2001-06

An Approach to Properly Account for Structural Damping, Frequency-Dependent Stiffness/Damping, and to Use Complex Matrices in Transient Response Presented at the MSC.Software Corporation 2001 Aerospace Conference and Technology Showcase by: Ted Rose Manager, MSC.Nastran Training and Support MSC.Software Corporation

Abstract:
Transient response analysis in MSC.Nastran is performed using real arithmetic. Structural damping requires a complex stiffness matrix to properly account for its effects. Unfortunately, these two facts mean that when performing transient response analysis in MSC.Nastran, structural damping must be approximated. Frequency-dependent elements (such as the CBUSH) are available in Frequency Response solutions, but must be linear in Transient Response. Adding a complex matrix into a transient response solution is not allowed, meaning that the user must come up with a way to "convert" their complex matrices into equivalent real matrices. This paper provides an approach to properly account for structural damping, frequency-dependent elements, and to include complex matrices in transient response by using the Fourier Transform approach. In addition, the approach in this paper allows for a simple solution of a structure with multiple harmonic inputs acting simultaneously (for example a car, which has multiple rotating bodies, each with a different steady-state frequency, acting simultaneously. A list of some of these is: a) engine crankshaft

b) camshaft(s) c) the wheels,

d) driveshaft e) f) alternator power steering pump

and many more. Solving for a steady-state solution of this model would require a transient solution integrating over an extended period of time. Using Fourier transformations and frequency response can do it quickly and efficiently.

-1-

Introduction:
This paper uses standard features of MSC.Nastran to apply Fourier Transforms during a solution for transient response analysis. By doing this, the program will be able to solve steady-state problems with multiple simultaneous input frequencies, problems with structural damping (since transient response analysis in MSC.Nastran uses real arithmetic and structural damping requires a complex stiffness matrix, there is a conflict whenever you wish to apply structural damping in a transient response solution), problems including frequency-dependent elements, and problems with complex matrices. The approach is based on the information provided in Dean Bellingers 1995 paper1 . In his paper, he describes the approach in MSC.Nastran to use a Fourier Transform to perform a transient response solution in the frequency domain. Since the frequency response solutions in MSC.Nastran use complex arithmetic and support frequency-dependent elements, the above items are handled correctly in the solution.

Getting the Load Right (Selecting the Frequencies for the Fourier Transform)
As mentioned in Dean Bellingers paper1 and the Application Note on Fourier Transforms in MSC.Nastran4 , performing a Fourier Transformation is very simple to do. Simply set up the file for a transient response solution, then change the solution to SOL 108 (direct frequency response) or SOL 111 (modal frequency response) and add a FREQ case control command with the associated FREQi entries in the bulk data section. The proper selection of the frequencies is the hard part. The following (in addition to the information in the two references) will help. I talked to Dean and got the following recommendations from him: 1) 2) 3) 4) Use FREQ1 entries to define the loading frequencies. Use a constant DF (maximum value = 1/T , where T = the duration of the transient event) when defining the frequencies. This is easiest on the FREQ1 Set T2 on the TLOAD2 entry to T (if you are using a TLOAD2). IF YOU DO NOT HAVE A TSTEP (inverse Fourier Transform), set the A field on the DAREA entry to 2*DF*transient_DAREA (DF = frequency step size on the FREQ1, transient_DAREA = EXCITEID on the DAREA entry in the transient analysis).

In selecting the frequencies, it is very important to be sure that you are able to represent the loading properly when performing the Fourier transformation, so I recommend that you make up a simple model first to verify the loading and Fourier transformation. The simple model is a single point with a mass of 1.0 on it (no stiffness or damping). The acceleration response will be equal to the input loading, so this is a good model use to verify the method. To demonstrate this, I will use a simple loading, which is a time-decaying 1.0 hz sine wave. Although this loading is very simple to define in the time domain (TLOAD2 entry), it is a very complicated loading in the frequency domain, as there will be contributions from many frequencies. The input file to verify the input loading for transient response is:
SOL 109 CEND TITLE = TRANSIENT RESPONSE - PLOT INPUT LOADING $ DLOAD = 10 TSTEP = 20 DISP(PLOT)=ALL ACCEL = ALL OLOAD = ALL

BEGIN BULK TLOAD2,10,25,,,0.,10.,1.,-90. ,.5 DAREA,25,1,1,1. TSTEP,20,1000.,.01, GRID,1 CONM2,100,1,,1. ASET1,1,1 ENDDATA

