A GENERAL PROCEDURE FOR FINITE ELEMENT MODEL CHECK AND MODEL IDENTIFICATION Y.T. Chung & L. L. Kahre Space Station Division ‘McDonnell Douglas Aerospace Company - West 5301 Bolsa Avenue Huntington Beach, Ca 92647 ABSTRACT Dynamic loads analysis using finite element models is a major task in the structural design process. An error free model to predict accurate loads or responses is essential for designing a vehicle to meet its pesformance requirements. A systematic approach employing MSC/NASTRAN direct matrix algorithm program is developed to check the residual loads and the rigid body strain energy for identifying improper modeling. The modal effective mass of each system mode is also computed using rigid body modes for statically determinate structures or constraint modes for statically indeterminate structures to identify the significant modes of the structure with respect to the constrained interfaces. These significant modes are the target modes to be identified in the payload structural qualification modal survey for generating a test-verified dynamic model used in the verification loads analysis. Furthermore, the modal kinetic energy distribution per degree of freedom and the kinetic energy fraction for each superelement are computed to identify the important subsystem local modes in additional to the standard strain energy fraction provided by MSC/NASTRAN. An example is presented to demonstrate this systematic approach for analytical model check and model identification.INTRODUCTION Dynamic loads analysis using finite clement models is a major task in the structural design process, An error free model to predict accurate loads or responses is essential for designing a vehicle to meet its performance requirements. Several techniques, such as a 1-G check, enforced displacement check, and Cholesky check are available to verily analytical models. However, these techniques are applied at the analysis set level (Le. at the A-set level in terms of MSC/NASTRAN terminology) and are unable to identify the problem area at the upper stiffness assembly levels. In 1985, an MSC/NASTRAN multi-level rigid body strain energy check routine [1] was developed to provide a means for checking an analytical model at three model stiffness assembly levels. For large complicated structures which are modeled as superelements, this routine needs to be extended in order to pin-point the modeling problems for each superclement and to identify the important dynamic characteristics ofthe structures. Several addtional techniques to detect modeling problems which result fom improperly applied constraints and stiffness distribution as well as the algorithms to identify the dynamic characteristics of large complex dynamic models have been developed for MSCINASTRAN version 67.5 structural superelement analysis. ‘The residual loads and the rigid body strain energy concept are applied on the three levels of the stiffness matrices associated with MSC/NASTRAN analysis sets to identify the modeling errors. The total weight of the ‘model and the weight of each up-stream superelement is calculated for comparison with the ‘Mass Property Report. The modal effective mass [2] of each system mode is computed using the rigid body modes for statically determinate structures or the constraint modes for statically indeterminate structures. ‘The modal effective mass is used widely to identify the significant modes of the structure with respect to the interfaces [3,4]. ‘These significant ‘modes are the target modes to be identified during the structural qualification modal survey. ‘The accumulated modal effective mass is usually used to establish the frequency range of interest for predicting the dynamic responses using the mode transformation approach. Furthermore, the modal kinetic energy distribution per degree of freedom [5] and the kinetic energy fraction for each superelement are computed to identify the important subsystem local ‘modes in addition to the standard strain energy fraction provided by MSC/NASTRAN. ‘An example using the Version 67.5 application direct matrix abstraction program (DMAP) is presented in this paper to demonstrate this systematic approach for analytical model check and model identification, MODEL CHECK Several techniques, such as a I-G check, enforced displacement check, and Cholesky check are available to verify the analytical model. However, these techniques are applied at the analysis set level and are unable to identify the problem area at the upper levels. ‘The