Sie sind auf Seite 1von 19

Implications of Vertical Mass Modeling Errors on 2D Dynamic Structural Analysis

Timothy M. Whalen*, Graham C. Archer, and Kishor M. Bhatia$


School of Civil Engineering, Purdue University, West Lafayette, IN, 47907-1284, USA.

ABSTRACT
The use of a diagonal (lumped) mass matrix is common in finite element analysis, particularly for structural dynamics problems. The assumptions made when performing the lumping procedure, however, are not always consistent with the underlying physical behavior of the system and can therefore significantly modify the dynamic properties of the modeled structure. The influence of errors in the modeling of lumped masses for vertical degrees of freedom on dynamic behavior and seismic response is illustrated in the context of the well-known Los Angeles SAC building models. We demonstrate that over-estimation of the lumped masses associated with vertical displacements in 2D frame models of these structures leads to inaccurate estimates of modal periods and associated modal participation factors for dynamic response. The ramifications of these inaccuracies on seismic response quantities are shown, and implications for such errors in other dynamic analysis situations are discussed.
Keywords: diagonal mass matrices, vertical mass modeling, dynamic analysis, 2D modal analysis, modal participation factors, SAC buildings.

* $

Corresponding author. Phone: (765) 494-2225. Fax: (765) 496-1105. E-mail: whalen@purdue.edu Now at Lin Engineering, Chatham, IL, 62629, USA.

1. INTRODUCTION
Finite element modeling has long been a favorite tool of engineers seeking to analyze the dynamic behavior of a structure subjected to earthquake or wind excitation. This approach requires the engineer to make a variety of approximations of the structures properties so that a tractable yet physically reasonable model results. One such approximation is the representation of a continuous distribution of structural mass as discrete entries in a global mass matrix. Traditionally, the lumped mass matrix is preferred for use in practical analysis problems. The diagonal form of this matrix is convenient for eigenvalue analysis, and certain explicit and higher-order implicit time integration methods depend upon the presence of a diagonal mass matrix for efficient computation. Other advantages and disadvantages exist for this approach to mass modeling; see [1], for example, for further discussion. Various schemes exist for deriving a diagonal mass matrix via lumping; see [2] for examples and discussion. As indicated in [1], mass lumping techniques are often guided by intuition and physical insight, sometimes assigning mass terms (especially rotary inertia terms) in an arbitrary manner. While often effective, this approach raises concerns since intuition can be risky [1]. Even in the relatively simpler case of mass lumping for translational degrees of freedom, experience has shown that errors in mass modeling can le ad to unusual physical behaviors for the system under study. If not diagnosed properly, such unusual physical responses can lead to erroneous conclusions about the system, potentially hiding problems in the design of the system or its safety features. In this paper, we illustrate some of the problems associated with improper mass modeling by considering mass lumping for vertical displacement degrees of freedom in a two-dimensional finite element building model. We focus upon the well-known SAC buildings [3], which are widely employed by researchers and designers seeking to illustrate features of structural response or control during earthquakes. We will show that a common mass modeling assumption for these displacements leads to

incorrect response behavior, and we shall discuss how this error could cause improper interpretation of results for structural control algorithms and other applications.

2. VERTICAL MASS: ASSUMPTIONS AND GENERAL CONSEQUENCES 2.1. Assumptions


A typical element used to model buildings in two dimensions is a plane frame element, a combination of a bar element and a beam element. In a typical lumped-mass scheme for bar elements, where only horizontal axial translation is allowed at the nodes, the horizontal degrees of freedom are assumed to have the same lumped mass, each being half of the total element mass. For beam elements, the lumped masses for each vertical degree of freedom are also set to be half of the element mass. (The rotational mass is often ignored or set as the beams mass moment of inertia.) Since the masses for the horizontal and vertical degrees of freedom at each node of a plane frame element are those of a bar and beam element, respectively, any structure modeled by 2D plane frame elements will obtain a global mass matrix that has equal horizontal and vertical mass at every node. Thus, modeling a building via plane frame elements inherently assumes that each element has equal amounts of vertical and horizontal mass. The mass modeling for the 2D models of the SAC buildings follows from the mass modeling for plane frames described previously. As shown in Figure 1, for the north-south direction of the floor plans for the various buildings, two exterior moment-resisting frames resist the horizontal motions. Thus, one would typically assign half the mass of each floor to each frame as total horizontal mass. Based on the procedure described above, one would then assign a total vertical mass for each frame equal to half of the floor mass. This is precisely what the originators of the SAC building models prescribed, and these models were ultimately adopted into many different dynamic analyses of the SAC buildings (e.g., [4,5]). Unfortunately, this will lead to a gross over-prediction of the mass in the vertical direction. While the horizontal inertial forces can be assumed to be equally distributed to the two exterior frames, the vertical

