By Mladen Chargin
CDH AG Advanced Engineering Am Marktplatz 6 79336 Herbolzheim Germany Phone: Fax: E-Mail: URL: +49 (0) 7643-9377-26 +49 (0) 7643-9377-29 david.bella@cdh-ag.com http://www.cdh-ag.com
CDH Aktiengesellschaft Vorstand: Dr Leo W. Dunne Aufsichtsratsvorsitzender: Dr Theodor Seitz Sitz und Registergericht: Augsburg, HRB 19468
November 2008
Table of Contents
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. Introduction.......................................................................................................... 3 Background.......................................................................................................... 3 EIGRL Specification............................................................................................ 4 Accuracy Issues ................................................................................................... 4 Accuracy Issues (solid models) ............................................................................ 6 Enforced Motion .................................................................................................. 8 Use of CDH/AMLS for Component Modes Synthesis (CMS) Calculation............ 10 Structural Damping Definition ............................................................................. 10 Modal/Grid/Panel Participation Factor Calculation............................................... 11 Residual Vector Issues ....................................................................................... 11 CDH/AMLS SMP Parallel ................................................................................. 13 I/O Enhancement for IBM and SGI Computers .................................................. 13 Use of MFLUID................................................................................................. 14 Rigid Body, Massless Mechanism Modes, and Singular Stiffness....................... 16
References: ................................................................................................................. 19 Appendix A: Enforced Motion for Modal Frequency Response................................ 20 Appendix B: CDH/AMLS Parameter List ................................................................ 22
CDH/AMLS
1. Introduction
This document will describe the software product CDH/AMLS (Automated Multi-Level Substructuring), its interaction with Nastran, and the most effective way to use it in a variety of analysis situations. CDH/AMLS supports all the modal solution in Nastran, sol103, 110, 111, 112 (normal modes, modal complex eigenvalue, modal frequency response, modal transient response) as well as optimization, sol200. A number of references listed at the end of this document may be of interest to some users. They may be obtained by request to mladen.chargin@cdh-ag.com. Another purpose of this document is to describe some of the parameters that may be specified by the user. For a typical response analysis, the user needs to be concerned about very few details. However, there are some specific cases that one must be aware of, because the solution depends on a proper understanding of these cases. Any discussion of CDH/AMLS capability relates to version 3.2.r159 (even though some, but not all, were included in earlier versions) or later. The only necessary requirement is that the CDH/AMLS Delivery Data Base be used. CDH provides the customers with the appropriate DB for MSC (2001, 2004, 2005,, 2007, 2008, MD) and NX (5, 6) Nastran versions for various computers. CDH/AMLS is a very efficient approximation method for analyzing vibration of large finite element models of structures. This fact is evident from the speed with which CDH/AMLS has been adopted by nearly the entire international automobile industry within the past several years.
2. Background
CDH/AMLS begins by dividing a finite element model into two substructures, then dividing each of these into its own substructures, and recursively continuing until thousands of substructures have been defined in a tree hierarchy. Then modes of substructures, whose natural frequencies are below a certain cutoff value, are computed and the model is transformed from the finite element representation to one in terms of substructure modes using the Craig-Bampton method. Modes of the overall structure are approximated in terms of the substructure modes. Finally, the response of the overall structure is represented in terms of these approximate global modes, and exact static response vectors if the residual flexibility technique is used.
CDH/AMLS There are only two required items that the user must provide in the Nastran input file, an include statement in the FMS Nastran section and an EIGRL data in the Bulk Data deck. There are other user options which will be discussed in subsequent sections. The include statement is: include /full/path/name/amls_assign The file, amls_assign, is part of the AMLS delivery package. EIGRL requirement is discussed in the next section. In order to facilitate its adoption as a production analysis tool, CDH/AMLS has been implemented in a form that is as transparent to users as possible and requires very little training for productive use. However, some explanation of the issues related to various aspects of CDH/AMLS, such as accuracy, will be helpful for users and will be discussed in Sections 4 and 5.
3. EIGRL Specification
The only requirement for using AMLS (besides the use of the AMLS Delivery Data Base) is that one must define an EIGRL in the Bulk Data deck and request it in the Case Control deck. The reason for this is that AMLS needs to know the cutoff frequency below which it will calculate the eigenvalues and eigenvectors. The necessary value is the F2 entry on the EIGRL. Only specifying the ND value (number of desired roots) is unacceptable and CDH/AMLS will issue a fatal message.