multi level rigid body strain energy check provides a means to check the model thoroughly. The basic concepts and algorithm which have been coded in the DMAP presented are discussed in the following sections, Residual Loads and Residual Displacements Check MSCI/NASTRAN is well laid out in the various analysis sets: the G-set level contains all the degrees of freedom (DOF) of the model; the N-set contains all the DOF's after removing, from the G-set all the dependent DOF's due to multiple-point-constraint equations (MPC’s) and rigid elements; the F-set contains all the DOF’s after the single-point-constraint ‘equations (SPC's) have been removed from the N-set. ‘The basic principle applied in the residual loads and the residual displacements (separation ratio) check technique uses theigid-body modes [®;p] to post-multiply the stiffness matrix [Ks] of the model for determining the residual constraint forces {Pgrmd] associated with the rigid-body motion as shown in Eq. (1), (KM 51= [Pyne] Q Note that the rigid body modes {®;b] are determined from the structural geometry and have nothing to do with the rigid body modes computed in the later eigensolution process. Naturally, an ideal unconstrained model without errors would not have any constraint force associated with the rigid-body motion. The residual displacement, [Sra], in a specific direction is defined as the normalized displacement with respect to the stiffness in that direction with the effects of the other DOF's, ignored as shown in Eq. (2), P. (eo @ [Diag x 10] where 5 is the element-by-element division operator (1) is aunit column vector with I's in all rows Diag is a column vector containing the terms from the diagonals of the matrix enclosed in the parentheses ‘To minimize the numerical round-off error caused by dividing two small numbers in Eq. (2) the diagonal terms of [Ks] less than a pre-selected stiffness threshold (currently set to 10°5 ) are reset to zero. By computing the residual loads and the residual displacements at each unique level, the type ofthe modeling problems and the DOF's associated with it can be identified, In general, the residual loads and the residual displacements at each DOF should be small (typically <10-5 for translational DOF's and <10-3 for rotational DOF's) for an error-free model. Therefore, a relatively large value at different levels indicates certain modeling errors. In addition, comparison of the residual load and the residual displacement at the same DOF may provide some indication of the numerical round-off problem. The possible ‘modeling problems at each level are summarized as follows: G-lvel: + Coordinate coupling, e.g. RBE of grid points with different output coordinates + I-conditioning due to short beams + Element coordinate for pin flag DOF's offsets from grid point output coordinate + CELAS element between non-coincident points Nelevel: + ‘Improper MPC equations Felevel: + Over-constrained due to SPC's ‘+ Normal rotational stiffness is non-zero for CQUADS or CTRIA3 elements while the parameter K6ROT iss tured onIt is emphasized that the model is GROUNDED (or CONSTRAINED) only if non-zero values not in the G-set or the N-set occur in the F-level. Any non-zero values in the G-level or the N-level indicate geometry modeling error or numerical round-off error. Multi-Level Rigid-Body Strain Energy Check ‘The multi-level rigid body strain energy check is a similar technique as that of the residual displacement check. It simply determines the strain energy of the model associated with the ‘igid-body motion at different levels, (K, ©, KILO) @) A large value (typically >10°3 for translational direction and >10! for rotational direction) indicates modeling errors. ‘The causes are the same as those summarized in the previous Section. ‘The rigid body strain energy check is considered less important than the residual loads and the residual displacements check. ‘The residual displacements and the rigid-body strain energy computed by Eqs. (2) and (3) will be printed out for each superelement automatically. ‘The residual loads of Eq, (1) will be printed out per user's request (parameter PGPR). The multi-level superelement approach to locate the possible modeting errors of smaller components is more efficient than the model check of the complete model, Mass Property Check ‘The mass properties (total mass, inertia, and the location of the center of gravity) listed in the weight generation table of the MSC/NASTRAN output should agree with those specified in the