Sie sind auf Seite 1von 148
Advaiced Nodal Methods for MOX Fuel Analysis by Scott P. Palmtag M. S., Nuclear Engineering, Massachusetts Institute of Technology (1995) B. S., Nuclear Engineering, University of Missouri-Rolla (1993) ‘Submitted to the Department of Nuclear Engineering in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPRY at the TGASSAGHUSETTS INSTITUTE ‘OF TECHNOLOGY MASSACHUSETTS INSTITUTE OF TECHNOLOGY August 1997 © Scott P. Palmiag, 1997. All rights reserved, LIBRARIES ‘The author hereby grants to MIT permission to reproduce and distribute publicly pay electronic copies of this thesis document in whole or in part, and to grant others the right to doso, @RCHIVES Author... coo : eens Department of Nuclear Engineering ‘August 8, 1997 Contified bY oo... ceececseceesererecsererseree saceeweee tee sensi veneers ‘Adan Hency Professor of Nuclear Engineering ‘Thesis Supervisor Cenified by... eee - veces Kord 8. Smith Vice-President, Studsvik of America ‘esis Supervisor Certified by... PO ede ee wee Kent F. Hansen Professor of Nuclear Engineering ‘Thesis Reader Accepted by 6.6... cece ees ener ‘ettreych. Fre idberg, Chairman, Departmental Committee on Graduate Students Advanced Nodal Methods for MOX Fuel Analysis by Scott P. Palmtag Submitted to the Department of Nuclear Engineering on August 8, 1997, in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Abstract ‘One method being considered for the disposal of surplus weapons grade plutonium is to burn the plutonium as mixed oxide fuel (MOX) in commercial light water reactors (LWR’s). The introduction of MOX fuel into LWR’s poses several challenges for the reactor analyst. The spectrum difference between MOX aiid UO> assemblies creates a large thermal flux gradient at the interface between these assemblies. Current nodal methods, which use a transverse leakage approximation, have difficulty modeling this gradient. In this thesis, an advanced nodal method is introduced which uses a two-dimensional, non-separable expansion of polynomial and hyperbolic functions to represent the two-group flux, rather than the transverse leakage approximation. ‘The steep thermal flux gradient also introduces errors in the spatial cross section homogeniza- tion. The spatial flux shape used to generate two-group cross sections does not account for the steep gradient, and therefore, the cross sections may be inaccurate. In this thesis, the cross sections are ‘dynamically homogenized using the actual flux shape found in the reactor configuration. This is done by using a consistent method of spatial homogenization derived directly from the assumption ‘of homogeneous and heterogeneous flux separability. From this assumption, discontinuity factors, homogenized cross sections, and adjusted diffusion coefficients are derived which preserve hetero- geneous reaction rates and currents in the nodal solution. The flux separability assumption also leads to a pin-power reconstruction method which is completely consistent with the homogenization procedure. Because MOX and UO2 assembly cross sections are homogenized using infinite-medium spectra, MOX and UO: spectrum interactions introduce error into the spectral cross section homogenization. In the actual reactor configuration, the spectrum is a combination of MOX and UO> spectra, and single-assembly cross sections may be inaccurate. This problem is addressed by correcting the homogeneous cross sections with empirical correlations which account for spectral interactions. ‘The methods outlined in this thesis are applied to several benchmark problems containing MOX and UO; assemblies. These benchmark problems show that the advanced nodal and homogenization methods derived in this thesis produce more accurate results than current analysis methods. With these advanced methods, reactor cores containing MOX fuel can be accurately analyzed using two-group nodal methods and single-assembly calculations. ‘Thesis Supervisor: Allan F. Henry Title: Professor of Nuclear Engineering ‘Thesis Supervisor: Kord S. Smith Title: Vice-President, Studsvik of America Acknowledgments I would like to thank my advisor, Professor Allan F. Henry, for the guidance and support that he has given me during my thesis research and education at MIT. His insight into a broad range of problems, as well as his never-ending enthusiasm, continues to amaze me. T would also like to thank Kord Smith for the guidance and suggestions he has given me during my thesis research. Further, 1 would like to thank all of the friends I've made at MIT. They have made MIT an interesting place to live and lear, and have helped me realize that there is more to life than work. Finally, 1 would like to thank my fellowship sponsor for making my studies at MIT possible and for allowing me the flexibility to perform research on a topic of my choice. This research was performed under appointment to the Nuclear Engineering/Health Physics Fellowship Program administered by the Oak Ridge Institute for Science and Education under the contract number DE-AC0S-760R00033 between the U. S. Department of Energy and Oak Ridge Associated Universities. Contents Acknowledgments Table of Contents List of Figures List of Tables 1 Introduction and Background 1.1 Introduction. 1.2 Background and Problem Description - 1.2.1 Nodal Methods : 1.2.2. Homogenization and Reconstruction... . 2.22... 1.2.3. Spectral Interactions 1.3 Research Objectives . 1.4 Thesis Organization . Derivation of Nodal Equations 21 Introduction ©... 20.0... 0.200% nn 22 Flux Expansions . . 23. Neutron Balance Condition in Fast Group 24 Neutron Balance Condition in the Thermal Group 2.5. Surface Continuity Conditions 2.6 Comer Flux Continuity Conditions 2.7 Comer Balance Condition... . 2.8 Boundary Conditions 2.9 Summary Solution of Nodal Equations 3.1 Introduction 3.2. Non-Linear Iteration . . nae 3.3 Coarse-Mesh Finite-Difference Solution 3.3.1 Bigenvalue Shifting . . . 3.3.2 Inmerlterations..... 6... eee 3.4 Subdomain Solution... . 2 12 2 12 13 15 16 16 7 7 "1 21 26 28 30 32 35, 35 7 37 7 38 4l 4 43 35 3.6 Numerical Example. . Summary Homogenization and Reconstruction ad 42 43 44 45 46 47 48 Results 5.1 5.2 53 54 55 56 87 Spectral Correction 6.1 62 63 64 65 66 67 68 69 Implementation of Spectral Correction 7 72 13 Introduction Homogenization . Discontinuity Factors Adjusted Current Cross Section Rehomogenization Baffle Homogenization. ........ Pin-Power Reconstruction Bo ‘Summary cnn ee Introduction . . . Computer Code Comparison of Results LRA 2-D BWR Static Benchmark EPRI-9 Benchmark : 5.5.1 Single-Assembly Calculations . 5.5.2. EPRI-9 Benchmark Results . . . NEACRP Benchmark _ 5.6.1 Single-Assembly Calculations 5.6.2. NEACRP Benchmark Results Conclusion . Introduction Spectral Errors... . Cross Section Sensitivities : Spectral Corrections .... 2... 6.4.1 Leakage Correction 6.4.2 Asymptotic Spectrum Correction... - Coefficient Values . . . Results . . . Simplified Coefficient Set Uranium/Uranium Interactions ‘Summary Introduction . . . . Spatially Dependent Cross Sections... 7.2.1 Quartic Polynomial Cross Sections 7.2.2 Quadratic Polynomial Cross Sections Improved Constraint... . 4... 73.1 Quartic Polynomials... . 45 47 48 48 50 34 56 58 60 61 62 62 62 62 68 n 2B 15 9 80 a1 92 92 92 97 98 98 99 103 104 107 108, 109 109 109 109 mW 116 73.2 Quadratic Polynor oe see 16 14 Discussion of Results... 2... 222 e eee eee nN7 7.5 Implementation of Spatially Dependent Cross Sections into the Nodal Method 119 75.1 Nodal Balance Equations .....-..-+ ++ s+ 0+ pene eee HY 7.5.2. Pin-Power Reconstruction 9 7.6 Summary... 120 8 Summary 121 8.1 Summary... . vette ete e ene eee cece WL 8.2 Recommendations for Future Work. 122 8.2.1 Improvements in Speed . . . - 122 822 Extension of Nodal Method to Three-Dimensions : ve 122 8.2.3. Extension of Nodal Method to Multigroups . . . 122 82.4 Improvements to Spectral Corrections... 2-0-2 eevee ees 123 iography 124 A. Solution of Thermal Polynomial Expansion Coefficients 128 B_ Description of One-Dimensional 2-Group Problem 136 B.1 Cross Sections rrr err 137 B.2_ Flux Calculations 137 B.3_ Interface Current Calcul 137 C_ Description of One-Dimensional Multigroup Problem 139 C.1 Two-Group Cross Sections rere . it C2 Reference Results... 2. 00s se eee beeen e eee 142 C3. Cross Section Plots. - 143 List of Figures 53 63 64 TA 72 Coordinate system for atypical node ij... . 2.0... Diagram showing notation of node surfaces. . . . . Box surrounding a comer used for the comer balance. condition. 3x3 subdomain problems for arbitrary nodes nandn+1...... « Plot showing non-zero entries in matrix formed by nodal quations for a 3x3 problem, Convergence Results for LRA Benchmark Problem... ... . . « Diagram of a single-assembly domain and colorset domain. . . . ‘Thermal flux in a one-dimensional UO2 / MOX problem. Extended core calculation for baffle homogenization. ..... 2.2... 0-005 Two Normatized assembly power distribution and errors in STENCIL, solution for 2-D LRA Benchmark. . 2... .....-..05 cnn eee EPRI-S/9R core configuration. . Fuel Assembly for EPRI-9 benchmarl Notation used for describing discor assembly. Normalized assembly power distribution for EPRI-9 benchmark problem. . Normalized assembly power distribution for EPRI-9R benchmark problem. UX and UA Fuel Assembly, PX Fuel Assembly... 0.0... k problem. . y factors in calculations with 4 nodes per One-Dimensional Multigroup Configuration. Fractional change in exact fast absorption cross section relative to single assembly fast absorption cross section. Fractional change in the thermal-to-fast flux ratio from the single-assembly ratio, . Fractional change in fast absorption cross section from single-assembly value. Fractional change in fast absorption cross section using quartic polynomial fit with (1) edge constraint and (2) weighted residual constraint... . . . . Fractional change in fast absorption cross section using quadratic polynomial fit with (1) edge constraint and (2) weighted residual constraint. ..... 0.0... One-dimensional 2-group configuration. 18 28 33 43 44 46 49 52 59 64 67 68 ro) 72 74 74 16 n 93 94 97 100 C3 One-dimensional multigroup configuration. Fractional change in scattering cross section from single-assembly value. Fractional change in fast absorption cross section from single-assembly value. Fractional change in fast fission cross section from single-assembly value. Fractional change in thermal absorption cross section from single-assembly value. Fractional change in thermal fission cross section from single-assembly value. 139 144 145 146 147 148 List of Tables 41 5.1 52 53 54 55 56 57 58 59 5.10 S.A 5.12 5.13 5.14 5.15 5.16 5.17 5.18 5.19) 5.20 5.21 5.22 5.23 5.24 5.25 5.26 5.27 61 62 63 64 65 66 Rehomogenized Cross Section Results . ; 58 Assembly cross sections specified in LRA BWR benchmark problem... . . . 65 Results for 2-D LRA BWR benchmark problem using 1 node per assembly... .. 65 Results for 2-D LRA BWR benchmark problem using 4 nodes per assembly. ... 66 Pin-cell cross sections specified in EPRI-9 benchmark problem........... 70 EPRI-9 Assembly Calculation Results... .. . « n EPRI-9 Results... B Pin-cell cross sections specified in NEACRP benchmark problem, B ‘Assembly Homogenization Results. eee 9 Reference eigenvalues and normalized assembly powers. ..... 2.0... - 81 Assembly Results for Configuration Cl. . 82 Pin Power Results for ConfigurationCl.. 0.0.20. v eve ee eee ee 82 Assembly Results for Configuration C2... . . 83 Pin Power Results for Configuration C2... . « 83 ‘Assembly Results for Configuration C3. vee 84 Fin Power Results for Configuration C3. 84 Assembly Results for Configuration C4, 85 Pin Power Results for Configuration C4. 85 Assembly Results for Configuration CS. . ; 86 Pin Power Results for ConfigurationC5..... 0.2... 0.2 86 Assembly Results for Configuration C6. . . . 87 Pin Power Results for Configuration C6... . . 87 ‘Assembly Results for Configuration C7. 88 Pin Power Results for Configuration C7... . . . 88 Assembly Results for Configuration C8... . . . -. 9 Pin Power Results for ConfigurationC8............ eee 89 Relative Pin Power Error in the Peak Pin(%). ...... 90 Relative Pin Power Error in the Peak Pin (%). 90 2-group single-assembly cross section resulls. ... 2.2... « . 93 Exact 2-group cross section results. 95 U4/MB Results using exact cross sections EXCEPT: . 96 Leakage Coefficients. 101 Reference Pa Coefficients. . 101 Reference Cz, Coefficients for UOp Assemblies. . . 102 10 67 68 69 6.10 6.1L 6.12 6.13 6.14 1 12 13 14 Bl B2 B3 cl C2 C3 ca Reference Ca Coefficients for MOX Assemblies. Fitted 2-group cross section results with reference coefficients. Simplified P,, Coefficients. Simplified Cz Coefficients. . Fitted 2-group cross section results with simplified coefficients. . U3/U4 Results... . . « U3/US Results. U4NUS Results. . . Quartic polynomial cross section results with edge constraint... . . Quadratic polynomial cross section results with edge constraint... . . . . Quartic polynomial cross section results with weighted residual constraint. Quadratic polynomial cross section results with weighted residual constraint, Pin-cell cross sections for One-Dimensional 2-group Problem. Pin-cell cross sections for One-Dimensional 2-group Problem. Estimates of Neutron Current at UO2/MOX interface .............5 Material Compositions for Multigroup Problem. 