ERASMUS Teaching (2008), Technische Universität Berlin

Excitation response relations for

linear systems
y
Assoc. Prof. Dr. Pelin GUNDES BAKIR
Istanbul Technical University
[email protected]
[email protected]
d b ki @ h
 ERASMUS Teaching (2009), Technische Universität Berlin

Introduction
• We turn
W t now to
t the
th response characteristics
h t i ti off physical
h i l systems.
t Before
B f
considering what happens when a system is subjected to random
excitation, we must deal first with various different methods for describing
the response of a general system to deterministic (non‐random)
(non random)
excitation.

• The general system may be a vibrating structure or machine, or a

complete building, or a small electric circuit. Whatever it is, we assume
that there are a number of inputs x1(t),x2(t),x3(t) etc., which constitute the
excitation
i i and d a number
b off outputs y1(t),y
( ) 2(t),y
( ) 3(t)
( ) etc., which
hi h constitute
i
the response.

• The x(t) and y(t) may be forces, pressures, displacements, velocities,

accelerations, voltages, currents, etc., or a mixture of all these.

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Introduction
• We shall
h ll restrict ourselves
l to linear
l systems for
f which
h h each h response
variable y(t) is related to the excitation by a linear differential equation of
the form:

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Introduction
• Th
These equations
ti are linear
li and
d the
th principle
i i l off superposition
iti
applies. The coefficients a,b,c, and d may in general be functions of
time, but we shall consider only cases when they are constant
which means that the vibrating system does not change its
characteristic with time.

• O
On accountt off the
th principle
i i l off superposition,
iti our problem
bl is
i greatly
tl
simplified because we can consider how each output variable
responds to a single input variable alone, and then just add
together the separate responses to many input variables in order to
obtain the response to the combined excitation at many points. Of
course, the assumption of linearity is a bold one, but since
vibrations usually only involve small displacements from
equilibrium, it is very often not too far from the truth.

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Introduction
• We can therefore
h f simplify
i lif the
h system we should
h ld consider
id on the
h left
l f to that
h shown
h
in the figure on the right.

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Classical approach
• If the equations of motion for the constant parameter linear system can be determined
determined,
then there is a known linear differential equation relating y(t) and x(t) with the form:

• For given excitation x(t) and given initial conditions, this equation can be solved by
classical methods to give a complete solution for y(t). However, such an approach is not
usually helpful for random vibration problems for two reasons. First the above
differential equation is seldom obtainable directly because there is inadequate data
available and simple experimental methods for finding the coefficients a and b are not
available. Secondly, even if the differential equation is known, a complete time history
for y(t) can only be calculated if we have a complete time history x(t), and for random
vibration
ib i problems,
bl this
hi data
d iis not off course available.
il bl IIn order
d to calculate
l l average
values of the output variables, it is more convenient to concentrate on alternative ways
of representing the relationship between y(t) and x(t).

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Frequency response method

• A completely
l l different
d ff method
h d off d
describing
b theh ddynamic characteristics
h off
a linear system is to determine the response to a sine wave input. If the
input is a constant amplitude sine wave of fixed frequency

• Then from

• The steady state output must also be a sine wave of fixed amplitude, the
same frequency ω and phase difference φ so that

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Frequency response method

• Information
f about
b the
h amplitude
l d ratio yo/x / o and
d the
h phase
h l φ defines
angle d f
the transmission characteristics or transfer function of the system at the
fixed frequency ω. By making measurements at a series of closely spaced
frequencies, amplitude ratio and phase angle can be plotted as a function
of frequency, and in theory if the frequency range extends from zero to
infinityy then the dynamic
y characteristics of the system
y are completely
p y
defined.
Example: Determine the amplitude ratio and phase angle for the
transmission of sine wave excitation through the spring‐damper
spring damper system of
the figure. The excitation is the force x(t) and the response is the
displacement y(t).
 ERASMUS Teaching (2009), Technische Universität Berlin

Frequency response method

• FFor a lilinear spring
i off stiffness
tiff k and
d a lilinear viscous
i damper
d off coefficient
ffi i t c,
the equation of motion is

and so when x(t) is a constant amplitude sine wave

and y(t) is the response given by:

then:

which gives collecting terms:

For this to be true, the terms in brackets must be separately zero, so that the
amplitude ratio:

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Frequency response method

and the phase angle:

