Analysis and Design of Linear Control Systems, 05th.

10th July, 5th Lecture

Analysis and Design of Linear Control System Part2Spring, 2012 Instructor: Prof. Masayuki Fujita (S5-305)
Lecture Time and Place: Tuesday, 10:4512:15, S224 Schedule: 10th, 24th, 31st, July TA: Tatsuya Ibuki, Yasuaki Wasa (S5-302)

10 PID Control
Keyword : PID Control, Ziegler Nichols Tuning

11 Frequency Domain Design

11.4 Feedback Design via Loop Shaping Keyword : Lead and Lag Compensation

Basic Feedback Loop

d
C ( s)
1

PID Control

F (s)

u
y

P( s)

F =1
Controller

ki s k
p

Controller

P( s)

kd s
Process
Fig. 11.1

P( s) : Process

r : Reference signal d : Load disturbance

n : Measurement noise

C ( s) : Feedback block ( F ( s ): Feedforward block) e : Output error u : Control variable

: Process output : Measured signal

de de 1 t u = k p e + k i e ( ) d + k d e + e ( ) d + Td = kp 0 0 dt Ti dt (10.1) k p : proportional gain k i : integral gain Ti : integral time k d : derivative gain Td : derivative time
t

e=ry

Fig. 10.1 (a)

C ( s) = k p +

ki + kd s s

(10.4)

PID Control

ki s k
p

Response of the process output to a unit step

Controller

P( s)

kd s

1
Past Present Future
Fig. 1.17

Fig. 10.1 (a)

integral proportional derivative

Based on a survey of over eleven thousand controllers in the refining, chemicals and pulp and paper industries, 97% of regulatory controllers utilize PID feedback. [L. Desborough and R. Miller, 2002]

proportional gain k p increases error decreases but more oscillatory

integral gain ki increases steady-state error is eliminated but more oscillatory

Fig. 10.2

derivative gain k d increases more damped

Analysis and Design of Linear Control Systems, 05th.

Ziegler Nichols Tuning (Step response method)

1 e + K u = kp T i de 0 e ( ) d + T d dt
t

Ziegler Nichols Tuning (Frequency response method)

= c

L = kc P( s) u = k p e + T i

t 0

e ( ) d + T d

de dt

(10.1)

(10.1)

T
Fig. 10.7 (a) Tab. 10.1 (a) Fig. 10.7 (b) Tab. 10.1 (b)

: time delay

a / : steepest slope

From a unit step response K K : steady state value e s (10.10) P( s) = K / T : slope 1 + sT

a / = K / T

a = K / T

Step1 Integral and derivative gains are set to zero Step2 Increase proportional gain kc until the system starts to oscillate Step3 From fig. 10.7(b) c is the frequency which L = kc P( s ) intersects the critical point Tc = 2 / c Relay Feedback

[Ex. 10.4] Atomic force microscope in tapping mode 1 e sTn P( s) = sTn ( s + 1) Tn = 2n ( = 0.002 , n = 20)
Ziegler Nichols Tuning
Fig. 3.14 (a)

kc = 21.7 Tc = 0.48
PI Controller

k p = 0.4kc = Ti = 0.8Tc = kp = 23.1 ki = Ti Many versions

1 u = kp e + T i

t 0

e ( ) d + T d

de dt

(10.1)

Tab. 10.1 (b)

- Nathaniel B. Nichols (1914-1997) Born in Michigan B.S degree from Central Michigan University in1936 M.S degree from the University of Michigan in1937 An Honorary Doctor of Science degree from Central Michigan University in1964 An Honorary Doctor of Science degree from Case Western Reserve University in1968 Professional Automatic control, automatic radar tracking Fire control computers, power-driven servomechanisms Industrial process controllers ,spacecraft attitude controls Well known for Ziegler Ziegler and Nichols PID tuning Nichols chart (created to facilitate the calculation of Closed-loop frequency response in Aerospace Corporation)

Windup

The control variable reaches the actuator limits Output is saturated The integral term and the controller output become very large (windup)
Fig. 1.17

Implementation Filtering the Derivative Ideal derivative High gain for high-frequency sig. High-frequency measurement noise will generate large variations
[dB]

Anti-Windup

Add extra feedback path

Low-pass filter Filtering ideal controller

kd s

k d s /(1 + sT f )

20 log k d / T f 20 log k d

kd s
+ 20dB/dec

k d s /(1 + sT f )
1 T f [ rad / s]

es : error signal
es = 0
No saturation When the actuator saturates, the signal es is fed back to the integrator in such a way that es goes toward zero
Fig. 10.11

1 + T s 1 C(s) = k p 1 + sT d 2 i 1 + sTf + (sTf ) / 2

(10.13)
[dB]

Low-pass filter Weighting

peak can be avoided

20dB/dec

| C(j) |

Extra feedback path

Controller output is kept close to the saturation limit

t dr dy u = k p (r y ) + k i (r ( ) y ( ) )d + k d 0 dt dt

: reference weight

: setpoint weight

1T I

20dB/dec 1 T D T f [ rad / s]

Analysis and Design of Linear Control Systems, 05th.

