Professional Documents
Culture Documents
Dynamics of Structures: Jorma K. Arros
Dynamics of Structures: Jorma K. Arros
3
Dynamics
of Structures
3.1 Introduction
3.2 Single-Degree-of-Freedom System
Equation of Motion · Solution of the Equation of Motion
· Evaluation of Damping in SDOF Systems · Response to
Impulsive Loads · Approximate Analysis of Impulsive-Load
Response · Duhamel Integral · Response to Periodic Loading
3.3 Multidegree-of-Freedom Systems
Equations of Motion · Vibration Mode Shapes and Frequencies
· Modal Equations of Motion · Earthquake Response Analysis
Jorma K. Arros · Nonlinear Analysis
ABS Consulting Defining Terms
Oakland, CA References
3.1 Introduction
The essence of earthquake effects on structures is the dynamic nature of earthquake loading. This section
provides an overview of the dynamics of structures, as a foundation for succeeding sections.
Mechanics as a branch of physics is subdivided into statics and dynamics. Statics studies systems in
static equilibrium, i.e., in a state where the system internal forces counterbalance external forces acting
on the system. Static refers to the fact that the state of the system and the applied forces do not vary in
time; they are time-independent. Dynamics is the study of systems subject to time-varying applied forces.
As a consequence of the time variability of the applied forces, the system’s internal forces and its state
(defined in terms of displacement and deformation) also vary with time — the system’s response involves
motion. While a static problem has a single time-independent solution, the solution of a dynamic problem
involves a description of the system’s state at every time point within the period of study. The appearance
of inertia effects associated with mass in motion is another key distinction of dynamic problems.
Structural dynamics can be considered as the study of a body or structure in dynamic equilibrium.
The mathematical expression of this equilibrium is the equation of motion. While the static equilibrium
equation expresses the balance between the structure’s internal forces and externally applied forces, the
equation of motion expresses the equilibrium of internal and external force terms (which are exactly the
same as in the static equilibrium equation) and the mass inertia and damping effects. As the inertia term
involves the second derivative and the damping term the first derivative of the displacement with respect
to time, the equation of motion is a second-order differential equation with constant coefficients. Theory
for this type of differential equation is well established in mathematics and provides ready tools, both
analytic and numerical, for solution of structural dynamics problems.
This chapter begins the discussion with the dynamics of a single-degree-of-freedom (SDOF) system
in Section 3.2. Dynamics of multidegree-of-freedom (MDOF) systems is discussed in Section 3.3. Typical
aspects of dynamic analysis specific to earthquakes are discussed in Section 3.3.4, while dynamic analysis
of nonlinear systems is reviewed briefly in Section 3.3.5. The discussion of dynamics in this chapter
specifically addresses response of structures to seismic ground motion.
f1 + f D + fS = p (t ) (3.1)
Each of the forces represented on the left side of this equation is a function of the displacement u, or
of its derivatives as follows.
Per d’Alembert’s principle, the inertia force is:
f1 = mu˙˙ (3.2)
f D = cu˙ (3.3)
Spring force:
fS = ku (3.4)
FIGURE 3.1 Single-degree-of-freedom system and a free body diagram of the mass block.
By substituting Equations 3.2, 3.3, and 3.4 in Equation 3.1, the equation of motion of the SDOF system
is written as:
f1 + f D + fS = p (t ) (3.6)
where the damping and elastic forces may be expressed as in Equations 3.3 and 3.4, except that the
expressions are now in terms of the relative displacement ur , i.e., displacement relative to a frame of
reference attached to the support points that are moving relative to the reference axis. The inertia force
in this case is given by:
f1 = mu˙˙t (3.7)
where ut represents the total displacement of the mass from the reference axis and is equal to the sum
of the relative displacement and the ground motion:
ut = ur + u g (3.8)
Substituting for the inertia, damping, and elastic forces in Equation 3.6 yields:
FIGURE 3.2 (a) Support point excitation and (b) the free body diagram of the mass.
In this equation, peff (t) denotes the effective support excitation loading. By comparing Equations 3.5
and 3.10, it can be concluded that the relative motion of the system, ur(t), excited by support point
motion üg(t), will be the same as the total motion of a fixed base system, u(t), acted upon by a force
equal to peff (t) = –müg(t).
u (t ) = uc (t ) + u p (t ) (3.11)
Since the equation of motion is a second-order differential equation, the complementary solution has
two constants of integration that are evaluated based on the initial conditions, i.e., u(0) and u· (0).
The complementary solution is the solution of the homogeneous equation, i.e., the equation of motion
with the right-hand side set equal to zero:
Motions that satisfy Equation 3.12, with no forcing function, are referred to as free vibrations. The
solution of Equation 3.12, as of any linear homogeneous ordinary differential equation with constant
coefficients, is of the form:
u (t ) = Ce st (3.13)
Substituting this into Equation 3.12, and after dividing by Cest, leads to the characteristic equation:
ms 2 + cs + k = 0 (3.14)
k
ω2 = (3.15)
m
s is solved:
2
c c
s=− ± −ω
2
(3.16)
2m 2m
It is evident from Equation 3.16 that the nature of the solution will depend on the damping value c.
If c > 2mω, s will be real valued, but if c < 2mω, s will be a complex number. How this affects the response
will be explored in the following sections.
3.2.2.1 Undamped Free Vibrations
If the system is undamped, i.e., if c = 0, s becomes, by Equation 3.16:
s = ±iω (3.17)
and the response given by Equation 3.13 is:
As both of the terms on the right-hand side, irrespective of the values C1 and C2, are solutions to
Equation 3.12 and the equation is linear, the sum is also a solution. As we are looking for the general
solution, both terms are included.
By utilizing Euler’s equation:
This type of motion is called a simple harmonic motion. The quantity ω is the natural angular
velocity of the undamped system (sometimes also referred to as the natural angular frequency) and is
related to natural frequency f as:
ω
f = (3.21)
2π
The reciprocal of f is called the natural period T:
2π 1
T= = (3.22)
ω f
Constants A and B are determined based on the initial conditions: the displacement u(0) = B and
·
velocity u(0) = Aωt at time t = 0, resulting in:
u˙ (0)
u (t ) = sin ωt + u (0) cos ωt (3.23)
ω
By utilizing the relationships of the trigonometric sine and cosine functions, the motion u(t) of
Equation 3.23 can be recast into:
The response is given by the real part, or horizontal projection, of the two rotating vectors. Thus, the
amplitude of motion is given by the resultant:
u˙ (0)
2
[u (0)]
2
R= + (3.25)
ω
u˙ (0)
θ = tan −1 (3.26)
ωu(0)
2
c c
s=− ± −ω
2
(3.16)
2m 2m
c c = 2 mω (3.27)
cc
s=− = −ω (3.28)
2m
u (t ) = (C1 + C 2t ) e − ωt (3.29)
which is easily verified by substituting in Equation 3.12. By imposing the initial conditions the response
is written as:
[ ]
u (t ) = u (0) (1 + ωt ) + u˙ (0) t e − ωt (3.30)
It is readily observed from Equation 3.30 that the critically damped response does not involve oscil-
lations about the zero deflection point and the displacement returns to zero in accordance with the
exponential decay term. Critical damping is the smallest amount of damping that keeps a SDOF system
from oscillating during free response.