14-Group Energy Structure... . . « : ‘Two-group Single Assembly Cross Sections. Reference 14-group Results. " 102 103 105 105 106 107 107 107 ut 13, 17 1g 136 137 138 140 140 141 142 Chapter 1 Introduction and Background 1.1 Introduction The addition of mixed oxide fuel (MOX) into light water reactors poses several challenges for the reactor analyst. One difficulty, in the area of steady-state reactor physics, is that the energy spectrum of the MOX fuel is significantly different than the energy spectrum of ordinary UO> fuel. Specifically, the thermal flux in the MOX fuel is approximately half to a fourth of the thermal flux in UO> fuel. This spectrum difference creates several possible sources of error in reactor analysis. One difficulty is that the difference between the thermal flux in MOX and ordinary UO> fuel creates a large spatial thermal flux gradient at the interface between these assemblies. Modem nodal methods have difficulty modeling this gradient accurately. Another difficulty is that the two-group cross sections used in the nodal method are usually obtained from infinite-lattice calculations for a single type of fuel assembly (referred to as lattice- physics or single-assembly calculations). The single-assembly calculations do not account for the large thermal flux gradient at the interface between MOX and UO>, and therefore, the two-group cross sections from the single-assembly calculation may be inaccurate. Finally, two-group cross sections from single-assembly calculations are collapsed with infinite- medium energy spectra. MOX cross sections are collapsed using infinite-medium MOX spectra, and UO> cross sections are collapsed using infinite-medium UO; spectra. In the actual reactor configuration, the spectrum is neither “purely” UO2 nor MOX, and therefore, the two-group cross sections from single-assembly calculations may be inaccurate. In this thesis, advanced nodal and homogenization methods are introduced to solve the problems introduced by MOX fuel. These advanced methods include a semi-analytic nodal method which uses a two-dimensional, non-separable flux expansion; advanced homogenization methods which include cross section rehomogenization and adjusted diffusion coefficients; and a spectral correlation to correct two-group cross sections for local spectral effects. 1.2. Background and Problem Description 1.2.1 Nodal Methods In most modem core analysis, the neutron diffusion equations are solved using nodal methods. ‘Nodal methods solve the diffusion equations over large mesh sizes (referred to as nodes). The size 12 of each node is typically the size of a single fuel assembly. Most nodal methods use a transverse integration procedure to solve the multi-dimensional dif- fusion equations. In the transverse integration procedure, the multi-dimensional diffusion equations are integrated over the directions transverse to each direction, to reduce the multi-dimensional equa- tions to.a set of coupled one-dimensional equations. The neutron leakage in the transverse directions (ie., the transverse leakage) is usually assumed to be in the shape of quadratic polynomials. Several methods are used to solve the one-dimensional diffusion equations. Analytic methods include the Analytic Nodal Method (ANM) [1] and the Nodal Green's Function Method [2, 3] Another method of solving the one-dimensional equations is to represent the flux with finite polyno- ‘ial expansions and then use weighted residual methods to determine the coefficients... Examples of polynomial nodal methods include the Nodal Expansion Method (NEM) [4] and the QPANDA (5). Polynomial nodal methods are convenient to use because they are faster than analytic methods and allow spatially dependent cross sections to be implemented easily. For most LWR problems, the flux is a relatively smooth function and polynomial methods give accurate results. However, for problems with large flux gradients (such as at the interface between UO and MOX), polynomial ‘methods may be inaccurate. Recently, a combination of the analytic and polynomial nodal method, referred to as the Semi- ‘Analytic Nodal Method (SANM), has been developed. In this method, the fast flux is represented by a polynomial expansion and the thermal flux is represented by an expansion containing both polynomial and hyperbolic terms [6]. The hyperbolic terms in the thermal flux expansion represent analytic solutions to the thermal diffusion equation, and can model the steep thermal flux gradient between UO, and MOX fuel better than strict polynomial expansions [6]. The SANM method is now being used in several commercial nodal codes, including codes produced by Studsvik of ‘America [7] and Westinghouse (8). ‘Multi-dimensional affects in the transverse integration approximation are modeled by assuming, shape for the leakage in the transverse directions. This leakage is usually assumed to be in the shape of quadratic polynomials (i.e., quadratic leakage). However, for non-separable two-dimensional cases, for instance at the comer between one MOX assembly and three UO assemblies, the quadratic leakage assumption may not be valid. Several nodal methods which do not use the transverse leakage approximation are the Analytic Function Expansion Nodal Method (AFEN) [9] and ILLICO-HO [10]. AFEN represents the neutron flux as a two-dimensional expansion of non-separable analytic functions. ILLICO-HO uses a transverse integration procedure, but includes interface net current moments in the leakage term. In this thesis, a nodal method is developed which represents the neutron flux with a tw. dimensional expansion of non-separable, semi-analytic functions. The semi-analytic expansion able to model the steep thermal flux gradients associated with MOX fuel, and the two- expansion is able capture the two-dimensional effects created when MOX fuel is placed in LWR's. 1.2.2. Homogenization and Reconstruction In addition to the nodal method, two separate auxiliary calculations are normally performed. The first auxiliary calculation is the homogenization, or lattice-physics calculation. In this calculation, homogenization parameters, including homogeneous cross sections, discontinuity factors, and reho- mogenized cross sections, are defined for each heterogeneous fuel assembly, such that homogeneous 13 results from the nodal solution will reproduce results from a heterogeneous calculation, ‘The second auxiliary calculation that must be performed is pin-power reconstruction. In this calculation, the heterogeneous pin-powers are “reconstructed” using the homogeneous results from anodal calculation, and form-functions from lattice-physics calculations. Each of these calculations will be discussed below. For the assembly homogenization, two methods are commonly used to perform the lattic: physics calculations. These include the single-assembly method and the colorset method. In the single-assembly method, a two-dimensional multigroup transport calculation is performed for each type of fuel assembly, modeling the geometry explicitly. ‘The single-assembly calculation uses reflective boundary conditions on the outer edges of the assembly, and the results of the calculation are used to define average, homogeneous, two-group cross sections for the assembly, as well as, discontinuity factors and other homogenization factors (to be discussed in Chapter 4). Inaccuracies can arise in single-assembly calculations when the flux shape in the actual reactor configuration is different than the flux shape in the single-assembly calculation. This is especially true for problems containing UO2 and MOX, because of the large thermal flux gradient at the interface between the two assemblies. ‘The second type of lattice-physics calculation is the colorset calculation. This calculation is similar to the single-assembly calculation, but instead of solving the multigroup transport problem over a single-fuel assembly, the problem is solved for four quarter-assemblies which are adjacent toeach other. The b ndary conditions for the colorset problem are reflective at the fuel assembly centerlines. A separate colorset problem must be performed for each unique set of four assemblies. The colorset calculations automatically account fer the steep gradients at the interface between MOX and UO. However, because of the large number of colorsets that must be run foreach practical configuration, colorset problems are impractical for most reactor analysis. This is especially true when considering several fuel loading patterns or assembly shuffling, In this thesis, the spatial homogenization errors introduced by single-assembly calculations are eliminated with cross section rehomogenization [11]. In cross section rehomogenization, the homogeneous cross sections are recalculated at each iteration by collapsing the two-group cross sections using the actual flux shape computed for the reactor. Using cross section: r-homogenization with single-assembly calculations gives accurate two-group cross sections while retaining the ease- of-use of single-assembly calculations. Concerning the homogenization parameters used in the nodal methods, it was first recognized by Koebke that additional degrees of freedom must be introduced into the homogeneous problem to reproduce the reaction rates from heterogeneous problems. Koedke introduced additional degrees of freedom by adding heterogeneity factors to the homogeneous problem and adjusting the diffusion coefficient [12, 13, 14]. Smith generalized this method method by introducing discontinuity factors for each surface of a node [15]. Both Koebke and Smith use an arbitrary value for the diffusion co- efficients and introduce heterogeneity/discontinuity factors simply as additional degrees of freedom in the problem. In this thesis, a consistent derivation of the discontinuity factors is given, as well as ‘a consistent definition of surface diffusion coefficients. These definitions are derived directly from the heterogeneous flux separability approximation used for pin-power reconstriction ‘The second auxiliary calculation that must be performed, in addition to the nodal method, is pin-power reconstruction. The results from the nodal solution yield the homogeneous flux shapes in the reactor. In order to calculate the true pin-powers, the heterogeneous flux shapes in each assembly must be calculated. One difficulty is that, in nodal methods, the transverse integration procedure is 14 used and the only information known about the intranodal flux shape is from the one-dimensional transverse integrated fluxes. Therefore, in codes that used the transverse integration procedure, 2 separate calculation must be performed to estimate the intranodal flux shape. To estimate the intranodal flux shape additional constraints are needed for the problem. Usually these constraints, are the values of the flux at the comer points of each node [16, 17, 18, 19]. ‘Once the comer point flux values have been estimated, the intranodal flux is calculated by fitting anon-separable expansion to each node using the surface-averaged flux and comer point flux values from the nodal solution as constraints. The non-separable expansion is either a combination of polynomials and hyperbolic functions, or an analytic solution to the local problem (16, 17, 18, 20, 19, 21). In this thesis, the flux at the comers of each node and the intranodal homogeneous flux shape are integral parts of the nodal method. Therefore, auxiliary calculations do not need to be performed to calculate these values, and the values are completely consistent with the nodal calculation. In addition, the methods of reconstruction are completely consistent with the methods of ho- mogenization. This is different than other nodal methods which treat the homogenization and reconstruction separately. 1.2.3 Spectral Interactions Another possible source of error with two-group cross sections cbtained from single-assembly calculations is that the cross sections may not be collapsed with the same energy spectrum found in the actual reactor configuration. In single-assembly calculations, two-group cross sections are obtained for each assembly by collapsing multigroup cross sections using the infinite-medium spectrum for that assembly. For most LWR problems which contain only UO» fuel, this is usually aot a problem because the multigroup spectra for all UO2 assemblies are similar. However, for MOX assemblies, the infinite-media spectra is significantly different than the UO2 infinite-media spectra. Therefore, in an actual reactor configuration, where a UO2 assembly is adjacent to a MOX. assembly, the actual spectrum is much different than the infinite-media spectra to obtain UO and MOX cross sections. ‘One way of overcoming the spectral homogenization problem (and for that matter, the spatial homogenization problem) is to calculate two-group cross sections from colorset problems. In colorset problems, two-group cross sections are collapsed using a multigroup spectrum obtained from an actual interface problem. The difficulty is that using colorsets greatly increases the complexity of the problem because a separate colorset problem must be run for each interface in a reactor. Single-assembly calculations are more efficient because they are run for each type of assembly, regardless of where the assembly is placed in a reactor. Another method of accounting for spectral interactions is to adjust the two-group cross sections for local spectral effects. In Reference [22], the assembly downscatter cross section is tabulated as a function of the actual spectrum found in the reactor. However, this method gives erroneous results when applied to problems where no MOX is present, and in Chapter 6, it will be shown that to obtain accurate results, it is important to correct the downscatter, fast absorption, and thermal fission cross sections for local spectral effects. In this thesis, the two-group cross sections are correlated to account for local leakage and spectrum effects. The spectrally dependent cross sections are then converted to spatially dependent cross sections by defining polynomial expansions for each cross section. The polynomial cross 15 sections can be used directly in the nodal method. 1.3 Research Objectives ‘The objective of this research is to derive methods better able to model MOX fuel in light water reactors. These methods include an advanced nodal method, advanced homogenization methods, and a spectral correction model. 