inertial forces are resisted unequally by the columns. As shown in Figure 1, the tributary areas of the vertical mass for the moment-resisting frame are much smaller than the tributary areas of the horizontal mass. Thus, using the horizontal mass value in both directions artificially increases the vertical mass by a factor of 3.75 in the 20-story building, 5.0 in the 9-story building, and 6.0 in the 3-story building. This overestimation of vertical mass has a noticeable effect upon the dynamic properties of the SAC building models and thus upon the results of any dynamic analyses performed on these models. Specific effects for these buildings are discussed in Section 3, while general consequences of improper vertical mass modeling are detailed below.

2.2. Consequences
An obvious consequence of improper modeling of vertical mass is to change the modal properties of the structure as compared to its real properties. More precisely, one finds that some natural frequencies of vibration calculated using the over-estimated vertical mass are artificially shifted downward. It has been shown [6] that rescaling all of the mass terms by a factor b causes all of the vibration frequencies to change according to

= b 1 / 2 ,

(1)

where and are frequencies associated with the rescaled and unscaled problems, respectively. This
result is not directly applicable to the issue considered here, since only the vertical masses are rescaled. However, if one assumes that the mode shapes of the unscaled problem have their response primarily in either the horizontal degrees of freedom (mostly horizontal) or in the vertical degrees of freedom (mostly vertical), one can demonstrate approximately that the frequencies associated with the mostly vertical modes change according to Equation (1) when the vertical mass is rescaled by b. Furthermore, the frequencies of the mostly horizontal modes are unaffected (to first order) by the vertical mass

rescaling. Since over-estimating the vertical mass corresponds to taking b > 1, one concludes that the frequencies of the mostly vertical modes are shifted down relative to the mostly horizontal modes. The consequences of the shift described above on 2D dynamic analyses can now be detailed. The MDOF dynamic equation of equilibrium for the model is given as

&& & mv t + cv + kv = 0 ,

(2)

where the total displacement v t consists of a horizontal base displacement v gh , a vertical base displacement v gv , and a relative nodal displacement v . We express these contributions in vector form as

v t = v + rv vgv + rh v gh ,

(3)

where rv and rh represent rigid-body base displacement vectors in the vertical and horizontal direction respectively. Substituting Equation (3) into Equation (2) gives

&& & & & mv + cv + kv = mrv v&gv (t) mrh v&gh(t) .


Solving the undamped free-vibration eigen-problem for N mode shapes and frequencies gives

(4)

v = nYn .
n =1

(5)

Substituting equation (5) into (4) provides N uncoupled equations of motion, each of which can be generally written as
T T & & & & M nY& + CnYn + K nYn = n mrv v&gv (t) n mr h v&gh(t) . n

(6)

By defining the Duhamel integrals as

& Vnv (t ) = v&gv ( )e n n ( t ) sin n ( t ) d


0

(7)

and

&& Vnh ( t ) = v gh ( )e n n ( t ) sin n ( t ) d ,


0

(8)

the solution of equation (6) is

Yn ( t ) = FpnvVnv ( t ) + FpnhVnh ( t )

(9)

where Fpnv and Fpnh are the modal participation factors for the vertical and horizontal ground motions defined by

Fpnv =

nT mr v M n n

, Fpnh =

nT mr h . M n n

(10)