3.2.2.2.2 Underdamped Systems
If the damping is less than critical, i.e., c < 2mω, it is customary to express the damping as a ratio to the
critical damping value:
c c
ξ= = (3.31)
c c 2 mω
where ξ is called the damping ratio. Substituting Equation 3.31 into Equation 3.16 leads to:
s = −ξω ± iω D (3.33)
ωD is called the damped vibration frequency. Note that for typical structures damping ratios rarely exceed
about 10% (ξ < 0.10), and ωD differs very little from the undamped natural frequency.
By substituting to Equation 3.13, the response is written as:
(
u (t ) = e − ξωt C1e iωDt + C 2e − iωDt ) (3.34)
The second term in Equation 3.35 is of the same form as the simple harmonic motion of the undamped
oscillator, Equation 3.20, except now at the damped, slightly lower frequency. The first term in Equation
3.35 indicates exponential attenuation of the oscillations. Constants of integration A and B are again
·
determined based on the initial conditions u(0) and u(0) as before.
3.2.2.3 Harmonic Loading
The particular solution is a solution that satisfies the equation of motion with the loading included on
the right-hand side of the equation, Equation 3.5:
mu˙˙ (t ) + cu˙ (t ) + ku (t ) = p (t )
The particular solution is particular to the loading function. As the first case, the response of the SDOF
system to a harmonic forcing function is studied. The equation of motion is written as:
The complementary solution of this equation is the free-vibration response of Equation 3.20.
In this case, the particular solution can be found by “guessing” that the solution is in the form of
harmonic motion, i.e.,
u p (t ) = C sinω pt (3.38)
p0 1
C= (3.39)
( )
2
k 1− ω ω
p
p0 1
u p (t ) = sin ω p t (3.40)
k 1− ω ω 2
p ( )
The general solution is the combination of the complementary solution, Equation 3.20, and the
particular solution, Equation 3.40:
p0 1
u (t ) = uc (t ) + u p (t ) = A sin ωt + B cos ωt + sin ω pt (3.41)
( )
2
k 1− ω ω
p
Again, constants A and B are solved based on initial conditions. For the specific case of a system that
·
is initially at rest, i.e., u(0) = 0 and u(0) = 0, the constants are readily solved as:
p0 ω p ω 1
A=−
( )
2
k 1− ωp ω (3.42)
B=0
and the expression for the response becomes:
u (t ) =
p0 1
(sin ω t − ω ω sin ωt ) (3.43)
( )
2 p p
k 1− ω ω
p
Note that the first term on the right-hand side of Equation 3.43, p0 /k, is the displacement, ust , which
force p0 would cause if applied statically. The rest of the right-hand side of Equation 3.43 is termed the
response ratio, R(t), and expresses the momentary ratio of the total displacement to the static displace-
ment, i.e.,
u (t ) u (t )
R(t ) = = (3.44)
ust p0 k
Dividing by m, and using notation of Equation 3.31, i.e., c/m = 2ξω, leads to:
p0
u˙˙ (t ) + 2ξωu˙ (t ) + ω 2u (t ) = sin ω pt (3.45)
m
The complementary solution of this equation is the damped free-vibration response given by Equation
3.35 (assuming that the structure is less than critically damped, as is the case for all practical structures):
The particular solution can be found by substituting in Equation 3.45 the following “trial” solution:
(or, alternatively, up (t) = C3 sin (ωp t – θ). Note that, in general, the response of a damped system is not
in phase with the loading.)
Coefficients C1 and C2 (or, alternatively, C3 and θ) can be solved readily by equating (1) those terms
that multiply sin ωpt and (2) those terms that multiply cos ωpt. The general solution is the combination
of the complementary solution, Equation 3.35, and the particular solution, Equation 3.47. After denoting
ωp/ω = α, the general solution is written as:
As before, the constants A and B could be evaluated based on the initial conditions. However, as the
complementary solution is attenuated by the exponential function e−ξωt, and therefore typically diminishes
quickly, the exact determination of the complementary solution is often not called for. Also, consistent
with the fact that it typically diminishes quickly, the complementary solution is often referred to as the
transient response. The second term in Equation 3.48 does not diminish with time and is referred to as
the steady-state response. The steady-state response involves harmonic motion at a frequency equal to
that of the applied loading but out of phase with it as indicated by the presence of both the sin ωpt and
cos ωp t terms.
By manipulation of the trigonometric terms, the steady-state response of Equation 3.48 can be cast
to the following alternative form, directly indicating magnitude R and phase angle θ:
(
u (t ) = R sin ω pt − θ ) (3.49)
where
−1/ 2
p0
( )
1 − α 2 + (2 ξα )
2 2
R= (3.50)
k
2 ξα
θ = tan −1 (3.51)
1 − α2
The ratio of the response amplitude, R, to the static displacement, p0/k, caused by the static application
of force p0 is the dynamic amplification factor, D:
−1/ 2
R
( )
= 1 − α 2 + (2αξ)
2 2
D≡ (3.52)
p0 k
Figure 3.4 shows plots of the dynamic amplification factor as a function of ωp/ω for various levels of
damping.
3.2.2.3.3 Response at Resonance
The condition where the frequency of harmonic loading is equal to the undamped natural frequency of
an oscillator, i.e., α = 1, is called resonance. We readily observe from Figure 3.4 that the peak steady-
state response occurs near the resonance frequency for lightly damped systems and at slightly decreasing
frequency ratios with increasing damping. Based on Equations 3.43 and 3.44, the response ratio of an
undamped system grows to infinity when the frequency ratio α = 1. Equation 3.52 indicates that for a
damped system at resonance, the dynamic amplification factor is inversely proportional to the damping
ratio:
1
Dα =1 = (3.53)
2ξ
FIGURE 3.4 Dynamic amplification factor for harmonic force at varying damping levels.
As indicated above, Dα = 1 of Equation 3.53 is not exactly the maximum value of the amplification
factor, but is close to it. The actual maximum value can be easily determined by using basic differential
calculus. The frequency ratio that maximizes D is found by setting the first differential of the expression
in Equation 3.52 with respect to α equal to zero, resulting in:
α peak = 1 − 2 ξ 2 (3.54)
1
Dmax = (3.55)
2 ξ 1 − ξ2
For typical levels of damping, the difference between Equation 3.55 and the simpler Equation 3.53 is
negligible.
Where proper estimation of damping is important, experimental methods may be useful in determin-
ing the damping ratio. Three simple methods to estimate damping based on measurements are introduced
in the following.
3.2.3.1 Free-Vibration Amplitude Decay
Consider the damped SDOF response expressed by Equation 3.35:
It is evident that rate of decay of the oscillatory response reflects the level of damping, suggesting that if
the decay rate can be measured, the level of damping can be determined. Recognizing that for low to
moderate damping levels, ωD ≈ ω, and consequently TD ≈ T, the ratio from one peak value to the next,
un/un+1 (or any two response values one period apart), are related as:
2π
ξ T
un / un+1 = e ξωT = e T = e 2 πξ (3.56)
ln(un / un+1)
ξ= (3.57)
2π
and if the peaks are measured m cycles apart, i.e., un and un+m, an estimate of the damping ratio can be
computed from:
ln(un / un+m )
ξ= (3.58)
2mπ
The vibrations can be initiated by any convenient method, and only the relative displacement amplitudes
need be measured.
3.2.3.2 Amplification at Resonance
Measurement of amplification at resonance is based on determination of steady-state harmonic response
amplification at (essentially) resonance and requires application of harmonic excitations to the structure
at selected frequencies.