• Instead of thinking of amplitude ratio and phase angle as two separate

quantities it has become customary in vibration theory to use a single
quantities,
complex number to represent both quantities. This is called, the
(complex) frequency response function H(ω) which is defined so that its
magnitude is equal to the amplitude ratio and the ratio of its imaginary
part to its real part is equal to the tangent of the phase angle. If

where A(ω) and B(ω) are real functions of ω, then

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Frequency response method

• Using complex
l exponentiall notation, we can now write using
eiθ = cos θ + i sin θ
that if the input is a sine wave of amplitude xo

then the corresponding harmonic output will be:

• The proof that this is correct is easily obtained by substituting

into the above equation for y(t) to give after using the eqns on the rhs:
 ERASMUS Teaching (2009), Technische Universität Berlin

Frequency response method

• I summary, th
In therefore,
f if a constant
t t amplitude
lit d hharmonic
i iinputt given
i b
by:

is applied to a linear system, the corresponding output y(t) will be given

by:

where H(ω) is the system

system’ss complex frequency response function
evaluated at angular frequency ω. In interpreting the above two
equations, we mean either the “real part” or “the imaginary part” of the
right hand side of the equations (but not both) according to the
convention agreed on. The magnitude of H(ω) gives the ratio of the
amplitudes of y(t) and x(t) and its argument gives the phase angle
between y(
y(t)) and x(t).
( ) Notice that the q
quantityy xo in the above equations
q
need not necessarily be real; its magnitude gives the amplitude of x(t)
while the magnitude of H(ω)xo gives the amplitude of y(t).

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Example
• Calculate the frequency response function for the system of the figure.

• The equation of motion is

• Putting (where either “the imaginary part of” or “the real part of” is
understood) and

gives
or

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Example
• The
h amplitude
l d ratio

and the phase angle is given by:

which checks the previous example. Notice that H(ω) has dimensions (in
this case those of displacement/force) and that the output lags behind the
input by angle φ.

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Impulse response method

• The ffrequency response function
Th f ti H(ω) H( ) gives
i th
the steady
t d state
t t response off a
system to a sine wave input. By measuring H(ω) for all frequencies, we
completely define the dynamic characteristics of the system.
• An alternative
l approachh is to measure transienti response, initiated
d by
b a
suitable disturbance. If we measure the transient response for all times,
until the static equilibrium has been regained after the disturbance, then
thi iis another
this th method
th d off d
defining
fi i completely
l t l a system’s
t ’ ddynamic i
characteristics.
• It is usual to consider the result of exciting the system by an input
di
disturbance
b which
hi h is
i short
h and d sharp
h and
d iis present for
f a very short
h
(theoretically zero) time interval. The transient response is then not
complicated by removal of the disturbance. Using the delta function
notation,
t ti we can representt such h a disturbance
di t b by
b the
th equation:
ti
where I is a constant parameter with the dimensions (x)x(time).

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Impulse response method

• For the
h case when
h x(t)
( ) represents a force
f

describes a hammer blow or impulse of magnitude:

• This terminology is carried over to the general case when x(t) may
represent any input parameter, whether a force or not, and the impulse
response of a system is defined as the system
system’ss response to an impulsive
output of the form
where I has the proper dimensions. The excitation is described as a unit
i
impulse
l when
h I isi numerically
i ll unity
i iin the
h above
b equation.
i

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Impulse response method

• In response to such
h an impulsive
l input, the
h initially
ll dormant
d system
suddenly springs to life and then gradually recovers its static equilibrium
position as time passes.
• The response to a unit impulse at t=0 is represented by the (unit) impulse
response function, h(t) as shown in the figure.

• Notice that h(t)=0 for t<0 because y(t)=0 before the impulse occurs.

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Example
• Determine the
h impulse
l response function
f ffor the
h system shown
h in the
h ffigure.

• Starting from the equation of motion

x(t) δ(t), i.e.,
h(t) is the solution y(t) when x(t)=δ(t),
• For t>0, δ(t)=0 and so
whose solution is:
where C is a constant to be determined from the conditions at t=0. In order to
find what happens when the hammer blow falls, we can make use of the fact
that δ(t) is zero everywhere except at t=0.

