Anda di halaman 1dari 14

Frequency Domain Analysis of Control System

By the term frequency response, we mean the steady-state response of a system to a sinusoidal
input. Industrial control systems are often designed using frequency response methods. Many
techniques are available in the frequency response methods for the analysis and design of control
systems.
Consider a system with sinusoidal input r (t ) = A sin ωt . The steady-state output may be
written as, c(t ) = B sin(ωt + φ ) . The magnitude and the phase relationship between the sinusoidal
input and the steady-state output of a system is called frequency response. The frequency response
test is performed by keeping the amplitude A fixed and determining B and Φ for a suitable range of
frequencies. Whenever it is not possible to obtain the transfer function of a system through
analytical techniques, frequency response test can be used to compute its transfer function.
The design and adjustment of open-loop transfer function of a system for specified closed-
loop performance is carried out more easily in frequency domain. Further, the effects of noise and
parameter variations are relatively easy to visualize and assess through frequency response. The
Nyquist criteria is used to extract information about the stability and the relative stability of a
system in frequency domain.

Correlation between time and frequency response

The transfer function of a standard second-order system can be written as,


C ( s) ωn2
T (s) = = 2 .
R ( s) s + 2ζωn s + ωn2
ωn2 1
Substituting s by jω we obtain, T ( jω ) = = .
( jω ) + 2ζωn ( jω ) + ωn (1 − u ) + j 2ζ u
2 2 2

Where, u = ω / ωn is the normalized signal frequency. From the above equation we get,
1
T ( jω ) = M =
(1 − u ) + (2ζ u )2 .
2 2

∠T ( jω ) = φ = − tan −1[2ζ u /(1 − u 2 )]


The steady-state output of the system for a sinusoidal input of unit magnitude and variable
frequency ω is given by,
1  2ζ u 
c (t ) = sin  ωt − tan −1 .
(1 − u ) + (2ζ u )
2 2 2
 1− u2 
It is seen from the above equation that when,
u = 0, M =1 and φ =0
1
u = 1, M = and φ = −π / 2

u = ∞, M →0 and φ → −π
The magnitude and phase angle characteristics for normalized frequency u for certain values of ζ
are shown in figure in the next page.

The frequency where M has a peak value is called resonant frequency. At this point the slope of the
1  −4(1 − ur )ur + 8ζ ur 
2 2
dM
magnitude curve is zero. Setting = 0 we get, − = 0.
du u =ur 2 (1 − u 2 ) 2 + (2ζ u )2  3/ 2
 r r 
Solving, ur = 1 − 2ζ 2 or, resonant frequency ωr = ωn 1 − 2ζ 2 . ………… …… (01)
1
The resonant peak is given by, resonant peak, Mr = . ……………… (02)
2ζ 1 − ζ 2
1
• For, ζ > (= 0.707) , the resonant frequency does not exist and M decreases
2
monotonically with increasing u.
1
• For 0 < ζ < , the resonant frequency is always less than ωn and the resonant peak has a
2
value greater than 1.
From equation (01) and (02) it is seen that The resonant peak M r of frequency response is
indicative of damping factor and the resonant frequency ωr is indicative of natural frequency for
a given ζ and hence indicative of settling time.
1
For ω > ωr , M decreases monotonically. The frequency at which M has a value of is called the
2
1
cut-off frequency ωc . The range of frequencies over which M is equal to or greater than is
2
defined as bandwidth, ωb .
The bandwidth of a second-order system is given by,
1/ 2
ωb = ωn 1 − 2ζ 2 + 2 − 4ζ 2 + 4ζ 4  ………….(03)
 

Figure below shows the plot of resonant peak of frequency


response and the peak overshoot of step response as a
function of ζ .

It is seen that the two performance indices are correlated


as both are the functions of the system damping factor ζ
1
only. For ζ > (= 0.707) the resonant peak does not
2
exist and the correlation breaks down. For this range of ζ ,
M p is hardly perceptible.

From equation (03) it is seen that the bandwidth is


indicative of natural frequency and hence indicative of
settling time, i.e., the speed of response for a given ζ .

Polar Plot
The polar plot of a sinusoidal transfer function G ( jω ) is a
plot of the magnitude of G ( jω ) versus the phase angle of
G ( jω ) on polar coordinates as ω is varied from zero to
infinity. An advantage of using polar plot is that it depicts
the frequency response characteristics of a system
over the entire frequency range in a single plot.
The polar plot of
1 1
G ( jω ) = = ∠ tan −1 ωT is shown in
1 + jωT 1+ ω T2 2

figure below.
The polar plot of the transfer function,
1 is shown in figure
G ( jω ) =
jω (1 + jωT )
above.