The method is based on the relationship expressed by Equation 3.53, i.e.,
1
Dα =1 = (3.53)
2ξ
1
ξ= (3.59)
2Dα =1
As was pointed out in Section 3.2.2.3.3, Dα = 1 is not exactly the maximum amplification factor, which
is easier to determine by measurement, since the maximum amplification factor is related to the damping
factor as:
1
Dmax = (3.60)
2ξ 1 − ξ 2
ω p = ω 1 − 2ξ 2
For low and moderate damping levels, the square root expression in Equation 3.60 is essentially unity,
and the damping ratio can be expressed as:
1
ξ= (3.61)
2Dmax
The method requires that the frequency-response curve, i.e., the plot of D = D(ωp), for the structure
be constructed by applying a harmonic load at closely spaced frequencies extending up to the resonance
frequency. A typical frequency-response curve for a moderately damped structure is shown in Figure 3.5.
3.2.3.3 Half-Power (Bandwidth) Method
Examination of Equation 3.52 and Figure 3.5 readily reveals that damping affects not only the height of
the peak of the D vs. α curve, but also the width of the peak. The half-power bandwidth method of
estimating damping value is based on a particular relationship between the width of the peak of the D
vs. α curve. Specifically, by using the relationship of Equation 3.52, it can be shown that the difference
between the two frequencies on either side of the peak of the curve, α1 and α2, at which the response
amplitude is reduced to 1 2 times the peak value, is related to the damping ratio as:
ξ≅ 1
2 (α 2 − α1 ) (3.62)
To illustrate the use of this method of estimating the damping ratio, a horizontal line is shown across
( )
the curve at 1 2 times the resonant-response value. According to Equation 3.62, the difference between
the frequencies at which this line intersects the response curve is equal to twice the damping ratio.
u (t ) =
p0 1
k 1 − α2
(
sin ω pt − α sin ωt ) (3.63)
During phase II, for t = t – t1 ≥ 0, the free-vibration motion depends on the displacement u(t1) and
velocity u· (t1) at the end of phase I, and may be expressed as follows (see Equation 3.23):
u˙ (t1 )
u (t ) = sin ωt + u (t1 ) cos ωt (3.64)
ω
umax can be found using the standard approach of differential calculus, i.e., as the value of u at the
point where the first derivative of u(t), i.e., velocity u· (t) is equal to zero.
The computations reveal that if the ratio of the loading and system natural frequency, α = ωp/ω, is
less than 1, umax will occur during phase I, i.e., while the impulsive load is acting. If α >1, umax will occur
during free vibration in phase II with the amplification factor, D, given by:
umax πt
D≡ = 2 sin 1
p0 k T
p0
up = (3.65)
k
The general solution with the integration constants set to satisfy the at-rest initial conditions is easily
found to be:
p0
u (t ) =
k
(1 − cos ωt ) (3.66)
In phase II, the free vibration, with t = t – t1 ≥ 0, is again given by Equation 3.64:
u˙ (t1 )
u (t ) = sin ωt + u (t1 ) cos ωt (3.67)
ω
The maximum response umax can again be found using the approach outlined above for the sine-wave
impulse case, and the amplification factor is as follows:
For
t1 1
≤
T 2
(3.68)
u πt
D ≡ max = 2 sin 1
p0 k T
and for
t1 1
>
T 2 (3.69)
D=2
3.2.4.3 Decaying Triangular Impulse
Response to the decaying triangular impulse shown in Figure 3.8 is solved again in two steps using the
same approach for the rectangular impulse above.
For phase I, the decreasing triangular loading is p0(1 − t/t1), and the particular solution to this loading is:
p0 t
u p (t ) = 1− (3.70)
k t1
p0 sin ωt t
u (t ) = − cos ωt − + 1 (3.71)
k ωt1 t1
For phase II, Equation 3.71 and its first derivative at the end of phase I (t = t1) gives:
p0 sin ωt1
u (t1 ) = − cos ωt1 (3.72)
k ωt1
which can be substituted into Equation 3.67 to obtain the free-vibration response in phase II.
The maximum response is found again from the zero-velocity condition. For very short duration
loading (t1/T < 0.4), the maximum response occurs during the free vibrations of phase II; otherwise it
occurs during the loading interval (phase I).
For phase I, during the application of the loading, for 0 ≤ t ≤ t1, with zero initial conditions, the general
solution is:
p0 t sinωt
u (t ) = − (3.74)
k t1 ωt1
During application of loading in phase I, for t = t – t1 ≥ 0, is again given by Equation 3.64, which
after determining u(t1) and u· (t1) at the end of phase I and substitution becomes:
p0 1 sin ωt1
u (t ) = (1 − cos ωt1 ) sin ωt + 1 − cos ωt (3.75)
k ωt1 ωt1
Figure 3.10 shows the dynamic amplification factor D as a function of t1/T for the four idealized
impulsive load shapes.
t1
∫ 0
˙˙ = m∆u˙
mudt (3.76)
∫ [ p (t ) − cu˙ (t ) − ku (t )]dt
t1
m ∆u˙ = (3.77)
0
where ∆u̇ represents the change of velocity from time zero to t1.
The velocity u˙ (t1) and displacement u(t1) developed over the period from time zero to t1 are very
small, the damping and elastic energy terms in Equation 3.77 can be ignored1 and Equation 3.77 rewritten
as:
t1
∫
1
m ∆u˙ = p (t ) dt (3.78)
m 0
The response after the termination of the loading is a free vibration expressed by Equation 3.67:
u˙ (t1 )
u (t ) = sin ωt + u (t1 ) cos ωt (3.79)
ω
1Note that as the left-hand side and the integrated impulse term are of the order of t1, while the integrated term
associated with damping is of the order of t12 and the elastic energy term of the order of t13, the latter two terms
become negligible when t1 approaches zero.
FIGURE 3.10 Dynamic amplification factor for four types of impulse loads as a function of t1 /T.
–
where t = t – t1.
· and t ≅ –t, the following approximate relationship
· ) = ∆u,
As u(t1) is negligibly small and velocity u(t1
may be used:
1 t1
u (t ) ≅
mω ∫
0
p (t ) dt sin ωt
(3.80)
p (τ) dτ
du (t ) = sin ω (t − τ) (3.81)
mω
du(t) represents the differential response contribution of the impulse p(τ)dτ to the total response, which
is obtained by integrating Equation 3.81 as:
∫ p (τ) sin ω (t − τ) dτ
1
u (t ) = (3.82)
mω 0
Equation 3.82, known as the Duhamel integral, can be used to obtain the response of an undamped2
SDOF system to any dynamic loading p(t).
∞ ∞
2πn
p (t ) = a0 + ∑
n=1
an cos
Tp
t+ ∑ b sin 2Tπn
n=1
n
p
(3.84)
2 For a damped system, the derivation is identical except that the free-vibration response initiated by the differential
load impulse decays exponentially. The Duhamel integral for a damped SDOF system is:
t
∫ p ( τ) e
1
u (t ) = − ξω(t − τ )
sin ω D (t − τ) dτ (3.83)
mω D 0
Tp
∫
1
a0 = p (t ) dt (3.85)
Tp 0
Tp
2πn
∫
2
an = p (t ) cos t (3.86)
Tp 0 Tp
Tp
2πn
∫
2
bn = p (t ) sin t (3.87)
Tp 0 Tp
The response to the loading of Equation 3.84 is the sum of the responses for each of the terms. The
SDOF response to the type of harmonic (and the constant) terms of Equation 3.84 has been discussed
in the previous sections.
