Vous êtes sur la page 1sur 11

2010/8/9

Asia-Pacific Summer School on Smart Structure Technology

Introduction of Control Competition


& Lecture on PID Control

Overview

Control object and tool for Competition

Introduction to classical control

Toru Watanabe
Nihon University, JPN

Introduction of DCMCT unit and QICii software


Laplace Transform
Dynamical modeling
Transfer Function
Block Diagram and Feedback
Poles and Zeros

Introduction to PID control

Control Object

Control Object and Tools for


Competition

DCMCT (DC Motor Control Trainer unit)

QICii (Interactive interface software)

Outlook of DCMCT Unit

Equipped with DC motor with flywheel, amplifier,


microchip controller and digital/analog interface
Interactive modeling, feedback controller design and
evaluation can be carried out
Able to upload controller to DCMCT / download time
history

Block diagram of the DCMCT


1. DC Motor
2. Flywheel
3. Amp.
4. Encoder
7. Analog I/O
9. Microchip
10. USB port
11. CTL Reset
12. Interrupt SW
15. Power port

Schematic view of open loop

Control
Error of angle
Driving
command
or angular
Current
velocity
MicroAmpliMochip
fier
tor

Friction
Torque

Torque

Angle

Flywheel

Angle

Encoder

Controller is downloaded from PC by using QICii

Controller on 9 drives 1 through 3 based on signal from 4

2010/8/9

Window of QICii

Tasks in Control Lab & Competition

Educational Labs

Control Lab Competition

Identify parameters in the control object


Study basic properties of P, I, and D controls
Supervisors add extra weights to flywheel
Design PID controller to above perturbed system
Investigate robustness and evaluate performance

Universal Interface for the operation of DCMCT

Why DCMCT & QICii?

What is Classical Control?

In the field of Control Engineering, there


are two major stream of formulation

Introduction to Classical
Control

Electromagnetic motor is commonly used in


structural control to drive mechanisms such as
mass dampers
General idea of feedback control
control, controller
design procedure and characteristics of gains
are essentially identical over any dynamical
systems
They are so well-organized that fair competition
can be done within such limited time

State space equation essentially in time domain


Transfer function essentially
y in frequency
q
y domain

Both formulations are important

State space model is widely used in modern (or LQ)


control, while transfer function is the basis of
classical control.
Post modern control (e.g. H-infinity) uses both.

State-Space vs. Transfer Function (1)

Example of State Space Formulation


cx(t)
kx(t) u(t)
mx(t)
y(t) x(t)
X x(t)
(t) v(t)
(t)

0

x(t)
k
v(t)

1
0
Dam
x(t)
1 u(t) ping
c

v(t)
c
m
m
x(t)

y(t) 1 0 v(t)

Disp. x(t)
Mass m
u

Spring
k

2010/8/9

State-Space vs. Transfer Function (2)

Example of transfer function formulation


cx(t)
kx(t) u(t)
mx(t)
Replacing derivative and time
with Laplace operator s
s

ms 2 X(s) csX(s) kX(s) U(s)

X(s)
1
G(s)

U(s) ms 2 cs k

Dam
ping
c

Modeling

Disp. x(t)

Mainly in frequency domain s domain

Controller Design

Spring
k

Utilizing transfer functions to describe the


dynamics of control objects

Analysis

Mass m

Take output/input ratio

Framework of Classical Control

Generally for single-input single-output system


Manual calculation based design (practically by
try-and-error approach)

Advantage of Classical Control

Simple

Analysis and controller design can be done without


computer, nor MATLAB of course they are helpful
Simple controller such as PID controller is easy to tune
manually, and easy to understand intuitively

Disadvantage

Hard to deal with MIMO (multi-input multi-output) system


The optimality of controller is not guaranteed

Laplace Transform

Identification

F s = f t e-st dt = L f t

Laplace Transform and Differential


Equation

Laplace transform of
integrated function

d n x(t )
d n1 x(t )
dx(t )
1
n1
n u (t )
dt n
dt n1
dt
n 1
d x(t )
dx(t )
n1
n
y (t ) 1
dt n1
dt

L f t sF ( s )