updated mass property report. The weight table is generated for each superelement and can be used to verify the weight of tip superelements. If the superelement contains up- sments the application DMAP will generate an additional weight table which includes the weights of all up-stream superclements. A special parameter (WTONLY) is available to perform the separation ratio and weight calculation without computing the frequencies of the components to minimize the computation cost. Rigid-Body Frequency Check ‘The rigid body frequency check can be done on the residual superelement only. This is because the frequencies of the up-stream superelements are computed with respect to the boundary DOF's (similar to the Craig-Bampton modal model). However, the constrained ‘normal modes computed for the up-stream superelement can be used to verify the minimum frequency requirement for that subsystem. For an error-free unconstrained model, the first six modes should be rigid body modes. The frequencies of the six rigid-body modes in general should be five orders of magnitude less than the first clastic mode (the SUPORT entry should not be used in the Bulk Data when checking the rigid-body modes). MODEL IDENTIFICATION Determination of the structural fundamental frequency is a very important step in the design process. However, the identification of the mode shapes of a complex structure is always Gifficult. In addition, there is always a concern about how many modes should be kept if the mode transformation approach is to be used in determining the dynamic responses. ‘Theeffective mass which represents the contribution of each mode to the system response may be used as an indicator for mode selection. Effective Mass Using Rigid Body Modes ‘The conventional approach is to compute the modal effective mass with respect to the rigid body modes [®,,] assuming the load path can be ignored. This requires the calculation of the modal participation factors matrix (MPF) as defined in Eq, (4) [MPF] =[0, (M15) @) where [®,] is the orthonormal mode shape matrix [M,] is the system mass matrix ‘Then the modal effective mass of cach mode ( Meff,,%) with respect to the rigid body modes ccan be obtained as follows: 02.71, 1%,)@ (0.710, 0] eff - = *100 5) (Meo) Ty Diagl FMA DF] ® ® is the clement-by element multiplication operator E is the element-by-clement division operator * is the multiplication operator {1} is a unit column vector with 1's in all rows DiagQ _isacolumn vector containing the terms from the diagonals ‘of the matrix enclosed in the parentheses. where Effective Mass Using Constraint Modes ‘On the other hand, if the structure is statically indeterminate then the modal effective mass computed using rigid body mode approach is incorrect because the interface load path has not been taken into account. In this case the constraint mode set instead of the rigid body mode set has to be used. The constraint mode {@,,.] is determined from Eq. (6), [,,] [ © where [K,] _ is the stiffness corresponding to non-interface DOF's [Kg] is the stiffness corresponding to interface DOF's Ug] is an identity matrix corresponding to interface DOF's ‘The modal effective mass of each mode ( Meff.,,%) with respect to the constraint modes can. be obtained from Eq. (7),C.F 4,1.) O10, 1%.) [Me %) *100 (7) (Meh 0 =" Ty Diag FOTO, A Ideally, if all the modes were included in the prediction of the dynamic response and the structure was linear, the total modal effective mass and inertia would be equal to the physical ‘mass and inertia with respect to the reference point. In reality, the significant modes within the frequency range of interest are considered to have been included in the analysis if the ratio of the total modal effective mass to the physical mass is higher than 85%. Two levels of modal effective masses, G-level and J-level, are computed for the superelement approach. The G-level modal effective mass, GMBFF%, includes the ‘contribution from the up-stream superelements while the J-level modal effective mass, IMEFF%, is the contribution of that superelement only. By examining the modal effective mass of each mode in the translational and rotational directions, the motion of each mode can be identified and the significant modes of the structure can be selected. The total modal effective mass of the significant modes can be determined to check if they indeed contribute ‘most to the dynamic responses of the structure. The calculation of modal effective mass is, controlled by the parameter