Introduction to the Concepts of

PID AUTO-TUNING

Prof.Dr.ir. Robin DE KEYSER

Ghent University – EeSA Department of Electrical energy, Systems & Automation
[email protected]

Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 1
 Part 1:
AUTO-TUNERS

Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 2
 RELAY Experiment

ÅSTRÖM-HÄGGLUND (AH) & KAISER-CHIARA (KC) KAISER-RAJKA (KR)

Σ
e d u y Σ
e d u y
P(s) P(s)
τd
Relay Process Relay Delay Process

-1 -1

u
d=1

0 KR: ITERATIVE procedure

-1 select τ d as:
PM − 37 *
y τd = Tc
0.02 Tc 360
0 2a PM = desired Phase Margin
-0.02

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 3
 TUNING Values
4d
period oscillation = Tc ; amplitude oscillation = a ⇒ Kc =
πa
(Tc : critical period; K c : critical gain)
AH : UCM (relay) KC : PMM (relay) KR : PMM (relay + delay)
Mode
Ultimate Cycle Method Phase Margin Method Phase Margin Method
P control K p = 0.5 K c adapt PID as for AH adapt PID as for AH
K p = 0.45 K c
PI control adapt PID as for AH adapt PID as for AH
Ti = 0.83Tc
K p = K c cos PM
K p = 0.6K c K p = 0.8K *c
PID 1 + sin PM
= =
Ti 0.5T T T = T 0.64T*

π *cos PM
c i c i c
control
=
Td 0.25T Td 0.25Ti
i Td = 0.25Ti
Notice: Ti=4Td ⇒ PID has 2 identical real zeros
Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 4
 AH: tuning principle (1/2)

2π output amplitude a πa 1
ωc = 2π f c =  G ( jωc ) = Φ = −180° G ( jωc ) = M = = = =
Tc input amplitude 4d 4d Kc
π

Critical point
Im G(jω) ω = ωc 1
NYQUIST (polar plot) Kc

x
-1  G ( jω ) Re G(jω) -1
G ( jωc ) = Me jΦ
G ( jω )
1 − j180°
= e
Kc
ω 1
= −
Kc
G(s) = (unknown !) process transfer function
Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 5
 AH: tuning principle (2/2)

 1   1 
PID: R( s )= K p 1 + + Td s  ⇒ R( jω c )= K p 1 + + Td jω c 
 Ti s   Ti jω c 
  
  2π 1   π 1 
R( jω c ) = K p 1 + j  Td −   = 0.6 K c 1 + j  −   = K c (0.6 + 0.28 j )
  Tc T 2π     4 π 
  i 
  Tc 

−1
LOOP frequency response: R( s )G ( s ) ⇒ R( jω c )G ( jω c ) =
K c (0.6 + 0.28 j ) *
Kc

NYQUIST
-0,6 ROBUSTNESS:
-1
ωc -0,28 0.42 + 0.282 =
0.5

Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 6
 KC: tuning principle (1/2)

1
Given: G ( jωc ) = − (ref. AH)
Kc
Task: find the controller parameters so that the Phase Margin = PM
 1    1 
R( jωc ) = K p 1 + + Td jωc  = K p 1 + j  Td ωc − 
 Ti jωc    T ω
i c 

R ( jωc )G ( jωc ) = 1* e j ( −180°+ PM ) = cos(−180° + PM ) + j sin(−180° + PM ) = −a − jb

b

Typical PM =40° … 70° 1

PM larger: x -a
a
• more robustness -1

• less overshoot M=1

-b PM
• larger settling time ω = ωc
Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 7
 KC: tuning principle (2/2)

cos(−180° + PM ) + j sin(−180° + PM ) = −a − jb ⇒ a = cos PM and b = sin PM

  1   1 
R ( jωc )G ( jωc ) =K p 1 + j  Td ωc −   *  −  =−a − jb =− [ co PMs + j sin PM ]
  Tiωc    K c 
Kp Kp  1 
⇒ cos PM and =  Td ωc −  sin PM
Kc Kc  Tiωc 
1
put Ti =4Td ⇒ Td ωc − =tan PM ⇒ (Td ωc ) − tan PM * Td ωc − 0.25 =
2
0
4Td ωc

( tan PM ) + 1 tan PM
2 2
tan PM  tan PM  tan PM 1
Td ω=
c ±   + 0.25= ± = ±
2  2  2 2 2 2 cos PM
sin PM + 1 sin PM + 1
Td ωc =
(only + gives a positive result for Td ωc ) ⇒ Td Tc
2 cos PM 4π cos PM
Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 8
 KR: tuning principle (1/2)

BODE: Magnitude (dB); Phase (deg)