In this file, T is 20., so we will use DF of .05 in the frequency response file. The file converted into a frequency response run to use the Fourier transformation is:
SOL 108 diag 8,15,56 compile sedfreq alter 'frlg.*ppf' message //' ' $ message //' loading transformed into the frequency domain' $ message //' ' $ matprn ppf// $ tabprt frl/////1 $ CEND TITLE = FREQUENCY RESPONSE - PLOT INPUT LOADING $ DLOAD = 10 TSTEP = 20 FREQ=99 DISP(phase,PLOT)=ALL ACCEL(phase) = ALL OLOAD = ALL OUTPUT (XYPLOT) $ XGRID=YES YGRID=YES XYPLOT ACCE / 1(T1) BEGIN BULK freq1,99,.05,.05,60 TLOAD2,10,25,,,0.,10.,1.,-90. ,-.5 DAREA,25,1,1,1. TSTEP,20,1000,.01, GRID,1,,,,,,3456 Spcoff,1,1 CONM2,100,1,,1. $ $ small spring for limitation 41260 $ celas2,1111,.00001,1,2 ENDDATA

The DMAP in the above file is to print out the loading transformed into the frequency domain (NOTE: if you remove the TSTEP case control command, the output will be in the frequency domain). Plots from the transient run and the frequency response run follow.

Figure 1. Time History of loading from the transient run

Figure 2. Time history of load from frequency response run

Figure 3. Plot of the loading in the frequency domain (Fourier transformation) Figures 1 and 2 compare well, indicating that the Fourier transformation worked well for this example. Figure 3 shows the loading converted into the frequency domain (the second file without the TSTEP case control request). NOTE: There is no checking in the program to determine whether your Fourier Transform is valid. This step is up to you!

Multiple Simultaneous Harmonic Loads

When you have multiple simultaneous harmonic loads, which are not all acting at the same frequency, performing a solution requires either a very long transient integration (to get the steady-state solution) or using Fourier Transforms and frequency response analysis. For this situation, I will use a simple model with multiple loads. The model is a cantilever beam with three simultaneous harmonic loads. All of the loads will represent rotating imbalances and each will have a different frequency associated with it. This might compare to a turbine with a transmission. This structure has multiple components, each rotating at a different rate. In the example, the 3 input loads will be out-of-phase with each other and one will be rotating in the opposite direction of the others. This is accomplished by using the phase angle on the TLOAD2 entries. Each load consists of a force (mr 2 where m= mass, r= radius, = rotational frequency) which acts in the radial direction as the mass is rotating. In order to describe this for the program, two TLOAD2 entries will be used for each rotating imbalance with FORCE entries to provide the two components of the load ( for

this model Y and Z), which will be 90 degrees out of phase. A single DLOAD entry will be used to apply all of the components simultaneously. The model used for this is a simple cantilever beam. It is shown below . The three imb alanced loads are on GRIDs 3, 6, and 9.

. Figure 4 - Cantilever Beam Model For this example, the loads are defined as rotating at 1.0, 2.0, and 3.0hz. The input file for a transient solution is shown below:
$ multiple_rotating_imbalances_trans.dat - cantilever beam model $ with multiple imbalances - transient solution until steady-state $ $ for this model use modal transient with 1% of critical damping on $ each mode sol 112 cend title = cantilever beam model with rotating imbalances subtitle = transient analysis $ spc = 1 dload = 1000 method = 10 disp(plot)=all tstep = 2000 sdamp = 1 $ OUTPUT (XYPLOT) XGRID=YES YGRID=YES xtitle applied load XYPLOT OLOAD / 9(T2) xtitle acceleration

XYPLOT ACCE / 9(T2) xmin 15. xmax 20. xtitle applied load 15 to 20 sec XYPLOT OLOAD / 9(T2) xtitle acceleration 15 to 20 sec XYPLOT ACCE / 9(T2) begin bulk $ model include 'cantbeam.dat' $ $ multiple rotating imbalances $ param,resvec,yes dload,1000,1.,1.,1002,1.,1003,1.,2002 ,1.,2003,1.,3002,1.,3003 $ tload2,1002,12,,,0.,100.,1.,-90. tload2,1003,13,,,0.,100.,1.,0. force,12,9,,10.,,1., force,13,9,,10.,,,1. $ tload2,2002,22,,,0.,100.,2.,90. tload2,2003,23,,,0.,100.,2.,0. force,22,6,,10.,,1., force,23,6,,10.,,,1. $ tload2,3002,32,,,0.,100.,.5,0. tload2,3003,33,,,0.,100.,.5,90. force,32,3,,10.,,1., force,33,3,,10.,,,1. $ eigrl,10,,,10 tabdmp1,1,crit ,0.,.01,10000.,.01,endt $ tstep,2000,20000,.001,10 enddata