The plot is asymptotic to the vertical line


passing through the point (-T, 0).
Polar plots are useful for the stability study of
systems. The general shapes of the polar plots
of some important transfer functions are given
in figure below.
From the plots above, following observations are made,

1. Addition of a nonzero pole to the transfer function results in further rotation of the polar plot
through an angle of -90° as ω → ∞.
2. Addition of a pole at the origin to the transfer function rotates the polar plot at zero and
infinite frequencies by a further angle of -90°.
Bode Plots
jφ (ω )
The transfer function G ( jω ) is represented by, G ( jω ) = G ( jω ) e .
Taking natural logarithm of both sides, ln G ( jω ) = ln G ( jω ) + jφ (ω ) ………………………(04)
The unit of real part is called neper.
Similarly, log G ( jω ) = log G ( jω ) + 0.434 jφ (ω ) ……………………………………..(05)
The standard procedure is to plot 20 log G ( jω ) and phase angle φ (ω ) vs. log ω . The unit of
magnitude 20 log G ( jω ) is decibel. These two plots are called Bode plots in honor of HW Bode.
1 1
Example G ( jω ) = = ∠ tan −1 ωT .
1 + jωT 1+ ω T 2 2

The log-magnitude is, 20 log G ( jω ) = −10 log ( 1 + ω 2T 2 ) .


For low frequencies ( ω = 1/ T ), the log magnitude is approximated as,
20 log G ( jω ) = −10 log1 = 0 db. (01)
For high frequencies ( ω ? 1/ T ), the log magnitude is approximated as,
20 log G ( jω ) = −20 log ω − 20 log T . (02)
The logarithmic plot of equation (01) is a straight line coincident with the horizontal axis. The plot
of equation (02) is also a straight line with a slope -20 db per unit change in log ω . A unit change of
log ω means
log(ω2 / ω1 ) = 1 or, ω2 = 10ω1 .
This range of frequencies is called a decade. The slope of the equation (02) is -20 db/decade.
• The range of frequencies ω2 = 2ω1 is called an octave. Since -20log 2 = - 6 db, the slope -20
db/decade can also be expressed as -20 db/octave.
Further at ω = 1/ T the plot has a value of 0 db. The plot is shown in figure below.

The straight line approximation holds good for ω = 1/ T and ω ? 1/ T . With some loss of accuracy
these could be extended for frequencies ω ≤ 1/ T and ω ≥ 1/ T . The frequency ω = 1/ T at which the
two asymptotes meet is called the corner frequency. The corner frequency divides the plot in low
and high frequency regions.
The actual log-magnitude plot can be obtained by applying corrections for the errors
introduced by asymptotic approximation. The error at the corner frequency ω = 1/ T is,
−10 log(1 + ω 2T 2 ) + 10 log1
= −10 log(1 + 1) + 10 log1 = −3 db

The error at the corner frequency


ω = 1/ 2T is,
−10 log(1 + ω 2T 2 ) + 10 log1
.
= −10 log(1 + 1/ 4) + 10 log1 = −1 db
For 1/ T ≤ ω < ∞ , the error in log-magnitude
is given by,
−10 log(1 + ω 2T 2 ) + 20 log ωT .
The error caused by the asymptotic plot is shown in figure above.

Simple Rules for Plotting Bode Diagrams


The open-loop transfer function for a linear system can be written in the form,
K (1 + jωTa )(1 + jωTa ) L
G ( jω ) =
  ω   ω   .
2

( jω ) (1 + jωTa )(1 + jωTa )L 1 + 2ζ  j  +  j   L


n

  ωn   ωn  
Bode diagram can be sketched for any general system by simply adding the effects of each pole and
each zero in order to determine the angles and intersection points of the asymptotes.

1. Factors of the form K /( jω ) r


K
The log magnitude of this factor is 20 log = −20r log ω + 20 log K and the phase is,
( jω )
r

φ (ω ) = −90or . With log ω as abscissa, the plot of above equation is a straight line having a slope
of -20r db/decade and passing through 20log K at ω = 1 . This is shown in figure below for r = 0,
1, 2 and 3.

