See discussions, stats, and author profiles for this publication at: [Link]

net/publication/235962585

Computation of Rayleigh Damping Coefficients for structures with large degrees

of freedom

Article in Electronic Journal of Geotechnical Engineering · April 2003

CITATIONS READS

6 1,779

2 authors:

Indrajit Chowdhury Shambhu Dasgupta

Independent Researcher Indian Institute of Technology Kharagpur
84 PUBLICATIONS 630 CITATIONS 46 PUBLICATIONS 726 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Indrajit Chowdhury on 16 May 2014.

The user has requested enhancement of the downloaded file.

 Computation of Rayleigh Damping Coefficients for Large
Systems
By

Indrajit Chowdhury
Chief Discipline Supervisor
Civil & Structural Engineering
Petrofac International Limited
Sharjah, U.A.E.

and

Shambhu P. Dasgupta
Professor
Department of Civil Engineering
Indian Institute of Technology
Kharagpur 721302, India.
e-mail: dasgupta@[Link]

Summary

For systems with large degrees of freedom, it is difficult to guess meaningful

values of Rayleigh damping coefficients α and β at the start of an analysis. A number of
general-purpose commercially available have the provision of providing the value of α
and β for calculation of Rayleigh damping matrix for dynamic analysis of systems with
multi-degree of freedom. Since an engineer may not be in a position to pre-assess the
same at the beginning, has no option but to assume an unrealistic constant damping ratio
for all modes. Based on the present technique it is very simple to develop a spreadsheet
and arrive at a rational value of α and β which develops a damping ratio sequence
increasing progressively with each of the subsequent modes and one can furnish input
data for the dynamic analysis.

The present paper outlines a procedure, which ensures a rational estimate of α and
β even for a system with large degrees of freedom. The results obtained have been
checked against different class of real life structure and foundation systems and the
results are presented graphically.

Key-Words: Dynamics, damping, eigen values, finite element method, natural

frequency, soil-foundation system.
 Computation of Rayleigh Damping Coefficients for Large
Systems
Introduction

In dynamic analysis of structures and foundations damping plays an important role.

However due to the limitation in our knowledge about damping the most effective way to
treat damping within modal analysis framework is to treat the damping value as an
equivalent Rayleigh Damping in form of

[C] = α [M] + β [K] (1)

in whch [C] = damping matrix of the physical system; [M] = mass matrix of the physical
system; [K] = stiffness matrix of the system; α and β are pre-defined constants.

The major advantage gained in converting the damping matrix into an equivalent
Rayleigh damping lies in the fact that using orthogonal transformation a structure having
n degrees of freedom can be reduced to n-number of uncoupled equations. However, for
systems with large degrees of freedom, it is difficult to guess meaningful values of α and
β at the start of the analysis.

As such in most of the practical engineering analysis the analyst makes

simplifying assumptions in selecting damping ratios (constant for all significant modes)
based on his experience or standard literature that would hopefully be valid for the
overall system. It is a fact that modal mass participation decreases with increase in modes
e. g. say, for first mode mass participation be 45%, second mode 20%, third mode 10%
etc… till nearly 100% mass participation is achieved. Based on above, one can infer that
[ω = √(k/m)], as mass participation decreases with higher modes, the frequency increases
and it is indeed an observed phenomenon. Considering cc (critical damping) = 2√(km),
we can conclude that with reduction in modal mass for successive modes, cc will decrease
with increase in mode. Overall damping of a system being a constant (since k and m are
constant for a system), the damping ratio, D, is given by D = c/cc. As cc decreases with
increase in modes, D will increase with increasing modes. Thus the main digression from
reality in such case is that while damping goes on increasing with each mode with a
guessed unique value of damping ratio at the advent of the analysis the damping ratio
remains constant for all modes. For a particular system where higher mode contribution is
significant the results obtained based on the presumptive damping ratio will surely not be
realistic.