4. Accuracy Issues
The accuracy of the CDH/AMLS approximation depends mainly on the cutoff frequency used for the substructure modes calculation. This cutoff frequency is determined as the product of the cutoff frequency for the global modes (set as the EIGRL upper frequency limit F2) and the Bulk Data parameter ss2gcr (substructure-to-global cutoff ratio). param, ss2gcr, 5.0 [default] ss2gcrs default value of 5.0 can be changed by adding the line param, ss2gcr, 7.5, for example, to the Bulk Data file, to increase accuracy. The default value has been found to give excellent accuracy in typical structures (such as car bodies) composed of mostly shell elements and where the structure behaves as a shell structure. A later section will
CDH/AMLS describe the appropriate value of solid structures. Increasing ss2gcr to improve accuracy only modestly increases the cost of analysis. The lowest-frequency global modes are approximated very accurately by AMLS, but the highest-frequency global modes are less accurate because the truncation of substructure modes has the greatest effect at the highest frequencies. The accuracy of the global modes whose natural frequencies are within the range of excitation frequencies is more important to the accuracy of the computed frequency response than the accuracy of the global modes above the highest excitation frequency. Typically, global modes are computed to a cutoff frequency fifty percent higher than the highest excitation frequency when residual flexibility is used. Then the highest computed global modes are far enough from resonance that the effect of the small error in their natural frequencies is difficult to detect in the frequency response functions. If only transfer mobilities are needed, keep in mind that residual flexibility does not improve the accuracy of transfer mobilities nearly as much as it improves the accuracy of point mobilities. CDH/AMLS makes it possible to analyze complex structures at much higher frequencies than were feasible with previous methods. An important point to remember, however, is that there are so many modes involved in the response at higher frequencies that individual modes do not have as much influence on the shape of the frequency response functions as at lower frequencies. CDH/AMLS produces Rayleigh-Ritz approximations of the natural frequencies and modes, so the approximate natural frequencies are always greater than or equal to exact ones. Because natural frequencies are very closely spaced at higher frequencies, Rayleigh-Ritz approximations of the exact modes tend to be combinations of the modes that are very close in frequency. This makes it difficult to match approximate global modes with exact global modes. The use of MAC (Modal Assurance Criteria) to compare the exact and AMLS mode shapes is discouraged because it leads to very erroneous results. A much better technique using SVD (Singular Value Decomposition) is a more appropriate method of comparing these two sets of mode shapes. Nevertheless, the computed frequency response can be very accurate even if individual modes do not agree well with the exact modes, because the modes combine to form the frequency response. It is more important to approximate well the subspace of global modes in a particular frequency range than to approximate the individual modes very
CDH/AMLS accurately. Another way in which CDH/AMLS differs from the Lanczos eigensolution method is that with the Lanczos method, if the modes in a narrow frequency range are needed, it is more economical to solve for only the modes in that range and not the modes below that range in frequency. In contrast, with AMLS, very little is saved by not calculating the modes whose natural frequencies are below the lower frequency limit (the EIGRL frequency limit F1) of the frequency range of interest. Therefore, it is generally more advisable to set the lower frequency limit to zero (or not specify it at all, i.e., leave it blank). Because of the importance of residual flexibility to frequency response accuracy, CDH/AMLS computes exact static responses for residual flexibility calculations. If the structure is unrestrained, it is not necessary when using AMLS to supply Nastran support information (in fact it should not be used), because it is ignored by AMLS. It should be noted that AMLS takes the applied loads, residual loads, and requested response points into consideration when defining the substructure tree to improve accuracy. As a result, some slight differences may be observed between results from different CDH/AMLS analyses if there are differences in the definitions of applied loads or the selection of response points. One can also improve the accuracy by simply increasing the value of F2 on the EIGRL. Thus, if one uses a rule-of-thumb of requiring modes 1.5 times the highest excitation frequency, simply use a value of 1.75, for example. This will only slightly increase the CDH/AMLS CPU time as will be seen in the next section. However, this approach will also increase the calculation time of the modal frequency response, especially if the modal equations are fully coupled.
CDH/AMLS increasing the value of ss2gcr has a small influence on the overall solution time. This is owing to the fact that in Craig-Bampton method, the substructure modes consist of two distinct sets of shapes. The constraint modes are obtained by a static analysis and the dynamic modes are obtained by an eigenvalue analysis of each substructure. Note that the constraint modes must be calculated irrespective of the excitation frequency of interest. In true 3D structures, as compared to 2D structures, the calculation of constraint modes dominates the solution time, thus it is does not make much difference if the value of ss2gcr is 5, 10 or 15. However, the result accuracy improves with the larger value of ss2gcr. At some value one reaches a point of diminishing returns, thus a good compromise value should be between 10 and15.