In this file, modal transient response (SOL 112) is used with residual vectors to obtain a solution. The following file is used to do the same thing in a modal frequency response (SOL 111) run:
$ multiple_rotating_imbalances_freq.dat - cantilever beam model $ with multiple imbalances - transient solution until steady-state $ sol 111 cend title = cantilever beam model with rotating imbalances subtitle = frequency response analysis $ spc = 1 dload = 1000 method = 10 disp(plot)=all tstep = 2000 freq = 500 sdamp = 1 OUTPUT (XYPLOT) XGRID=YES YGRID=YES xtitle applied load XYPLOT OLOAD / 9(T2) xtitle acceleration XYPLOT ACCE / 9(T2) begin bulk include 'cantbeam.dat' $ $ multiple rotating imbalances $

param,resvec,yes dload,1000,1.,1.,1002,1.,1003,1.,2002 ,1.,2003,1.,3002,1.,3003 $ tload2,1002,12,,,0.,10.,1.,-90. tload2,1003,13,,,0.,10.,1.,0. force,12,9,,10.,,1., force,13,9,,10.,,,1. $ tload2,2002,22,,,0.,10.,2.,90. tload2,2003,23,,,0.,10.,2.,0. force,22,6,,10.,,1., force,23,6,,10.,,,1. $ tload2,3002,32,,,0.,10.,.5,0. tload2,3003,33,,,0.,10.,.5,90. force,32,3,,10.,,1., force,33,3,,10.,,,1. $ eigrl,10,,,10 tabdmp1,1,crit ,0.,.01,1000.,.01,endt tstep,2000,5000,.001 freq1,500,.1,.1,19

enddata In these runs, the Y-direction acceleration for GRID 9 is plotted vs time. The following plot is the complete time history from the SOL 112 run.

Figure 5 - Complete Time History of Transient Response with 3 Rotating Imbalances From this plot, it is apparent that the structure has reached a steady-state response by 10 seconds into the integration. For safety, the following plot shows the last 5 seconds of the time history.

Figure 6 - Steady State Response from Transient Response with 3 Rotating Imbalances For comparison, here is the plot from the SOL 111 (modal frequency response) run.

Figure 7 - Steady State Response from Frequency Response with 3 Rotating Imbalances

From the above, it becomes obvious that we can use frequency response with transient loads to get the steady-state response for a structure undergoing multiple simultaneous oscillating loads.

Structural Damping
Handled properly, structural damping is included in the equations of motion as complex terms in the stiffness matrix. In MSC.Nastran, structural damping is defined using material damping Ge on the MATi entries or PARAM,G to get overall structural damping. This is described in references 2 and 3. When properly accounted for, these terms are added into the stiffness matrix as follows: [Ktotal ] = [K](1+iG) + iKeGe Where: [K] = stiffness due to elements +K2GG matrix G = overall structural damping coefficient PARAM,G Ge = structural damping value from MATi entries Ke = element stiffness matrix Which gives the equation of motion: [M]{a} + [B]{v} + [Ktotal ]{u} = {p} where: {a} = acceleration as a function of time {v} = velocity as a function of time {u} = displacement as a function of time {p} = force as a function of time [M] = Mass matrix [B] = viscous damping matrix In frequency response the above equation is transformed into the frequency domain and solved. Since transient response analysis does not include complex terms, the structural damping has to be converted into an equivalent viscous damping. Details of this conversion (and an explanation of how it works) are presented in the dynamics users guide2 . For transient analysis, the viscous damping matrix is: [Btotal ] = [B] + [K]G/W3 + keGe/W 4 Where W 3 and W 4 are user-provided parameters. They represent the dominant frequency of the response in radians per second. If the response is a steady state response and the user-provided parameters W 3 and W 4 are set to the response frequency, this approximation is exact. Unfortunately, in transient response, there is normally no single dominant frequency of response and the above is simply an approximation, which may or may not be reasonable. Whenever structural damping is used in a transient response, you have introduced an approximation in the damping, which is hard to measure. This approximation might result in unacceptable results without any indication. If the structural damping could be properly handled in transient response, this approximation could be avoided.

10

There are two obvious solutions to this problem; 1) 2) use complex arithmetic in transient response transform the loading from the time domain into the frequency domain and solve using frequency response