The steady-state response to the constant load component (which is equal to the average load) is the
static deflection, i.e.,
a0
u0 = (3.88)
k
The steady-state response of an undamped SDOF structure for each term of the series is written, by
analogy with Equation 3.40, as follows:
Sine terms:
bn 1
un (t ) = sin nω pt (3.89)
k 1 − αn2
where
nω p nω p
αn = = (3.90)
ω ω
Cosine terms:
an 1
un (t ) = cos nω pt (3.91)
k 1 − αn2
The total response of the undamped system is then expressed as the sum of all the individual responses
above:
1
∞
u (t ) = a0 +
k
∑ 1 −1α (a cos nω t + b sin nω t )
n=1 n
2 n p n p (3.92)
The response of a damped SDOF system subject to the periodic loading of Equation 3.84 is obtained
by using the damped-harmonic-response expressions of the form of Equation 3.48 in lieu of the
undamped-response expressions, Equations 3.89 and 3.91 above.
As a precursor for response analysis in the frequency domain (in the next section), Equations 3.84
through 3.87 are written in exponential form by utilizing Euler’s equation:
(
sin x = − 1 2 i e − ix − e − ix ) (3.94)
cos x = 1
2 (e − ix
+ e − ix ) (3.95)
p (t ) = ∑c
n=−∞
n exp inω pt ( ) (3.96)
where
( )
Tp
∫
1
cn = p (t ) exp −inω pt dt (3.97)
Tp 0
Note that while cn is complex, in the series of Equation 3.96 all the terms with n ≠ 0 can be arranged
in complex conjugate pairs such that the imaginary parts cancel and the sum is real.
To examine the response in a case where the loading is expressed in the exponential form of Equation
3.96, first consider the response to a single complex term (n = 1) of Equation 3.96:
( ) ( )
u (t ) = H ω p exp iω pt (3.99)
( )
H ωp = 2
1
−ω p m + iω pc + k
(3.100)
Denoting the frequency ratio by α and the damping ratio by ξ, this becomes:
( ) k ( −α
H ωp = 2
1
+ 2iαξ + 1 )
(3.101)
( ) k ( −n α
H nω p = 2 2
1
+ 2inαξ + 1 )
(3.102)
Note that H(nωp) and H(−nωp) are complex conjugates, as are, consequently, each (+n) and (−n) term
of the summation, and hence the sum is real.
∞
p (t ) =
1
2π ∫ ω p =− ∞
( ) ( )
c ω p exp iω pt dω p (3.104)
∞
( ) ∫
c ωp =
t =−∞
( )
p (t ) exp −iω pt dt (3.105)
Equations 3.104 and 3.105 are known as a Fourier transform pair and Equation 3.104 can be considered
as an extension of the finite sum of Equation 3.96 to an infinite integral that represents the loading as
an infinite sum of harmonic components. Analogous to Equation 3.103, the response u(t) can be written
(omitting some details) as:
∞
u (t ) =
1
2π ∫ ω p =− ∞
( )( ) ( )
H ω p c ω p exp iω pt dω p (3.106)
Equation 3.106 can be considered as the basic equation for the analysis of response through the frequency
domain.
3.2.7.1.1 Numerical Analysis in the Frequency Domain
For practical problems, the integrals of Equations 3.105 and 3.106 are evaluated numerically. The key
elements in this process are the concept of discrete Fourier transform (DFT) and the fast Fourier
transform (FFT) numerical technique used for evaluating the DFTs.
The DFT converts the integrals of Equations 3.105 and 3.106 back to finite series and thereby invokes
the periodicity assumption. However, the period can be taken sufficiently long, and the number of terms
in the series sufficiently high, as to not pose an unwanted limitation on the accuracy of the solution of
the practical problem.
The selected load period, Tp , defines the lowest frequency that can be captured in the analysis, i.e.,
2π
ω p = ∆ω p = (3.107)
Tp
The load period is divided into N equal time increments ∆t = Tp/N, and the load is defined for the
discrete times tm = m∆t and frequencies ωpn = n∆ωp. The exponential term in Equation 3.104 can be
written:
( )
(
exp iω pnt m = exp in∆ω pm∆t = exp 2πi
nm
N
) (3.108)
3 For the Fourier integral to exist, the only condition for the loading function is that the following integral must
be finite: ∞
∫−∞
p (t ) dt
N −1
∆ω p
p (t m ) =
2π ∑ c (ω ) exp 2πi nm
n=0
pn
N
(3.109)
N −1
( )
c ω pn = ∆t ∑ p (t m ) exp −2πi nm
N
(3.110)
m=0
Equations 3.109 and 3.110 are the DFT pair corresponding to the continuous integral transforms of
Equations 3.104 and 3.105.
3.2.7.1.2 Fast Fourier Transform Analysis
In the current practice of frequency domain analysis, the sums in the DFT equations are computed
utilizing the powerful FFT technique. The details of this method are beyond the scope of this discussion,
and the reader is referred to the many texts on this topic [e.g., Newland, 1975; Clough and Penzien, 1975].
FI + FD + FS = P (t ) (3.113)
The inertia forces are given simply as the product of the story mass and the story acceleration, which
may be represented in matrix form as:
FI M1 0 u˙˙1
FI = 1 = (3.114)
FI2 0 M 2 u˙˙2
or simply:
FI = Mu˙˙ (3.115)
in which FI is the inertia force vector, ü is the acceleration vector, and M is the mass matrix.
It is important to note in Equation 3.114 that the inertia force corresponding to any degree of freedom
depends only on the acceleration in that degree of freedom. Such mass description is referred to as
lumped mass and the mass matrix is diagonal. In general, the mass matrix for a MDOF system is not
necessarily diagonal. For example, if the mass matrix is developed using the consistent mass matrix finite
element formulation, the mass matrix will not be diagonal and inertia coupling will be introduced
between the displacement coordinates. In some cases, the lack of mass coupling is desirable, and the
lumped mass representation is used in many practical analyses.
The elastic forces in Equations 3.111 and 3.112 are expressed as:
FS = Ku (3.116)
where Fs is the elastic force vector, u is the displacement vector, and K is the stiffness matrix of the
structure. The stiffness matrix K is generally not diagonal. The off-diagonal terms indicate that elastic
forces for a given coordinate depend on displacements of the other coordinates; there is stiffness coupling
between the coordinates. In practical applications, the stiffness matrix is most commonly constructed
using computer programs based on the finite element method.
Conceptually, the damping forces in Equations 3.111 and 3.112 can be expressed as:
FD = Cu˙ (3.117)
In the practical methods developed for MDOF analysis, individual elements of the damping matrix are
not computed based on “local” damping properties in the vicinity of the associated degrees of freedom.
Instead, damping is usually expressed in terms of damping ratios as discussed in Sections 3.2.2.2.2 and 3.2.3.1.
Substituting Equations 3.115 through 3.117 to Equation 3.113, the equations of dynamic equilibrium
may be written in matrix form as:
Equation 3.118 represents the equations of motion of an arbitrary structural system having any number
of degrees of freedom. The similarity of this matrix equation to the corresponding SDOF equation
(Equation 3.5) is noteworthy.