The modal participation factors and the associated modal periods for both the correct and incorrect vertical mass modeling are shown in Table 1(a) for the mostly horizontal modes of vibration of the 20-story SAC building and Table 1(b) for the mostly vertical modes of vibration of the same building. The mode shapes are shown in Figures 2 and 3. From Table 1(a), it is clear that the periods of vibration for the mostly horizontal modes do not change noticeably when the vertical masses are rescaled. The mode shapes and participation factors are similarly unaffected. As can be seen from Table 1(b) and Figure 3 however, the vertical modes are greatly affected by the error in the vertical mass modeling. Predictably, the vertical mass modeling error shifted the primarily vertical modes to longer periods (higher frequencies). This results in the higher modal participations for the vertical modes. As evident from the very low Fpnv numbers in the mostly horizontal modes and the equally low Fpnh values for the mostly vertical modes in Tables 1(a) and 1(b), the regular, orthogonal building geometry produces very little coupling between the vertical and horizontal masses. Thus the horizontal modes were not affected by the vertical mass modeling error.

3. EFFECT OF THE VERTICAL MASS MODEL ON SAC BUILDING DYNAMICS


The previous discussion makes clear that improper vertical mass modeling produces inaccuracies in the modal properties of the structural model. These inaccuracies may have a noticeable effect upon any

dynamic analysis being performed. Some of these inaccuracies are discussed for the specific case of a 2D modal analysis of the SAC buildings in the following section. Letting r represent the response quantity of interest for a given element (e.g., moments, shears, or axial forces), Ke the linear stiffness matrix for that element, Te a transformation used to extract the element end displacements from the set of global displacements v, and t T a vector used to extract the 1 chosen value, we have that
T r = t 1 K e Te v .

(11)

In modal coordinates this becomes

r = rnY n ,
n =1

(12)

where rn, the value of the response quantity resulting from mode shape n, is given by
T rn = t 1 K e Te n .

(13)

The pseudo-velocity spectrum for the vertical and horizontal ground motions can be defined as

S vv ( n , n ) = Vnv ( t ) max
and

(14)

S vh ( n , n ) = Vnh ( t ) max .

(15)

The vertical and horizontal pseudo-velocity spectrums for the two quakes used in this modal analysis, the 2001 Nisqually event measured at the Tacoma VA Hospital and the 1989 Loma Prieta event measured at Corralitos Eureka Canyon Rd., are shown in Figures 4 and 5. Substituting Equations (8), (9), and (10) into (13), using Equations (14) and (15) to maximize the result for each mode, and combining the resulting equations using the SRSS rule gives

r=

[(r F
N n =1 n

pnv

S vv ( n , n )) 2 + ( rn FpnhS vh ( n , n )) 2

(16)

The percentage change in some selected response quantities for the 20-story L.A. SAC building under the two quakes is shown in Table 2. Clearly, the bending moments in the floor beams were not affected by the vertical mass modeling error when using either the horizontal ground motions or the combined horizontal and vertical ground motions. This is due to the low rn values for the mostly vertical modes. Since it was only the participation factors for the mostly vertical modes that were affected by the vertical mass modeling error, the bending moment responses were little affected. Interestingly, the axial forces in the columns were affected by the vertical mass modeling error not just for the combined horizontal and vertical ground motion, but also for the horizontal ground motion alone. For the Nisqually and Loma Prieta horizontal ground motions, the vertical mass modeling error increases the predicted axial forces in the columns by about 4 to 12%. For the combined ground motions, the vertical mass modeling error increases the predicted axial forces in the columns by 14-22% for the Nisqually quake and decreases them by 10-30% for the Loma Prieta quake. The axial forces in columns contain significant modal response quantities for both the mostly horizontal modes and the mostly vertical modes. Since the periods of vibration for the mostly vertical modes of vibration were shifted by the vertical mass modeling error, the pseudo-velocity value for the modes also changed. This, coupled with the changes in the modal participation factors for the horizontal ground motions, created the observed changes in the axial forces in the columns under the horizontal ground motion. The relatively large effect of the vertical mass modeling error on the axial force response for the combined horizontal and vertical ground motions is to be expected since the modal participation factors for the vertical component of the ground motion in the mostly vertical modes of vibration were greatly affected, and the axial forces in the columns will clearly be sensitive to the vertical modes. The overall increase in axial force under the Nisqually quake and the decrease under the Loma Prieta quake arise from the changes in the pseudo-velocities brought about by the shift in the periods of the mostly vertical modes of vibration.