SS2GCR 15 10 7 5 7 7 Lanczos ROOTS 270 268 266 261 854 1387 271 ELAPSED(min) 296:41 287:49 283:33 281:20 296:221 307:412 363:47
Fmax=15810 (3x max freq of excitation) Fmax=21080 (4x max freq of excitation)
As was already mentioned, an alternative to increasing the ss2gcr value is to simply increase the F2 value on the EIGRL. The following example illustrates the computational cost associated with the increase in the value for F2. All jobs run on an IBM P5 1.9GHz, 4CPUs, and 4GB memory. Problem size: 2348611 Grids, 6.89M DOFs Engine Block Model.
CDH/AMLS
Tota l Elap sed (min ) 248:59 259:33 291:35 448:55 430:28 943:05
Clearly, by increasing the number of modes from 237 to 5704 (a factor of ~24) the elapsed time increases less than a factor of two. Similarly, for modes below 2500 Hz, CDH/AMLS calculates only two modes less than Lanczos and for modes below 5000 Hz, eight modes less. Also note that more than doubling the number of modes from 237 to 672 increased the elapsed time only ten minutes. It was simply impossible to calculate modes at the other two cutoff frequencies (10000 Hz and 20000 Hz) with Lanczos because the computation time would have been beyond any reasonable wait time.
6. Enforced Motion
In Nastran, there are a number of ways of specifying enforced motion. The original schemes from the beginning days of Nastran were the Large Mass/Spring methods and the Lagrange Multiplier method. Large Mass (LM) method was the preferred choice. However, in the early 1990s CDH discovered that the Large Mass method, in combination with Lanczos (the preferred eigenvalue extraction technique), lead to unreliable dynamic response results. Consequently, CDH developed an exact technique that was eventually incorporated into both MSC and NX Nastran. This method required an additional decomposition of the stiffness matrix, a FBS (Forward-Backward Substitution) with the number of right-handsides equal to the number of enforced DOFs, as well as a number of matrix multiplies
8
CDH/AMLS (see Eqs. (4) and (7) in Appendix A). The algorithmic details can be seen in Appendix A. However, since the advent of CDH/AMLS it became very obvious that the use of LM for enforced motion is again the preferred choice owing to the fact that it has absolutely zero additional computational cost. The reason that the LM technique failed is that Lanczos was used to extract the modes. In Lanczos the default block size is seven. This number was derived based on the expected multiplicity of eigenvalues, six being a typical number because of six rigid body modes. However, when using the LM method, the multiplicity of eigenvalues may be relatively large, e.g., 40 or more. Thus, when using Lanczos in Nastran, the results may be unpredictable. CDH/AMLS does not use the Lanczos method; rather it uses the Householder method whose accuracy is not sensitive to the multiplicity of eigenvalues. This document will not describe the LM method because it is already well documented in various Nastran publications, such as Dynamics Handbook. The only remaining question concerns the size of the large mass used in LM method. In the past, it was recommended that the value be between 103-106 times the total mass of the structure. This value took into consideration the fact that Lanczos was being used. When using CDH/AMLS, the recommended value is anything greater than 106 (no problems were encountered even when using a value of 1010). Another item of interest when using the LM technique occurs when the user specifies or requests residual vectors by defining (param, resvec, yes) or resvec = yes (MSC V2004+). By default, the residual vectors are obtained from the applied loads, which in the case of enforced motion, refers to all the loads on the large masses. If one simply accepts the default, then one would obtain two sets of almost identical residual vectors. The first set would correspond to the zero frequency mode shapes for each large mass and the second set would be obtained from the static analysis subject to applied loads at the large masses. The duplicity of almost identical residual vector could cause numerical difficulties. To avoid this condition, a user defined Bulk Data parameter, lgmass, can be specified. The function of this parameter is to eliminate any loads on DOFs which have a mass larger than the value of lgmass during the calculation of residual vectors,
param, lgmass, 1.e+5 [default]
The user will be informed regarding which residual loads have been eliminated by a message in the f06 file such as:
CDH/AMLS
================================================ RESIDUAL VECTOR DOFS THAT WILL BE ELIMINATED BECAUSE THEY HAVE A LARGE MASS ================================================ MATRIX MSVEC POINT VALUE POINT COLUMN 1 1001 T1 1.00000E+08 1001 T3 1504 T1 1.00000E+08 1504 T3
In the above example, the value of 1.000000E+08 represents the value of the large mass at the particular grid and DOF and since the default value of lgmass is 1.e+5, the loads on these DOFs have been removed during the residual vector calculation.