1.4 Thesis Organization Chapter 2 presents a two-dimensional, non-separable nodal expansion method and gives a descrip- tion of the necessary conditions needed to solve for the flux expansion coefficients Chapter 3 presents a method for solving the nodal equations using a non-linear iteration be- tween a corrected coarse-mesh finite-difference (CMFD) calculation and a domain decomposition calculation, Chapter 4 outlines a consistent method of defining homogenization parameters such that the results from the homogeneous nodal method reproduce results for heterogeneous fuel assemblies, Assembly cross sections, discontinuity factors, rehomogenized cross sections, and adjusted cur- rents are derived from the basic assumption that the heterogeneous flux can be represented by a superposition of the homogeneous flux from the nodal method and a heterogeneous form func- tion from single-assembly calculations. This basic assumption is also used to derive a pin-power reconstruction method which is completely consistent with the homogenization procedure. ‘Chapter 5 includes results for several two-group benchmark problems. These results demonstrate that the nodal and homogenization methods outlined in the thesis are more accurate for MOX analysis, than current nodal methods. Chapter 6 presents an empirical spectral correction to the two-group cross sections to account for local leakage and spectral effects. The cross section correction is correlated to the local leakage to-removal fraction and thermal-to-fast flux ratio. Chapter 7 provides a method of applying the spectral correction within the nodal method. This is done by converting the spectrally corrected cross sections from Chapter 6, to spatially dependent cross sections which can be included directly in the nodal method. The spatially dependent cross sections are in the form of either quartic or quadratic polynomial expansions. Chapter 8 summarizes this thesis and discusses several recommendations for further research, Chapter 2 Derivation of Nodal Equations 2.1 Introduction In this chapter a nodal method is derived for solving the two-dimensional diffusion equations by representing the spatial shapes of the two-group homogeneous fluxes with a non-separable expansion of polynomial and hyperbolic functions. ‘Several conditions are then imposed on the expansion to form a set of equations, which when solved, will yield flux expansion coefficients providing an approximate solution to the neutron diffusion equation. ‘This method explicitly accounts for the flux at the comer points and the homogeneous flux shape can be used directly in heterogeneous flux reconstruction. 2.2. Flux Expansions ‘The first step in the derivation of the two-dimensional nodal expansion method is to divide the radial plane of the reactor (region V) into rectangular nodes, each of area V'%, such that UV# =v, and 1) VINV™ =O for ij # mn. (2.2) A typical node, denoted by the coordinates ij, is shown in Figure 2-1 and is defined by the volume 2 € [ei,t¢41), and ¥ € [yyy Each node is typically the size of a single fuel assembly (15-20 cm) or a quarter fuel assembly (7.5-10 cm). It is assumed that each node is square and all nodes are the same size. The width of each node is defined as ha win UH — Uy (2.3) ‘The energy dependence of the neutron flux is represented with two energy groups, fast and thermal. is? Node | Node | Node i-Ljtt |] jet |ietjea y Ue Node | Node | Node Lo A inj | ij | itty uy Node | Node | Node i-1j-1] ij-1 fitd,j—a nt te a Figure 2-1: Coordinate system for a typical node ij. ‘The fast flux in the interior of each node is approximated by a two-dimensional, non-separable expansion of polynomial functions of the form olen = yy Ath im (StH) 5, (em w), 24 ™ ayeVve, where the functions fm,(€) are polynomial expansion functions given by [4, 24] folé) = 1, Wl) = & Alé) = 38-4, (5) AlE) = 46(E + 3) (E- 3), and LG) = (aH) (E+HE-$). ‘The thermal flux in the interior of each node approximated by a two-dimensional, non- separable expansion of polynomial and hyperbolic functions of the form #8 (eu) = OF (zu) + (ay) (2.6) where $f is the polynomial expansion a4 oP ay) = Lhe im (= S212) (ogy : en m=on=0 and @#! is the hyperbolic expansion a4 Mays Yo Gnom (AE (2.8) mona ‘The functions gm(€) are hyperbolic expansion functions given by golf) = 1, (2.9) m() = cosh(n €), (2.10) wie) = sinh(n €), Qu F 93(€) = cosh (39): (2.12) Pe w(§) = sins (5a), (2.13) and x is defined as ne (2.14) Not all of the expansion coefficients are used in the flux expansions. The fast flux expansion uses 15 coefficients in (2.4) and the thermal flux expansion uses 19 polynomial coefficients in (2.7) and 8 hyperbolic coefficients in (2.8). The non-zero expansion coefficients are given in matrix form as ay a a2 0 0 a a a2 0 0 , (2.15) 0 bao bar boo 0 bog |, and 2.16) 0 cm a 0 07” co 0 0 0 0 a 0 0 0 0]. 2.17) 0 0 0 ey ew 0 0 0 c43 cas ‘The reason for using polynomials in the fast flux expansion is that the fast flux is a relatively smooth function and can be accurately represented with polynomials. The thermal flux, on the other hand, can have large gradients which cannot be accurately represented with polynomials. ‘This is particularly true at the interface between MOX and UO, assemblies. Hyperbolic functions are added 19 to the thermal flux expansion to capture the steep gradients. Physically, the hyperbolic functions represent the strong attenuation of the thermal flux found in the analytic solution to the diffusion equation. Mathematically, the « term in the hyperbolic expansion functions is defined so that the hyperbolic functions are particular solutions to the homogeneous part of the thermal diffusion equation. This point will be described in more detail in Section 2.4 when the thermal diffusion ‘equation is discussed. The use of polynomial and hyperbolic functions to represent thermal flux ‘expansions is referred to as the semi-analytic method {6}, ‘The polynomial expansion functions are similar to expansion functions used in NEM to represent the one-dimensional transverse integrated flux [24]. These expansion functions are chosen so that 4 1 m=0, fi mioae={5 Sera 18) so the volume-averaged fast flux in each node, denoted with a double-overline, is simply Pe ae [ae [aye = a 19) ‘The volume-averaged thermal flux is more complicated because of the addition of the hyperbolic terms and is equal to zo pm pin @ = al, ae [” dy dole,y) 2.20) 2 inn (02) (eid ef 8 wi) +5 sin (2) +68) + psi (9) oh (2.21) ‘Together, the fast and thermal flux expansions have a total of 42 flux expansion coefficients per node. In order to obtain a unique solution for the expansion coefficients, 42 linearly independent conditions must be applied to each node. These conditions are derived using: ‘© 7 weighted balance equations in the fast group, ‘* 19 balance equations in the thermal group, ‘© 8 surface continuity equations (4 per group), and © 8 comer continuity equations (4 per group). ‘These equations are derived in the following sections. A method to solve the equations is described in Chapter 3, It should be noted that a separate flux expansion is defined for each node in the problem and therefore, each node will have its own set of expansion coefficients. In the following sections, the conditions necessary to solve the problem will be derived for a single node ij. In order to simplify the notation, the superscripts will be left off the individual flux expansion coefficients, and unless otherwise stated, expansion coefficients will refer to the ar 2.3. Neutron Balance Condition in Fast Group The first set of conditions to be applied to the flux expansion is neutron balance in the interior of each node. This condition is derived by first writing the few-group neutron diffusion equations in two-dimensional Cartesian geometry with constant cross sections as [23] « “ 1 asst a ¥ dele) +2 oes = 3 [yer +28] ote, 022) Iyle) = -DPVég(as), 9=1.G, nye VY, 223) where J,(z,y) = netneutron current in group g, Go(zyy) scaler neutron flux in group g, ER = total macroscopic cross section for group 9, A = the reactor eigen, Pt fraction of fission neutrons appearing in group 9, VEY, = number of neutrons emitted per fission multiplied by the macroscopic fission cross section for group 9, 21, = macroscopic scattering cross section scattering from group g/ to 9, DY = diffusion coefficient for group 9, and G = total number of energy groups. In this derivation, two energy groups are used and the group structure is chosen such that the thermal group fission spectrum is zero (x2 = 0) and there is no upscattering from the thermal group to the fast group (Z}2 = 0). By defining the group removal cross section as the total cross section minus the in-group scattering cross section (Zy = Ezq ~ Zgq), the two-group diffusion equations for node ij are 1 V-dn(x,y) +E) dn(2,0) = 5 [PEP du(zu) + 22h orlau)] 2.24) V-Ja(z,y) + EY dalt,y) = Ey bias), (2.25) J,(z,y) = -DP Ve,(z,y), ty eV. (2.26) We can now apply the weighted residual method to the fast diffusion equation (2.24) to derive conditions for the fast flux expansion. This is done by multiplying (2.24) by an arbitrary weighting function, wp(z,y), and integrating over the volume of node ij. ‘This leads to the weighted fast 2 neutron balance equation for node ij: fives J" ty aplz.0) [0 ee) +28 (0, ol Loptiee 4 1 =f def" dyugte,y) 5 PA (x0) +122, dol, v)- — ean hy Defining the inner product of two functions over the velume of node 4 as (nd) = Fs fae" ay uplav) dtu, 2.28) ys ‘equation (2.27) can be written in the simplified form (up V+) 439 the flux is being approximated by a truncated expansion which is usually not general enough to Satisfy the diffusion equation at every point. In fact, the ‘runcated expansion function will only Satisfy (2.29) for a limited number of weight functions!. However, if we choose the weight function, we ean force (2.29) to be true for that choice of weight function and thereby generate a condition on the fan expansion. This method is referred to as the weighted residual method because the residual of (2.29) is set to zero for a particular choice of weight function. ‘The weighted residval method is typically used with two choices of weight functions, The first choice, moments weighting, uses low-order expansion functions as weight functions. The Second choice, Galerkin weighting, uses higher-order &xPansion functions as weight functions [24]. Numerical studies have found that the most accurate ‘esulls are obtained with moments weighting (4), Using moments weighting with 7 weight functions, we can generate 7 conditions on the fast flux expansion, The weight functions we use are: wolzy) = 1, (2.30) wi(zy) = f, Beas), 31 2y—y = wn(z,y) = (oun), (2.32) 20 — ist — ay y — inlay) = Maas) g (amy 3) wiley) = sn(*agu-*), 234) "The sumber of linearly independent weight functions for which (2.29) will be satisfied is equal to the number of degrees of freedom in the flux expansion, 22 ws(2,y) wo(ts¥) ‘The first choice of weight function leads simply to the neutron balance condition. (2.35) (2.36) Substituting the weight functions and the fast flux expansion from (2.4) into the definition of the inner product gives the following fast lax moments: (wo, 61) = 400 (wi) = (wr, 1) Wo 12 1 Jato > 35! 360 1 = yuan 4 (ws,$1) = an + F503 (ws, 1) (ws, 1) (wes 1) (237) (2.38) (2.39) (2.40) (241) (2.42) (2.43) Substituting the weight functions and the thermal flux expansion from (2.6) into the definition of the inner product gives the following thermal polynomial and hyperbolic flux moments: (wo, #5) (wi, (we, (ws, #8) (ws, 62) (us, 62) #) (we, ) #) oe) ) ) ) (wo, 63") (wi, 08!) (ws, 64!) (ws, off) 00, 1 1 77010 — 35h 1 1 jqbot — yqhos 2 4 bu — 5 (brs + bat) + sebss 35° ~ B75 aye + 292 ion (5h) « 1 9 + 09 za) ® aye + 28 in (5) 0 1602 + 9 = aa) ™ 144.03 cas alba + ba) + oho 2 k = sinh “(3 (cor + ¢10) + sinh? ( 5 23 aon" (2.44) (2.45) (2.46) (241) (2.48) (2.49) (2.50) 2st) (2.52) (2.53) (2.54) (wa, $f) = 5 [sinh (§) - ar} co 16 in +25 sinh (a) [inn G& )- a} en 255) (ws, 6!) = ‘ [snn (5) ~ 6a] oo 16 +3 sinh (ss) [sinh (354) -S] «0 (2.56) 32 7 (wo!) = [inn (fa) ~ 6a] on (2.57) where a and az are defined as 1 K 2. a1 = Leosh (5) - 3 sinh 4) and (2.58) a = cosh (s4)-4 -4 sinh (5"5) (2.59) ‘Substituting the fast flux expansions into Fick’s law (2.26) and using the definition of the inner product, gives the following fast current-derivative moments: (wo,0-Ji) = — 2! (6000+ 60% + 204 +2 2.60) (us, 0-5) = — Ft (6am + 6030+ 200 + 0 0) 2.60) (wyV edi) = G an + 2a) (261) (w2,V-Ji) = (Jen +200) (2.62) (w3,V -Ji) Fas (2.63) 6 4 2 (waV edi) = (Jen + $auo - am) (2.64) Dy (6 4 2 (ws, V -Ji) EE (Gon + 5008 — a) (2.65) _ D1 8 (w6Vodi) = +53 gygaue (2.66) Finally, substituting the moment equations from (2.37) thru (2.66) into the fast weighted neutron balance equation (2.29) gives the 7 weighted neutron balance equations we need. These equations are listed below. The fast weighted neutron balance equation using wo(, y): Dy 1 PE (S00 + 6000-4 Fa + F000) + (2 5221) oo pen [to + 2 < sinh (3 ) (eo ei) + si? (a) on (267) 24 ‘The fast weighted neutron balance equation using w(x, y): Dl a yn ( a ‘The fast weighted neutron balance equations using w2(z, y): D I 1 d (3 on *0) + @ vp) (ae - 7%) B 1 ai SPE [Feb — syn ton con + 2M a si ( xa) 4] ‘The fast weighted neutron balance equations using w3(2, y) D, 9% 4 tar zat (a - x) (au + 5) 1 = jn (bu Hb +a) + hon + M4afeu) ‘The fast weighted neutron balance equations using wa( x, y): Dy 4 1 1 1 eR (Sea + 5am a) * 7 ~ wn) (jem- TB, =) 1 1 e = Lp (om - Tabor! & [sinh (3) ~ 6a] co «Sou(ca) aa) -eo) ‘The fast weighted neutron balance equations using ws(zx, y): D 4 1 1 1 BR (Sea+5 5am ast) + 920) (5 ny 75%) 1 = Lp (jm - 7 a [sin 6) = sai] co ta ( 5) om (a) -o]) 25 (2.68) (2.69) (2.70) 7m (2.72) The fast weighted neutron balance equations using 25 w6(:z,y) Di (8 I I +58 (54) + (fete) (oa + beau) 1 1 I = FH (ba Hlbas tba) + ma 800 6 7 cy {sn (5) - Sen] «) 2.73) 2.4 Neutron Balance Condition in the Thermal Group The second set of conditions applied to the flux expansion is a neutron balance in the thermal ‘group. Substituting Fick's Law for the neutron current in the thermal neutron diffusion equation (2.25) yields ~ DIVea(e,) + Eh dale,y) = 24 di(ey), zy e V5. (2.74) If the constant x? from the thermal hyperbolic expansion functions is defined as WE an De then the hyperbolic expansion functions become solutions to the homogeneous thermal diffusion ‘equation and cancel from the left side of (2.74). Equation (2.74) then reduces to a form which only involves polynomial terms, or « (2.75) ~ DEVO (tu) +E OP (ty) = 29 di(2,y), zy eV. (2.76) ‘Since this equation is composed entirely of polynomials, the thermal polynomial coefficients can be found in terms of the fast expat coefficients such that (2.76) is true at every point in the node. ‘This derivation is included in Appendix A and the results aie listed below. ‘This derivation also explains why the thermal polynomial expansion has 19 coefficients while the fast expansion has only 15 coefficients. The additional coefficients are due to the Laplacian term in (2.76). The Laplacian of the expansion function corresponding to bys leads to polynomials with Powers corresponding to the terms by3 and 631. Likewise, the Laplacian of the expansion function corresponding to bas is a polynomial with powers corresponding to the terms. by4 and bay. Therefore, in order to match the coefficients, the thermal polynomial expansion includes the additional terms. bus, bat Bag and dap. 00 = te ao + 6 Rts (a02 + a29) +72 Rt, ar + (a4n4 3) Rt, (aso + a4) + (2456 Ry % R+ Fa) Rts aus 2.7 bo = tag +24 Rt, aos +6 Rty ary (2.78) 26 bon = tram + (2888+ S) Rta + 6Rigan +4 Rtseas bos = teas by = tooo (24+ 3) Rau bio = tsa +24Rtsay +6 Rt, ay bn = tran + 1152 Rt ass be = tear bis = 24Rtyass bo = that (2882+ 2) Rts au + Rt an + ARCs fy = tran boy = tear + 32K? teas by = 4Rtaw by = tsa by = 24Rtzaxy b33 = tsa3 by = tran + (24R+ by = 4Rtyagy (2.94) buy = thous 2.95) ‘The constants R and t, are defined as (2.96) (297) ‘The results in (2.77) to (2.95) can be used to eliminate the 6's from all of the equations given in this chapter. This immediately reduces the number of unknown per node from 42 to 23. 