3.3.1.1 Equation of Motion for Support Excitation Loading
Figure 3.13 illustrates the case where the dynamic loading is due to support excitation. The horizontal
ground motion relative to the fixed reference axis is indicated by the displacement ug. Instead of Equation
3.115, the inertia force for this case is given by:
FI = Mu˙˙ t (3.119)
i.e., for each mass point, the inertia force is the product of the mass and the total, absolute acceleration,
which can be written as:
In the above example each mass point has only one translational degree of freedom and each of the
two elements of vector üg is simply üg(t), that is, the ground acceleration time history. In a more general
case each mass point has more than one degree of freedom, e.g., all three translations and three rotations.
In the general case the ground acceleration vector is written as:
)
u˙˙ g = I u˙˙g (3.121)
)
where I defines the direction of the support excitation and consists of ones for those degrees of freedom
in the direction of excitation, and zeros otherwise.
After moving the ground motion acceleration terms to the right-hand side, the equation of motion
for support excitation becomes:
)
Mu˙˙ + Cu˙ + Ku = −MI u˙˙g (t ) (3.122)
On the left-hand side, the acceleration ˙u̇ , velocity u̇, and displacement u are relative to the fixed base.
)
Therefore, the term Mü does not represent the full inertial loading on the mass points [ M (u˙˙ + I u˙˙g )
does], while Cü and Ku do represent the full damping and elastic forces acting on the mass points.
It is worth noting that the solution of Equation 3.122 for the relative displacement vector is the) same
as the solution of Equation 3.118, in this case for total displacement vector, for P (t ) = Peff (t ) = − MI u˙˙g (t ).
This is illustrated in Figure 3.13. Note that the “effective” inertia force acts at every mass point.
Mu˙˙ + Ku = 0 (3.123)
Postulating that a solution can be found as harmonic motion with all the degrees of freedom moving
in phase, the displacement vector is written as:
u = ϕ sinωt (3.124)
where ϕ represents the amplitude vector of the vibratory motion and ω is the circular frequency.
By differentiating Equation 3.124 twice with respect to time, the acceleration vector is:
Substituting Equations 3.124 and 3.125 into Equation 3.123 and canceling the sin ωt term leads to:
−ω 2Mϕ + Kϕ = 0 (3.126)
or rearranging:
(K − ω M) ϕ = 0
2
(3.127)
Equation 3.127 is a form of an eigenvalue problem. An eigenvalue problem associated with square
matrices of dimension n typically has n solutions, each consisting of an eigenvalue ωi2 and the eigenvector
ϕi , referred to as a mode shape in structural dynamics.
Before proceeding to discuss the response of the MDOF system, the properties of the mass and stiffness
matrices are briefly explored in the following section.
3.3.2.1 On Properties of Mass and Stiffness Matrix
The mass and stiffness matrices of a MDOF system have characteristics that have a strong bearing on
the mathematical analysis of the dynamic response of these systems. Some of these characteristics are
identified in the following, but a detailed discussion of these aspects is beyond the scope of this summary
presentation.
1. Both stiffness and mass matrices for linear elastic systems are symmetric.
2. Mass matrices are positive definite and stiffness matrices are positive semidefinite, i.e.,
and
Another important property of the eigensolutions is the orthogonality of the eigenvectors with respect
to both the mass and stiffness matrices M and K. This orthogonality property is expressed mathematically
by the following two equations:
T
ϕ i Mϕ j = 0 (3.130)
and
T
ϕ i Kϕ j = 0 (3.131)
Furthermore, the n orthogonal eigenvectors are linearly independent, and consequently any and all
vectors of the n-space can be constructed as a linear combination of these eigenvectors, i.e., any displace-
ment vector u can be expressed as:
u (t ) = ∑ ϕ η (t ) = Φη(t )
i =1
i i (3.132)
where Φ is a matrix containing the vectors ϕi (i 1, … n) as its columns and η can be considered a
generalized coordinate vector.
(K − ω M) ϕ = 0 2
(3.133)
is termed a natural mode of vibration and is defined by an eigenvalue ωi, termed the natural angular
frequency, and the corresponding eigenvector ϕi, termed the natural mode shape. It is readily observed
from Equation 3.133 that any eigenvector multiplied by a constant is also an eigenvector, hence the
magnitude of the eigenvector is not unique. However, it is customary to normalize the model shape
vectors with respect to the mass matrix such that:
T
ϕ i Mϕ i = 1 (3.134)
When the eigenvectors are arranged to a square matrix, with the vectors as the columns of the matrix,
the following relationships are written based on Equation 3.134 and the orthogonality properties:
ΦT MΦ = I (3.135)
ΦT KΦ = Ω (3.136)
u˙ = Φη˙ (t ) (3.137)
˙˙ (t )
u˙˙ = Φη (3.138)
˙˙ + CΦη˙ + KΦη = P (t )
MΦη (3.139)
˙˙ + ΦT CΦη˙ + ΦT KΦη = ΦT P (t )
ΦT MΦη (3.140)
Based on the orthogonality conditions of Equations 3.130 and 3.131, the first and third terms of the
left-hand side are diagonal matrices. If we further assume that same type of orthogonality conditions
also applies to the damping matrix, i.e., that:
T
ϕ i Cϕ j = 0 (3.141)
the normal coordinate equation of motion may be written more conveniently by introducing the following
symbols for the generalized coordinate properties of each mode i:
T
˙˙ i + ϕ i Cϕ i η˙ i + ϕ i Kϕ i ηi = ϕ i P (t )
ϕ i Mϕ i η
T T T
(3.142)
˙˙ i + Ci* η˙ i + K i* ηi = Pi* (t )
M i* η (3.143)
T
Generalized mass: M i* = ϕ i Mϕ i (3.144)
T
Generalized damping: Ci* = ϕ i Cϕ i (3.145)
T
Generalized stiffness: K i* = ϕ i Kϕ i (3.146)
Thus, by expressing the equations of motion, Equation 3.118, in terms of the modal coordinates, the
system of n coupled equations have been decoupled to n independent equations of motion that can be
written as:
Pi* (t )
˙˙ i + 2ξi ω i η˙ i + ω i2 ηi =
η i = 1,..., n (3.148)
M i*
These n independent equations of motion are of exactly the same form as the SDOF equation of
motion discussed above. Each of the equations in Equation 3.148 can be solved for the modal coordinate
ηi(t) using methods discussed in Section 3.2.2. The modal coordinates can then be combined to obtain
the displacement vector u per Equation 3.132.
The method of solving the MDOF equation of motion by transforming the problem to the modal
coordinate system, solving the uncoupled equations, and finally superposing the modal contributions is
referred to as the modal superposition method.
3.3.3.1 Rayleigh Proportional Damping
An assumption was made above that the damping matrix C has similar orthogonality properties as the
mass and stiffness matrices. If the damping matrix is represented as Rayleigh proportional damping,
i.e., as a linear combination of the mass and stiffness matrices:
then the damping matrix does possess the same orthogonality properties as the mass and stiffness
matrices, i.e.,
T
ϕ i Cϕ j = 0 (3.150)
Using notations of Equations 3.144 through 3.146, damping of the ith mode is expressed as:
and as for the SDOF system, the damping ratio for mode i is written as:
Ci*
ξi = (3.152)
2M i*ω i
a0 1 a1
ξi = + ω (3.153)
2 ωi 2 i
Equation 3.153 indicates that the contribution from the mass related damping term is inversely
proportional to frequency, while the contribution from the stiffness related term is proportional to
frequency. Figure 3.14 shows the Rayleigh damping variation with natural frequency.