10
CDH/AMLS This operation may be quite expensive, depending on the number of non-zero columns in K4AA. If one had a single MAT1 record in the Bulk Data and that particular MAT1 had a ge specification, then the K4AA matrix would look just like the structural stiffness matrix KAA and the above operation would consume significant CPU time. To avoid this situation the following approach is highly recommended. Find all the MAT1s that have the same value of ge or rather that set of MAT1s that provide structural damping for the largest number of DOFs with the same value of ge . Delete that value of ge (assume it is x.xx) from that set of MAT1s and specify a parameter, param, g, x.xx For any other value of ge specified on other MAT1s or CELAS2, PELAS, or PBUSH, replace their ge value by ( ge x.xx).
11
CDH/AMLS of residual vectors, which by default is 500. The default value was selected based on the years of experience with residual vectors and it exceeds the number in any practical application. If this number is exceeded then the user will get a message in the f06 file:
=================================================== FATAL MESSAGE: TOO MANY DOFS ARE IN THE RESVEC AMLS PUTS ALL RESVEC DOFS IN THE RESIDUAL STRUCTURE MOST PROBABLY LOADS HAVE BEEN APPLIED TO DEPENDENT DOF ON RBE3. MAKE REFGRID INDEPENDENT, IF POSSIBLE! NUMBER OF ALLOWABLE RESVEC DOFS = 500 NUMBER OF RESVEC DOFS = 666 ===================================================
As mentioned in the message above, owing for the need to calculate the static response for the residual loads exactly, CDH/AMLS places all the DOFs loaded by residual load in the last (residual) substructure. Since CDH/AMLS works primarily with very small substructures (approximately several thousand DOFs), a larger value than 500 for the residual load DOFs may cause some problems. If the user insists on having a number greater than 500, it can be changed, but only on request to CDH. The above message is usually caused by applying a load to a dependent DOF of an RBE3, which has over 500 independent DOFs. A simple UNIX script, rbex, provided in the CDH/AMLS delivery package, will read the messages in the .f06 file to help the user identify which of the RBE3 elements are connected to how many independent grid points. There are other reasons that all RBE3s should be written such that the REFID is an independent grid point. This means that each RBE3 should have a um continuation record that will indicate which of the independent DOFs will become dependent. Please refer to Remark 3 in the Nastran QRG for the rule governing this action. An alternative approach available in both MSC and NX Nastran is to use the AUTOMPC option. In either case, the user must provide additional input. It has been a well-accepted fact that adding any number of residual vectors will never cause bad results. Some vectors may not help much, but they should never generate a worse solution. That, of course, is true in a perfect numerical world with infinite precision. However, there have been a number of cases where adding some residual vectors indiscriminately has, in fact, generated a nonsense solution. Thus, one should always keep in mind the following statement: residual loads are a MUST for accurate point mobility response, and in general, do not help much the transfer mobility accuracy. Only the addition of more modes will improve the transfer mobility accuracy.