30

1
25
Kp=1!!!
1+ + Td s for Ti =
4Td 20
Ti s 15

1 10
(2 real zeros at s = − ) 5
2Td 1.25 (1.94 dB)
0
0.5 1.0
90

Tc* 60
take ωTd ≡ 1 or Td = 30 37°
2π
0
*
Kc
take K p = and Ti 4Td -30
1.25 -60
ωTd
-90
-2 -1 0 1
10 10 Frequency (rad/sec) 10 10

Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 9
 KR: tuning principle (2/2)
*
K
R( jωc* ) K=
p *1.25e
j 37°
=c
*1.25e j 37° K c*e j 37°
1.25
1 j ( −180°+ϕ )
LOOP frequency response: R( s )G ( s ) ⇒ R( jω c )G ( jω c ) = *
Kc e * * e * * j 37°

Kc
τd
PM − 37 *
⇒ Phase Margin= 37° + ϕ= 37° + 360° *= PM because τ d= Tc
Tc 360

ω = ω c* ϕ
Typical PM=40° … 70°
 τd 
PM larger: = ω=*
τ
c d 2π 
 Tc* 
• more robustness x
-1
• less overshoot 1 1 j ( −180°+ϕ )
G ( jω *
c ) = e
• larger settling time K c* K c*

Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 10
 Part 2:
APPLICATIONS
• Double Integrator (project)
• Time Delay (example)

Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 11
 Double Integrator / Dead-time

DIFFICULT control problem (double integrator!)

(angular) position control
(servo, robot, antenna, disk drive, …)
32
nominal system:
s(s+4)(s+16)
DIFFICULT control problem (dead-time / time-delay!)
heat exchanger: temperature control
(concentration regulation, material transport effects, …)
2e −3 s
nominal system:
(1+10s)(1+5s)
Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 12
 Time Delay: PI and PID FR-designs

Ro=0.45 Ro=0.45
OS<40% OS<40%
Ts minimum (100 shown) Ts minimum (50 shown)

FR-PI FR-PID

Zero: -0.06 Zeros: -0.12 -0.12

Kp=1.0 Ti=16.7 Kp=1.82 Ti=16.64 Td=4.16
Gain: 1.0*0.06=0.06 Gain: 7.57*0.12^2=0.11

Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 13
 Time Delay: KR and AH autotuners

Ro=0.45 Ro=0.45
OS<40% OS<40%
Ts: 50 shown Ts: 50 shown

KR 10° AH

Zeros: -0.135 -0.135 Zeros: -0.192 -0.192

Kp=1.95 Ti=14.84 Td=3.71 Kp=1.73 Ti=10.4 Td=2.6
Gain: 7.2*0.135^2=0.13 Gain: 4.5*0.192^2=0.17

Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 14
 Time Delay: PID parameters

Kp Ti Td K z1 z2
FR-PI 1.00 16.7 0 1.00 −0.06 −
FR-PID 1.82 16.6 4.2 7.57 −0.12 −0.12
AH 1.62 10.5 2.6 4.26 −0.19 −0.19
KC 1.84 17.0 4.2 7.80 −0.12 −0.12
KR 1.85 15.4 3.8 7.10 −0.13 −0.13

AH : a =0.47; Tc =21 KC : PM =47° KR : PM =47°;τ d =0.65; a =0.55; Tc =24

Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 15
 Time Delay: Setpoint Step
1.4

AH
PI
1.2

KR
0.8 PID KC

0.6

0.4

0.2

0
0 5 10 15 20 25 30 35 40 45 50
Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 16
 Time Delay: Load Disturbance
0.9

0.8

0.7

0.6
PI
0.5
AH
0.4

0.3
PID
KC
0.2

KR
0.1

-0.1
0 5 10 15 20 25 30 35 40 45 50
Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 17
 Robot Controlled Motor

φ
Dynamic Model for:
V=19.6 m/s; c=0.98 m; L=h=0.49 m
Design 5 PID-controllers:
- a controller tuned by yourself
- a controller designed with FRtool
- a controller using the AH autotuner
- a controller using the KC autotuner (PM=47°)
- a controller using the KR autotuner (PM=47°) L
Compare the results and discuss.
Check setpoint & disturbance response

32
??? s(s+4)(s+16)

1 (very fast)
Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 18
 CONCLUSIONS

1) If a process model is available:

Computer Assisted Control System Design!

The use of a CACSD-tool (such as FR-tool)
turns the controller-design-problem into
an interactive and graphically-based
easy exercise.
2) If a process model is NOT available:

Auto-Tuning!
The use of an autotuner
turns the controller-design-problem into
an automatic design based on a simple experiment.
Robin De KEYSER, Ghent University / Belgium, EeSA-department of Electrical energy, Systems & Automation 19