27 2.5 Surface Continuity Conditions The third set of conditions that will be applied to the flux expansion is that the surface-averaged flux and current are continuous across surfaces of adjacent nodes. uF ve ba) Node py ij a v5 ut 7 x Node 4 iit ye yo Pe et % Tie Figure 2-2: Diagram showing notation of node surfaces. Referring to Figure 2-2, the z+ and y+ surfaces of a node are denoted as the surfaces on the right and upper sides of a node. Likewise, the surfaces on the left and lower sides of a node are referred to as the x— and y~ surfaces of a node. Using this notation, the surface-averaged flux on the 2+ and y+ surfaces of node ij, denoted with a single overline, is defined as aT, 1 put Ber) =p f veers and 2.98) Bur) = 5 [ aedaleu) 2.99) = Mite ‘The surface-averaged flux for the 2— and y-- surfaces are defined in an similar manner. Using similar notation, the surface-averaged current across the z+ and y+ surfaces of node ij is defined as _ pus = f Adlai] and (2.100) 1 pr Ftv) = jf dedyglavw) @.101 yan Using these definitions, the continuity equations for the flux on the four surfaces of node ij are org 4(z+) = $23(x-), (2.102) 28 Pict) = 55" (2-), (2.103) oat) = dbly-), 2.104) Au) = HW, (2.105) and the continuity equations for the current are 2.106) 2.107) Tut) = Tey), and (2.108) Tht) = Te), (2.109) for a total of 8 surface continuity conditions per node per group. However, each surface is shared by 2 nodes, so only 4 of the surface continuity equations “belong” to the node ij. So ar, ithas been assumed that each node is a homogeneous material, In Chapter4, the continuity equations will be corrected with “discontinuity factors” to account for effects of hetcrogencous nodes. ‘Substituting the flux expansions from (2.4) and (2.6) into the definitions of the surface-averaged flux and current give the surface-averaged expressions in terms of the expansion coefficients, These expressions are listed below. Fast surface-averaged flu (at) = 2.110) (2.111) PAlyt) = ‘Thermal polynomial surface-averaged flux: GPa 1 1 22 (a) = boo 5b10 + 5br0 (2.112) OPP (ust) = boo Shor + 32 ans) ‘Thermal hyperbolic surface-averaged flux: (at) = 2 sinh 3) co + cosh (5) cio sinh (§)e en +72 cosh (85) sh (5%) en 2v2 « BE aint (Fa) en (2.14) cosh (3) cot 2 sinh (§) yo + sinh (5) con 29 (ys) + 208 ain? (2.115) Fast surface-averaged current: Ties) = 2 2.116) = =F (cot 30m +200 + 117) ‘Thermal polynomial surface-averaged current: =~ 2 (tno 300+ 2b0 + 5 bu) (2.118), va)= Bi caste vama ln) (2.119) ‘Thermal hyperbolic surface-averaged current: Fee) = Pe [ee sinh 5) cio + cosh (5) oo ( £2 sink? (ss) en + 2 sinh (fa) cosh (f) ea] 2.120) ( ya? (ut) 2 [es sinh 5) cor + cosh (5) cn £2 sinh? (sa) oss + 2sinh (35) cosh (<5) ou] 2.121) ‘The expressions for the thermal polynomial expansion coefficients in (2.77) to (2.95) can be used to eliminate the b's from the thermal polynomial surface-averaged equations. The resulting expressions which give the thermal polynomial surface-averaged expressions in terms of the fast flux coefficients is shown in Appendix A. 2.6 Corner Flux Continuity Conditions ‘The fourth set of conditions to be imposed on the flux expansions is the requirement that the flux is continuous at the corners of adjacent nodes. Referring to the comer (2, y;) in Figure 2-2, four flux continuity conditions can be written at the comer. These are G3 (tru) = % 9 (2isys) (2.122) 65 (anys) = oy '9Maiys) (2.123) (eis) = O° \(xnys) (2.124) 30 OP \(uus) = $5) leirys) (2.125) However, by inspection, it can be seen that only three of the four continuity equations are linearly independent. Therefore, a flux continuity condition at a comer point gives only three linearly independent equations per comer. A fourth condition at the comer point will be given in the next section. ‘InChapter 4, the comer continuity equations will be corrected with “comer discontinuity factors” to account for the effects of heterogeneous nodes. Using the flux expansions to solve for the flux in the four comers of node ij yields the comer point flux equations listed below. Fast comer point flux for node ij: 1 (2:49) = ao0 + 5(—a01 ~ a10 + G02 + 20) 1 + Glan ~ a12 — a21 + a2) (2.126) ‘ 1 PPC wss1) = 400+ 5 (001 ~ a10 + a02 + 420) 21 — aya + a1 + azn) (127) r 1 PP (a414Y5) = aon + 5(—ao1 + 10 + ao2 + a20) 1 + g(-au +12 ~ a2 + x2) (2.128) . 1 OP (@is1sUse1) = 00 + 5 (aor + a10 + a02 + 220) 1 + gla +412 +421 + a2) (2.129) ‘Thermal polynomial comer point flux for node éj : Pci ay) = bao + (bo bia + bor + b20) + bu ~ bia ~ ba +b) (2.130) EP %aisuiss) = beat Z(bo1 ~ bro + bo + bn) +4 (bu ~ bia +b + be) 131) OP %(aistsu5) = bon-+ 3(—bor + b+ bra + bo) + E(u + bia — bn + baa) 0.1%) OP (asst asst) = Boot 3(bo1 + bio + bon + bo) 1 + glow + biz + bai + baa) (2.133) 31 ‘Thermal hyperbolic comer point flux for node ij : eeu) = (oo +e1)eosh ($) ~ (con + eo) sinh (5) +emscosh? (5 ~ (cx4-+ e3) cosh (Fx) sinh (&) a(_* +44 sinh? (& (2.134) 8% Ceysar) = (c+ e1)cosh ( H ( * + escost? (55 ~ aa sinh? (a = 8s a + (eon ~ cao) sinh 4] + (crs — en) cosh (53) sinh (ss) (2.135) al 8) 5 (xituy) = (eo + ero) cosh (§) ~ (x - co) sian (5) sn () toe) 0 (5) cas sinh? (x55) (2.136) Pf Caisryses) = (6+ c1) cosh (5) + (en + a0) sinh (5) so (2) von) (5) 2(*_ “+ ca sinh (5) 2.137) ‘The b's in the thermal comer flux equations can be eliminated using expressions for the thermal polynomial expansion coefficients given in (2.77) to (2.95). This gives the thermal polynomial ‘comer flux equations in terms of fast flux expansion coefficients. Results for this substitution are shown in Appendix A. 2.7. Corner Balance Condition In the previous section it was found that only 3 linearly independent equations could be derived at a comer by imposing flux continuity. The fourth comer condition used is a neutron balance at each comer point [16, 25, 26}. ‘The comer balance condition is derived by drawing a square box of width 26 centered around a ‘comer point as shown in Figure 2-3. If the dimensions of the box are reduced to zero in a limiting case, there will be no volume, and hence no absorptions or fissions inside the box. Therefore, by using a neutron balance condition in the box, there is no net leakage from the box. If we let L43(zi, ys) represent the group g leakage from the box located at comer (:, yj) into 32 Ut Node Node ij ij y I s uy | x Node il, jel Yin ay Tit Figure 2-3: Box surrounding a comer used for the comer balance condition node ij, then around the comer (2,43) the comer balance condition gives LE" (zisyy) + £53 5) + LS '9"(xi,y;) + LiI“"(x, yy) = 0, (2.138) ‘here the leakage from the box at comer (7, yj) into node ij is defined as. 1 pats 1 puts 23am) = tim if dedytorny +8145 dy Jeg(x;+5,y)| (2.139) = Seg tinuy) + Sid (zisys) (2.140) Likewise, the leakages into the other nodes from the comer (ziyyj) are = Sa! (ainys) + Fis" (xi,y5) (2.141) Jeg! Maisus) ~ Sig's) (2.142) = dF einyy) — Jai, ys) (2.143) Substituting the flux expansions into the leakage equations, and using Fick’s Law to represent the SxiTenis, gives the four comer leakages into node ij from the four comers of node ij. ‘These expressions are listed below. Fast comer leakage equations: i D: Li (xinys) FP [~eo1 ~ 0 + a1 +3 (a+ 039 +02) 1 2 (a03 + 30) ~ 2 (a2 + a2) + 5 (201 +)} (2.144) 4 D; Biteuyss) = F200 ~ a0 ~ au +3 (0m +430 + a2) 33 +2 (aos ~ a30) + 2(—ay2 + a1) + 5 (a04-+ a (2.145) D. Lites) = FE[~ ao tao ~ an +3(am + 020+ 02) +2(—aos + a30) + 2(a12 - a1) + 3 (ao4 + aa] (2.146) n D. Wns) = Fn + ayo + an +3 (aq + ax + a22) +2 (ao3 + 030) +2 (a12 + an1) + ; (04 +au)] (2.147) Thermal polynomial comer leakage equations: LE%(aow) = a — bor ~ bio + bis +3 (bon + bao + bra) ~ 7 (bos + bx) — 2 (bia + bai) + bis + bat + 5 (bo + bu) +75 (bu + ba)] 2.148) LP xiapas) = 8b ~ bio bs +3 (bra + bo +b) +2 (bos ~ yo) +2(—bia + bat) — bis — bat + 5 (bo + bo) + 75 (bu + ba] 2.149) D; LP ais) = FE [~ tar +b — 011 +3 (ben +b +b) 42(-to +b) + 2(b2~ bi) bs ~ bo +} los +o) + 75 (bu +b] (2.150) 12% aspnsei) = FE [ba + 0+ bu + 3( boa + bn + bm) +2 (bos + b30) + 2 (bi2 + bai) + bis + bar +5 lou + bo) + 7 ba + be] as ‘Thermal hyperbolic comer leakage equations: LE (ei,y) = a sinh (5 ) (cor + 10) + neosh (5 J eer) + Vin cosn (355) sinn (55) (cx + 04) + Vin feos? (545) ~ 4] (ou ~ el] (2.152) 34 HEP uyser) = FE [asian (3) (a tem) + weosh (3) (cee) + Vin coh (55) sh (55) ee 22, + Vin [cos ( [Ail > [nl 2 > 1. B.19) ‘At each outer iteration, the error in the eigenvalue is reduced by a factor approximately equal to the dominance ratio. For most calculations, the dominance ratio is approximately 0.9-0.99 and the iterations are slow to converge. 3.3.1 Eigenvalue Shifting (One effective method of accelerating the convergence of the outer iterations is to apply Wielandt’s eigenvalue shift to reduce the dominance ratio (31, 1, 34]. With Wielandt's eigenvalue shift, a fraction of the fission source is subtracted from each side of (3.16), or 1 1 1 -tree s(t _ [aos] #? = fren With this change, the dominance ratio is reduced to (3.20) > G21) which is less than the unshifted dominance ratio for A, > A1. However, the shifted problem only converge to the correct eigenvalue solution if A, is closer to Ao than to any other eigenvalue. Therefore, A, is always chosen to be greater than Ag so the method will converge to the correct solution. ‘Two tradeoffs occur as the eigenvalue shift approaches the reactor eigenvalue (Ay —> Ao). The first effect is the dominance ratio approaches zero and the outer iterations converge rapidly. The second effect is (3.20) becomes singular and the inner iterations take longer to converge. An eigenvalue shift must be chosen to balance these effects. Also, the problem is solved using a non-linear iteration aad the converged value of Ag is different at each CMED iteration. Therefore, an optimal shift will be different at each CMED iteration, and a single, optimal value of X, does not exist [34]. It has been found that the best choice of eigenvalue shift isto let A,=At, 3.22) where 2 is the latest “best estimate” of the reactor eigenvalue obtained from a previous CMFD solution. Smith and Zerkle performed several optimization studies and found the best choice of 6 to be in the range 0.03 to 0.05! [1, 29}. 3.3.2 Inner Iterations ‘The tradition method used to solve (3.20) is the Cyclic Chebyshev Semi-iterative method (CCSI) LI, 29, 30]. However, recent advances in numerical analysis have shown that Krylov subspace ‘Note that this range of optimal shift values is only valid for assembly sized nodes. If fne-mesh calculations or pin by pin calculations are used, the optimum shift value will be much larger. 4 methods are more efficient at solving this type of problem (35, 36]. Of the different Krylov solvers, the Bi-Conjugate Gradient Stabilized Method (Bi-CGSTAB) seems the most promising (37, 38, 39]. In this implementation, the traditional Gauss-Seidel method (GS) is used to solve (3.20). While CCSI and Bi-CGSTAB are much faster that GS, less than 1.5% of the computational time is spent in the inner iterations. ‘Therefore, the added complexity of CSI and Bi-CGSTAB has not been implemented. ‘The values on the right-hand-side of (3.20) are from a previous outer iteration and remain constant in the inner iterations. Rewriting (3.20) in a standard format for the discussion of iterative solvers gives Ax=b, (3.23) where (3.24) 3.28) Fee. (3.26) In the Gauss-Seidel method, the matrix A is decompose Jower matrix B, and a strictly upper matrix F, such that io a diagonal matrix D, a strictly A=D+E+F. 27) ‘The Gauss-Seidel iteration is then x) = —(D +B)! Px + (D+E)"'b (3.28) where n is the inner iteration index. The Gauss-Seidel iterations will converge for any initial guess of x if and only if A is strictly diagonal dominant or is an irreducible diagonally dominant matrix 135}. ‘The convergence rate of the Gauss-Seidel method is determined by the spectral radius of the iteration matrix.r The iteration matrix is defined as G=-(D+5)'F, 3.29) and the spectral radius, denoted p(G), is the maximum modulus of the eigenvalues of G. Since A is diagonally dominant, (G) is less than unity and the iterations converge. The closer p(@) is to unity, the slower the iterations converge. ‘The spectral radius of G is sensitive to the mesh spacing. As the mesh spacing increases, p(@) decreases. Therefore, p(G) is smaller for acoarse-mesh calculation than for a fine-mesh calculation, and coarse-mesh calculations converge faster than fine-mesh calculations [24]. Generally, the inner iterations do not need to be converged completely at each outer iteration. This is because the fission source is not converged completely, and any convergence of x beyond a certain “error reduction” is unnecessary. The approach taken is to run a set number of inner iterations per outer iteration. 10 to 20 inner iterations per outer iteration has been found to be adequate. 