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Example
• Iff t=0‐ is a point just to the
h left
l f and
d t=0+ a point just to the
h right
h off the
h
origin, then

and integrating both sides of the equation of motion from t=0‐ to t=0+
gives:

• To go further, we need to employ some physical reasoning about what

happens
pp to the system
y when the impulse
p is applied.
pp The massless trolleyy
responds to the impulse with a sudden movement and its velocity will be
infinite instantaneously. Because is infinite at t=0, the integral
will be finite
finite, even though the integration time from
0‐ to 0+ approaches zero in the limit. However, h(t) is not infinite at t=0,
and so the value of the integral must be zero in the limit.

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Example
• We therefore
h f obtain:
b

• so that
The full solution for h(t) is:

as sketched in the figure. Notice that the units of h(t) are those of
(displacement)/(impulse) oror, equivalently,
equivalently of (velocity)/((force)
 ERASMUS Teaching (2009), Technische Universität Berlin

Relationship between the frequency response

and impulse response functions
• Since complete
l information
f about
b either
h the
h frequency
f response ffunction
or the impulse response function fully defines the dynamic characteristics
of a system, it follows that we should be able to derive one from the other
and vice versa. The Fourier transform method of breaking an aperiodic
function into its frequency spectrum provides the necessary link.

• As we are dealing with stable systems which are dormant before they are
excited and for which motion dies away after an impulse, we know that

• And we may therefore take Fourier transforms of both the impulse input
x(t)=δ(t) and the transient output ).

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Relationship between the frequency response

and impulse response functions
• If we d
do this,
thi we obtain:
bt i

and

• The first of these equations may be simplified by expanding the complex

exponential to obtain:

• and then using the property of a delta function

+∞

∫ δ (τ − T ) f (τ )dτ = f (τ = T )
−∞

• We can show that the first integral is unity and the second integral is zero
(since sinωt=0 at t=0) to obtain:

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Relationship between the frequency response

and impulse response functions
• We have
W h therefore
th f obtained
bt i d the
th Fourier
F i transforms
t f off an impulsive
i l i input
i t
x(t)=δ(t) in

and of the corresponding impulse response function y(t)=h(t) in

• Furthermore, the Fourier transforms are related by the frequency

response function H(ω). The nature of this relationship can be seen by the
following argument.
argument We know that when a linear system is subjected to
steady state harmonic excitation at frequency ω, it responds with a steady
harmonic output at the same frequency. It therefore seems reasonable to
expect that signal frequency components X(ω)d ω
that, for an aperiodic input signal,
in the frequency band ω to ω+d ω in the input will correspond with
components Y(ω)d ω in the same frequency band in the output. In this
case, if wee had
ad a harmonic
a o c inputpu oof thee form
o

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Relationship between the frequency response

and impulse response functions
• The
h corresponding
d h harmonic output would
ld be
b given b
by:

• But from

• We also know that

• And so, by comparing these two expressions for y(t), we obtain:

which is a very important relation between the Fourier transforms of the

input
p and output,
p , X(ω)
( ) and Y(ω),
( ), and the frequency
q y response
p function
H(ω).

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Relationship between the frequency response

and impulse response functions
• Finally,
ll substituting
b ffor X(ω)
( ) ffrom

• and for Y(ω) from

into
gives:
or:

expressing
i the
h result
l that
h theh frequency
f response function
f i H(ω)
H( ) iis the
h
Fourier transform of the impulse response function h(t). Actually, by
comparing the above equation with the definition of a Fourier transform
given in 1
+∞

∫ x(t )e
− iωt
X (ω ) = dt
2π −∞

it can be seen that there is a ½π missing from the above equation.

equation
Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics
 ERASMUS Teaching (2009), Technische Universität Berlin

Relationship between the frequency response

and impulse response functions
• However, as already
l d known,
k in Fourier transforms,
f the
h position off this
h
factor is optional so long as it appears in either the Fourier transform
equation or the inverse Fourier transform equation. From
+∞

∫ X (ω )e
iωt
x(t ) = dω
−∞
the inverse transform equation corresponding to

is:

and the impulse response equation is shown as a Fourier integral of the

frequency response function H(ω).