1
L f d F ( s )
0
s

Differential equation in time domain


0

Laplace transform of L d f t sF ( s ) f ((0))


d i d ffunction
derived
ti
d
dt

Neglecting initial value

Introduction to Laplace
Transform

Advantage

Transformed differential equation

s
0

1s n1 n1s n X s U( s )

Y( s ) 1s n1 n1s n X s

Transformed equation is no longer differential, but algebraic

2010/8/9

Laplace Transform to Solve Differential


Equation Operator Method
Tough to solve
t-domain

Differential
Equation

Laplace transform
s-domain

Algebraic
Equation

solution

Laplace Transform and Fourier


Transform

Identification of Fourier Transform

F j =

(t-function)

Inverse transform

Easy to solve

Algebraic
solution

Infinite function (e.g.


unit step function)
cannot be transformed

f (t ) e- jt dt

Introduce convergence factor


F j =

f (t ) e- jt e- t dt

Infinite function (e.g.


unit step function) can
be transformed!

Replacing +j with s, Laplace transform


F j =

(s-function)

- j t

f (t ) e

dt =

f (t ) e- st dt F s

Representing Operator s

Therefore, the transformed function must be


g as Real/Imaginal
g
ratio
complex to denote damping

In Laplace transform, operator s can be


understood as frequency and damping

Therefore, the transformed function simply


became rational function because the existence of
damping is already taken into consideration as s

Modeling Dynamical System (1)

Modeling of Dynamical
Systems as Transfer Function

In Fourier transform, operator j can be


simply understood as frequency

First, identify all elements concerning dynamics


of control object

Electric Circuit including DC motor


Resistance Rm
Voltage
Vm(t)

Electromotive
coefficient Km

Flywheel including rotor


Motor
torque

Moment of
Inertia Jeq(t)

Tm(t)
Current Im(t)

Inductance Lm

Armature (rotor)
rotation speed

m(t)

Friction Torque Td(t)

Modeling Dynamical System (2)

Describe all elements as differential equations

Each element is supposed to possess single input and


single output

Electric Circuit including DC motor

Vm (t) R m I m (t) L m Im (t) K m m (t)


t


Flywheel J eq
m (t) K m I m (t) Td (t)
t
Electromotive / Back
electromotive Force

Tm (t) K m I m (t), Vbe (t) K m m (t)

2010/8/9

Modeling Dynamical System (3)

Modeling Dynamical System (4)

Transform all equations into s domain

Electric Circuit including DC motor

Electric Circuit including DC motor

R m L ms Im (s) Vm (s) K mm (s)

Vm (s) R m I m (s) L msI m (s) K m m (s)

I m (s)

Flywheel

Flywheel

J eqs m (s) K m I m (s) Td (s)

Flywheel & Electric Circuit with motor


2

Km
Km
V (s)
(s) Td (s)
R m L ms m
R m L ms m

Km
Km
V (s) Td (s)
J eqs
(s)
R m L ms m
R m L ms m

Transfer
Function

m (s)

Vm (s)

Substituted in

Transfer Function & State Equation (1)

Transform the equation into transfer function


J eqs m (s)

1
Km
Vm (s)
(s)
R m L ms
R m L ms m

J eqs m (s) K m I m (s) Td (s)

Modeling Dynamical System (4)

Erase non-essential variables

Km
2

Km
R m Lms J eqs

R
L
s

m
m

Here, Td(s)
is neglected

You may be able to obtain transfer function


from state equation

Vm (t) R m I m (t) L m Im (t) K m m (t)


t

J eq m (t) K m I m (t) Td (t)


t

Here, Td(s)

Rm

I(t) L m

K
(t) m

J eq

is neglected
Km
1
L m I(t)
I(t)

L m Vm (t) , y(t)= 0 1

(t)
(t)
0
0

Transfer Function & State Equation (2)

Matrix equation can be dealt with as


t AX t Bu t , y t = CX t
X
Transform all equations into s domain

sX s AX s Bu s , y s = CX s

Transfer Function and


Frequency Response Function

Transpose terms,

y s = C sI - A Bu( s ), G s
1

Transfer
Function

y s
1
C sI - A B
u( s )

m (s)
Km

Vm (s) J eqs R m L ms K m 2

NOTE: Denominator is reduced and compound fraction is turned to simple

2010/8/9

Analogy: Frequency Response Function

Transfer Function is quite similar to


Frequency Response Function

s j

Di
Disp.
x(t)
(t)

u(t) U()e jt
x(t) X()e

j t

Mass m

and take output/input ratio

Spring
k

Dam
ping
c

X()
1
G()