42 3.4 Subdomain Solution ‘The second calculation performed in the non-linear iteration is the subdomain solution. The subdo- main problem consists of solving the nodal equations using an overlapping domain decomposition method [40]. In the domain decomposition method, the nodal equations are not solved over the entire problem, but are instead solved for many overlapping regions of the problem. Specifically, for each node in the problem, a 3x3 subdomain “stencil” is solved consisting of the node of interest and the 8 surrounding nodes. Figure 3-1 shows “stencil” subdomains for arbitrary nodes n and n+l ‘Subdomain Subdomain / for node n ! for node n+1 Figure 3-1: 3x3 subdomain problems for arbitrary nodes n and n+ 1. ‘The boundary conditions for each subdomain are the surface and comer flux values on the outer ‘edge of the subdomain obtained from a previous iteration. The eigenvalue used in each subdomain calculation is from the latest CMED calculation and is assumed to be constant. Using a fixed eigenvalue and known flux boundary conditions, each subdomain problem reduces to a fixed source problem, The nodal equations for each subdomain are linear and can be written in matrix form. Each ‘matrix includes 23 equations per node for 9 nodes, fora total of 207 equations. Since each subdomain problem is a source problem, the flux coefficients are found directly by solving the matrix using an LU Decomposition method [41] Over 92% of the elements in euch subdomain matrix are zero. The non-zero elements of the matrix are shown graphically in Figure 3-2. To take advantage of the zero elements and reduce the 43 number of computations needed to solve the matrix, a sparse LU decomposition method is used. ‘The sparse LU decomposition solver used is from the Unsymmetric-pattern Multifrontal Package (UMFPACK) [42]. The UMFPACK solver is approximately 28% faster than the LAPACK [43] solver used for dense (non-sparse) matrices. Figure 3-2: Plot showing non-zero entries in matrix formed by nodal equations for a 3x3 problem. ‘As the non-linear iterations converge, the boundary conditions for each subdomain problem converge, and the solution of the nodal equations becomes identical to the solution that would be obtained if the nodal equations for all of the nodes are solved simultaneously. Once the expansion coefficients for a subdomain are known, the correction factors for the CMFD problem can be calculated for the center node of the subdomain. Using the currents and average flux values from the subdomain solution, (3.6) and (3.7) are used to define D as: hJey(2+) + D9 (oh ) Resolet) + Daley" s') 2. ae ro (3.30) 631) where @ and Jzg are the flux and currents from the subdomain solution. For each subdomain, D 44 calculated for the 4 surfaces of the center node. Using this definition of D, the currents in the converged CMFD solution will exactly match the currents in the converged domain decomposition solution. From (3.30), it can be seen that D is a function of the diffusion coefficient of the node. However, if an arbitrary value of the diffusion coefficient is used consistently in the definition of D and in the CMFD equations, the correct current will be preserved and the actual value of the diffusion coefficient does not have to be used. Also, if the same arbitrary diffusion coefficient is used for nodes on both sides of an interface, the D value for the interface will be the same for both nodes. ‘Therefore, a value of D need only be stored for each interface, instead of for each interface of ‘each node. In this thesis, a constant value of the diffusion coefficient is used consistently in the definitions of D and in the CMFD equations. ‘Since a subdomain problem is run for each node in the problem, and the subdomains overlap, there are two separate values of the coupling coefficients calculated for each interface. The D values used in the CMFD calculation are averages of the two estimates. In addition to calculating the correction factors in the CMFD equations, the expansion coeffi- cients from the subdomain solution are also used to update the boundary conditions for the next subdomain calculation. The expansion coefficients from each subdomain solution are used to find new estimates for the flux at the four surfaces and four comers of the center node of the subdomain. Since the subdomains overlap, there will be two estimates of the flux at each surface and four esti- mates of the flux at each comer point. The averages of the different estimates are used as boundary conditions in the next subdomain calculation. ‘To summarize the subdomain solution, for each node in the problem, a subdomain is formed and the following steps are taken: 1. The nodal equations for the subdomain are used to form a subdomain matrix. The boundary conditions in the nodal equations are from a previous subdomain iteration and the eigenvalue is from the latest CMFD solution. 2. The subdomain matrix is solved using a sparse LU decomposition solver. 3. D values are calculated for the four surfaces of the center node. 4, Surface-averaged fluxes and comer point fluxes are calculated for the center node. ‘Then, for each surface in the global problem, the two estimates of D are averaged to use in the ‘CMFD calculation, For each surface and comer point in the global problem, the new flux estimates are averaged to use as boundary conditions in the next subdomain calculation, 3.5 Numerical Example ‘The convergence properties of the non-linear method can be demonstrated with a numerical example. To summarize, the non-linear iteration consists of the following algorithm: 1. B values are initialized to zero and an initial CMED calculation is performed, 2. Results from the initial CMFD calculation are used to estimate the boundary conditions and eigenvalue for the subdomain calculation, 45 3. A subdomain calculation is performed to solve the nodal equations over overlapping subdo- mains, 4, New D and boundary conditions are calculated, and 5. ACMFD calculation is performed. Steps 3 to 5 are repeated until the problem converges. ‘The numerical example is the LRA Benchmark Problem. The LRA Benchmark is a simplified two-dimensional BWR quarter-core (the complete problem description is given in Chapter 5). This problem is solved using the following convergence parameters: ‘¢ 5 inner iterations per outer iteration, ‘ an eigenvalue convergence of 10-8 AK in each CMFD iteration, a maximum of 50 outer iterations per CMED iteration, and © an eigenvalue shift of 6 = 0.04, ‘The maximum error in the node averaged flux, value of D, and reactor eigenvalue is shown in Figure 3-3 as a function of the CMED iteration number. The maximum errors are relative to a fully converged problem using 15 CMED iterations. 1 2 3 5 6 7 + (CMFO tteatione Figure 3-3: Convergence Results for LRA Benchmark Problem, 46 From this figure, the non-linear method converges quickly. The eigenvalue is converged to less than 10-$ AK in only 4 CMED iterations. The run-time for 6 CMED iterations is 21.9 seconds on a DEC AlphaStation 400 4/233. 98.2% of the run-time is in the subdomain solution and 1.4% of, the run-time is in the CMED solution. 3.6 Summary In this chapter an efficient solution method is developed to solve the nodal equations. This method includes a non-linear iteration between a CMED calculation and a higher-order nodal calculation, ‘The CMFD equations are solved using a fission source iteration and the nodal equations are solved using a domain decomposition method. 47 Chapter 4 Homogenization and Reconstruction 4.1 Introduction ‘A method for obtaining equivalent homogenized cross sections and discontinuity factors for hetero- geneous fuel assemblies is derived in this chapter. The homogenized cross sections are defined from single-assembly calculations such that the reaction rates and currents in the homogeneous nodal solution will match those from an equivalent heterogeneous solution. ‘Additional homogenized parameters are derived from the basic assumption that the heteroge- neous flux shape is equal to the superposition of the homogeneous flux from a nodal solution and a heterogeneous form function from a single-assembly calculation, From this assumption, several parameters are derived. These include discontinuity factors to correct the continuity of the homo- geneous flux, a homogenized diffusion coefficient to correct the continuity of the homogeneous current, and rehomogenized cross sections to find homogeneous cross sections averaged with the actual heterogeneous flux shape found in the reactor. Finally, flux superposition assumption is used to reconstruct individual pin powers from the homogeneous nodal solution. ‘A method of obtaining equivalent homogenized parameters for reflector/baffle nodes using an extended core calculation is also presented. 4.2. Homogenization In modern LWR analysis, homogeneous cross sections are obtained from lattice physics calculations. Lattice physics calculations are high-order, usually transport, multigroup calculations which model a single heterogeneous lattice. ‘The results from this calculation are used to obtain equivalent homogeneous, two-group cross sections defined such that the average reaction rates in a two-group, homogeneous solution for the lattice match the average reaction rales in a detailed heterogeneous solution. ‘Two types of lattice physics calculations are usually performed. A single-assembly calculation is over a single fuel assembly with zero-current boundary conditions on the edges of the assembly. A colorset calculation is over 4 adjacent quarter-assemblies with zero-current boundary conditions at the centerline of assemblies. The domains for a single-assembly and colorset calculation are shown in Figure 4-1 for the 4 assemblies labeled A, B, C, and D. In both the single-assembly and colorset calculations, the assembly geometry is explicitly modeled and many groups are used to 48 Single-Assembly Domain Figure 4-1: Diagram of a single-assembly domain and colorset domain. capture intranodal transport and spectrum effects. Colorset calculations also capture spectrum and leakage effects between assemblies. However, colorsets are expensive to obtain because a separate calculation must be run for each combination of assemblies in the problem. Each time a new core configurations is analyzed, and assemblies are shuffled, new colorset calculations must be performed. Single-assembly calculations are less expensive to obtain because calculations only need to be performed for each type of assembly in the problem, regardless of where the assembly is placed. ‘When assemblies are shuffled, single-assembly calculations do not need to be recalculated. For this reason, the cross section homogenization methods in this thesis will be from single- assembly calculations. In addition, methods will be introduced which account for spectrum and leakage effects between assemblies using single-assembly calculations. Equivalent homogeneous cross sections are defined such that the average reaction rates in a two-group, homogeneous solution match the average reaction rates in a detailed heterogeneous solution. The equivalent homogeneous cross sections are found by setting the heterogencous and homogeneous reaction rates equal to each other, or 1) where a is either the transport, absorption, or fission cross section, and $i) is the homogeneous flux. In the single assembly calculation, the homogeneous flux is spatially constant and equal to the average heterogeneous flux, or =z e 75 op, f,, eveeen, a2) 49 ‘The equivalent homogeneous scattering cross section from group g to 9’ is defined as 1 yo |, ] aE |] dedyd,(E > E',2,y) H(E,2,y) vi “ ¥,= 75 Je a 8 fy WEE > Bie En) = he ; (43) ‘a9 = oor 7 However, in the nodal method derived in Chapter 2, no upscattering cross sect Therefore, to account for the neutron upscattering, a net downscatter cross section is used in the nodal method and is defined as — 5 Hiner =H) 54S. (44) a ‘The equivalent homogeneous cross sections are defined to preserve the average reaction rates in an assembly with no net leakage. However, in most reactor configurations, there is net leakage from an assembly and the average reaction rates and net leakage will not be preserved without introducing additional degrees of freedom into the homogenization procedure [44]. ‘The additional degrees of freedom, provided by discontinuity factors, are discussed in the next section. In order to obtain equivalent homogencous cross sections for the entire problem, a single- assembly calculation must be run for each type of assembly in the reactor core, as well as for each reactor condition (e.g. fuel temperature, moderator temperature, boron concentration, void), and each depletion timestep. Examples of lattice physics codes which perform single-assembly and depletion calculations are CASMO [45] and PHOENIX [46]. 4.3 Discontinuity Factors ‘Additional equivalent homogeneous parameters are derived using a basic assumption that the actual, heterogeneous flux can be expressed asa superposition of the homogeneous flux and a heterogeneous form function, or 1U) = G91OM (x,y) 634(z,y)- 45) ‘The homogeneous flux in this equation is from a nodal calculation and the form function is from a single-assembly calculation. in fact, the form function is defined from a single-assembly calculation as OUETSA(a, y) een = Gfomar yy ‘The homogeneous flux in a single-assembly calculation is the average of the heterogeneous flux, so the form function is equal to the single-assembly heterogeneous flux divided by the average of the single-assembly heterogeneous flux. ‘Anexample of the superposition approximation is demonstrated in Figure 4-2. This figure shows the thermal flux profile for a one-dimensional problem containing a UO2 and MOX fuel assembly'. Both assemblies are heterogeneous with a single fuel enrichment and 5 water holes. The top graph in Figure 4-2 shows the thermal flux form functions obtained from single-assembly calculations. ‘The bottom graph shows the thermal heterogeneous flux obtained from a heterogeneous calculation (46) "This problem is described in Appendix B ‘over the entire problem. Using the heterogeneous flux and assembly form functions, the pointwise homogeneous flux is calculated directly from (4.6). The resultant pointwise homogeneous flux is shown in the bottom ‘graph of Figure 4-2. SI MOX Thermal UO2 Thermal Form Function Form Function 200-15 0-10-58, ° 5 10 15 20 ~ = Heterogeneous Flux — Homogeneous Flux 5| 5 a, a Es é 2 1 U02 Assembly MOX Assembly oO 200-15 10-5 0 5 10 15 20 Position (cm) Figure 4-2: Thermal flux in a one-dimensional UO, / MOX problem. 