The present paper outlines a procedure, which ensures a rational estimate of α and
β even for a system with large degrees of freedom. The results obtained have been
checked against different class of real life structure and foundation systems and the
results are presented graphically.
 Basic Formulation on Rayleigh Damping

A system having multi-degrees of freedom, the equation of motion under externally

applied time dependent force is given by

{}
&& + [C] X
[M] X {}
& + [K ]{X} = {P }
t (2)

in which {Pt } force vector which is a function of time.

By orthogonal transformation, the above equation reduces to

{} {}
{φ }T [M ]{φ } ξ&& + {φ }T [C]{φ } ξ& + {φ }T [K ]{φ }{ξ } = {φ }T {Pt } (3)

Eqn. (3), subsequently reduces to an n-uncoupled equations of the form

{ξ&&j}+ 2ζ jω j{ξ&j}+ ω 2j {ξ j} = {Pj(t)} (4)

in which {ξ } = displacement of the structure in the transformed co-ordinate; ζ = damping

ratio in uncoupled mode; ω = natural frequency of the system; {P(t )} = modified force
vector in transformed co-ordinate; {φ } = normalised eigen vector of the system.

The above orthogonal transformation is valid only when the damping matrix is
proportional i.e. it is some function of the mass and stiffness matrix [M] and [K]. It is for
this reason that the damping in the form, shown in eqn. (1), is advantageous as on
orthogonal transformation the damping term in eqn. (3) reduces to

α + βω 2 0 . . 0 
 1 
 0 α + βω 2 2 . . 0 
{φ }T [C]{φ } =  . . . . .
 (5)
 
 . . . . . 
 
 0 . . . α + βω n 2 

Again, from symmetry, it can be inferred that

2ζ1 ω1 = α + β ω12
2ζ2 ω2 = α + β ω22
……………………………
……………………………
2ζn ωn = α + β ωn2 (6)

When the system has two degrees of freedom eqn. (6) reduces to

2ζ1 ω1 = α + β ω12
2ζ2 ω2 = α + β ω22 (7)
To find the values of α and β, one has to solve eqn. (7).
 However, while solving a system having a large degrees of freedom, say 400 or
1000 equations, the analyst is in some difficulty to arrive at the values of Rayleigh
coefficients, which shall be valid for all the n degrees of equation (which could be any
values like 400, 800, 1000 etc) or shall be valid for all significant modes. Surely there is
no straightforward solution to arrive at these values. An iterative solution is possible and
this can be obtained possibly from the best-fit values of α and β in a particular system. A
method is described in the following through which one can arrive at the unique values of
Rayleigh coefficients and they will be valid also for systems having large degrees of
freedom.

Computation of coefficients α and β for large systems

As it is shown in eqn. (5), the orthogonal transformation of the damping matrix reduces
the matrix [C] to the form

2 ζiωi = α + β ωi2 (8)

This, on simplification reduces to

α βω i
ζi = + (9)
2ω i 2
From eqn. (9) it can be observed that the damping ratio is proportional to the natural
βω
frequencies of the system. A typical plot of the equation α + i is as shown in
2ω i 2
Fig.1. The figure demonstrates some interesting result. For the first portion (frequency
range 0.5-8.5 rad /sec) the curve shows marked non-linearity and beyond that the
variation is linear.

0.3

0.25
Damping Ratio(C/Cc)

0.2

0.15

0.1

0.05

0
frequency 6 12 18 24 30 36 42 48
Natural Frequency

Fig. 1 Variation of damping ratio with natural frequency of a system

One need not measure ζn, where n could be 100, 500,1000 depending on the degree of
freedom. What is relevant here is a first few modes for which there is a significant mass
participation. Beyond this, results are of no practical consequence. For instance, for a
steel structure of 500 degree-of-freedom if it is found that 100% mass participation
occurs in the first 15 modes (say), instead of starting with 5% constant damping for all
modes, one can start with a minimum 2% damping in the first mode and define at 15th
mode, ζ= 5% and this is the zone of relevance.

From the curve shown in Fig.1, it can be concluded that for some y = a/x + bx,
when x is small, the first term is a/x dominates at the initial stage and as x increases the
value a/x diminishes and approaches zero and the term bx starts dominating the equation.
In other words for if it is very flexible and have a very low fundamental frequency will
show non linear damping properties in the beginning with respect to frequency and would
converge to a linear proportionality with frequency as the eigen values increases with
each subsequent mode. Flexible antennas, very long piles, or tall chimney (height > 275
m) would possibly show this type of behaviour at the outset.