U() m2 cj k

As shown, operator s is a complex variable

External
force u

cx(t)
kx(t) u(t)
mx(t)
Suppose

Difference between TF and FRF

As we had already dealt with,


with Fourier
transform (basis of FRF) is quite similar to
Laplace transform (basis of TF).
FRF can be obtained by simply replacing s
of TF to j

Laplace Transform and Frequency


Response Function

Block Diagram and


Connection

Tough to derive
t-domain

Frequency Response
Function (-function)

System

Laplace transform
s-domain

Replace s with j

System

Transfer Function

(s-function) Easy to derive (s-function)

Block Diagram Graphical Description

Each block denotes transfer function


Input

Parallel Connection = Addition or


Subtraction

Transfer function

Output

In parallel connection, total transfer function is


equal to the sum or difference of two functions
G1(s)

Signals are added


added, subtracted or distributed
Input A

Input B

Distribution

Addition

Subtraction

Output
C=A-B

Input A
Output C

Input

Output B
A=B=C

Output

Input

Output

G(s)=G1(s) - G2(s)

2(s)

2010/8/9

Cascade Connection = Multiplication

Feedback Connection = Fractional


Function

In cascade connection, total transfer function is


equal to the product of two functions
Input

G1(s)

Input

2(s)

G(s)=G1(s)G2(s)

Input u

Output

Output

Input

Why Feedback Connection makes


Fractional Function?

signal e
signal c

Output y
G1(s)
2( )
2(s)

Proof

Output y

signal e

G1(s)

signal c

2(s)

Output
G(s)=G1(s)/{1+G1(s)G2(s)}

DCMCTs Block Diagram

Input u

Feedback connection makes fractional function

Element-by-element block diagram

Vm

Td

Im
1
S Lm+Rm

Km

Tm

1
Jeq

1
S

Km

y = G1(s) e = G1(s) { u c } = G1(s) { u G2(s) y }


{ 1+G1(s) G2(s) } y = G1(s) u
Therefore, y / u = G1(s) / {1+G1(s) G2(s) }

1
Km
I m (s)
Vm (s)
(s)
R m L ms
R m L ms m

=Tm

J eqs m (s) K m I m (s) Td (s)

DCMCTs Transfer Function according


to Block Diagram

Ignore Td
Let G1(s)= Km / { S Jeq (S Lm +Rm)}
Let G2(s)
G2(s)= Km
According to Feedback Connection, G(s) =
G1(s) / { 1+G1(s)G2(s) }, Then
Transfer
Function

Feedback Compensation and


Cascade Compensation

m (s)
Km

Vm (s) J eqs R m L ms K m 2

2010/8/9

Where to Put Controller?

Difference in Transfer Function

Feedback compensation (common in modern control)


Input u

Suppose control object as G1(s), while


controller as G2(s)
Feedback Compensation

Cascade Compensation

Output y
Control Object

Controller

Cascade compensation (common in classical control)


u

Error e

Controller

Control Object

G(s) = G1(s) / { 1+G1(s)G2(s) }


G(s) = G1(s) G2(s) / { 1+G1(s)G2(s) }

Looks not a big difference just in numerator

No, it is a big difference zeros are different!

Definition Poles and Zeros

Poles and Zeros

Suppose G(s) = N(s)/D(s)

Poles (Pi) = Operator s


s that gives D(s)=0

Poles in classical control is identical to eigenvalues


in modern control
According to transfer function, poles of DCMCT are
given by solving the following equation

Poles denotes the behavior of the system

According to state equation, eigenvalues of DCMCT


are given by solving the following equation
Rm

s L
m
det sI A det
K m
J
eq

On zeros, G(s) possess zero transmission

Poles and Behavior of System

J eq s R m L m s K m 0

On poles, G(s) possess infinite transmission

Zeros (Zi) = Operator s that gives N(s)=0

Poles and Eigenvalues

N(s): Numerator polynomial


D(s): Denominator polynomial

K m
2
Lm

s s R m K m 0
L m J eq L m

If any Re(Pi) > 0, the system possesses instable


Magnitude of Pi (=Re (Pi)* Re (Pi) + Im(Pi)*Im(Pi))
q
y
is identical to natural frequency
Angle of Pi ( = arctan(Re (Pi)/Magnitude of Pi) ) is
identical to damping ratio
See Prof. R. Christensons slide No. 30-34.