Computer Implementation
dy 1 u = k p ( r y ) + k i (r ( ) y ( ) )d + k d 0 dt (1 + sT f )
t

Feedback Design via Loop Shaping Basic Feedback Loop

d
n

Performance
y

k p ( r y ) P(t k ) = k p (r (t k ) y (tk ) )

C ( s)
1

u
y

P(s)

L ( j )

1.7 p gc 0.6 z

1.2

k i (r ( ) y ( ) )d
t 0

Controller
h (sat (v) v ) Tt Tt = h / kt : anti windup term
Fig. 10.11

Process

Fig. 11.1

I (t k +1 ) = I (t k ) + ki he(t k ) +

PID Controller
r
ki / s

Robust Stability

h : sampling time

1 dy kd dt (1 + sT f ) Tf k D(t k ) = D (t k 1 ) d ( y (t k ) y (t k 1 ) ) Tf + h Tf + h

kp

P(s)

(a) Frequency response ( L ( s ))

gm = 2 5 m = 30 60 sm = 0.5 0.8
Fig. 11.8

kd s

Controller

1
Fig.10.1 (a)

Time Delay, Smith method

(from backward difference)

Lag Compensator and PI Controller

pole: b , zero: a s+a corner freq.: a , b (11.12) C (s) = k (break point) s+b b < a : Lag compensator b = 0 PI controller 40 Lag

pole: b , zero: a s+a corner freq.: a , b (11.12) C (s) = k (break point) s+b a < b : Lead compensator If s of a denominator equals to
20 10
10 20

Lead Compensator and PD Controller

: PD controller PD controller

20 0

PI

20

Disturbance attenuation at low frequencies Improve tracking performance b a

Lead compensator
Lead

0 30 Lag compensator 60 PI controller 90 a b

Frequency response ( C ( s ) )

PD

Fig. 11.9 (b)

90 60 30 0

Improve phase margin Fig. 11.9 (a) b a Frequency response ( C ( s ) )

Lead and Lag Compensator s + a1 s + a 2 C (s) = k s+b a1 < b1 1 s + b2 a 2 > b2 Lead compensator + Lag compensator PID Conditional Stability Lag
Integral Windup Derivative Factor, etc. Derivative has a high gain for high freq. signals Filtering

Lead compensation Higher-order lead compensator

C (s) = k

higher-order lead compensator can provide larger phase lead than 90

Slope 20n [dB/dec]

(s + a )n (s + b )n

a<b

Lead
Frequency response ( C ( s ) )

kd s
Tf =

Alternative: (Ideal Controller) + (Noise Filter)

kd / k , N 2 20 N

kd s 1 + sT f

n=3
Frequency response ( C ( s ) )

k 1 e.g. C ( s ) = k p + i + k d s 2 2 s T f / 2s + T f s + 1

Analysis and Design of Linear Control Systems, 05th.

[Ex. 11.5] Atomic force microscope in tapping mode

(3.5, Exe. 9.2)

[Ex. 11.5] Atomic force microscope in tapping mode

System dynamics a(1 e s ) P( s ) = s ( s + a) Integral controller k Ci ( s ) = i ki = 8.3 s Phase margin: m =
40 20 0 -20 -40
1 100 (by 10 attenuation 10 1 10k 10 2) disturbance i /s

1.2

gc = 4

-60

Fig. 3.14 (a) Schematic diagram

(b) AFM image of DNA

system dynamics (dynamics of the vertical motion) a 1 e s P( s) = s ( s + a ) a = 0 = 1 = 2n / 0 = 0.25

PI controller k p s + ki C pi ( s) = k p = 3.5 s improve phase margin (by k p ) m = 45 bound on the crossover frequency

-90 -180 -270

: damping ratio 0 : undamped natural frequency

gc = 4 [rad/s] < 1.2 = 1.2 = 4.8 [rad/s] 0.25

102

10 1

100

101

10 2

Fig. 11.10 (a) Loop shaping

[Ex. 11.5] Atomic force microscope in tapping mode PI controller System dynamics k p s + ki s C pi ( s) = a 1 e P( s) = s s ( s + a ) ki = 8.3 k p = 3.5

sensitivity

(M s < 2 )
1.6

(M T
T

1.3

< 1.25)

[Ex. 11.5] Atomic force microscope in tapping mode PI controller Integral controller k p s + ki ki C pi ( s) = Ci ( s ) = s s k = 8.3 k = 3.5
i p

complementary sensitivity Feedforward (2 Degree of Freedom)

Closed system with Ci

load sensitivity

PS
Fig. 11.10 (b) Gang of Four

0.3 P

CS 6
P 1

3.5 C noise sensitivity

Closed system with C pi

Step response

10th July, 5th Lecture

10 PID Control
Keyword : PID Control, Ziegler Nichols Tuning

(17th July) Friday Class

24th July, 6th Lecture
11 Frequency Domain Design
11.4 Feedback Design via Loop Shaping Keyword : Lead and Lag Compensation 11.2 Feedforward Design Keyword : Feedforward 2 Degree of Freedom 11.3 Performance Specifications Keyword : Time Domain Analysis Step Response

11 Frequency Domain Design

11.4 Feedback Design via Loop Shaping Keyword : Lead and Lag Compensation