12
CDH/AMLS
CDH/AMLS
14
CDH/AMLS message:
=================================================== ACTUAL VIRTUAL MASS: COLUMN 1 (X,Y,Z) AMLS VIRTUAL MASS: COLUMN 2 (X,Y,Z) =================================================== MATRIX VMASS (GINO NAME 101 ) IS A DB PREC 2 COLUMN X 3 ROW RECTANG MATRIX. COLUMN 1 ROWS 1 THRU 3 ------------------------ROW 1) 1.8523D-02 3.4813D-02 5.2897D-02 0COLUMN 2 ROWS 1 THRU 3 ------------------------ROW 1) 1.7791D-02 3.6005D-02 4.5756D-02
The message above indicates the total VM in the X, Y, and Z directions and the actual VM1 used by CDH/AMLS after the modification by scalec. Clearly, the values are very similar in magnitude. The structure eigenvalues are determined by CDH/AMLS using the matrix VM1. Subsequently, the eigenvectors and eigenvalues returned from CDH/AMLS are modified by the so-called modal correction method if param, mfluid, modes. Otherwise, if param, mfluid, mfreq, the matrix ! T VM2! is added as a direct input matrix (M2PP in Nastran syntax) to the modal response equations. The question one may ask is: Which value of parameter mfluid should one use? The first decision to be made concerns the kind of an analysis one wants to perform. If it is SOL111 then the following line of reasoning applies. Assume that some kind of damping exists, BJJ matrix (cdamp2, etc.) or the structural damping matrix, K4JJ, (caused by elemental damping, such as Ge on MAT1 data or Ge for PELAS). (Note, the presence of param, G, ALONE does not matter). The choice is now obvious. Set param, mfluid, mfreq. This means that the VM matrix from MFLUID will be added to the modal mass matrix. Note that in this case the modal mass matrix will be absolutely full, i.e., 100% dense, but that is not a problem because the modal equations are already fully coupled owing to the presence of BJJ or K4JJ. This is the cheapest solution by far and just as accurate as any other approach. However, if one is running SOL 111 and there is no damping as described above (modal equations are uncoupled) or if one is using SOL 103 then the choice becomes more problematic. Now one can specify either param, mfluid, [amls, modes]
15
CDH/AMLS If one chooses modes then CDH/AMLS will calculate the structural modes with VM1 (thus making AMLS very efficient, in other words, there is very little change in the AMLS run time with or without MFLUID) and after the eigenvectors are returned to Nastran, a further eigenanalysis will be performed (modal correction method) to modify the modes by adding VM2. This approach works well if the number of output DOFs is relatively small, <100000, irrespective of the number of modes that were calculated by AMLS. However, if one sets DISP=all, this could become very expensive if there are many modes, e.g., (> several thousand). If there are few modes (< 100) and DISP=all, then this approach would still be acceptable. Alternatively, if one has DISP = all and many modes (> several thousand), then the only choice is param, mfluid, amls. This means that the full VM matrix is passed on to CDH/AMLS. However, this will significantly increase the elapsed time for CDH/AMLS.
This is the simplest to explain. It means that there is one or more (most likely one) DOF associated with grid point 4362 that has both a singular stiffness and mass and CDH/AMLS will automatically apply an SPC to that DOF (very similar to autospc in Nastran). There are occasions where Nastran does not manage to constrain all singular DOF. Typically, this will be a rotational DOF. The second warning message that may be in the log file is:
**********************AMLS WARNING********************** Singularity found in the stiffness matrix ******************************************************** Grid point Solution to problem 143548 Treat as a mechanism
16
CDH/AMLS This message means that for some DOF of grid point 143548 (probably a rotational DOF) there exists a vector x such that xT Kx ! 0 , but xT Mx ! 0 , thus CDH/AMLS treats this DOF as a mechanism. There is no need for any user intervention; however, one should review the model near this grid point. The user should check if this is a mechanism one intended, such as a steering wheel mechanism or is a spurious mechanism that should not exist. The third possible warning message is:
**********************AMLS WARNING********************** Rigid body mode(s) found for a non-root substructure ******************************************************** One or more rigid body modes were found for a non-root substructure in the substructure tree. This indicates that: - there is a portion of the structure that is free to rotate and/or translate relative to the rest of the structure (a mechanism), or - the finite element model does not represent the structure, as it should. For each rigid body mode, the eigenvalue and the grid point associated with the largest entry in the associated eigenvector are printed below in case this information is useful for addressing modeling deficiencies: Eigenvalue Grid Point ------------------------------7.408036E-06 146435 Modes in non-root substructures may lead to an AMLS error. If an error related to these modes is found, try one of the following: 1. Try setting k6rot equal to 0., and snorm equal to 45.: param, k6rot, 0. param, snorm, 45. 2. In the Case Control section, before any subcases, create a SET containing the above grid points and set DATAREC equal to the SET's setid. As an example, if the above grid points are 12345 and 23456, choose a SETid of, e.g., 98765, and insert these lines in Case Control before any subcases: set 98765 = 12345, 23456 DATAREC = 98765 3. Contact mladen.chargin@cdh-ag.com for assistance with this problem.
