3 Linear-System PDF
3 Linear-System PDF
Dipankar Sanyal
Mechanical Engineering
Department
Jadavpur University
Presentation layout
• Introduction to feedforward-PID (FFPID) control through a simple example of motion dynamics of a mass
excited by a solenoid motor
• Characteristic function and equation for a linear system
• Poles as roots of a characteristic equation
• Block diagram and transfer function
• Solution of transfer function for stability assessment
• Stability assessment by forming Routh/Routh-Hurwitz array
• Steady error
• Post-transient periodic response for sinusoidal demand – “steady response” → Bode plot/Gain-phase plot
• Gain and phase margins and crossover frequencies
Feedforward-PID (FPID) Control of Nonlinear Dynamic System – an example
Response r to a voltage excitationV 0 to a nonlinear dynamic system N 0 with initial gap g 0 is given as
r − a1r + a0 r − bV 2 /( g 0 − r ) 2 − = 0 , where a1 , a0 , b are dimensional positive constants, is known &
major part of an uncertain disturbance. Excitation voltage V = V f + Vb involves a feedforward voltage V f &
PID feedback given as Vb = k p e + ki edt + k d e with Vb V f and error e = d − r , the difference between
demand 0 d (t ) g 0 . The controller keeps the error bound close to zero implying e r .
N 0, r = d − e, e r & Vb V f
d − a1d + a0 d − bV f2 /( g 0 − d ) 2 − = 0 Feedforward formulation aiming explicit algebraic equation for Vf
& e − a1e + a0 e − bV f2 /( g 0 − d ) 2 − bV f2 (1 + 2Vb / V f ) /[( g 0 − d ) 2 {1 + 2e /( g 0 − d )}] − = 0 : L 0
V f = ( g 0 − d ) | (d − a1d + a0 d − )1/ 2 | uncertainties & approximations linearized error dynamics
a) all positive coefficients to error and its derivatives k p max [V f /{b( g 0 − d )} − a0 ( g 0 − d ) 2 /( 2bV f )] |max ,0 ,
k d a1 /{2bV f /( g 0 − d ) 2 } |max , ki 0.
t t s
b) integral gain ki = /{2V f edt /( g 0 − d ) 2 } ensuring lower ts for higher ki .
0
Characteristic Function & Equation for a Linear System
s n + an −1s n −1 + ... + a2 s 2 + a1s + a0 = 0 is called the characteristic equation, since its root characterizes the
system behavior in terms of a first or second-order system or their combination for n equal to 1, 2 or greater than
2 respectively. RHS is called the characteristic function. Roots of characteristic equations are called poles.
Poles and Transients for Characteristic Equations of Different Orders
First order: s + 1 = 0 → Initial transient growth of error inflicted by uncertain part of Im()
1/2,(15/4)
disturbance eventually decays asymptotically to 0 for negative real pole. 2, 2 j
-2 -1
Second order : s 2 + 2 2 s + 22 + 22 = ( s + 2 + j2 )( s + 2 − j2 ) = 0 Re()
1
→ Initial growth of error eventually decays asymptotically to 0 for negative real part 2, -2 -j unstable
stable
of the complex conjugate poles together characterizing a second order system. 1/2,-(15/4)
+ .... + a2 s 2 + a1s + a0 i =1 ( s − zi ) = 0; zi = i + ji is stable,
n −1 n
Order n: s + an −1s
n
e 1st Order Characteristics
if real parts i of all poles zi are negative, requiring all coefficients ai to be positive. 12<11
Examples third order:
11
(E1) s 3 + s 2 + s + 1 = ( s + 1)( s 2 − s + 1) + s ( s + 1) = ( s + 1)( s 2 + 1) = ( s + 1)( s + j )( s − j ) t
(E 2) s 3 + s 2 + 2 s + 8 = ( s + 2)( s 2 − s + 4) = ( s + 2)( s − 1 / 2 + j 15 / 4 )( s − 1 / 2 − j 15 / 4 )e 2nd Order Characteristics
→ In addition to necessary condition of positive coefficients, (E2) makes it evident that
additional sufficiency condition is necessary for stability to avoid complex conjugate poles 22<21
of the second-order part falling on the positive half-plane excluding the imaginary axis.
(E1) with a pair of poles lying on the imaginary axis implies bounded stability for periodic t
Linear Linear
d(t) e(t) Feedback V(t) r(t) D(s) E(s) Feedback V(s) R(s)
+ = Dynamic + = Dynamic
− Controller, c(t) − Controller, System, G(s)
System, g(t)
C(s)
Sensor, Sensor, S(s)
s(t) Control System in Time Domain and after Laplace Transformation
Block Diagram for Linear
Appropriate C(s) would absorb
R ( s ) = G ( s )V ( s ) = G ( s )C ( s ) E ( s ) = G ( s )C ( s ){D( s ) − R ( s ) S ( s )} approximations & uncertainties
Numerator/Denominator form; in G(s) by making response
R ( s ){1 + G ( s )C ( s ) S ( s )} = G ( s )C ( s ) D( s )
numerator polynomial roots are dependent only on sensor
R ( s ) / D( s ) = G ( s )C ( s ) /{1 + G ( s )C ( s ) S ( s )} called zeros. accuracy, if product of constant
R ( s ) / D( s ) = 1 / S ( s ) for G ( s )C ( s ) S ( s ) 1 terms in G(s)C(s)S(s)>>1.