Since using complex arithmetic in transient response would require major DMAP or re-writing of the program (not to add, this would still not provide support for frequency-dependent elements), this paper deals with the second solution. Using the model with 3 rotating imbalances, I will remove the modal damping (1% of critical damping) and use structural damping. In an "ideal" world, at resonance, the ratio between the critical damping ratio and the structural damping is: = g/2 where: = critical damping ratio g = structural damping coefficient Using this relationship, applying g=.02 should give the same damping at resonance. Unfortunately, the loads are not being applied at resonance, so the relationship will not be exact for this problem. In order to demonstrate the potential for error in applying structural damping in transient response, I will apply an imbalanced load at the natural frequency of the first mode (6.3hz), and to exaggerate the potential for error, I will set PARAM,W3 to 1256.6 (200hz) by "mistake". This will result in the damping being much lower than it should be in the transient solution. PARAM,G,.02 will be added to both input files (transient and frequency response).
$ structural_damping_freq.dat - cantilever beam model $ with a single load a resonance for mode 1 $ $ for this model use Structural damping - G=.02 $ sol 111 cend title = cantilever beam model with structural damping subtitle = frequency response analysis spc = 1 dload = 1000 method = 10 disp(plot)=all tstep = 2000 freq = 500 OUTPUT (XYPLOT) XGRID=YES YGRID=YES xtitle applied load XYPLOT OLOAD / 9(T2) xtitle displacement XYPLOT DISP / 9(T2) begin bulk $ model include 'cantbeam.dat' $ $ structural damping $ param,g,.02 $ param,resvec,yes $ $ load at 6.3 hz $ dload,1000,1.,1.,1002,1.,1003 $ tload2,1002,12,,,0.,10.,6.3,-90. tload2,1003,13,,,0.,10.,6.3,0.

11

force,12,9,,10.,,1., force,13,9,,10.,,,1. $ eigrl,10,,,10 tabdmp1,1,crit ,0.,.01,1000.,.01,endt $ tstep,2000,5000,.001,10 freq1,500,.1,.1,69 enddata

Note that this run has a single steady-state harmonic load, so it can be run as frequency response using RLOAD1 entries. Replacing the TLOAD2 entries by RLOAD1 entries and loading only at 6.3 hz gives a steady state response with a maximum amplitude of 12.13797 at GRID 9, while the run above gives 12.13799. The difference is in the 7th significant figure. Setting PARAM.W3,1256.6 (200hz) gives a maximum response in SOL 112 of 81.66032, which is noticeably different. NOTE that this difference is caused mainly by the incorrect value of W3. As the response should be at the loading frequency, W3 should be set to 39.584 (6.3hz). The purpose of this example was simply to demonstrate how wrong the answers could be if a transient response has the wrong value for W3 when structural damping is used. The fact that the loading is at a resonant frequency of the model exaggerates the difference.

Frequency-dependent Elements
MSC.Nastran has several frequency-dependent elements, including the CBUSH. In frequency response, the properties of these elements are updated to the proper values for each loading frequency. In transient response, these elements use the nominal properties provided by the user. Therefore, it is up to the user to select an acceptable set of properties for the frequency-dependent elements, which might lead to incorrect results. Using a Fourier Transform and solving the problem in the frequency domain allows the program to use the "correct" properties at each frequency. To demonstrate this, I will add the following bushing element, which has frequency-dependent stiffness, into the sample files used for the multiple rotating imbalances.
CBUSH,100,100,9,,,,,0 PBUSH,100,K,10.,10.,10. PBUSHT,100,K,102,102,102 TABLED1,102, ,.5,5.,1.,10.,2.,15.,endt

Adding this element into the model will use the "nominal" stiffness of 10.0 for the transient run, but will use a stiffness of 5.0 when the load is at .5hz, 10.0 when the load is at 1.0hz, and 15. when the load is at 2.0hz in the frequency response run. The resulting XYPLOTs for acceleration at steady state are shown below. The one on the left is from the frequency response (fourier transform) run :and the one on the right is from the transient solution. The results from the transient response run are approximately 10% higher at the peaks, indicating that accounting for the frequency-depenent stiffness terms made a noticeable difference in the solution.

12

Figure 8 - Frequency Response (fourier transform) and transient results including a frequency-dependent spring with multiple rotating imbalances.

Matrices with Complex terms

This one is easy, simply select the matrices using the K2PP, M2PP, B2PP, or TF Case Control commands and the program will include them in the solution. Note that the "G" matrices (K2GG, etc) must be symmetric and real, while the "P" matrices do not have this requirement. In transient response analysis, complex terms are not allowed in the matrices, so the inclusion of complex coefficients would result in a FATAL message, while frequency response uses complex arithmetic, so complex coefficients are allowed.

Summary and Conclusions

. This paper expands on the information presented by Bellinger in his 1995 paper. In this paper, I show how to apply fourier transforms for specific cases and demonstrate how to include structural damping, frequency-dependent elements, and complex terms in matrices properly into a transient solution

References
1. 2. 3. 4. Bellinger, Dean, Dynamic Analysis by the Fourier Transform Method in MSC/NASTRAN, Proceedings 1995 MSC World Users Conference. Sitton, Grant, MSC/NASTRAN V69Basic Dynamic Analysis Users Guide, MSC.Software, Los Angeles, Ca, 1998 Herting, Dave, MSC/NASTRAN Advanced Dynamics User's Guide, MSC.Software, Los Angeles, Ca, 1997 Fourier Transform Behavior and Usage in MSC/NASTRAN, Application Note, November, 1985, MSC/NASTRAN Application Manual, MSC Software, Los Angeles, Ca

13