MEFF. The rigid body mode approach is the default method. ‘The constraint mode approach is selected automatically by the DMAP if BSET cards defining the interface DOF's are used in the model. ‘Occasionally, the modal effective mass of a subsystem constrained at its interfaces is desired for validating the minimum frequency requirement for the secondary structure and for supporting the subsystem stand alone modal survey. A parameter, CBMEFF, can be turned on for each superelement to compute the modal effective mass corresponding to the ‘component Craig-Bampton model. Modal Kinetic Energy per DOF ‘The difficulty of mode identification of a dynamic model increases substantially as the model becomes larger and more complex. In order to assist the analyst in differentiating system ‘modes and local modes, a concept of determining the modal kinetic energy per degree of freedom is developed. This modal kinetic energy check is bascd on the orthogonality principle as shown in Eq, (8), [oF 1e,] = ® ‘where [M,] is the mass matrix of the system, [@,] is the orthonormal mode shape matrix, and [J] is the identity matrix. Equation (8) can be expanded by using a special element-by-clement multiplication operation and normalizing the diagonal terms to 100 to determine the percent modal kinetic energy distribution per DOF for each mode, [KENG%] =[©,]@ CM, 1,D o where ® represents the element-by-element multiplication operator. In the modal kinetic energy matrix, [KENG%], the rows represent the DOF's of the model and the columns represent the calculated modes. The value at any location of [KENG%] is the percent contribution a particular DOF makes to the total kinetic energy of that mode. ‘The summation of any column should be equal to 100. ‘The modal kinetic energy per DOF exposes thespecific directional behavior of the response. This is a remarkable improvement over standard NASTRAN clement strain energy output which lumps all the DOF contribution for an element and does not distinguish between directional effects. The calculation of modal kinetic energy pet DOF i activated by speifying te parameter KE inthe bulk data or ease ‘control deck. Similar to the effective mass calculation, two levels of kinetic energy per DOF, G-level and F-level, are computed for the superelement approach. The G-level kinetic energy per DOF, KENG‘%, includes the contribution from the up-stream superelements while the J-level, JKENGZ, is the contribution of that superelement only. For superclement with up-stream superelements, the actual kinetic energy contribution of the DOF's are those listed in the JKENG%. The J-level kinetic energy per DOF, JKENG%, is also used widely in the industry to determine accelerometer locations for modal survey testing. Theoretically, the DOF's with higher kinetic energy respond more effectively to the dynamic environment. Consequently, the data collected at these locations are less biased and are better suited for use in modal parameter estimation, Modal Kinetic Energy Fraction ‘The modal effective mass criteria used to identify the significant system modes is usually less sensitive for determining important local modes. To identify the significant local modes of subsystem for improving the dynamic response prediction, the modal kinetic energy fraction of that subsystem, [KEF] is calculated. [KEF] is a measurement of the amount of system kinetic energy contained within each subsystem for any given mode. ‘The modal kinetic energy fraction is computed based on Eq. (10), Diag (1 IM,1®.))] did CHITA) ao) where [®,] is the component partition of the system mode shape matrix [®,] is the system mode shape matrix [M,]_ is the component mass matrix [M,] is the system mass matrix = isthe element-by-element division operator Diag0 is a column vector containing the terms from the diagonals of the matrix enclosed in the parentheses ‘The modal kinetic energy fraction is usually employed in conjunction with the modal effective mass for selecting important modes. A rule of thumb is that if a mode has more than 50% of the modal kinetic energy of the whole system in any single subsystem, that mode is considered to be a significant local mode for that component. “The calculation of modal kinetic energy fraction is controlled by the parameter SEKEF. Modal Strain Energy Fraction ‘The modal strain energy fraction, [SEF], is similar to [KEF] in its calculation. [SEF] is a measurement of how much of the strain energy of the system is in each subsystem for any given mode. The modal