The coefficients a0 and a1 can be determined from specified damping ratios at two independent modes
k and l. Expressing Equation 3.153 for these two modes will lead to the following equations:
a0 1 a1
ξk = + ω (3.154)
2 ωk 2 k
a0 1 a1
ξl = + ω (3.155)
2 ωl 2 l
2ω kω l
a0 = ξ (3.156)
ωk + ωl
2
a1 = ξ (3.157)
ωk + ωl
Note that at frequencies between ωk and ωl, damping is less than ξ. In practical problems, the
parameters ξ, ωk, and ωl should be set such that all the modes with significant contribution to the response
are assigned reasonable damping ratio values. Especially when used in nonlinear analysis for systems
where motions resembling rigid-body motion may occur, mass proportional damping is to be used with
caution, as unreasonably high damping may result.
By defining
) a participation factor for mode i for earthquake excitation in the direction defined by the
vector I as:
)
βi = ϕTi MI
The response of the ith mode can be expressed in terms of the Duhamel integral:
βi 1 t
ηi (t ) =
∫ u˙˙ (τ)e sin ω i (t − τ) dτ
− ξ i ω i (t − τ )
(3.161)
M i* ω i 0
g
∫ u˙˙ (τ)e
1
U i (t ) = − ξ i ω i (t − τ )
sin ω i (t − τ) dτ (3.162)
ωi 0
g
βi
ηi (t ) = U (t ) (3.163)
M i* i
The complete displacement of the structure at time t is then obtained by superposing the contribution
of all modes evaluated at this time by Equation 3.132 as:
u (t ) = ∑ ϕ η (t ) = Φη(t )
i =1
i i (3.132)
Typically, the modes with the lowest natural frequencies contribute most to the overall response, (1)
as the participation factors of these modes for earthquake excitation are high in comparison to those of
the higher modes, and (2) due to the presence of ωi in the denominator of the Duhamel integral (Equation
3.161). Therefore, an adequate approximation of the response is often obtained by including only a
number of the lowest modes in the summation of Equation 3.132. The method of MDOF earthquake
response analysis outlined above, culminating in superposition of modal responses, is referred to as the
modal superposition method. While the time integration of modal responses was formulated above in
terms of analytic functions, the time integration could be, and in practice usually is, performed using
numerical time integration methods that are discussed in Section 3.3.4.2.
3.3.4.1 Response Spectrum Analysis
The response spectrum method allows an approximate determination of the maximum response of an
MDOF system without performing a time history analysis, e.g., without employing the modal superpo-
sition method of the previous section or the direct time history analysis discussed in Section 3.3.4.2.
Response spectrum analysis consists of three steps:
1. Modal analysis, i.e., determination of the natural modes of vibration and the associated frequencies
as outlined in Section 3.3.3
2. Determination of the maximum response in each mode, “modal response,” included in the analysis;
this involves determination of maximum value of the modal coordinate, ηimax, over the duration
of the earthquake loading for each mode
3. Combination of the modal responses to obtain approximate maximum response
The maximum modal response for mode i, ηimax , can be expressed (see Equation 3.163) in terms of
the displacement response spectrum, Sd (ωi)(see Chapter 4):
βi
ηimax = S (ω ) (3.164)
M i* d i
The (relative) displacement vector uimax associated with the maximum peak response of mode i, i.e.,
ηimax, is:
βi
u imax = ϕ i ηimax = ϕ i S (ω ) (3.165)
M i* d i
Maximum values of response quantities such as relative displacement, element forces, and reaction
forces due to response in mode i can readily be determined based on the deformation defined by uimax.
In general, the maximum response uimax does not occur at the same point in time for all the modes
i = 1, …, n, and any combination of the modal maxima can only be an approximation of the actual
maximum response. A relatively simple and practical method, supported to an extent by probabilistic
considerations, for computing approximate total maximum response quantities is the square-root-of-
the-sum-of-the-squares (SRSS) procedure. If qimax is the maximum value of a response quantity, e.g., an
element shear force, due to mode i, an approximate total maximum of this quantity due to responses in
all modes included in the analysis, i = 1, …, N, is computed as:
qmax = ∑q
i =1
2
imax (3.166)
Several other schemes of combining modal responses have been developed, for example, the complete
quadratic combination (CQC) rule (see Chopra [2001] for details), some of them particularly to predict
total response more accurately than the SRSS procedure in situations where the system has closely spaced
modes (i.e., some of the natural frequencies are very close to each other).
A key benefit of the response spectrum method is the ability to obtain a generally good approximation
of the earthquake response of a multidegree structure without having to perform a full-time history
analysis. The method is a standard feature in various structural analysis software packages. Response
spectra for design purposes are specified in the design codes and response spectra for major earthquakes
are customarily computed from accelerogram records and made available for public use.
3.3.4.2 Numerical Time Integration
The Duhamel integral is a useful conceptual tool and can be used to obtain a closed-form solution of
the equation of motion in cases where the integrand has an analytic integral function. In cases where an
analytic integral function cannot be developed, the integral can be evaluated numerically based on
approximation of the definite integral by a summation utilizing any of the classical summation rules,
e.g., the trapezoidal or Simpson’s rules.
In practice, however, step-by-step numerical integration schemes derived by expressing the relation-
ships between displacement and its time derivatives, velocity, and acceleration, utilizing approximate
discretized (with respect to time) expressions, and substituting these to the equation of motion, are most
commonly used. The Newmark family of time integration methods illustrates these numerical schemes.
The Newmark family contains as special cases many widely used methods. The method is based on the
following formulas:
∆t 2
u t + ∆t = u t + u˙ t ∆t +
2
[
(1 − 2β)u˙˙ t + 2βu˙˙ t +∆t ] (3.168)
[
u˙ t + ∆t = u˙ t + ∆t (1 − γ ) u˙˙ t + γu˙˙ t + ∆t ] (3.169)
where u t , u˙ t , and u˙˙ t are approximations of u (t ), u˙ (t ), and u˙˙ (t ), respectively. Equations 3.167, 3.168, and
3.169 can be thought of as three equations for three unknowns, u t + ∆t , u˙ t + ∆t , and u˙˙ t + ∆t , assuming that
u t , u˙ t , and u˙˙ t are known from the previous step. By solving ˙u̇ t +∆t from Equation 3.168 in terms of u t +∆t ,
then substituting for ˙u̇ t +∆t in Equation 3.169, expressions are obtained for ˙u̇ t +∆t and u̇ t +∆t in terms of
u t +∆t as:
1
2 ( t + ∆t
− ut ) −
1 1
u˙˙ t + ∆t = u u˙ t − − 1 u˙˙ t (3.170)
β∆t β∆t 2β
When these expressions are substituted in Equation 3.167, the resulting equation can be written as:
K *u t + ∆t = Pt*+ ∆t (3.172)
where
1 γ
K* = M+ C +K (3.173)
β∆t 2 β∆t
and
1 1 1
Pt*+ ∆t = Pt + ∆t + M u + u˙ t + − 1 u˙˙ t +
β∆t β∆t 2β
2 t
(3.174)
γ γ ∆t γ
C ut + − 1 u˙ t + − 2 u˙˙
β∆t β 2 β t
The response can be solved step by step using the above equations starting from the initial conditions
u 0 , u˙ 0 , and u˙˙ 0 and over the period of time of interest. Convergence to the true solution requires that a
sufficiently short time step, ∆t, be used. The specific choice of values for the parameters β and γ affect
the accuracy and stability of the algorithm. A method is called unstable for a given time step size if the
solution, ut , grows out of bounds at some point even if the true solution, u(t), does not. It can be shown
that when:
γ ≥ 0.50 (3.175)
the method defined by Equations 3.170 through 3.174 is stable provided that the time step is less than
the threshold value that can be expressed in the form:
∆t ≤ αTmin (3.176)
where Tmin is the period of the highest natural mode of the system (and α is a coefficient that depends
on β and γ as well as on damping). The constraint imposed by Equation 3.176 may lead to a far shorter
time step than required to adequately resolve the low modes that dominate the response, and thereby
potentially cause an unreasonable computational burden.