4. CONCLUSIONS AND DISCUSSION


The accuracy of 2D dynamic analysis of a structure clearly depends upon the accuracy of the modeled dynamic properties. We have shown that improper vertical mass modeling, brought about by a simple but inappropriate assumption about mass lumping, can cause noticeable changes in estimated seismic performance of a structure. The shifts in modal periods described above create significant changes in modal participation factors for the affected modes, which in turn impacts the modal responses. These shifts also move the mostly vertical modes into new regions of the earthquake response spectrum, with unpredictable effects upon the response spectrum amplitudes. In the case of the SAC buildings studied here, the main consequence of this mass modeling problem was to change the estimated axial responses of the columns. Undertaking a modal analysis with the i clusion of the vertical component of acceleration n resulted in incorrectly magnified responses in the vertical modes and thus a distorted picture of the overall behavior. Even when the vertical acceleration component was excluded from the analysis, some coupling of the horizontal acceleration to the mostly vertical modes was present (via the small horizontal components of the modal response), leading to similar inaccuracies in the results. The effect of incorrect vertical mass modeling on bending response was negligible in the case of the SAC buildings due to their highly orthogonal splitting of horizontal and vertical modes. Such a splitting is not unusual, but it is not a universal feature of all 2D building models. This is particularly true if the building being analyzed possesses irregular features, such as removed base columns, that tend to produce more coupling between the two types of motions. Other consequences of this mass modeling error follow from this problem with modal frequencies. For example , it is known that extreme axial loads in the columns of building frames are underestimated when vertical motions are present, especially when plastic hinges form in the frames [7]. Shifting of the modal frequencies for the vertical vibration modes complic ates this problem, since the first vertical vibration mode has been shown [8] to play a crucial role in determining the maximum absolute column

axial force and maximum absolute beam moment in building frames. Also, evaluation of structural control devices and strategies can be compromised by this modeling error. Any control device that produces vertical forces would act as an excitation source for these mismodeled modes, potentially producing exaggerated motions in the analysis. In addition, the small horizontal motion coupling to these vertical modes could allow earthquake energy to hide in these modes and possibly escape detection in a typical performance measure for control systems (i.e., peak base shear or horizontal drift ratio), leading to an erroneously high evaluation of the systems dissipation capacity. Finally, certain control strategies are very sensitive to the modal content of the structure under investigation. The authors have found an example [9] in which the presence of an artificially lowered vertical vibration mode had a dramatic influence on the magnitude and direction of the control forces along the height of the structure, due to ill-conditioning of a matrix inverse used in force evaluation. All of these consequences have the effect of allowing misinterpretation of the structural behavior and thus the adequacy of its design.

REFERENCES
1. Cook, R.D., Malkus, D.S., and Plesha, M.E., (1989). Concepts and Applications of Finite Element Analysis, John Wiley & Sons, New York. 2. Zienkiewicz, O.C. and Taylor, R.L. (1989). The Finite Element Method, McGraw-Hill, London. 3. Gupta, A. and Krawinkler, H. (1999). Seismic demands for performance evaluation of steel moment resisting frame structures (SAC Task 5.4.3), The John A. Blume Earthquake Engineering Center, Stanford, CA, Report No. 132, June 1999. 4. Spencer, B.F., Jr., Christenson, R.E., and Dyke, S.J. (1999). Next generation benchmark control problems for seismically excited buildings, Proceedings of the 2 nd World Conference on Structural Control (T. Kobori et al., eds.), Volume 2, p. 1135-1360. 5. Barrosa, L.R. (1999). Performance Evaluation of Vibration Controlled Steel Structures under Seismic Loading, Ph.D. Dissertation, Stanford University, Stanford, California, USA. 6. Neuss, C.F. and Maison, B.F., (1987). Modeling variations in response spectrum analysis, Earthquake Spectra, 3(1), February 1987, p. 199-207. 7. Salazar, A.R. and Haldar, A., (2000). Structural responses considering the vertical component of earthquakes, Computers and Structures, 74, p. 131-145. 8. Ju, S.H., Liu, C.W., and Wu, K.Z., (2000). Analyses of buildings under vertical component of earthquakes, Journal of Structural Engineering, 126(10), October 2000, p. 1196-1202.