Solution of Transfer Function – Example of a 2nd order system
s c1 c2 c1 ( s + p2 ) + c2 ( s + p1 )
R( s) = = + =
s 2 + s + 1 s + p1 s + p2 {s 2 + ( p1 + p2 ) s + p1 p2 }
p1 + p2 = 1; p1 p2 = 1 p1 − p2 = (12 − 4)1/ 2 = j 3
p1 , p2 = (1 j 3 ) / 2
c1{s + (1 − 3 j ) / 2} + c2 {s + (1 + 3 j ) / 2} = s
c1{−(1 + 3 j ) / 2 + (1 − 3 j ) / 2} = −(1 + 3 j ) / 2 & c2 {−(1 − 3 j ) / 2 + (1 + 3 j ) / 2} = −(1 − 3 j ) / 2
c1 = (1 + 3 j ) /( 2 3 j ) = ( 3 − j ) /( 2 3 ) & c2 = −(1 − 3 j ) /( 2 3 j ) = ( j + 3 ) /( 2 3 )
r (t ) = exp(−t / 2)[( 3 − j ) exp{−( 3 / 2) jt} + ( j + 3 ) exp{( 3 / 2) jt}] /( 2 3 )
= exp(−t / 2)[( 3 − j ){cos( 3t / 2) − j sin( 3t / 2)} + ( j + 3 ){cos( 3t / 2) + j sin( 3t / 2)] /( 2 3 )
= exp(−t / 2){cos( 3t / 2) − sin( 3t / 2) /( 2 3 )}
Denominator of a closed loop transfer function is the product of the denominator functions of all the individual
blocks the controller, plant and sensor, whereas the numerator is a product with blocks in the feedback path
omitted. This makes the denominator order higher than the numerator. Only zero of R(s) lies at origin of the
complex plane ̶̶̶ -.
Stability Assessment from Routh or Routh-Hurwitz (R-H) Array
R( s) G (s) sD ( s )
D(s) E(s) ess = Lim[ s{D( s ) − R ( s )}] = Lim[ sD ( s ){1 − }] = Lim[ sD ( s ){1 − }] = Lim{ }
+ = G(s) R(s) s →0 s →0 D( s) s →0 1 + G (s) s →0 1 + G ( s )
−
Example: Unit step input d(t)=1 D(s)=1/s and G(s)=kp/(s2+as+b2)
1 s (1 / s ) b s 2 + as + b
ess = Lim{ } = Lim{ }= with increase in k p
Closed loop with unity feedback s →0 s →0 s + as + b + k p
2kp b+kp
1+ 2
s + as + b
If proportional gain kp is changed to ki /s for integral gain These inferences for diminishing steady error with
s (1 / s ) s 3 + as 2 + bs increase in proportional gain kp and zero steady error in
ess = Lim{ } = Lim{ }= 0
s →0 k /s s →0 s + as + bs + k i
3 2 presence of integral gain ki remain valid for any rational
1+ 2 i
s + as + b transfer function with order of numerator polynomial
lower than the denominator polynomial.
Amplitude and Phase of Steady-State System Response to Sinusoidal Input
D(s) =
G(s) R(s) Steady-state response rs(t) is the stable periodic response to a sinusoidal input demand
/(2+ s2)
d(t) = sin(t) D(s) = /(s2+2) after the subsidence of the initial transient.
R ( s ) = D( s )G ( s ) = G ( s ) /( s 2 + 2 ) = K1 /( s + j ) + K 2 /( s − j ) + terms from poles of G(s) with decaying response for
being left of imaginary axis with demand poles on it
K1 ( s − j ) + K 2 ( s + j ) = G ( s ) K1 = −G (− j ) /( 2 j ) & K 2 = G ( j ) /( 2 j )
rs (t ) = [− exp(− jt )G (− j ) + exp( jt )G ( j )] /( 2 j ) with G ( j ) = Re{G ( j )} + j Im{G ( j )} = M ( ) exp{ j ( )},
M ( ) | G ( j ) |= [Re 2 {G ( j )} + Im 2 {G ( j )}]1 / 2 =| G (− j ) |,
( ) G ( j ) = tan −1[Im{G ( j )} / Re{G ( j )}] = −G (− j )
rs (t ) = M ( )[− exp{− j (t + ) + exp j (t + )] /( 2 j ) = M ( ) sin(t + )
Of course, G(s) separating the demand and response emerges as a rational function through loop reduction algebra.