A method that is stable only when a condition for the time step, of the type of Equation 3.176, is
satisfied is conditionally stable and a method that is stable regardless of time step size is unconditionally
stable.
If, in addition to the condition of Equation 3.175, coefficient β satisfies the condition:
β ≥ 0.25 ( γ + 0.5)
2
(3.177)
∆t
Method Type β γ TMIN
u t + ∆t − 2u t + u t − ∆t
˙u̇ t = (3.178)
∆t 2
u t + ∆t − u t − ∆t
u̇ t = (3.179)
2∆t
and (3) M (and C if present) is diagonal, the matrix K* multiplying the unknown vector ut + ∆t is diagonal
(the nondiagonal K only multiplying the known ut [see Equation 3.180 below]), and consequently the
equations are uncoupled, leading to fast solution for each time step:
1 1 2 1 1
2 M+ C u = Pt − K − 2 M u t − 2 M − C u (3.180)
∆t 2∆t t + ∆t ∆t ∆t 2∆t t − ∆t
When time integration is used for nonlinear problems with the stiffness matrix varying over time, the
reduction in the amount of computations, in comparison to implicit methods, is significant. Countering
the benefit of simple fast solution per time step is the requirement for short time step imposed by
conditional stability. Problems for which use of explicit methods are particularly beneficial, or in many
cases mandatory, are nonlinear problems that involve one or more of the following characteristics:
1. Large deformations
2. Large rotations
3. Contact/impact with or without sliding
These types of problems often require the use of very short time steps in any case to achieve reasonable
accuracy and, therefore, the condition for short time step imposed by the stability condition is not a
burden.
As is typical of numerical methods, the numerical time integration algorithms are approximate.
However, like any generally acceptable numerical algorithm, a properly implemented time integration
scheme is convergent, i.e., the computed solution approaches the exact value when time step size
approaches zero (i.e., ut → u(t) when ∆t → 0). For the time integration algorithms introduced above,
the errors in the computed response can be categorized as period elongation and amplitude decay.
Figure 3.15 illustrates these errors for a case where a free vibration response is integrated numerically.
Typically, the implicit methods elongate the period while the explicit central difference method reduces
the period relative to the exact value. Typically, all the methods except the Newmark constant average
acceleration method (i.e., β = 1/4 and γ = 1/2) cause amplitude decay that can be thought of as numerical
damping, which adds to any damping included explicitly in the model, e.g., through Rayleigh damping.
By shortening the time step, these errors can be limited to any desired level. For the Newmark family of
implicit methods, it is generally recommended that time step be limited to:
where Tcut-off is the period of the highest mode required to be resolved with a reasonable level of accuracy
in order to capture the response adequately. With the central difference method, any time step that
satisfies the stability condition typically results in adequate accuracy of the computed response for most
practical problems.
In the dynamic analysis of building structures, the mass matrix is generally assumed to be constant.
In many nonlinear building time-history analyses, it is convenient and in most cases adequate to model
damping with the Rayleigh proportional damping, Equation 3.149:
C = αM + βK (3.149)
where the stiffness matrix K is the original elastic stiffness matrix. It is important to note that nonlinear
modeling may introduce damping mechanisms in addition to the damping effected through the C matrix.
For example, elements with material descriptions including plasticity, as well as nonlinear link, cable,
and contact elements may introduce hysteretic deformation behavior that results in energy dissipation.
Care has to be exercised not to overestimate damping by “double counting” the energy dissipation through
the combined effect of nonlinear element behavior and damping with the global damping matrix C.
To introduce the nonlinear effects due to material nonlinear behavior, geometric nonlinearity, and
contact/impact, we first denote the structure internal force vector by fint, which for a linear elastic system
can be written as:
f int = Ku (3.182)
Assume that the displacements increase to a point beyond which response is nonlinear. Beyond this
point, the change in the internal force vector, ∆f int, due to a small additional displacement increment,
∆u, can be written as:
∆f int = K T ∆u (3.183)
where KT is the tangent stiffness matrix. (The superscript T denotes tangent, not transpose, of the matrix
K.) The elements of the matrix are defined by:
∂fi int
kijT = (3.184)
∂u j
+ ∆t = ft
ftint + KtT ∆u t + ∆t
int
(3.185)
where KTt is the tangent stiffness matrix at time t, and ∆ut + ∆t “updates” ut to ut + ∆t , i.e.,
u t + ∆t = u t + ∆u t + ∆t (3.186)
Direct time integration algorithms, such as those represented by the Newmark family, can be readily
adopted for use in nonlinear analysis. Using the Newmark Equations 3.170 and 3.171 for üt + ∆t and u· t + ∆t ,
respectively, and substituting expression 3.185 for ftint
+∆t
leads to:
˜ ∆u
t + ∆t = Pt + ∆t
*
K t (3.188)
where, when parameters β and γ are assigned values 1/4 and 1/2, respectively (i.e., the constant average
acceleration method is used):
4 2
K̃t = M + C + KtT (3.189)
∆t 2
∆t
4
Pt*+ ∆t = Pt + ∆t − ftint + M u˙ t + u˙˙ t + Cu˙ t (3.190)
∆t
Once ∆ut + ∆t is solved from Equation 3.188, ut + ∆t is computed from Equation 3.186. Because the
tangent stiffness was evaluated at ut (not at ut + ∆t) the solution is approximate. Often iterations are
required to achieve sufficient accuracy. One widely used iteration procedure is the modified Newton
iteration, which is formulated by replacing Equation 3.186 with:
and
u t(0+)∆t = u t (3.192)
˜ ∆u (k ) = P *
K (3.193)
t t + ∆t t + ∆t
The coefficient matrix K̃ t is unchanged, while the right-hand side Pt*+∆t is modified to:
4
( 4
)
Pt*+ ∆t = Pt + ∆t − ftint(k −1) − M 2 u (t k+−∆1t) − u t − u˙ t − u˙˙ t −
∆t ∆t
(3.194)
2
C
∆t
( )
u (t k+−∆1t) − u t − u˙ t
For each iteration, acceleration and velocity are computed from the following equations (Equations
3.170 and 3.171 with u (t +∆
k)
t
, Equation 3.191, substituted for ut + ∆t , and β and γ equal to 1/4 and 1/2,
respectively):
u˙˙ (t k+)∆t =
4
∆t 2 ( ) 4
u (t k+−∆1t) + ∆u t(k+)∆t − u t − u˙ t − u˙˙ t
∆t
(3.195)
u˙ (t k+)∆t =
2
( )
u (k −1) + ∆u t(k+)∆t − u t − u˙ t
∆t t + ∆t
(3.196)
Defining Terms
Angular velocity — A measure of rotational speed, for which the typical unit is radians/second.