However, most of the civil engineering structures are usually designed to have a
reasonable rigidity and would have a much higher value of the fundamental frequency,
the term containing β/ 2 will usually dominate. Moreover, considering the fact that the
non-linear range is very small for normal structures it will not be unrealistic to assume
that the damping ratio for each mode is linearly proportional to the frequency of the
system.

Thus, a set of values ω 1,ω 2 , ω 3 .......... ...ω n and ζ 1,ζ 2 , ζ 3 .............ζ n have been
assumed as the corresponding damping ratio for ith mode considering a linear relationship
and the damping ratio thus obtained is given by

ζ m − ζ1
ζi = (ω i − ω1 ) + ζ 1 (10)
ω m − ω1
in whichζi = damping ratio for the ith mode( for all i ≤ m ); ζ1 = damping ratio for the first
mode; ζm = damping ratio for the mth significant mode considered in the analysis; ωi =
natural frequency for the ith mode;ω1 = natural frequency for the first mode; ωm = natural
frequency for the mth significant mode considered for the analysis.

For structures having large degrees of freedom, it is only the first few modes,
which contribute to the significant dynamic behaviour. Now, how many modes will have
a significant contribution can be ascertained from

For most of the engineering structures, the number of significant modes by which
almost 95% of the mass has participated is usually around 3 at the minimum and about 25
at the maximum.

Based on an eigen value solution and modal mass participation result one can identify
the significant modes (= m) and follow the following procedure step by step as shown
hereafter. Select number of modes = 2.5 m and perform an eigen value analysis;
• Select ζ1, the damping ratio for the first mode of the system;
• Select ζm, the damping ratio for the mth significant mode;
• For intermediate modes i, where 1< i < m, obtain ζi from eqn. (10) based on linear
interpolation;
• For modes greater than m extrapolate the values based the expression

ζ m − ζ1
ζi = (ω m +i − ω m ) + ζ 1 , where m < i ≤ 2.5m (11)
ω m − ω1

• Select first set of data consisting of ζ1, ζm, ω1, ωm

• Based on the above sets of data obtain β from the equation

2ζ 1ω1 − 2ζ mω m
β = (12)
ω12 − ω m
2

Back-substituting the value of β in the expression

2ζ1 ω1 = α + β ω12 (13)

obtain the value of α
• Next select a second set of data consisting of ζ1,ζ2.5m, ω1, ω2.5m
• Find out α and β based on eqns. (12) and (13).

Now one has the three sets of data

a). based on linear interpolation
b). based on data set ζ1,ζm, ω1, ωm,
c). based on ζ1,ζ2.5m, ω1, ω2.5m ,
d).obtain a fourth set of data based on the averages of b) and c) as mentioned above

• Plot the four sets of data based on eqn. (9) and check which data fits best with the
linear interpolation curve for the first m significant modes.
• Select the corresponding value of α and β as the desired value, which will give the
incremental damping ratio based on Rayleigh damping.

In some cases it might so happen that values will show variation in higher modes
beyond m significant modes but this is irrelevant, so long as the values are closely
matching for the first m modes, since the contribution of higher modes greater than m are
deemed insignificant for the system, the above theory has been bench marked against
three real life cases consisting of different types of structures and foundations.

Case1: A 220m RCC chimney having 100 degrees of freedom has first six values of
natural frequencies as 3.0,4.0,7.0,8.0,12.0 and 20.0 rad /sec, respectively.
It is assumed that the significant dynamic response of the system will die down within
first six modes with damping ratios varying between 2% to 10% within the first six
modes.
Select suitable values of α and β.
Solution

Given below are the eigen values for the first 15 modes. Now, taking the first
modal damping ratio as 2% and sixth modal damping ratio as 10%, data for the full range
of eigen values interpolated / extrapolated are shown in Table I.