52 From this figure, two key points can be observed. The first point is that the homogeneous flux is a smooth function. This suggests that the superposition approximation is a good assumption and the detailed “shape” of the heterogeneous flux is absorbed by the form function. The smoothness also suggests that the polynomial and hyperbolic expansion used in the nodal method to represent the homogeneous flux should be accurate. ‘The second point from Figure 4-2 is that the homogeneous flux is discontinuous at the interface between the UO2 and MOX fuel assemblies. This can be explained by noting that the heterogeneous flux is the actual physical flux, and therefore, the heterogeneous flux must be continuous at the interface. Using (4.5) to represent the heterogeneous flux, the flux continuity at the interface is 8AU0%(2) gHOMUOr( 2) JBAMOX (2) gHOM MOX (n) F (47) Since the form functions for the UO and MOX are discontinuous atthe interface, the homogeneous flux must also be discontinuous at the interface. Koebke was the first to recognize that the homogeneous flux is discontinuous at assembly interfaces. Koebke accounted for the discontinuity by defining a single heterogeneity factor to correct the homogeneous flux at the surfaces of a node. For symmetric assemblies, the heterogeneity factor is equal to the surface-averaged form function. For unsymrmetric assemblies, the diffusion coefficient is adjusted until a single heterogeneity factor can be used for all of the surfaces of the assembly. Koebke’s method is referred to as Equivalence Theory (ET) [12, 13, 14]. ‘The derivation of Equivalence Theory is based on physical arguments. Koebke recognized that using only equivalent homogeneous cross sections does not preserve both the average reaction rates and the net currents in an assembly. He therefore introduced heterogeneity factors into the problem as additional degrees of freedom so that both reaction rates and net currents are preserved [44]. ‘Smith later recognized that ET could be generalized by defining a separate discontinuity factor for the surface of each node [15]. Smith's method, known as Generalized Equivalence Theory (GET), the discontinuity factor for each surface of a node is equal to the single-assembly form function averaged over the surface of the assembly. For a single node ij, the discontinuity factors for the four surfaces are defined as f 1 pos Sees ae age (48) 1 pe i f Pav diMav) 49) Shy i aedSMc,y)] and (4.10) es lyewsai i ae pS(z,y) (any ss lv=ws For the case of symmetric assemblies, the heterogeneity factors in ET are equal to the discontinuity factors in GET. Both ET and GET are derived from physical arguments of conservation of reaction rates and net currents. However, in the derivation of this thesis, the discontinuity factors are obtained directly from the superposition assumption of (4.5). ‘The nodal equations derived in Chapter 2 are for the homogeneous flux and are based on 33 nodes having constant, homogeneous cross sections. However, it has just been shown that when heterogeneous materials are present, the homogeneous flux continuity equations must be corrected with discontinuity factors. Applying discontinuity factors to the homogeneous surface-averaged flux continuity equations for the four surfaces of node ij, from (2.102) to (2.105), leads to at) Sat? = GHe-) Sp, 4.12) Pet) A, = 6% (e-) ft, (4.13) WH) fae! = Hl) SG» and (4.14) A+) £3, = oa (4.15) Another requirement used in the derivation of the nodal equations is continuity of flux at the ‘comers of adjacent nodes. However, since the homogeneous flux is discontinuous for heterogeneous ‘materials, the homogeneous comer flux continuity equations must also be corrected. Defining comer point discontinuity factors for the four comers of node i as, = o5%(2inyj), (4.16) = OFM ey), ain GBA (istyyy), and (4.18) = OF (eists jens 4.19) the comer point continuity equations for the four comers of node ij, from (2.122) to (2.125), become. Pew) = oF Mew) fee, (4.20) 95 Many) fae = ee aa "Ceo Soe pal 421) » and (4.22) (4.23) 4.4 Adjusted Current The flux superposition assumption can also be applied to find the heterogeneous current at the surface of a node. The heterogeneous current is obtained by substituting (4.5) into Fick’s Law. For the surface-averaged heterogeneous current in the x-direction, with a heterogeneous diffusion ‘constant, this is = pir a) é of e) (424 surface surface —DyPTa) 2 fayom a o3A(a] 25) leerface = ~DYFr ey [open Zegomcay + Mar Zozra|| 26 surface 54 wx; 2 foM(a) tue (4.27) since, from the boundary conditions of the single-assembly calculation, the derivative of the form function is zero at the surface of a node, Therefore, to preserve heterogeneous current, (4.27) must be satisfied in the nodal method. ‘The correct heterogeneous current is satisfied in the nodal method by defining a homogeneous diffusion coefficient for the surface of each node as D(z) = DEPT (x) g84 (2) (4.28) surface sure ‘And since the form function evaluated at the surface of each node is the discontinuity factor, the homogeneous diffusion coefficients for the four surfaces of node ij are Dyer = fjh4 DEPT (x) lees 4.29) Dye Sg DPT (2) (4.30) Dove = Sone Del *r| 31) ly=wt Dry = Inv D5'*P) 432) yay ‘This method of defining the homogeneous diffusion coefficient as the product of the heteroge- neous diffusion coefficient and the discontinuity factor is is known as the Adjusted Current Model (ACM). This method defines a separate diffusion coefficient for each surface of a node. ‘The ACM procedure must be applied everywhere the diffusion coefficient is used in the nodal ‘method. This includes comer balance equations, current-derivative moments, and the definition of kappa. For the comer balance conditions described in Section 2.7, corner homogeneous diffusion coefficients are defined using ACM as the heterogeneous diffusion coefficients at the comers of the node multiplied by the comer discontinuity factors. For the current-derivative moments in Section 2.3, Gauss’s Law is applied to the moment ‘equations to transform the volume integrals into surface integrals. The diffusion coefficients used in the resulting surface equations are the same as the diffusion coefficients used in the current continuity equations. Diffusion coefficients also appear in the definition of kappa in Section 2.4 and Appendix A. Kappa is defined so the thermal hyperbolic expansion coefficients are solutions to the homogeneous part of the thermal diffusion equation. Using this definition, the thermal polynomial expansion coefficients are solved in terms of the fast expansion coefficients to reduce the number of expansion coefficients in the problem (the complete derivation is in Appendix A). This procedure of solving the thermal polynomial expansion coefficients requires a constant value of kappa and a value of kappa consistent with the thermal balance equations. Therefore, a single diffusion coefficient is required for each node, and ACM can only be used for symmetric nodes. When symmetric nodes are used, the adjusted diffusion coefficients from (4.29) to (4.32) should be used to find kappa. 55 ‘This limitation of only being able to use ACM with symmetric nodes is because of the way kappa is defined in Appendix A. This limitation does not apply to other nodal methods which use a interfaces or near control rods. Ideally, the homogeneous cross sections should be averaged using the actual flux shape found in the reactor configuration. This can be done with Cross Section Rehomogenization (1 1]. Substituting the heterogeneous flux from (4.5) for the flux in (4.1) gives the expression: os [42 d0%ag(250) 8542.0) 620" 9) [eto OD (4.33) ‘The homogeneous flux is obtained from the nodal solution, and therefore the flux expansions from (2.4) and (2.6) can be substituted for the homogeneous flux, J AvP le0) *e0) as Sa 2) fal) i a a ———_ $$ _____. (4.34) [440089 Dean Slt) fold) * The complete description of the problem is given in Appendix B 56 Rearranging gives Dahan J, de dv 28 0) OF (eu) faa) Sal) (4.35) Detban [,, dz dv 634 (c,u) fn(2) fad) All of the terms inside the integrals of (4.35) are independent of the nodal solution and are available in the single-assembly calculation. Defining the cross section rehomogenization coefficient as FSn = [dy Zag(2s¥) O54 (2,0) fal) ald) 436 and the flux rehomogenization coefficient as an = [4240 054(2,9), fC) fald)s (43n the equivalent homogeneous cross sections for node ij can be expressed by the simple summation Fan Reomn roe Shek (4.38) Using (4.38) and rehomogenization coefficients from single-assembly calculations, the actual homogeneous cross sections can be computed if the flux expansion coefficients are known. In the nodal calculation, a non-linear iteration is used to solve for the expansion coefficients and an estimate of the flux expansion coefficients exists at each iteration. Therefore, at each iteration, the homogeneous cross sections are updated using the latest expansion coefficients, and as the flux solution converges, the homogeneous cross sections also converge [1]. Cross section rehomogenization provides homogeneous cross sections which, when used in conjunction with the corrected diffusion coefficients of (4.29) to (4.32), guarantee neutron balance for the assembly. Cross section rehomogenization was first developed by Smith [11] using a somewhat different form for the rehomogenized cross sections. In Smith's formulation, the denominator of equa- tions (4.33), (4.34), (4.35), and (4.38) is the average homogeneous flux and not the reconstructed heterogeneous flux. ‘The accuracy of cross section rehomogenization is demonstrated with a sample colorset problem. ‘The colorset problem consists of a controlied uranium and a regular uranium assembly arranged in a checkerboard pattern? Four sets of homogeneous cross sections are listed for this problem in Table 4.1. The first column in Table 4.1 contains cross sections from single-assembly calculations, the second column contains cross sections from rehomogenization, the third column contains cross sections from Smith’s version of rehomogenization, and the last column contains reference cross sections which are obtained by averaging the cross sections using the actual heterogeneous flux >This is Configuration CI ofthe NEACRP Benchmark Problem described in Chapter 5. 7 Table 4.1: Rehomogenized Cross Section Results ‘Smith's Cross Assembly Rehomogenized Rehomogenized | Reference Section | Values (error %) | _Values (error %) | Values (error %) | Values rolled Uranium Assembly Yar] 00012392 (0.29) | 0.013349 (-0.06) | 0.012365 (0.07) | 0.012356 Ear | 0.13862 (1.13) | 0.13699 (-0.06) | 0.13792 (0.62) | 0.13707 vEy, | 0,0045798 (0.17) | 0.0045887 (0.03) | 0.0045950 (0.16) | 0.0045875 Ep, | 0.11753 (0.29) | 0.11788 (0.01) | 0.11870 (0.70) | 0.11787 Ea |.0.019209 —(-0.06) | 0.019223 (0.01) | 0.019250 (0.15) | 0.019221 King [0.66029 (1.26) | 0.66927___(0.08) | 0.66937 (0.09) | 0.66874 ‘Uranium Assembly Eni | 0.009259 (0.16) | 0.092104 0.01) | 0.005294 (0.02) | O.00SDITT Zaz | 0.092663 (0.24) | 0.092437 (0.00) | 0.092545 (0.12) | 0.092438 vEy, | 0.0045700 (0.18) | 0.0045613_(-0.01) | 0.0045¢09 —_(-0.02) | 0.045618 vEp, | 0.11354 (0.32) | 0.11318 (0.00) | 0.11331 (0.11) | 0.11318 Za, | 0.020430 (0.04) | 0.020439 (0.00) | 0.020436 ——_(-0.01) | 0.020438 king | 0.99818 (0.03) | 0.99791 (0.00) | 0.99790 (0.00) | 0.99790 ‘The effects of cross section rehomogenization are most evident in the controlled uranium assembly where there isa large flux depression in the actual solution, The flux depression does not show up in the single-assembly calculation because the assembly is modeled as part of an infinite lattice. ‘The results of this problem indicate that the rehomogenized cross sections are more accurate than the single-assembly cross sections or the rehomogenized cross sections from Smith’s original method. 4.6 Baffle Homogenization For reflector nodes containing a core baffle, single-assembly calculations cannot be used to find equivalent homogenized parameters because no fissionable isotopes are present in the reflector. Instead, homogenized parameters for these regions are obtained with extended core calculations (25, 26, 47, 44, 48). The extended core calculation is a two-node problem which contains a single heterogeneous fuel assembly and heterogeneous baffle and reflector as shown in Figure 4-3. The boundary conditions used in the extended core calculation are zero-current on three sides and zero-flux on the far reflector side. The results from the extended core calculation are used to calculate equivalent homogeneous cross sections for the baffle/reflector region using (4.1) iscontinuity factors for the homogenized hafile region are obtained by calculating a homoge- ‘neous flux distribution in the baffle/reflector region and then by calculating the discontinuity factor at the interface as the ratio of the heterogeneous flux to the homogeneous flux. ‘The homogeneous flux distribution is calculated from a one-dimensional nodal expansion in the baffle/reflector region. The one-dimensional expansion is completely consistent with the two- 58 Fuel Region Reflector Region Figure 4-3: Extended core calculation for baffle homogenization. dimensional expansion derived in Chapter 2 and includes 5 polynomial fast flux expansion coeffi- cients, 5 polynomial thermal flux expansion coefficients and 2 hyperbolic thermal flux expansion coefficients. ‘The thermal polynomial coefficients in the one-dimensional nodsl expansion are solved directly from the thermal diffusion equation using the same technique as outlined in Appendix A. The remaining 7 coefficients are calculated using the following constraints ‘¢ known current for each group at the fueV/baffle interface (from the extended-core calculation), © zero-flux boundary condition for each group at the outer surface of the reflector, and © 3 weighted balance equations in the fast group Using the one-dimensional expansion coefficients, the homogereous flux can be solved for at the interface of the fuel and baffle and the discontinuity factor can be calculated as the ratio of the heterogeneous flux (from the extended core calculation) and the homogeneous flux (from the ‘one-dimensional nodal expansion), or ofa) 3A #8) = Frome (4.39) where 2; is the fuel/bafile interface. Baffle discontinuity factors found by this method will preserve the net current at the outer surface of the fuel as long as the appropriate boundary condition is applied to the outer surface of the baffle node. The zero-flux boundary condition used in the extended-core calculation and the ‘one-dimensional nodal expansion can be replaced by any appropriate boundary condition for the problem. 