Modal damping is kind of the it-spell in the dark art that is modeling viscous damping in structures. Although modal damping is pretty straightforward, you should be aware of an important aspect of its implementation in OpenSees.

The issue is that OpenSees assembles the dynamic tangent in to the matrix storage scheme you choose via the ‘system’ command. This is a problem for modal damping because the damping matrix is full even when the matrix storage, based on the connectivity of the stiffness matrix, is banded, profile, or sparse. As a result, some terms of the dynamic tangent will be zero even though they are non-zero in the modal damping matrix. Convergence problems will ensue–*even for linear problems*–because the dynamic tangent will be inconsistent with the residual.

Let’s describe this issue with a canonical shear frame model.

Using floor mass and story stiffness , the model has the following mass and stiffness matrices

Note that the equations are numbered 1-3-2-4 (use the ‘Plain’ equation numberer) so that the stiffness matrix is not tridiagonal–its bandwith is 3, the profile dips inside the band, and there’s sparsity within the profile. Accordingly, the ‘BandSPD’, ‘ProfileSPD’, and ‘UmfPack’ systems will store different numbers of non-zeros.

Now suppose we want to use 5% damping () in the first mode. Based on the mass and stiffness, the first mode frequency is rad/sec and the corresponding eigenvector and the resulting modal damping matrix are

Assuming Newmark time integration with constant average acceleration ( and ) and time step sec, the consistent dynamic tangent is

With banded storage, the dynamic tangent that OpenSees will use in the analysis is

Notice the zeros at the upper right and lower left corners of the matrix, outside the bandwidth of 3. Because of these zeros, the dynamic tangent used in the analysis will be inconsistent with the residual and your analyses won’t converge in one iteration for linear problems.

Consider the banded matrix solver ‘BandSPD’ in a free vibration analysis of the shear frame. The analysis takes four iterations to find equilibrium at each time step.

Although not shown here, the analysis will take three or four iterations per time step if you switch the solver to ‘ProfileSPD’ or ‘UmfPack’, or any solver that does not store the full matrix.

Because it stores the entire matrix, you can use the ‘FullGeneral’ solver to get the convergence in one iteration that you expect for a linear problem.

Don’t worry though, the analyses will converge to the same solution regardless of which solver you use. However, without getting in to spectral radii, it’s possible that the analysis will not converge at all due to inconsistent zeros in the dynamic tangent. But this is unlikely for realistic values of mass, stiffness, and damping.

For large models, using a solver with full matrix storage will be dreadfully slow. You’ll be better off taking the extra iterations with a fast sparse solver like ‘UmfPack’ than using the ‘FullGeneral’ solver to converge in one iteration. When using a fast sparse solver, don’t use the ‘NewtonRaphson’ algorithm–you don’t want to factorize the same inconsistent tangent over and over. Consider ‘KrylovNewton’ or ‘BFGS’ instead. And don’t use the ‘Linear’ algorithm because it won’t iterate to equilibrium.

Recognizing that one size does not fit all, OpenSees was designed to give you total control over the analysis. Therefore, it’s important you are aware of issues like this one with modal damping so that you can choose analysis options wisely.

Here is an OpenSees Tcl script of modal damping, without all the equation numbering and solver shenanigans.

```
#
# Free vibration analysis of shear frame with modal damping
# Using zero length elements for the model
#
wipe
model basic -ndm 1 -ndf 1
set k 610.0 ;# Story stiffness
set m 1.0352 ;# Floor mass
node 0 0.0; fix 0 1
node 1 0.0
node 2 0.0
node 3 0.0
node 4 0.0
mass 1 $m
mass 2 $m
mass 3 $m
mass 4 [expr 0.5*$m]
uniaxialMaterial Elastic 1 $k
element zeroLength 1 0 1 -mat 1 -dir 1
element zeroLength 2 1 2 -mat 1 -dir 1
element zeroLength 3 2 3 -mat 1 -dir 1
element zeroLength 4 3 4 -mat 1 -dir 1
# Impose displacement at roof
timeSeries Constant 1
pattern Plain 1 1 {
sp 4 1 1.0
}
# Do a static analysis for initial condition
integrator LoadControl 0.0
constraints Transformation
analysis Static
analyze 1
wipeAnalysis
# Remove displacement so it doesn't affect eigen calculations
remove loadPattern 1
set N 2 ;# Number of modes for modal damping
eigen $N
modalDamping 0.05 0.02 ;# 5% in mode 1, 2% in mode 2
# Switch to dynamic analysis
analysis Transient
# Record roof displacement
recorder Node -file roofDisp.out -time -node 4 -dof 1 disp
# Analysis time, step, and num steps
set Tf 5.0
set dt 0.01
set Nsteps [expr int($Tf/$dt)]
analyze $Nsteps $dt
```

There are articles on the effect of viscous damping modeling methods on seismic performance using OpenSees.

As an example, https://www.sciencedirect.com/science/article/abs/pii/S2352012418300730 .

Please write your opinion and experience in this field.

LikeLike

Hello MSB, thank you for the article!

It appears to address only Rayleigh damping and which stiffness matrix to use. There are additional papers on this topic:

https://doi.org/10.1002/eqe.541

https://doi.org/10.1061/(ASCE)0733-9445(2008)134:4(581)

https://doi.org/10.1002/eqe.2357

The following paper also addresses modal damping:

https://doi.org/10.1002/eqe.2622

LikeLike

Thanks.

LikeLike

Hi, can you show me an example with tcl script to explain how can I use modal Damping in OpenSees?

Thank you.

LikeLike

Hello Kris,

1. Define your model

2. Perform an eigenvalue analysis

eigen(N) ;# N is the number of eigenmodes to use in modal damping

3. Issue the modal damping command

modalDamping(0.02) ;# 2% damping in all N modes

# or

modalDamping(0.02,0.05,…,0.04) ;# 2% damping in mode 1, 5% in mode 2, …, 4% in mode N

For the second option, you need to give N values

4. Analyze your model

PD

LikeLike

Thanks.

LikeLike