2. Pole or zero on the real axis


The pole factor 1/(1 + jωT ) has explained in the previous example. The phase angle for this
factor is φ = − tan −1 ωT . At corner frequency, the phase angle of this factor is -45°. At zero
frequency it is 0° and at infinity it is -90°.
The bode plot for the zero factor (1 + jωT ) has a slope of +20 db/decade and a phase angle of
+ tan −1 ωT . The db correction is added to the asymptotic plot.

3. Complex conjugate poles


The quadratic factor for a pair of complex conjugate poles may be written in normalized form as
1 1 ω
= ; u=
 ω   ω 
2
1 + j 2ζ u − u 2
ωn .
1 + 2ζ  j  +  j 
 ωn   ωn 
1 1/ 2
20 log = −20 log  (1 − u 2 ) 2 + (2ζ u )2 
The log-magnitude of this factor is, 1 + j 2ζ u − u 2
.
= −10 log (1 − u 2 )2 + (2ζ u )2 
1
For u = 1 , 20 log ≈ −10 log1 = 0 .
1 + j 2ζ u − u 2
1
For u ? 1 , 20 log ≈ −10 log u 4 = −40 log u .
1 + j 2ζ u − u 2

Thus, the log-magnitude curve of the quadratic factor consists of two straight-line asymptotes,
one horizontal line at 0 db for u = 1 and the other, a line with a slope -40 db/decade for u ? 1 .
The two asymptotes meet on 0-db line at u = 1. The asymptotic and the actual plots are shown in
figure right.
The error between the actual magnitude and the asymptotic approximation is as given below:

For 0 < u ≤ 1, the error is


−10 log  (1 − u 2 )2 + 4ζ 2 u 2  + 10 log1 For 1 < u ≤ ∞, the error is
−10 log  (1 − u 2 )2 + 4ζ 2 u 2  + 40 log u

The error versus u curves for different values of ζ are shown in figure below.

The phase angle of the quadratic factor is


given by,

 2ζ u 
φ = − tan −1  2 .
 1− u 
The phase angle plot are shown in figure
above. The phase angle curve also
depends on ζ .

General Procedure for Constructing Bode Plots


The following steps will be used in constructing the bode plot for a given G ( jω ) .
1. Write the sinusoidal transfer function in time-constant form.
2. Identify the corner frequencies associated with each factor of the transfer function.
3. Knowing the corner frequencies, draw the asymptotic magnitude plot.
4. From the error graphs, determine the corrections to be applied to the asymptotic plot.
5. Draw a smooth curve through the corrected points such that it is asymptotic to the line
segments. This gives the actual log-magnitude plot.
6. Draw phase angle curve for each factor and add them algebraically to get the phase plot.

64( s + 2) 4(1 + s / 2)
Example G(s) = = .
s ( s + 0.5)( s + 3.2 s + 64) s (1 + 2 s )(1 + 0.05s + s2 / 64)
2

The sinusoidal transfer function in time-constant form is,


4(1 + jω / 2)
G ( jω ) = 2
ω  ω  .
jω (1 + 2 jω )(1 + 0.4 j   −  
8 8
Factor fc Log-magnitude characteristic Phase angle
characteristic
4 / jω - Straight line of slope -20 db/decade, Constant
passing through 20 log 4 = 12 db point -90°
at ω = 1 .
1/1 + 2 jω ω1 = Straight line of 0 db for ω < ω1 , 0 to -90o ,
0.5 straight line of slope -20 db/decade for −45o at ω1 .
ω > ω1 .
1 + j 0.5ω ω2 = Straight line of 0 db for ω < ω2 , 0 to +90o ,
2 straight line of slope +20 db/decade 45o at ω2 .
for ω > ω2 .
ω  ω 
2
ω3 = Straight line of 0 db for ω < ω3 , 0 to − 180o
1 + j 0.4   −   ; ω = 8, ζ = 0.2
−90o at ω3 .
n

8 8 8 straight line of slope -40 db/decade for


ω > ω3 .