For the first six modes the range values are

ω2 = 20 rad/sec and ζ2 = 0.10 and ω1 = 3 rad/sec and ζ1 = 0.02.
Based on the above values

2 × 0.1× 20 − 2 × 0.03 × 3
β= = 0.009923 and α = 2 × 0.1 × 20 − 0.0009923 × 400 = 0.0307.
400 − 9

Table I._ Values natural frequency versus damping ratio

Sl No Natural Damping Ratio

frequency based on
(rad/sec) Linear
Interpolation
1 3 0.02
2 4 0.024705882
3 7 0.038823529
4 8 0.043529412
5 12 0.062352941
6 20 0.1
7 25 0.123529412
8 32 0.156470588
9 38 0.184705882
10 47 0.227058824
11 62 0.297647059
12 75 0.358823529
13 110 0.523529412
14 135 0.641176471
15 140 0.664705882

For the full range of 15 modes

ω 2 = 140 rad/sec, and ζ 2 = 0.664705882 ; and ω1 = 3 rad/sec and ζ 1 = 0.02

Based on the above values

2 × 0.6647 ×140 − 2 × 0.02 × 3

β= =0.0095and α = 2 × 0.6647 ×1400 − 0.0095 ×19600 = 0.05549
19600 − 9

Thus based on the above two sets of data the average values α and β are: α = 0.04309
and β = 0.009708. Based on the above values, the damping ratios are found to vary as
shown in Table II.
Table II.- Interpolated damping ratios

No of modes Frequency Linear Damping Damping Damping

Damping upto 6th upto full with average
Mode appr- range app- data
oximation roximation
1 3 0.02 0.02 0.02348961 0.021745
2 4 0.024705882 0.023682864 0.025924489 0.024804
3 7 0.038823529 0.036923639 0.037192791 0.037058
4 8 0.043529412 0.041611253 0.041444351 0.041528
5 12 0.062352941 0.060818414 0.059276352 0.060047
6 20 0.1 0.1 0.096327637 0.098164
7 25 0.123529412 0.124654731 0.119785269 0.12222
8 32 0.156470588 0.159251918 0.152771618 0.156012
9 38 0.184705882 0.188946022 0.181116821 0.185031
10 47 0.227058824 0.233523426 0.223700165 0.228612
11 62 0.297647059 0.307868988 0.294762608 0.301316
12 75 0.358823529 0.372327366 0.356396269 0.364362
13 110 0.523529412 0.545919554 0.522424179 0.534172
14 135 0.641176471 0.669934641 0.641052911 0.655494
15 140 0.664705882 0.694738765 0.66478066 0.67976

On plotting the data, the variations are as given in Fig. 2. It can be observed that
damping with 15-mode approximation matches the best with the datum value based on
linear interpolation. Hence, the design Rayleigh coefficients are:α = 0.0095 and β =
0.05549.

Comparison of Damping Modes

0.8

0.7

0.6 Linear Damping

Damping ratio

0.5 Damping upto 6 th Mode

aproximation
0.4
Damping upto full range
0.3 approximation
Damping with average data
0.2

0.1

0
3 4 7 8 12 20 25 32 38 47 62 75 110 135 140
Frequency

Fig. 2_ Variation of damping ratio for a 220m high RCC chimney

Thus, if one takes an input value of α and β as 0.0095 and 0.05549 respectively, the
analysis would show a progressive increment of damping with each mode as shown in
Fig. 2. Without further elaboration it will be shown hereafter the two more case studies
and the variation of results based on the above technique.

Case 2: Dynamic analysis of a boiler feed pump foundation based using FEM (degrees of
freedom = 7926):
 Number of significant modes = 6; Damping ratio = 0.10 for first mode; Damping
ratio = 0.15 for the sixth mode.
The design value of α = 0.05549 and β = 0.0105 based on 15 mode interpolation.
It can be concluded from Fig. 3 that the frequencies are more or less same for the first six
modes showing the rigid body modes of the system and the damping ratio remains more
or less constant at the range of 0.1. Data for the full range of eigen values interpolated /
extrapolated are shown in Table III.