A rational function is viewed as a constant in the numerator along with a number of linear and quadratic factors in
product forms.
The magnification M() is represented as 20log10M(), as the logarithm operation forms additive groups for the
numerator factors and subtractive groups for the denominator factors.
The plot of variations of 20log10M() and against log10 is called Bode plot.
Bode plots (Gain-Phase plots) for first-order and second-order systems
First order system with time constant T1: Corner plot
Asymptotic plot
G ( s ) = 1 /{T1 ( s + 1 / T1 )} = tan −1 (−T1 ) 20log10M
0.1 1 10 T
0 1
& 20 log10 M = 20 log10 {1 /{(T1 ) 2 + 1}1 / 2 = −10 log10 {1 + (T1 ) 2 }
−20 20db/decade decay
A) For (1 / T1 ), 20 log10 M → 10 log10 (1) = 0 & → 0
B) For (1 / T1 ), 20 log10 M = −20 log10 (T1 ) & → −( / 2)
0.1 1 10 T1
0
C ) At = (1 / T1 ), 20 log10 M = −10 log10 2 & = −( / 4)
− /4
Second order system with natural frequency 2 (time constant T2):
− /2
G ( s ) = 1 /{(1 / 22 )( s 2 + 2s + 22 )} = tan −1[−(2 / 2 ) /{1 − ( / 2 ) 2 }]
20log10M
1
& 20 log10 M = 20 log10 0
0.1 1 10 /
2
[{1 − ( / 2 ) 2 }2 + (2 / 2 ) 2 ]1 / 2
−20
= −10 log10 [{1 − ( / 2 ) } + (2 / 2 ) ]
2 2 2
40db/decade decay
−40
A) For 2 , 20 log10 M → 10 log10 (1) = 0 & → 0 through tan −1[ 0]
0.1 1 10 / 2
B) For 2 , 20 log10 M = −40 log10 ( / 2 ) & → − 0
− /2
through tan −1[ 2 2 / ] : 3rd quadrant
C ) At = (1 / T1 ), = − / 2 for falling between A) & C) above −
Bode plot for first-order by second-order system
First order numerator with time constant T1 and second order system with time constant T2 = 10 T1 :
G ( s ) = {T1 ( s + 1 / T1 )} /{T22 ( s 2 + 2s + 1 / T22 )} 20log10M
Corner plot Asymptotic plot
{(T1 ) 2 + 1}1 / 2 20
20 log10 M = 20 log10 20db/decade rise
[{1 − (T2 ) 2 }2 + (2T2 ) 2 ]1 / 2 0 0.1 1 10 100 T1
= 10 log10 {1 + (T1 ) 2 } − −20
40db/decade decay
10 log10 [{1 − (T2 ) 2 }2 + (2T2 ) 2 ]
A) For (1 / T1 ), 20 log10 M → 10 log10 (1) = 0 /2
& → 0 /4
0.1 1 10 100 T1
0
B) For (1 / T1 ) (1 / T2 ), 20 log10 M → 20 log10 (T1 ) − /4
& → ( / 2) − /2
C) For (1 / T2 ), 20 log10 M → 20 log10 (T1 ) − 40 log10 (T2 )
& → ( / 2) − = − / 2 Any additional constant K in the numerator would
D) At = (1 / T1 ), 20 log10 M → 10 log10 2 lead to a shift of the magnification (also called
& → / 4 gain) corner plot along the ordinate by 20log10 K
without any change of phase plot.
Gain and phase margins and crossover frequencies
G ( s ) = {T1 ( s + 1 / T1 )} /[{T22 ( s 2 + 2s + 1 / T22 )}{T3 ( s + 1 / T3 )}]; T3 = 10T2 = 100T3
The frequency at which the system phase crosses
20log10M
Corner plot Asymptotic plot -1800 phase is called the phase crossover
20 20db/decade rise 40db/decade decay frequency and the deviation of the gain at this
0.1 1 10 a 102 b 103
0 T1 frequency from unity is called the gain margin. In
Gain Margin the figure, the gain margin is about 38db. Gain
−20
attenuation above gain margin of 6db is
−40
considered unacceptable and a proportional gain
20db/decade decay
block prior to the plant would be required to
/2 achieve such a value in the present case.
/4
0.1 1 10 102 103 T1 The frequency at which the system gain falls
0
− /4 Phase Margin below unity gain is called the gain crossover
− /2 frequency and the deviation of the phase at this
a: Gain crossover frequency
b: Phase crossover frequency frequency from zero is called the phase margin.
−
In the figure, the phase margin is 1800. A phase
margin up to 900 is acceptable from stability viewpoint and interchange of time constants in the denominator
block would provide an acceptable solution.
Thank you