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Calculation of response to an arbitrary

input
• We turn now how
h theh frequency
f response andd impulse
l response functions
f
can be used to calculate how a system responds to a prescribed excitation.
Suppose that a linear system has an arbitrary input x(t) which is defined,
which satisfies +∞

∫ x(t ) dt < ∞
−∞
and for which we wish to calculate the resulting output y(t).
y(t)
• As already described in the previous section, the frequency response
function can be used to relate the Fourier Transforms of the input and
output, so that if
then:

and taking the inverse transform to find y(t), we obtain:

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Calculation of response to an arbitrary

input
• which
h h is a formal
f l solution
l ffor the
h output y(t).
( ) The
h integrall with h respect to
ω is, in general, extremely difficult to evaluate, and so it is rarely possible
to use

to obtain a neat closed expression for y(t). We do not therefore pursue

this approach for calculating y(t) but turn to the impulse response method
as a more promising alternative.
• The impulse response function h(t) gives the response at time t to a unit
impulse applied at time t=0, i.e., after a delay of duration t. It follows that
h(t‐τ) is the response at time t to a unit impulse or “hammer blow” at time
τ,, i.e.,, after a delayy of duration t‐τ. Now think of an arbitraryy input
p
function x(t) as being made up of a continuous series of small impulses.

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Calculation of response to an arbitrary

input
• The
h ““impulse”
l ” corresponding
d to the h input x(t)
( ) between
b the
h time limits
l τ
and τ+dτ as shown in the figure has the magnitude x(τ)dτ as shown
shaded. The response at time t to this “impulse” alone is just the fraction

of the response to a unit impulse at t=τ, which is h(t‐τ). The shaded area in
the figure therefore contributes an amount

to the total response

p at time t.
 ERASMUS Teaching (2009), Technische Universität Berlin

Calculation of response to an arbitrary

input
• Furthermore,
h since the
h principle
l off superposition applies
l forf a linear
l
system, we may obtain the total response y(t) at t by adding together all
the separate responses to all the small “impulses” which make up the
total time history back to t=‐∞. We therefore integrate the response to
x(τ)d τ back to τ=‐∞ to give:

which is another formal expression for the response y(t) at time t as a

result of an excitation whose value at time τ is x(τ) and which may exist
from τ=‐∞ to the present time τ=t.

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Example
Example:
l Calculate
l l the
h response at time t>0 off the
h system shown
h in the
h
figure when it is subjected to a step input x(t)=xo at t=0.

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Example
• Ui
Using

• The superposition or convolution integral

is a most important input‐output relationship for a linear system

system.
Provided that the system is passive so that it only responds to past inputs
and h(t) decays eventually to static equilibrium, so that

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Calculation of response to an arbitrary

input
• Then
h applies
l ffor any input x(t)
( ) whose
h magnituded
|x(t)| is in general bounded by a finite level. Notice that x(t) does not
need to satisfyy +∞

∫ x(t ) dt < ∞
−∞
which is necessary in the classical Fourier transform (frequency response
approach).
h)
• There are three alternative versions of

which can easily be derived by physical reasoning. First, we must recall

that h(t‐τ) is the response to a unit impulse at (t‐τ)=0 that is at t=τ. For (t‐
τ)<0,
) 0 there
h iis no response, as there
h h
has b
been no iimpulse
l applied.
li d Hence,
H
for τ>t

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Calculation of response to an arbitrary

input
• W may th
We therefore
f extend
t d th
the upper lilimit
it off th
the iintegral
t l iin

from τ=t to τ=∞ without changing the result, since h(t‐τ) is zero in this
interval. The first alternative form of the above equation is therefore:

• Next, consider changing the variable in the first equation by putting

where θ may be interpreted as the time delay between the occurrence of

an impulse and the instant when its result is being calculated. The limits of
integration τ=‐∞ and τ=t now become θ= ∞ (notice that time t is a
constant) and θ=0 and dτ becomes ‐d θ, so that substituting into

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Calculation of response to an arbitrary

input
gives:

or changing over the limits of integration to dispense with the minus sign
in front of dθ

• Finally, the third alternative form can be obtained either by putting θ=(t‐τ)
in

or by noting that h(θ)=0 for θ<0, since there is no response before the
impulse occurs, and using the above equation, to obtain:

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Calculation of response to an arbitrary

input
• Collecting
C ll ti ttogether
th these
th important
i t t alternative
lt ti results,
lt th
the response y(t)
(t)
of a passive linear system to an arbitrary input x(t) can be calculated by
evaluating one of the superposition integrals:

where h(t) is the response at time t to a unit impulse x(t)=δ(t) applied at t=0.

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics

 ERASMUS Teaching (2009), Technische Universität Berlin

Calculation of response to an arbitrary

input‐Example
 ERASMUS Teaching (2009), Technische Universität Berlin

Example

Pelin Gündeş Bakır, Probabilistic and stochastic methods in Structural Dynamics