In this particular case, neither the stiffness nor the mass matrices are singular at the grid point mentioned above. However, there exists an eigenvalue whose value is -7.408036E06 and this particular mode is considered a rigid body mode. This fact will not cause an error in phase3 but it may later cause a problem in phase4. If it does, one would get a fatal message in the log file from phase4 that would indicate the following:
17
CDH/AMLS
***********************AMLS ERROR*********************** Mass matrix singularity associated with rigid body modes See preceding AMLS Error message. ********************************************************
Since this message would indicate that AMLS stopped with a fatal error, the user should follow the instructions in Steps 1 or 2 above. The most likely solution would be to use instruction in Step 2, which means that for the example above, the user would define the following two lines before all subcases in the Case Control deck:
set 666 DATAREC = 146435 = 666
18
CDH/AMLS
References:
1. J.K.BENNIGHOF and C.K. Kim, An adaptive multilevel substructuring method for efficient modeling of complex structures, Proceedings of the AIAA, 39TH SDM Conference, 33rd SDM Conference, 1631-1639, Dallas, TX, 1992. 2. J.K.BENNIGHOF and M.F. KAPLAN, Frequency sweep analysis using multilevel substructuring, global modes and iteration, Proceedings of the AIAA, 39TH SDM Conference, Long Beach, CA 1998. 3. J.K. BENNIGHOF, M.F. KAPLAN, M.B. MULLER, M. KIM, Meeting the NVH computational challenge: automated multilevel substructuring, Proceedings of the International Modal Analysis Conference XVIII , San Antonio, Texas, 909-915, 2000. 4. J.K. BENNIGHOF, M.F. KAPLAN, M. KIM, C.W. KIM, M.B. MULLER, Implementing automated multi-level substructuring in Nastran vibroacoustic analysis, Proc. of the SAE Noise and Vibration Conference Traverse City, Michigan, SAE paper 2001-01-1405, 2001. 5. J.K. BENNIGHOF AND R.B.LEHOUCQ, An automated multilevel substructuring method for the eigenspace computation in linear elastodynamics, SIAM Journal Sci. Comput., 25, 2084-2106, 2004. 6. H. Voss, Automated Multi-Level Substructuring (AMLS), Technische Universitt Hamburg-Harburg, Lecture Presentation. 7. M. Bennur, Superelement, Component Mode Synthesis, and Automated Multilevel Substructuring for Rapid Vehicle Development, 2008 SAE World Congress, Paper 08B 11.
19
CDH/AMLS
(1)
# M cc !" 2 % $ M bc
Assume that {ub } is a known function of frequency and {uc } is represented by the normal modes plus static shapes, i.e.,
(3)
[ K cc ][Gcb ] = ! [ K cb ]
The flexible modes are calculated from the c set, or
(4)
(5)
! {Fc }
(6)
(7)
20
CDH/AMLS Since this is a harmonic analysis, the acceleration and velocity may be obtained from
(8)
21
CDH/AMLS
Appendix B:
Parameter
SS2GCR
Scale factor for determining the cutoff frequency of the substructure modes. SS2GCR*cutoff frequency for the global modes (F2 on EIGRL entry) Used to eliminate any loads on DOFs that have a mass larger than the value of LGMASS during the calculation of residual vectors. Use CDH/AMLS for the eigenvalue calculation of superelement component modes. Set to NO for standard eigenvalue operation. Set to YES if panel participation output is requested. If set to NO (default), panel participation results will be incorrect. Set to the number of CPUs to be used for the CDH/AMLS calculations. Default set to the value of parallel=ncpus on the nastran submit command. Available computer memory for I/O buffering on IBM machines. The units for MIO are in MB. Available computer memory for I/O buffering on SGI machines. The reserved memory is FFIO*16 MB. FFIO is an integer parameter.
LGMASS
1.0E+5
AMLSCMP
YES
AHH
NO
AMLSNCPU
MIO
1000.0
FFIO
22
CDH/AMLS
MFLUID
MODES
Options are: AMLS - Same as specifying VMOPT=1 in for standard Nastran operation, virtual mass will be included in the mass matrix at the same time as all other mass elements MODES - Fluid virtual mass (VM) is split into two parts: VM= VM1 + VM2. VM1 is used in the CDH/AMLS modes calculation. After the eigenvectors and eigenvalues are returned from CDH/AMLS, they are modified by the modalcorrection method to add VM2. MFREQ - Fluid virtual mass (VM) is split into two parts: VM= VM1 + VM2. VM1 is used in the CDH/AMLS modes calculation. The matrix T ! VM2! is added as a direct input matrix (M2PP in Nastran syntax) to the modal response equations.
SCALEC
0.05
Used with PARAM, MFLUID to split the fluid virtual mass (VM) into two parts: VM= VM1 + VM2
23