Comparison of Damping Modes

5
4.5
4
Linear Damping
3.5
Damping ratio

3 Damping upto 6 th Mode

aproximation
2.5
Damping upto full range
2 approximation
1.5 Damping with average data
1
0.5
0
18.9 20 20.2 20.2 21.9 28.5 124 128 157 294 313 348 366 412 452
Frequency

Fig. 3_ Variation of damping ratio for a boiler feed pump based on finite element
analysis
Table III._Values natural frequency versus damping ratio

Sl No Natural Damping Ratio

frequency based on Linear
Interpolation
1 18.93 0.1
2 19.97 0.105462185
3 20.19 0.106617647
4 20.24 0.106880252
5 21.91 0.115651261
6 28.45 0.15
7 124.41 0.653991597
8 128.38 0.674842437
9 157.43 0.827415966
10 293.7 1.543119748
11 312.86 1.64375
12 348.45 1.830672269
13 366.48 1.925367647
14 411.57 2.162184874
15 452.34 2.37631302

Case-3: 3D Analysis of a powerhouse frame of structural steel with 18,000 degrees of

freedom:
Number of significant modes = 6; Damping ratio = 0.02 for first mode; Damping ratio =
0.05 for the sixth mode

Design value of α and β = -0.354 and 0.0162 respectively based on interpolation for first
six modes. It is observed in Fig. 4 that for modes beyond the 10th, damping shows slight
lower value than the datum (i.e. the linear damping curve) but in reality it is irrelevant
since 95% of the mass has already participated by the 6th mode. The variation curves are
as shown Fig. 4. Based on the above values, the damping ratios are found to vary as
shown in Table IV.

Comparison of Damping Modes

0.12

0.1 Linear Damping

Damping ratio

0.08 Damping upto 6 th Mode

aproximation
0.06
Damping upto full range
0.04 approximation
Damping w ith average data
0.02

0
11 3
11 4
24
4

6
3

3
8

11 1
22

6
07

60

84
23

69
77

92

26

99

.6
9.

.0

.1
.4
6.

10
6.

6.

7.
8.

8.
8.

8.

9.

9.

Frequency

Fig. 4_ Variation of damping ratio for 3D analysis of a steel power house

building having 18,000 degree of freedom

Table IV._ Values natural frequency versus damping ratio

Sl No Natural Damping ratio

frequency based on linear
Interpolation
1 6.074 0.02
2 6.22 0.021672394
3 6.601 0.026036655
4 7.846 0.040297824
5 8.233 0.044730813
6 8.693 0.05
7 8.778 0.050973654
8 8.927 0.052680412
9 9.263 0.05652921
10 9.8 0.062680412
11 9.991 0.06486827
12 10.61 0.071958763
13 11.053 0.077033219
14 11.164 0.078304696
15 11.424 0.081282932
 Conclusion and Remarks

A number of general purpose commercially available Finite Element Analysis package

like SAP, GTSTRUDL, STARDYNE etc have the provision of providing the value of α
and β for calculation of Rayleigh damping matrix for dynamic analysis of systems with
multi-degree of freedom. Since an engineer is not in a position to pre-assess the same at
the beginning he has no option but to assume a constant damping ratio for all modes,
which is unrealistic. Based on the present technique it is very simple to develop a
spreadsheet and arrive at a rational value of α and β developing a damping ratio sequence
which increases progressively with each of the subsequent modes and can furnish an
input data for the dynamic analysis.

The value furnished by this method gives a more realistic picture for the
behaviour of the structure under dynamic loading than the presumptive damping ratio
constant for all modes.

References

1. Hurty W. and Rubenstein M. Dynamics of Structures, Prentice Hall of India, New

Delhi, 1969.
2. Clough, R.W. and Penzien J. Dynamics of Structures, McGraw Hill Kogakusha,
1975.
3. Craig Jr., R.R. Structural Dynamics, John Wiley, NY., 1981.
4. Wilson, E. L.. and Penzien, J. Evaluation of orthogonal damping matrix, Int. J. Num.
Meth. in Engng. 4, 1972, 5-10.
5. SAP User Manual, University of California, Berkeley.
6. GTSTRUDL Users' Manual, Georgia Inst. of Tech., U. S. A.

View publication stats