strain energy fraction is computed using Eq. (11),Diag} 1K U®,D) (FP ing, FUK,IO,D] we where [®,] is the component partition of the system mode shape matrix [@,] is the system mode shape matrix [X,] _ is the component mass matrix LK] _ is the system mass matrix ~ _ isthe element-by-element division operator DiagQ) is a column vector containing the terms from the diagonals of the matrix enclosed in the parentheses ‘The calculation of modal strain energy fraction is controlled by the NASTRAN standard parameter SESE. A rule of thumb is that if a mode has more than 50% of the modal strain energy of the whole system in any single subsystem, that mode is considered to be a significant local mode for that component. DMAP IMPLEMENTATION. The application of the model check and model identification DMAP is straightforward. The diagnostic information of the model will be computed by simply inserting the DMAP in the MSC/NASTRAN Executive Deck. The application DMAP is shown in Appendix A. Users can define the following parameters in the Case Control or Bulk Data to activate the options, available in the DMAP. Parameters Default Description WIONLY No If this parameter is selected (PARAM, WTONLY. YES), only the model check and the weight generation procedures are executed. This parameter is very useful for checking a free-free model before the effective ‘masses are computed for the constrained configuration. BGPTB No Print coordinate systems and grid point locations in basic coordinate system. PGPR No Print residual loads. PGFILTER ——1.0E-3_ Suppress the printing of residual loads smaller than PGFILTER. MEFF No ‘Compute modal effective mass. Rigid body mode approach is the default method. Constraint mode approach will be automatically activated if the interface degrees of freedom are defined in the B- Set in the residual superelement. LMEFF% 20 Generate a column vector, MODEID, with a value of 1.0 at the rows whose modai effective mass is larger than LMEFF%. CBMEFF No Compute modal effective mass for subsystems constrained at interfaces.KE No ‘Compute modal kinetic energy per DOF. KEFILTER 1.0 Suppress the printing of any modal kinetic energy pet DOF (KENG®) less than KEFILTER. SEKEF No ‘Compute subsystem modal kinetic energy fraction. EXAMPLE ‘A Space Station integrated truss segment, S3/S4, as shown in Figure 1 was used to demonstrate the application of the model check and model identification DMAP. The segment is supported by flexure assemblics at the trunnions and keels to simulate the National Space Transportation System (NSTS) flight boundary conditions. The interface DOF are 60200X, 602002, 57526Y, 60100X, 60100Z, 232097Y, 2320912, and 2320942 and the Orbiter coordinate system is used to identify the direction. The system frequencies and modal effective mass, up to 50 Hz, for the segment S3/S4 are computed and compared for the rigid body mode approach and the constraint mode approach. Figure 1. Segment $3/S4 Supported at Flexure Assembly For the rigid body mode approach there are always six columns of modal effective mass, three translational and three rotational, associated with each mode. Table 1 shows the modal effective mass of the target modes (translational modal effective mass equal to or greater than 2%, or the kinetic energy fraction or the strain energy fraction greater than 50%) of segment 3/84 using the conventional rigid body mode approach. “Fourteen target modes are identified. ‘The primary X mode has 78.14% modal effective mass at 8.36 Hy, the primary Y mode has modal effective mass 86.47% at 6.67 Hz, and the primary Z mode has 40.92% modal effective mass at 9.58 Hz. The total modal effective masses in the X-axis, Y-axis, and Z-axis are 95.56%, 93.17%, and 76.90%, respectively. The total modal effective mass in the Z direction is lower than that in the X or Y direction because more constraints (primary and secondary trunnions) are imposed in that direction.KE No ‘Compute modal kinetic energy per DOF. KEFILTER 1.0 Suppress the printing of any modal Kinetic energy per DOF (KENG#) less tan KEFIL TER. SEKEF No ‘Compute subsystem modal kinetic energy fraction. EXAMPLE ‘A Space Station integrated truss segment, $3/S4, as shown in Figure 1 was used to demonstrate the application of the model check and model identification DMAP. The Segment is supported by Mesure assemblies atthe trunnions and keels to, simulate the ‘National Space Transportation System (NSTS) flight boundary conditions. ‘The interface DOF5 are 60200X, 60200Z, 57526, 60100X, 601002, 