10

rd 9. Whalen, T.M., Bhatia, K.M., and Archer, G.C., (2002). Semi-active vibration control for the 3 th generation benchmark problem including spillover suppression, Proceedings of the 15 ASCE Engineering Mechanics Conference, June 2-5, 2002, Columbia University, New York, NY.

11

Figure Captions
Fig 1. Tributary areas for horizontal and vertical mass: (a) LA 3 -story SAC building, (b) LA 9 -story SAC building, (c) LA 20 story SAC building. Fig 2. Fig 3. Fig 4. Fig 5. Mostly horizontal mode shapes. Mostly vertical mode shapes. Pseudo-velocity spectrum (2% damping) for the Loma Prieta (Corralitos) earthquake. Pseudo-velocity spectrum (2% damping) for the Nisqually (Tacoma VA Hospital) earthquake.

12

Table 1(a). Mostly Horizontal Modes of Vibration and Participation Factors Using Correct and Incorrect Vertical Mass Modeling Correct Mass Modeling Incorrect Mass Modeling Mode Period (s) Fph Fpv Period (s) Fph Fpv 1 3.98 3.4592 0.0000 3.98 3.4588 0.0000 2 1.36 0.4529 0.0000 1.36 0.4561 0.0000 3 0.79 0.1392 0.0000 0.79 0.1394 0.0000 4 0.56 0.0708 0.0000 0.56 0.0710 0.0001 5 0.43 0.0393 0.0000 0.43 0.0393 0.0002 6 0.34 0.0253 0.0000 0.34 0.0253 0.0001 7 0.28 0.0167 0.0000 0.28 0.0167 0.0007 8 0.24 0.0120 0.0000 0.24 0.0120 0.0000 9 0.21 0.0085 0.0015 0.21 0.0085 0.0000 10 0.18 0.0066 0.0001 0.18 0.0066 0.0000

Table 1(b). Mostly Vertical Modes of Vibration and Participation Factors Using Correct and Incorrect Vertical Mass Modeling Correct Mass Modeling Incorrect Mass Modeling Mode Period (s) Fph Fpv Period (s) Fph Fpv 1 0.20 0.0001 0.0872 0.40 0.0000 0.3271 2 0.19 0.0008 0.0001 0.36 0.0016 0.0001 3 0.17 0.0008 0.0018 0.29 0.0008 0.0164 4 0.15 0.0001 0.0041 0.23 0.0003 0.0000

13

Table 2. Average % Change in Response Quantities Resulting from Incorrect Vertical Mass Modeling (20-story L.A. SAC building) Nisqually quake Loma Prieta quake Horiz. motion Combined motion Horiz. motion Combined motion 2nd floor moments 10th floor moments 19th floor moments exterior column axial forces interior column axial forces 1.3 -0.2 0.4 5.5 12.3 1.3 -0.1 0.3 14.0 21.8 0.0 -0.1 -0.9 3.8 7.4 0.0 -0.1 -0.9 -10.2 -29.6

14

Tributary area for Tributary area for horizontal mass vertical mass

Tributary area for Tributary area for horizontal mass vertical mass

North-South Moment Frame (a) Tributary area for horizontal mass

North-South Moment Frame (b) Tributary area for vertical mass

North-South Moment Frame (c)

Fig. 1

15

Shapes for correct and incorrect vertical-mass modeling are indistinguishable

10

Fig. 2

16

Correct Vertical-Mass Modeling

Incorrect Vertical-Mass Modeling

Fig. 3

17

100

80

Horizontal Vertical

Sv (in/s)

60

40

20

0 0.1 1 10

Period (s)

Fig. 4

18

Horizontal Vertical

Sv (in/s)

0 0.1 1 10

Period (s)

Fig. 5

19

Das könnte Ihnen auch gefallen