59 Equivalent homogenized baffle parameters from extended core calculations automatically ac- ‘count for the leakage and spectral shape effects between the fuel and baffle regions. Therefore, cross section rehomogenization and spectral corrections do not need to be applied to the bafife region. 4.7 Pin-Power Reconstruction ‘After the homogeneous flux has been calculated from a nodal calculation, flux superposition can be used to reconstruct the heterogeneous flux in each fuel pin. The homogeneous flux is taken directly from the nodal solution, and no further calculation is necessary. ‘Most modem nodal methods use a transverse leakage approximation and the only properties known about the intranodal flux shape are the one-dimensional flux expansions. However, the ‘one-dimensional flux shapes are decoupled and contain transverse leakage sources, and cannot be used to reconstruct the intranodal flux directly. Instead, a separate calculation must be performed in which the intranodal flux is usually calculated by fitting a non-separable expansion to each node using the surface-averaged flux and comer point flux values from the nodal solution as constraints. ‘The non-separable expansion is either a combination of polynomials and hyperbolic functions, or an analytic solutions to the local problem [16, 17, 18, 20, 19, 21] ‘Also with the transverse leakage approximation, the comer point flux values needed to fit the non-separable expansion to the intranodal flux shape are not avilable directly from the nodal solution, Therefore, an additional calculation must be performed to approximate the comer points. ‘Two examples of comer point interpolation include the method of Koebke and the method of Smith (16, 17, 18, 19}. ‘The two-dimensional flux expansion used in the nodal method is a major advantage over the transverse leakage approximation for pin-power reconstruction. This is because the comer point flux values are an integral part of the method and do not need to be approximated. Also, the homogeneous flux shape is taken directly from the nodal solution, and no interpolation is needed. ‘Therefore, the homogeneous flux used to reconstruct the pin-powers is fully consistent with the homogeneous flux from the nodal rnethod. Flux reconstruction can be simplified even further by noting that the fast flux shape is a relatively smooth function. Therefore, a total power form function (instead of groupwise form functions) can bbe used to reconstruct pin powers [18]. The power form function is taken from a power superposition approximation given by PHET (x,y) = PHOM (x,y) PSA(z, 9) (4.40) where the homogeneous power form function is defined as PHOM (zy) = Zf,™ (2, y) of OM (x, y) + EfOM (x, y) oHO™ (x,y). (441) In Chapter 6, spatially dependent cross sections will be introduced to correct for spectrum and leakage effects between assemblies. Power form functions allow the spatially dependent cross sections to be applied directly to the homogeneous power shape. 48 Summary ‘A method of obtaining equivalent homogenized parameters for heterogeneous fuel assemblies is presented in this chapter. Equivalent cross sections are defined from single-assembly calculations such that reaction rates and currents from an equivalent homogeneous solution preserve those from an equivalent heterogeneous solution. Discontinuity factors are derived from a basic assumption that the heterogeneous flux can be represented by a superposition of the homogeneous flux from the nodal solution and heterogeneous form functions from single-assembly calculations. Discontinuity factors are applied to the flux continuity equations in Chapter 2 to correct for heterogeneous materials. ‘The flux superposition is used to derive cross section rehomogenization, which generates equiv- alent homogeneous cross sections averaged with the actual heterogeneous flux shape found in the reactor, rather than the heterogeneous flux shape from single-assembly calculations. ‘The flux superposition assumption is then applied to define an adjusted homogeneous diffusion coefficient which, when used with rehomogenized cross sections, preserves neutron balance in an assembly. Equivalent homogenized parameters for baffle and reflector regions are derived which preserve leakage from the core when appropriate boundary conditions are applied. Homogenized baffle cross sections and discontinuity factors are derived from extended-core calculations. Finally, the flux superposition approximation is used to reconstruct individual pin powers from the homogeneous nodal solution using single-assembly form functions. 6 Chapter 5 Results 5.1 Introduction In Chapters 2 and 3, a nodal method was derived which uses a two-dimensional, non-separable expansion of polynomial and hyperbolic functions to represent the two-group flux. ‘The nodal method uses homogeneous cross sections and returns homogeneous flux distributions sized nodes. In Chapter 4, methods were outlined to obtain homogeneous cross sect single-assembly calculations, and to reconstruct the heterogeneous flux from the homogeneous nodal solution. In this chapter, these methods are applied to several benchmark problems to determine the accuracy of the methods. ‘The first benchmark problem is a simplified BWR with homogeneous fuel assemblies (LRA Benchmark). The second benchmark is a small PWR core with heteroge- neous fuel assemblies surrounded by a steel baffle and water reflector (EPRI-9 Benchmark). The third benchmark contains several colorset problems with heterogeneous uranium and MOX fuel assemblies (NEACRP Benchmark). 5.2 Computer Code ‘The nodal method developed in this thesis has been incorporated into a computer code called STENCIL. STENCIL solves the two-group, two-dimensional static diffusion equations in Carte- sian geometry using a two-dimensional, non-separable expansion of polynomial and hyperbolic functions. STENCIL is written in Fortran 90 (with some C system calls), and is written primarily for Unix workstations. It has been compiled and tested on several Unix workstations including an IBM RS/6000, an SGI Power Challenge Array, a Sun Sparcstation, and a DEC AlphaStation, All floating point calculations are performed in double precision in order to maximize accuracy and take advantage of modem computer architectures. 5.3 Comparison of Results ‘The benchmark results in this chapter are compared to reference calculations and/or higher-order spatial calculations. To compare the results, the reactor eigenvalue, assembly powers, and individual 62 pin powers are examined. Several measures of error are used in the comparisons. These include the absolute error of the power in an assembly (or pin) labeled i as defined by 6.1) the relative error defined by (5.2) the maximum error defined by émaz = max {l«il} (5.3) and the average error defined by Dial-vi (5.4) yr where P* is the calculated power for assembly (or pin) i, Pj. is the reference power for assembly (or pin) i, Ppe7 is the average power for the entire problem, and V' is the volume of assembly (or pin). The error in the reactor eigenvalue is not normalized and is defined simply as the difference between the calculated eigenvalue and the reference eigenvalue, or A= Apes. (5.5) ‘The eigenvalue errors are given in pcm which stands for “percent mille” and is equal to 10~°. 63 5.4 LRA 2-D BWR Static Benchmark ‘The first benchmark examined is the two-dimensional LRA BWR benchmark problem from the Argonne Benchmark Book [49]. ‘This benchmark consists of a two-dimensional BWR with 312 fuel assemblies surrounded by a water reflector, as shown in Figure 5-1. The boundary conditions for the problem are zero-current on the left and bottom sides of the core and zero-flux on the top and right sides of the core. Y (cm) 16s 150 13s J=0 Figure 5-1: Two-Dimensional LRA BWR Benchmark Problem. Each fuel assembly in the core is considered a homogeneous material and the two-group cross sections are listed in Table 5.1. The width of each assembly is 15 cm and an axial buckling of 10-* cm? is used in both energy groups to account for leakage in the axial direction. Nine control blades 64 are present and are modeled as smeared absorbers in adjacent assemblies, Table 5.1: Assembly cross sections specified in LRA BWR benchmark problem, Region | Assembly Type | Group | Dy | =n Tap vy, 1 Fuel 1 t 1.2550 | 0.02533 | 0.008252 | 0.004602 (controlled) 2 0.2110 0.100300 | 0.109100 2 Fuel 1 1 1.2680 | 0.02767 | 0.007181 | 0.004609 (uncontrolled) | _2 | 0.1902 0.070470 | 0.086750 3 Fuel 2 1 1.2590 | 0.02617 | 0.008002 | 0.004663 (controlled) 2 0.2091 0.083440 | 0.102100 4 Fuel 2 T 1.2590 | 0.02617 | 0.008002 | 0.004663 (uncontrolled) |_2 | 0.2091 0.073324 | 0.102100 3 Reflector 1 | 1.2570 | 0.04754 | 0.006034 | 0.00 2_| 01592 o.o191100 | 0.00 Several nodal methods have been applied to the LRA benchmark. ‘These methods include the Analytical Nodal Method (ANM) used in QUANDRY [1] and the Nodal Expansion Method (NEM) used in both QUAGMIRE [29] and CONQUEST [30]. Both ANM and NEM use a quadratic transverse leakage approximation to obtain one-dimensional flux shapes. The one-dimensional flux shapes are solved analytically in ANM and are approximated with fourth-order polynomials in NEM. ‘The eigenvalue and assembly power results are compared in Tables 5.2 and 5.3. Table 5.2 lists results using | node per assembly (1 npa), and Table 5.3 lists results using 4 nodes per assembly (4 1npa). The normalized assembly power distribution is shown in Figure 5-2 along with the errors from ‘STENCIL using 1 and 4 nodes per assembly. The accepted reference solution for this benchmark is a nodal calculation by Shober using a spatial mesh of 16 nodes per assembly [50]. Table 5.2: Results for 2-D LRA BWR benchmark problem using | node per assembly. Parameter__| STENCIL | CONQUEST | QUAGMIRE | QUANDRY | x 0.996304 | 0.996329 | 0.996328 | 0.99641 © (pem) “6 -3 -3 +5 maz (%) 0.65 LAs 1.18 0.30 (%) 0.16 0.37 0.39 0.07 Gretmaz (%) | 0.46 1.36 141 0.19 Fat (%) 0.15 0.40 0.42 0.07 outer iterations | 6 2 4 4 run-time? 219 0.6 = = * Reference: 0.99636 [50] ° Execution time in seconds on a DEC AlphaStation 400 4/233 65 Table 5.3: Results for 2-D LRA BWR benchmark problem using 4 nodes per assembly. Parameter_[ STENCIL | CONQUEST | _ QUAGMIRE * 0.996370 | 0.996362 0.996376 © (pem) +1 0 +2 émaz(%) | 0.37 0.18 0.19 €(%) 0.04 0.05 0.04 fretmaz (%) | 0.28 0.17 0.14 Fel (%) 0.03 0.05 0.04 time? 93.4 Ll = * Reference: 0.99636 [50] ® Execution time in seconds on a DEC AlphaStation 400 4/233 All of the nodal methods give very good results for this problem. The range of eigenvalues from all of the methods is only 11 pem with I node per assembly, and 2 pem with 4 nodes per assembly. This shows that all of the codes are solving the same problem. ‘The most accurate assembly power results are given by QUANDRY. Since the only approxim: tion made in QUANDRY is a quadratic transverse leakage approximation, this shows a quadratic transverse leakage approximation is reasonable for this type of problem. ‘The second best assembly power results are with STENCIL and then with the NEM codes. Since the quadratic transverse leakage gives good results in the QUANDRY solution, the differ- ence between STENCIL and NEM is most likely because of the semi-analytic expansion used in ‘STENCIL compared to the polynomial expansion used in NEM. € os242 | 08672 | 08268 | osx0 | 093s | osnis | osss «@ o27 | om | oo8 | 006 | 017 | 025 | -o2s J -oos | oos | oor | oo | om | o | ais sas fotze foams | azar | sae | 6 | von | ee os | 02 | oo | oo | 2 | o2 | os | 06 oo | 01 | oo | oo | or | or | on | os to] iast | 09667 | 1022 | 1339 | 20st | 2161 or | 02 | os | on | or | os | 06 o1 | 00 | oo | 00 | oo | 0 | 00 tai | 9398 | 0.7826 | ose | 11s. | 1asz 01 | 017 | ow | a | 00 | os 2 | oo | oo | oo | 00 | oo 0.7902 | osmos | osisi | os7ez | oxoss 016 | 009 | oor | 003 | oor oor | oo | oo | oo | oo costs | 0490 | oasar | ossze o1s | 008 | 006 | oos 00 | oo | oor | oo os130 | 04067 | 04240 ors | 009 | oo8 A erage 00 | oo | oo Awol oasor | 0.3995 Reference Value 1,000 01 | 010 I npaerror(%) |] 016 00 | -o02 4 apaeror %) |] 0.04 siz 006 00s Figure 5-2: Normalized assembly power distribution and errors in STENCIL solution for 2-D LRA. Benchmark. 67 5.5 EPRI-9 Benchmark ‘The second problem examined in this chapter is the EPRI-9/9R benchmark. EPRI-9 is a small PWR quarter-core composed of 9 fuel assemblies surrounded by a stee! baffle and a water reflector [26}. EPRI-9R is a variation of EPRI-9 with control rods inserted into the central fuel element. Both core configurations are shown in Figure 5-3. Nn € 60 - o Water Reflector 45 FL FI o=0 F2 FI FI 1s | | — Baffle F2F2R | F2 FI 0 > 0 15 30 45 60 J=0 Figure 5-3: EPRI-9/9R core configuration, The fuel assemblies in this problem are a standard PWR design with 15x15 fuel pins as shown in Figure 5-4. Three types of fuel assemblies are present. These include two assemblies of different enrichments (FI and F2), and a rodded version of the second enrichment (F2R). The two-group cross sections for the assembly pin-cells are listed in Table 5.4. ‘This purpose of this benchmark is to model fuel assemblies on the outer edge of a core and methods of treating the baffle. 