To the asymptotic plot, corrections are to be applied to get the actual plot. The following list
shows the list of corrections obtained from the error versus log-magnitude curve (plots are given in
the previous pages).
Frequency Net Correction
Frequency Correction
0.25 -1 db
ω1 = 0.5 -3 db
0.5 -3 db
ω1 / 2 = 0.25, 2ω1 = 1 -1 db
1 0 db
ω2 = 2 +3 db
2 +3 db
ω2 / 2 = 1, 2ω 2 = 4 +1 db
4 + 3 db
ω3 = 8, ζ = 0.2 +8 db
8 + 8 db
ω3 / 2 = 4, 2 ω3 = 16 +2 db
16 +2 db

The phase angle curve may be drawn using the following procedure.
1. For the factor K / ( jω ) , draw a straight line of -90°r.
r

2. The phase angles of the factor (1 + jωT ) ±1 are


a. ±45o at ω =1/T
b. ±26.6o at ω =1/2T
c. ±5.7o at ω =1/10T
d. ±63.4o at ω =2/T
e. ±84.3o at ω =10/T
3. The phase angles for the quadratic factor are
a. −90 at ω =ωn
o

b. A few points of phase angles are read off from the normalized Bode plot for the
particular ζ .

Experimental Determination of Transfer Functions

The system transfer function within a certain degree of accuracy can be obtained from bode plots by
fitting an asymptotic log-magnitude plot to the experimental data using the procedure outlined
below.
1. Use the experimental data to plot the exact log-magnitude and phase angle versus frequency
curves on a semilog graph sheet.
2. Draw the asymptotes on the log-magnitude curve keeping slopes as a multiple of
±20 db/decade. Adjust the corner frequencies keeping in mind the correction factors.
3. If the slope changes by −20m db/decade at ω =ω1 , a factor 1/(1 + jω / ω1 ) will exist in the
m

transfer function.
4. If the slope changes by +20m db/decade at ω =ω2 , a factor (1 + jω / ω2 ) will exist in the
m

transfer function.
5. If the slope changes by −40 db/decade at ω =ω3 , either a double pole or a pair of complex-
conjugate pole will exist in the transfer function. If the error between the asymptotic and
actual curve is about −6 db , then a factor of the form 1/(1 + jω / ω3 ) is present and if the
2

error is positive, then a quadratic factor of the form 1/[1 + j 2ζ (ω / ω3 ) + ( jω / ω3 ) ] is


2

present. The value of ζ will be obtained from the error graph.


6. In the low frequency range, the plot is determined by a factor of the form, K /( jω ) r .
a. For type-0 system, the low frequency asymptote is a horizontal line at x db such that
20 log K = x. from where we can get K .
b. If the low frequency asymptote has a slope of −20 db/decade. , it will indicate a
factor of the form, K / jω . The frequency at which asymptote intersects the 0-db line
numerically represents the value of K also the asymptote has a gain of 20 logK at
ω =1.
c. If the low frequency asymptote has a slope of −40 db/decade , then the transfer
function has a factor of the form K /( jω ) 2 . The frequency at which asymptote
intersects the 0-db line is numerically equal to K also the asymptote has a gain of
20 logK at ω =1.

The log-magnitude curve of type-0, type-1 and type-2 systems are shown in figure below.

After obtaining the transfer function from the log-magnitude curve, the phase angle curve is
constructed and is then compared with the one obtained experimentally. If the two curves are in fair
agreement and the curves at very high frequencies tend to −90o(q − p ), then the transfer function is
of minimum phase type. If the computed phase angle is 180oless negative than the one obtained
experimentally, then the transfer function is of nonminimum phase type and one of the zeros of the
transfer function lies in the right-half s-plane.

The experimental log-magnitude and phase-angle curves of a system are shown in the next page.

First draw the asymptotes on the experimental curve. From the slope changes and the corner
frequencies, the transfer function may be written as,
5 1 1
G ( jω ) = (1 + jω /10)
jω (1 + jω / 2) 1 + j 2ζ (ω / 50) + ( jω / 50) 2
5(1 + jω /10)
=
jω (1 + jω / 2)1 + j 2ζ (ω / 50) + ( jω / 50) 2
At the corner frequency of 50 the error between the actual and the approximate plot is +4 db.
Corresponding ζ from the error graph is obtained as 0.3.
The phase angle at very high frequencies is -270° which is equal to −90 (4 − 1) = −270 . Therefore
o o

the Bode plot represents a minimum-phase transfer function.

Log-magnitude versus Phase Plots


It is a plot of log-magnitude versus phase angle with frequency ω as the running parameter. It is
obtained from the Bode plots by reading the values of db and Φ at different frequencies. The
relative stability of the closed-loop control systems can be determined quickly using this plot.