Characteristic equation — The polynomial equation resulting from substitution of an exponential trial
solution to a differential equation.
Complementary solution — Solution of the homogeneous differential equation; includes the coeffi-
cients of integration that are defined based on initial conditions.
Conditional stability — A numerical time integration method is conditionally stable if it is stable only
for a time step size shorter than some threshold value; otherwise it is unconditionally stable.
Convergence — A numerical time integration method is convergent if the computed solution
approaches the exact value when time step size approaches zero (i.e., ut → u(t) when ∆t → 0).
Critical damping — The lowest level of damping at which the free response will not involve oscillations
across the zero displacement condition.
Damping — Energy dissipation mechanism that leads to decay of the response; often modeled by linear
viscous dashpot.
Damping ratio — Ratio of damping to critical damping.
Diagonal — Property of a matrix, where only the diagonal terms (i.e., terms in which row and column
numbers are equal) are non-zero.
Direct method — A numerical time integration method that operates directly on the “original” MDOF
equations of motion, i.e., not on the “transformed” uncoupled modal equations of motion.
Discrete Fourier transform — A discretized approximation of the Fourier transform integral.
Duhamel integral — An analytic integral expression that yields the response of an SDOF system to an
arbitrary loading function.
Dynamic amplification factor — Ratio of the maximum displacement response amplitude to the static
displacement.
Eigenvalue problem — In MDOF dynamics context, a set of n simultaneous homogeneous linear equa-
tions for n unknowns that only has a solution for certain values (the eigenvalues) of a parameter.
The solution of an eigenproblem consists of the eigenvalue and the associated eigenvector.
Elastic spring — A spring that follows the same force-displacement curve for loading and unloading.
Explicit — In an explicit time integration method, only the diagonal mass matrix multiplies the
unknown displacement vector to be solved for the next time step; does not involve solution of
coupled equations.
Fast Fourier transform — A numerical technique for computation of the Fourier transform; utilizes
the discrete Fourier transform in the process.
Fourier series — A representation of a periodic function as a (generally infinite) series of sine and cosine
terms.
Fourier transform — A generalization of Fourier series to nonperiodic functions, the infinite series
becoming infinite integral expressions.
Free vibrations — Vibration of a system with no loading applied, after the system has been set in motion
by some means.
General solution — The complete solution of a differential equation, consisting of the complementary
solution and the particular solution.
Homogeneous equation — The version of a differential equation with the right-hand side, i.e., the
loading term, set to zero.
Implicit — In an implicit time integration method, the (nondiagonal) stiffness matrix multiplies the
unknown (displacement or acceleration) vector to be solved for the next time step; involves
solution of coupled equations.
Impulsive load — A short duration (relative to natural period) loading consisting of a single “spike.”
Inertia — The tendency of mass to maintain its momentum (both linear and angular momentum).
Linear — In a linear system, multiplication of the input, e.g., applied force, by a constant results in the
response being multiplied by the same constant. L(·) is linear if L(u) = f ⇒ L(αu) = αf.
Linearly independent — Vectors ϕi are linearly independent if
n
Lumped mass — A discretized representation, in a diagonal mass matrix, of the generally continuous
distributed mass inertia (translational and/or rotational) of a MDOF system with no inertia
coupling between degrees of freedom.
Mode shape — Often referred to as a natural mode of vibration, is an eigenvector (see eigenvalue
problem) of the eigenvalue problem of structural dynamics.
Multiple degree of freedom (MDOF) — System with multiple mass points; description of motion
requiring use of multiple displacement variables.
Natural angular frequency — Synonym for natural angular velocity.
Natural angular velocity — The angular velocity obtained as a solution of the characteristic equation
of an undamped system.
Natural frequency — Number of cycles per second, obtained by dividing natural angular frequency by
2π.
Natural period — The inverse of natural frequency.
Nonlinear — A system is nonlinear if it is not linear.
Numerical damping — Associated with most numerical time integration methods and causes ampli-
tude decay in the computed free vibration response of an undamped system.
Orthogonal — Vectors ϕi and ϕj are orthogonal with respect to matrix M if ϕiT Mϕ ϕj = 0.
Particular solution — Solution of the differential equation with the loading term on the right-hand
side of the equation.
Periodic loading — A loading history function with a particular pattern repeating at constant intervals,
i.e., P(t + ∆) = P(t), where ∆ is the period.
Positive definite — Matrix M is positive definite if ϕT Mϕϕj > 0 for every ϕ ≠ 0.
Positive semidefinite — Matrix K is positive semidefinite if ϕT Kϕ ϕj ≥ 0 for every ϕ ≠ 0.
Rayleigh proportional damping — Rayleigh proportional damping is of the form C = a0 M + a1 K.
Resonance — The condition where the frequency of the harmonic loading is equal to the undamped
natural frequency of the system.
Response ratio — The momentary ratio of the total displacement to the static displacement.
Response spectrum — A representation of the maximum value of a response parameter, such as accel-
eration or displacement, over the duration of earthquake loading as a function of frequency ω
for a given earthquake and damping level. For an acceleration response spectral ordinate Sa (ω)
is the maximum acceleration experienced by the mass of the SDOF system that has a natural
frequency of ω during the earthquake loading.
Response spectrum method — An approximate method of earthquake response analysis based on
modal superposition and response spectrum concepts.
Simple harmonic motion — Undamped motion described by a sine or cosine function with oscillations
at the natural frequency.
Single-degree-of-freedom (SDOF) — System with a single mass point whose motion can be described
with a single displacement variable.
Solution in the frequency domain — Refers to solution of the equation of motion by performing a
Fourier transform of the loading, finding the Fourier transform of the solution, and obtaining
the solution (i.e., the displacement function) through the inverse Fourier transform. In practice
the computations are implemented using the fast Fourier transform algorithm.
Stability — A numerical time integration method is unstable if at some time point the computed
response grows without bounds even if the exact solution does not; otherwise it is stable.
Steady-state response — The particular solution of a damped system to periodic loading.
Tangent stiffness — A “local” approximate linearized stiffness at a given deformation state during non-
linear response.
Transient response — The complementary solution for a damped system that decays exponentially and
is typically relevant only in the “beginning” of the response.
References
Bathe, K.-J. 1982. Finite Element Procedures in Engineering Analysis, Prentice-Hall, New York.
Biggs, J.M. 1964. Introduction to Structural Dynamics, McGraw-Hill, New York.
Chopra, A.N. 2001. Dynamics of Structures, 2nd ed., Prentice-Hall, New York.
Clough, R.W. and Penzien, J. 1975. Dynamics of Structures, McGraw-Hill, New York.
Harris, C.M. 1988. Shock and Vibration Handbook, McGraw-Hill, New York.
Hughes, T.J.R. 1987. The Finite Element Method, Linear Static and Dynamic Finite Element Analysis,
Prentice-Hall, New York.
Meirovitch, L. 1975. Elements of Vibration Analysis, McGraw-Hill, New York.
Newland, D.E. 1975. An Introduction to Random Vibrations and Spectral Analysis, Longman, New York.
Wiegel, R.L. 1970. Earthquake Engineering, Prentice-Hall, New York.