232097Y, 2320912, and 2320942, and the Orbiter coordinate system is used to identify the direction. The system frequencies and modal effective mass, up to 50 Hz, for the segment $3/S4 are computed and compared for the rigid body mode approach and the constraint mode approach. Figure 1. Segment $3/S4 Supported at Flexure Assembly For the rigid body mode approach there are always six columns of modal effective mass, three translational and three rotational, associated with each mode. Table 1 shows the modal effective mass of the target modes (translational modal effective mass equal to or greater than ‘296, or the kinetic energy fraction or the strain energy fraction greater than 50%) of segment '$3/S4_using the conventional rigid body mode approach. Fourteen target modes are identified. The primary X mode has 78.14% modal effective mass at 8.36 Fiz, the primary 'Y mode has modal effective mass 86.47% at 6.67 Hz, and the primary Z mode has 40.92% ‘modal effective mass at 9.58 Hz. The total modal effective masses in the X-axis, Y-axis, and Z-axis are 95.56%, 93.17%, and 76.90%, respectively. The total modal effective mass in the Z direction is lower than that in the X or Y direction because more constraints (primary and secondary trunnions) are imposed in that direction,On the other hand, Table 2 shows the modal effective mass computed from the constraint ‘mode approach where the load path is taken into account. Each mode has a modal effective mass value corresponding to the constrained degree of freedom. For this example, which is, constrained at eight interface degrees of freedom, there are eight columns of modal effective mass for each mode. ‘The energy of each mode is transmitted differently to the interface based on the load path. However, the total modal effective mass of the target modes in the same vibratory direction at different interface locations should converge to a similar value, For this example X direction has 94.44% and 93.13%, Y direction has 97.02% and 92.36%, and Z direction has 82.04%, 81.50%, 84.59% and 85.21%. A total of twenty-one target modes are identified for $3/S4 using the constraint mode approach. The primary X mode has a maximum of 66.30% modal effective mass at 8.36 Hy, the primary Y mode has a maximum of 91.95% modal effective mass at 6.67 Hz, and the primary Z mode has a maximum of 41.15% modal effective mass at 6.67 Hz. ‘The averaged total modal effective masses in X-, ¥-, and Z-axis are about 94%, 94%, and 83%, respectively. It ean be scen that the modal effective mass distribution for the target modes has been changed significantly and more target modes are selected for the constraint mode approach. The constraint mode approach is the recommended approach for computing the modal effective mass and selecting the target modes for statically indeterminate structures beeause it represents the actal Might conditions. CONCLUSIONS A systematic approach to identify the modeling errors and the dynamic characteristics of a structural model is developed and coded in DMAP for MSC/NASTRAN implementation. ‘The modeling of each superelement is checked in three analysis levels such that improper ‘modeling can be easily located. Two methods to select the significant structural modes for dynamic load prediction are included in the DMAP to accommodate different attachment schemes. The modeling and the dynamic behavior of a structural mathematical model can be assessed and fully understood by incorporating this DMAP. REFERENCES (1) Parker, G. R. and Brown, J. J., "DMAP Alter for Performing Basic Constraint Checks on Structural Modeling,” MSC/NASTRAN User's Conference, March 1985, Pasadena, CA. [2] Trubert, M., "Assessment of Galileo Modal Test Results for Mathematical Model Verification,” 25th AIAA SDM conference, paper no. 84-1066, April 1984, pp 528- Sal. [3] Osman, B.A. & Krausser, D. G., "A Generalized Form for the Modal Effective Mass," 63°4 Shock & Vibration Symposium, Las Cruces, NM, October, 1992. [4] Chung, Y. T. & Sernaker, M. L., "Assessment of Target Mode Selection Criteria for Payload Modal Survey," Proceeding of the 12th International Modal Analysis Conference, Jan. 1994, Honolulu, Hawaii. [5] Parker, G. R. and Brown, J. J., "Kinetic Energy DMAP for Mode Identification,” MSC/NASTRAN User's Conference, March 1982, Pasadena, CA. 10RESIN Sem Y xpusddy 12(gure ary arse meni SNS met * SEMGSaaouTih dv ans Bua) Worea+ monet A aM ON Ga OSE GAY OIE a A savanivsoeezorpma xALio 4: + sivawnevvnwvy a xb = xt ormos aTmoult msn a TRL 168