed Gaussian. In addition, resonance parameters and their uncertainties are available, allowing the results to cover the entire energy range. 2) In this step, the TALYS code will be used. The TALYS code can calculate parameters such as crosssections, energy spectra and angular distributions for reactions
involving neutrons, gammas, protons, deuterons, tritons, 3He and alpha particles, in the 105 eV to 200 MeV energy range. This is done based on a limited set of model parameters. All nuclear quantities necessary for the creation of ENDF6 files are calculated with the TALYS code a large number of times. For each run, all elements of the input parameter vector are randomly sampled from a normal distribution with a specific width for each parameter (see step 1). A similar action is performed for the resonance parameters using experimentally derived uncertainties as widths. 3) Correlations between the nuclear model parameters are introduced by a binary reject/accept method; only if all predicted results fall within the uncertainty ranges of experimental data, is the TALYS run accepted. Thus, only certain combinations of model parameters survive, and the parameter distribution is automatically determined numerically by experimental data, without having to resort to an a priori distribution. Every individual run produces a complete nuclear data library containing all cross sections, angular distributions, etc. By this, a large set of random, complete ENDF formatted nuclear data libraries for the resolved and unresolved resonance range and the fast neutron range is created. The libraries are all mutually different for all reaction channels and energies. As a bonus, after enough runs all statistical information to fill a full covariance matrix for the calculated data and the nuclear model parameters become available. Note however, that although all models, parameters, data, etc., are intrinsically correlated, explicit covariance matrices are no longer required since all information is simply available in the large set of Monte Carlo results. This is one of the key ideas; processing of covariance data files and perturbation codes is no longer necessary. 4) The generated nuclear data libraries are processed with NJOY [3] and then used in a transport code like MCNP [4] to calculate macroscopic reactor parameters. For this, a model of a reactor system with proper geometry, elements, etc., is used. This is done for all the “random” data libraries and the variations of the final results can be studied. These variations are due to a) the statistical uncertainties in to the Monte Carlo transport process and b) the differences in the nuclear data libraries. The latter can be regarded as systematic uncertainties that result from the variation in the nuclear model and resonance parameters, and the basic experimental data. In sum, only the basic nuclear physics model and the experimental data in the input side and the macroscopic quantities on the output side are considered. The path inbetween is straightforward and the entire route of data evaluation, formatting, processing, etc., is automated. Manual covariance analysis becomes obsolete using computer power. Through brute force computations, the influence of a certain model or resonance parameter on a macroscopic quantity can be studied.
3.
The MACRO project
The MACRO effort will be started at two fronts in parallel. On the one hand, the “production line” will be developed. This will be a semiautomatic computational chain starting from the randomised sampling of TALYS input parameters and ending with macroscopic reactor parameters, for example the neutron multiplication factor keff and the effective delayed neutron fraction eff. The calculations will be based on TALYS, TEFAL [5], NJOY and MCNP. On the other hand uncertainties of a limited parameter set will be studied, i.e., optical model and level density parameters, on a likewise limited set of Na, Fe and Pb isotopes. The choice of these elements is due to their importance in Gen IV reactors (Na and Pb) and since they are well studied in integral benchmarks (Fe). In this area, the collaboration with A. Koning at NRG will be a significant factor. The results from this effort will be statistical distributions of nuclear model parameters. Points will be sampled from these distributions and used as input data to TALYS in the computational chain. Once the computational chain is completed, and enough information about prioritised nuclear model parameters for prioritised isotopes has been obtained, the two efforts will be combined. To get reasonably good sampling of the nuclear model parameter uncertainties, about 5000 data libraries per isotope will be produced. On a one processor machine, the creation of a full ENDF6 file up to 20 MeV takes about 5 minutes, the data processing about 1 minute and the MCNP calculation about 15 minutes [2]. This indicates that, while the final mass production has to be run on a computer cluster, a standard desktop computer can be used for the initial studies. Finally, the obtained reactor parameter data will be linked back to the input model parameters and the sensitivity of the end result on the nuclear model parameters can be determined. This will allow the assigning of uncertainties to the macroscopic reactor parameters, based on the uncertainties in the nuclear model parameters. The cases used as basis for the first largescale runs will be selected in such a way that they give good indications of where future efforts within MACRO should be focused. This can mean which isotopes should be studies, or which reactor types are of the most interest. The final results of MACRO will give indications of where future experimental research should be concentrated. For example, knowing that a certain crosssection contributes significantly to the uncertainty of a critical parameter for a given application would highly suggest concentrating efforts on obtaining more accurate values for that crosssection. In extension this will of course also apply to the base model parameters. If that crosssection has a high dependency on a given model parameter, increasing the accuracy with which that model parameter is known will allow more accurate values of the mentioned critical parameter in simulations.
This will lead to the ability to conduct more accurate studies on current and future reactor designs, allowing for better knowledge of e.g. the safety parameters of a given design. Having more accurate knowledge of safety parameters can allow more optimizations to be performed without compromising safety.
4.
References
[1] Koning, A.J., Hilaire, S., Duijvestijn, M., “Talys1.0”, in: Bersillon, O., Gunsing, F., Bauge, E., Jacqmin, R., Leray, S.(Eds.), Proceedings of the International Conference on Nuclear Data for Science and Technology, 2227 April 2007, Nice, France, EDP Sciences (2008), 211214. [2] Koning, A.J., Rochman, D., “Towards sustainable nuclear energy: Putting nuclear physics to work”, Ann. Nucl. Energy 35 (2008) 202430. [3] Macfarlane, R.E., “NJOY99 – code system for producing pointwise and multigroup neutron and photon crosssections from ENDF/B Data”, RSIC PSR480 (2000). [4] Briesmeister, J.F. (Ed.), “MCNP – a general Monte Carlo nparticle transport code”, version 4C, Los Alamos Laboratory, Report LA13709M. [5] Koning, A.J., “Tefal1.05”, unreleased code. [6] International Handbook of evaluated Criticality Safety Benchmark Experiments, NEA/NSC/DOC(95)03/I (2004). [7] Fast Reactor Database: 2006 Update, IAEA Report IAEATECDOC1531, December 2006. [8] Tucek, K., Carlsson, J., Wider, H., Nucl. Eng. Des 236 (2006) 1589. [9] Rochman, D., Koning, A.J., “Pb and Bi neutron data libraries with full covariance evaluation and improved integral tests”, Nucl. Instr. Meth. Phys. Res. A 589 (2008) 85108. [10] Koning, A.J., Blomgren, J., Jacqmin, R., Plompen, A.J.M., Mills, R., Rimpault, G., Bauge, E., Cano Ott, D., Czifrus, S., Dahlbacka, K., Goncalves, I., Henriksson, H., Lecarpentier, D., Malumbu Mbala, E., Stary, V., Trakas, C., Zimmerman, C., “Nuclear data for sustainable nuclear energy – Coordinated action on nuclear data for industrial development in Europe (CANDIDE)”