68 at ": a BBE water Hote (Assemblies Ft and F2) Control Rod (Assembly F2R) Fuel | (Assembly Fl) Fuel 2 (Assemblies F2 and F2R) ‘Assembly Dimensions 21 em x 21 em Pin cell dimensions 1.4.em x 1.4.cm Figure 5-4: Fuel Assembly for EPRI-9 benchmark problem. 9 ‘Table 5.4: Pin-cell cross sections specified in EPRI-9 benchmark problem. Pin Type | Group | Dy | 221 | 2p | Esa Fuel 1 T [1.500 | 0.0200 | 0.0130 | 0.0065 2__ | 0.400 0.1800 | 0.2400 Fuel 2 T_ | 1.500 | 0.0200 | 0.0100 | 0.0050 2__| 0.400 0.1500 | 0.1800 ‘Water T | 1.700 | 0.0350 | 0.0010 | 0.0 2 | 0.350 0.0500 |_ 0.0 Baffle T |1.020{ 00 [0.0032 [00 2 | 0.335 0.1460 |_ 0.0 Control Rod | 1 | 1.113 [0.0038 | 0.0800 | 0.0 2 | 0.184 0.9600 | _ 0.0 70 5.5.1 Single-Assembly Calculations Homogenized cross sections and discontinuity factors for this problem are obtained from single- assembly calculations run for each type of fuel assembly. The single-assembly calculations are performed with STENCIL using | node per pin and zero-current boundary conditions. The results from the single-assembly calculations are shown in Table 5.5. DR, refers to the face discontim factor and DC, refers to the comer point discontinuity factor using { node per assembly. ‘The homogenized baffle/reflector cross sections and discontinuity factors are from an extended core calculation as described in Section 4.6. ‘Table 5.5: EPRI-9 Assembly Calculation Results. ‘Assembly Type Property [Fi F2 F2R_ | Baffle Ring | 1.006603 | 0.959145 | 0.656523 | N/A Dy | 1.513361 | 1.513355 | 1.462630 | 7.350955 Dz _ | 0.395034 | 0.395156 | 0.381349 | 0.348301 Za1__ | 0.012099 | 0.009325 | 0.015144 | 0.001853 Za2 | 0.168560 | 0.141419 | 0.183746 | 0.060460 Eq) | 0.021126 | 0.021125 | 0.018810 | 0.021436 vE,y, | 0.006012 | 0.004625 | 0.004633 | 0.0 vE yr _| 0.218881 | 0.164554 | 0.172501 | 0.0 DF, _ | 1.002984 | 1.003623 | 1.027292 | 1.159791 DC; _| 1.603601 | 1.004699 | 1.045404 | 1.159791 DF: | 0.929192 | 0.938652 | 1.199483 | 0.287086 DC2_| 0.913828 | 0.923939 | 1.261256 | 0.287086 ‘The discontinuity factors to use in the 4 nodes per assembly (4 npa) calculation can be illustrated with Figure 5-5. This figure shows a fuel assembly with octant symmetry and the nodes used with the 4 npa calculation. The following discussion does not apply when using the adjusted current model or for nodes without octant symmetry. n A B Cc Figure 5-5: Notation used for describing discontinuity factors in calculations with 4 nodes per assembly. For the surfaces on the exterior of the assembly (surfaces AB and BC), the discontinuity factors are the same in both I npa and 4 npa. This is because surfaces AB and BC are identical in octant symmetry. For the surfaces in the interior of the assembly (surface BD), the discontinuity factors for the nodes on both sides of the surface are equal (due to symmetry), and the discontinuity factors cancel from the flux continuity equations. Therefore, any arbitrary discontinuity factor can be used for this, surface. (This is not true if adjusted currents are used.) For the discontinuity factors at the comers of an assembly (comers A and C), the discontinuity factors are the same in both | npa and 4 npa. Comer B has a unique discontinuity factor which must be obtained from the single-assembly calculation. For comer D, the comer discontinuity factor is the same for all four nodes surrounding the comer, and thus, any arbitrary value can be used for the discontinuity factor. ‘The preceding discussion does not apply when the adjusted current model (ACM) is used. This is because, in ACM, the discontinuity factors also appear in the diffusion coefficients. n 5.5.2 EPRI-9 Benchmark Results ‘The eigenvalue and assembly power results for the EPRI-9 and EPRI-9R benchmarks are listed in Table 5.6. Results are listed using the codes STENCIL, QUANDRY, and CONQUEST. The normalized assembly power distribution and STENCIL errors are shown in Figures 5-6 and 5-7. ‘The reference results are from heterogeneous STENCIL calculations using 1 node per pin ‘The QUANDRY results are reported by Khalil in Reference [26]. In the QUANDRY results, the homogenized cross sections and discontinuity factors for the fueled regions were obtained from single-assembly calculations. The cross sections and discontinuity factors for the homogenized baf- fle were obtained from olorset calculations for each unique reflector configuration. The QUANDRY calculations use 4 nodes per assembly and are relative to QUANDRY reference solutions. ‘The STENCIL and CONQUEST results are obtained using homogenized cross sections and discontinuity factors from Table 5.5. ‘STENCIL results are listed using the normal homogenization method (equivalent homogenized cross sections and discontinuity factors) and with rehomogenized cross sections and adjusted cur- rents (RXS and ACM). The normal homogenization results should be used when comparing results between codes. Table 5.6: EPRI-9 Results. EPRE9 EPRIOR Solution Method || ¢{ [emer] @ || & [mar] @ ‘STENCIL (I npa) [| -23 | 0.58 | 0.33 || -106 | 0.36 [0.22 ACM +7 | 0.58 | 0.36 || -88 | 0.36 | 0.12 RXS -29 | 0.58 | 0.31 || 69 | 0.87 | 0.46 RXS, ACM +1 | 059 | 0.35 }) -s5 | 1.05 | 037 CONQUEST +14 | 0.33 | 0.18 || +23 | 1.66 | 0.42 STENCIL G npa) || -32 | 0.54 [0.26 || +140 | 0.60 [0.29 CONQUEST -31 | 0.48 | 0.24 | -83 | 1.12 | 0.37 QUANDRY. +47| 0.91 | 0.41 || +91 | 069 | 0.34 * EPRI-9 reference eigenvalue: 0.92755 © EPRIOR reference eigenvalue: 0.88873 With the normal homogenization option, STENCIL gives the lowest assembly power errors in EPRI-OR, and CONQUEST gives the lowest assembly power errors in EPRI-9. Adjusted currents (ACM) give better results in STENCIL, but rehomogenized cross sections (RXS) do not. Comparing the actual homogenized cross sections (averaged with the pin-by-pin results), the rehomogenized cross sections, and the single-assembly cross sections; it can be seen that the rehomogenized cross sections are more accurate than the single-assembly cross sections. ‘Therefore, the additional error caused by rehomogenized cross sections is most likely due to the removal of a cancellation of error term. BB 1.2063 | 06096 | Reference vale 008 | 053 | emor(m) 007 | -057 | error with ACM (%) € vor | 054 | enor with RXS(%) 1000 | 059 | error with RXS and ACM i%) Average tage | 12173 | oss 1.0000 +058 | 1037 | 009 033 voss | soa | 012 | node/assembly 036 voss | +034 | 007 031 ose | soar | on 035 € Figure 5-6: Normalized assembly power distribution for EPRL-9 benchmark problem. « 1.4009 | 09208 | Reference value 4036 | 032 | enor) 40.36 | 028 | enor with ACM (%) & 40.21 0.63 | error with RXS (6) 4022 | 056 | enorwithRXSand ACM) Average 05296 | 11388 1.0000 o2 | soz oz voor | +000 | node/assembly on war | 4039 046 suos | out o37 & Figure 5-7: Normalized assembly power distribution for EPRI-R benchmark problem. 74 5.6 NEACRP Benchmark ‘The third benchmark examined is the NEACRP benchmark calculation of power distributions within assemblies [51]. This problem consists of 3 types of heterogeneous fuel assemblies arranged in 5 mini-core configurations. The fuel assemblies include a uranium assembly, a controlled uranium assembly, and a MOX assembly. The five specified core configurations are checkerboard colorset problems as shown in Section 5.6.2. In addition to the specified configurations, three other core configurations are examined. These configurations are not checkerboard patterns and are typical of configurations found in a real reactors. ‘The additional configurations are labeled C6 to C8 and are also shown in Section 5.6.2. ‘The fuel assemblies in this benchmark are a typical PWR design with 17x17 fuel pins per assembly. Each assembly contains 24 water holes and a central fission chamber. The uranium assemblies are composed of a single fuel enrichment and the MOX assembly has 3 fuel enrichments, The controlled uranium asseinbly (UA) is identical to the regular uranium assembly (UX) with the exception of control rods inserted into the water holes. ‘The uranium assemblies are shown in Figure 5-8 and the MOX assembly (PX) is shown in Figure 5-9. ‘The two-group cross sections for the assembly pin-cells are listed in Table 5.7. Except for the thermal diffusion coefficient in the reflector, all of the group diffusion coefficients are the same. Therefore, the flux weighted homogeneous diffusion coefficients are equal to the heterogeneous diffusion coefficients. 5 met oa BBL Wace tote 00 or 0x sendy, Absobe-Pin(A) or UA Assembly Fission Chamber (C) [] vo2 Fuet Pin uy Assembly Dimensions 21,42 em x 21.42.¢m Pin Cell Dimensions 1.26 em x 1.26 em Figure 5-8: UX and UA Fuel Assembly, 16 a am U : oe | sme ee r r Bh veces 00 Lo Eahed MX Pi BB Fission chamber (cy Medium Enriched MOX Fuel Pin (P2) Low Enriched MOX Fuel Pin (P1) Assembly Dimensions 21.42 em x 21.42 em Pin Cell Dimensions 1.26 em x 1.26em Figure 5-9: PX Fuel Assembly. 7 ‘Table 5.7: Pin-cell cross sections specified in NEACRP benchmark problem. Pin Type. Group [Dy | Zap | By | Bn U~ UO; Fuel T | 1.2 | 0.010 | 0.0050 | 0.020 2_| 0.4 | 0.100 | 0.1250 PI Peripheral MOX Fuel T_ | 12] 0015 | 0.0075 | 0.015 2__| 0.4 | 0.200 | 0.3000 P2 Intermediate MOX Fuel 1 | 12] 0015] 6.0075 | 0.015 2} 04 | 0.250 | 0.3750 3 Central MOX Fuel T_ | 1.2 | 0.015 | 0.0075 | 0.015 2_| 04 | 0.300 | 0.4500 X Guide Tube 1 [12[ooor| 0 [0025 2 |04|0020| 0 R Reflector T |12]o0or] 0 | 0.050 2_|02|0040| 0 Movable Fission Chamber | 1 [1.2 [0.001 | TE-7 [0.025 2_| 0.4 | 0.020 | 3.86 ‘A Absorber (AIC) T [12[004% | 0 [0010 2 |o4}o800] 0 8 5.6.1 Single-Assembly Calculations Homogenized parameters for this problem are obtained from single-assembly calculations run for each type of fuel assembly. Single-assembly calculations are performed in STENCIL using | node per pin and zero-current boundary conditions. The results from the single-assembly calculations are shown in Table 5.8. DF, refers to the group face discontinuity factor and DC, to the comer point discontinuity factor with I node per assembly. Table 5.8: Assembly Homogenization Results Homogenized ‘Assembly Type Parameter_[ UX PX UA King (0.998181 | 1.026669 | 0.660288 Di 1.200000 | 7.200000 | 1.200000 D2 0.400000 | 0.400000 | 0.490000 Ear 0.009226 | 0.013791 | 0.012392 Ean 0.092663 | 0.231691 | 0.138618 Ey 0.004570 | 0.006852 | 0.004580 vEy. | 0.113537 | 0.344583 | 0.117526 Zar 0.020430 | 0.015864 | 0.019209 DF; ¥,004803 | 0.994570 | 1.052142 Dc; 1006895 | 0.991336 | 1.104590 DF: 0.951007 | 1.041629 | 1.282139 DC, __| 0.930603 | 1.046322 | 1.451569 The fuel assemblies in the NEACRP benchmark have octant symmetry, and the discussion on which discontinuity factors to use with 4 nodes per assembly, from Section 5.5.1, applies. 19 5.6.2 NEACRP Benchmark Results This section compares the NEACRP benchmark results calculated by STENCU. to the nodal codes PANTHER [52], SILWER [19], CONQUEST, and SIMULATE-3. PANTHER results are obtained from Reference (52). PANTHER is a nodal code from Nuclear Electric which uses the ANM. All of the PANTHER results are relative to the PANTHER reference solution. ‘The SILWER results are reported by Grimm in Reference [19]. SILWER is the nodal code in the ELCOS Core Analysis System used at the Paul Scherrer Institute. The purpose of Reference [19] was to test different methods of reconstructing the heterogeneous flux from the nodal solution. ‘The two methods examined by Grimm are the methods due to Koebke (Method K) and to Smith (Method S). These methods are described in Section 4.7. Grimm does not report eigenvalue or assembly power errors, and all of the pin power errors are relative to Grimm's reference solution. ‘The CONQUEST, SIMULATE-3, and STENCIL results were obtained using the homogenized ‘cross sections and discontinuity factors from Table 5.8. Both CONQUEST and this version of SIMULATE-3 use the NEM [30, 53]. CONQUEST does not have a repeating boundary condition ‘and some of the configurations could not be run with I node per assembly. SIMULATE-3 does not have a zero-flux boundary condition and could not run configurations C4 and C5. Pin power reconstruction was not performed for either the CONQUEST or SIMULATE-3 results. The reference solution for each configuration is a heterogeneous calculation performed with STENCIL using | node per pin. The reference eigenvalues and normalized assembly powers for each configuration are listed in Table 5.9. STENCIL results are reported with and without the adjusted current model (ACM) and reho- ‘mogenized cross sections (RXS). To be consistent with the other codes, the regular option should be used when comparing results between codes, However, it will be shown that ACM and RXS give the best results compared to reference solutions. Tables 5.10 to 5.25 list the eigenvalue, assembly power, and relative pin power results for each configuration. In configurations C4 and C5, the relative pin power error is not meaningful for the low power assemblies. This is because the absolute power in these assemblies is small, and therefore a small error in absolute power leads to a large error in the relative power. For these configurations, it is ‘more meaningful to look at the absolute error in pin power or the relative error in only the peak pin of the configuration. The relative error in the peak pin of each configuration is listed in Table 5.27. For the cases using the regular homogenization method and | node per assembly (I npa), STENCIL gives the lowest maximum assembly power error in every configuration except C5. The ‘maximum assembly power error from STENCIL is 1.81% in a controlled uranium configuration (C6), and 0.26% in a MOX configuration (C5). STENCIL also gives the lowest maximum relative pin power error in each configuration, except for configuration C2. The maximum relative pin power error from STENCIL in all configurations is 4.5% (C6). ‘The maximum relative pin power RMS error is 3.1% in a controlled uranium configuration (C1) and 1.0% in a MOX configuration (C2). ‘The best results are obtained when ACM and RXS are applied to the problem. Both the assembly power and RMS pin power error is improved. The maximum assembly power error drops to 0.35% in a controlled uranium configuration (C1) and 0.10% in a MOX configuration (C7). The maximum pin power RMS error drops to 1.1% in a controlled uranium configuration and remains at 1.0% in a 80

Das könnte Ihnen auch gefallen