2010/8/9

Zeros and Root Loci

We know that the position of poles of the system is


changed according to controller gain.
The map that shows the trace of poles according to
the increase of gain is called Root Loci
Classical control theory tells us that all poles move
to zeros or toward infinite as the gain is increased
to infinity.
Therefore, position or location of zeros matters to
the characteristics of control systems

Concept of PID Controller

Introduction to PID Control

Why PID? How PID used?

Prefix the structure of controller as combination


of Proportional-Integral-Derivative gains

Advantage of PID controller

Input
(target)

s
error

KD

output
t t

Kp

1
s

Control
Object

If the feedback gain is


proportional to displacement

Control
Force

Application

Proportional gain equivalent to spring

Output = displacement

KI

The Role of P, I and D gains (1)

Equivalent
to "Spring"

F kx

Simple only three gains are required to tune


Easy to understand roles of each gains can be
p
intuitively
y
represented
Practically useful PID controller generally
satisfies requirements.
Any simple control problem (e.g., AMD for SDOF)
Local feedback for actuator (e.g., servomotor)

The Role of P, I and D gains (2)

Derivative gain equivalent to damper


Output = displacement
Derivative = velocity

If the feedback gain is


proportional to velocity
Equivalent
to "Damper"

Control
Force

F cx

2010/8/9

The Role of P, I and D gains (3)

Aim of PID Controller

Integral gain Stationary disturbance canceller

Output = displacement
Integral = sum of drift

Suppose stationary
force is subjected

To let the output (rotational angle or speed)


follow the target

If the feedback gain is


proportional to integral

Control
Force

F I xdt

Target
R(s)

DCMCTs transfer function


m (s)
Km

Vm (s) J eqs R m L ms K m 2
Km
m (s)

Vm (s) s J eq s R m L ms K m 2

Transfer Function of DCMCT with PID

Y(s) = m (s)

G(s)

Output : rotational angle

G(s)

Km
K m
s
2
J eq R m
J eq R ms K m

s J eq s R m L m s K m

K sK
m

K I s2 K D

K m sK P K I s 2 K D

s 2 J eq s R m L m s K m

K sK
m

K I s2K D

Effect of PID on DCMCT (2)

Y(s) = (s) case, No control


Lm is neglected for brevity

K m sK
K P K I s2K D

Y(s) = (s)

PID controllers transfer function


Vm (s)
1
sK K I s 2 K D
K P K I sK D P
E m (s)
s
s

Y(s) = (s) case Lm is neglected for brevity

G(s)

G(s)
( )

G(s) = G1(s) G2(s) / { 1+G1(s)G2(s) }

Y(s) = (s)

Output : rotational speed

Cascade Compensation

Y(s) = m(s)

Effect of PID on DCMCT (1)

KI

Control force is
gradually increased
until drift is converged

Transfer Function of DCMCT and PID

DCMCT

Kp

1
s

D gain does not work.


P gain reduces drift, but
not perfectly cancel it.

Output Y(s)
m(s) or
m (s))

Command
V ( )
Vm(s)

KD

Error
E
Em(s)

K m sK P K I s 2 K D

s J eq sR m K m

K sK
m

K I s2K D

K m sK P K I s 2 K D

K m J eq R m s 2 K m K m K P s K m K I

Km KmKP
2

2
m

KmKP

4 K D K m J eq R m K m K I

2 K D K m J eq R m

Poles become complex by introducing PID feedback

10

2010/8/9

How to Design PID gains?

Each gains possess their own trade offs


Any excessive gain makes system unstable
P gain as increased

faster but
b t vibrating
ibrating response
more damping but too sensitive to noise and error

I gain as increased

resistant to drift, but slower response

Several design procedures for PID controller


are already presented

D gain as increased

Design Theories for PID Controller

Ziegler-Nichols
Critical sensitivity
Step response

All these methods are procedures, not


theories because they do not guarantee the
optimality

11

Vous aimerez peut-être aussi