Open Loop vs.

Closed Loop
Objective: to precisely locate the head over the disk Options: 1) Open-loop control (without feedback) 2) Closed-loop control (with feedback)
ME 475 Session 2: Principle of Feedback Control Xinyan Deng Purdue University Slide 1

Open-Loop Control
Desired Track Position Controller Motor Voltage Motor Actual Head Position

ME 475 Session 2: Principle of Feedback Control

Xinyan Deng Purdue University

Slide 2

Open-Loop Control (with Disturbance)

Controller Voltage Disturbance Motor Voltage Motor

Desired Track Position

+ +

Actual Head Position

ME 475 Session 2: Principle of Feedback Control

Xinyan Deng Purdue University

Slide 3

Introduce Feedback
R(s)
Controller

D(s) + +

Plant

Y(s)

+ + N(s) R(s): reference input signal for desired action D(s): disturbance signal acting on plant N(s): noise signal corrupting the measurement
ME 475 Session 2: Principle of Feedback Control Xinyan Deng Purdue University Slide 4

Closed-Loop Feedback Control

R(s)
Controller

D(s) + +

Plant

Y(s)

+ + N(s)

CG G CG Y (s) = R (s) + D (s) N(s) 1 + CG 1 + CG 1 + CG

ME 475 Session 2: Principle of Feedback Control Slide 5

Xinyan Deng Purdue University

Closed-Loop Feedback Control

Every transfer function between any one input and one output has the same denominator but different numerators. The controller C affects inputs R and N in a similar way, but affects input D differently.

ME 475 Session 2: Principle of Feedback Control

Xinyan Deng Purdue University

Slide 6

Closed-Loop Control Requirements

YR R , YD YN

CG YR ( s ) = R (s) 1 + CG G D (s) 1 + CG CG N(s) 1 + CG

0, YD ( s ) =

0, YN ( s ) =

ME 475 Session 2: Principle of Feedback Control

Xinyan Deng Purdue University

Slide 7

Feedback Control
The use of feedback provides a mechanism to compensate for disturbances. High-gain feedback essentially approximates plant inversion (the essence of control). High-gain proportional control provides good reference tracking and disturbance rejection, but aggravates noise rejection.

ME 475 Session 2: Principle of Feedback Control

Xinyan Deng Purdue University

Slide 8

Combine Feedback with Feedforward

GF R(s) + C D(s)
Plant

+ ++ +

Y(s)

+ + N(s)

ME 475 Session 2: Principle of Feedback Control

Xinyan Deng Purdue University

Slide 9

Disturbance Feedforward
D(s) GD R(s)
Controller Plant

+ +

+ +

Y(s)

+ + N(s)

ME 475 Session 2: Principle of Feedback Control

Xinyan Deng Purdue University

Slide 10

Control System Design Example

Friction Force Cutting Force

Desired speed

R (s) (Reference Input)

What is the voltage u(t) that should be applied to motor so that output speed y(t) follows r(t) irrespective of various disturbance forces ? Controller

Voltage input of Motor

U (s)

Industrial Drive System G(s) Plant

Output Speed

Y(s)

ME 475 Session 2: Principle of Feedback Control

Xinyan Deng Purdue University

Slide 11

Open-Loop
The control input u(t) (or U(s)) is synthesized based on the a priori knowledge of the physical system (plant) and the reference input r(t) (or R(s)). The control system does not measure the output, and there is no comparison of the actual output to the reference input when generating control input.
Reference Input (Command) R(s) D(s) Disturbance Input System Output

C(s)
Controller

G(s)
U(s) Control Input Plant or System

Y(s)

Q: Ideally, if we want Y(s) to follow R(s) (i.e. want Y(s) = R(s)), how would you design the controller C(s) for the above open-loop control system?

Want:

System inversion!

Q: Can we attenuate the effect of disturbance D(s) on system output Y(s)? Q: Can we attenuate the effect of variations of plant transfer function G(s) on system output Y(s)?
ME 475 Session 2: Principle of Feedback Control Xinyan Deng Purdue University Slide 12

Open-Loop Control Example

Speed Control of a DC Motor without Sensors
1 C (s) = Km

TL(s)

Coulomb Friction Torque

KL
U(s)
Motor Torque

K m 147, m 0.1
Net Torque

R(s)

Km ms +1
Motor Model

Y(s)

Open-Loop Controller

Ideal Open-Loop Controller

s +1 1 (s) = m Cideal = G ( s) Km
U (s) = Cideal ( s) R( s) =

An Implementable Open-Loop Controller

C (s) = 1 Km U (s) =

(neglect electrical dynamics)

m 1 sR( s) + R( s) Km Km 1 (t ) + In time-domain: = u (t ) m r r (t ) Km Km

Implementing the ideal open-loop controller needs the derivative of the reference input, which can be done with microprocessor-based control implementation with known reference trajectory

1 1 R ( s ) u (t ) = r (t ) Km Km

Output Speed with Open-Loop Controller

Km R( s ) + K m ( m s + 1)

Y ( s) =

K L Km s + 1 TL ( s ) m
Slide 13

ME 475 Session 2: Principle of Feedback Control

Xinyan Deng Purdue University

Open-Loop Control Example

Steady-State Output Speed for Constant Desired Speed Reference Inputs (no friction) 1 Time constant (no controller) m 0 Y ( s) = TL ( s ) = R( s) ( m s + 1) Time constant (with controller) m For constant desired speed R(s)=r/s: No Change 1
yss lim sY ( s ) lim = =
s 0 s 0

ms +1

sR ( s ) r =

Steady-State Output Speed for Constant Desired Speed Reference Inputs (with a constant friction L)
= yss = lim = sY ( s )
s 0

Km KL 1 lim sR ( s ) sTL ( s ) s 0 1 1 s s + + ( ) m m

There is no attenuation on the effect of friction ! Q: In reality, the friction on a motor may change quite significantly. Will the customer be happy with such an openloop controller ?
ME 475 Session 2: Principle of Feedback Control Xinyan Deng Purdue University Slide 14

r K m K L L

Closed-Loop Control Example

Speed Control of a DC Motor with Speed Sensors
TC(s) Feedforward Controller
Coulomb Friction Torque

KL

+ U(s)

R(s)

Feedback Controller Closed-loop Controller

Km ms +1
Motor Model
(neglect electrical dynamics)

Y(s)

A Simple Closed-Loop Controller

Feedforward Controller Feedback Controller
ME 475 Session 2: Principle of Feedback Control Xinyan Deng Purdue University Slide 15

Closed-Loop Control Example

Output Speed with Closed-loop Controller
Y (s) KL K m 1 + Km Kc 1 TL ( s ) R ( s ) + m m s +1 s + 1 1 + Km Kc 1 + Km Kc
GYR ( s ) GYT ( s )

Closed-loop time constant

1 + Km Kc

Steady-State (SS) Output Speed for Constant Desired Speed Reference Inputs ofr (no friction) yss G = r No SS speed error ! TL ( s ) = 0 = YR ( 0 ) r Steady-State Output Speed for Constant Desired Speed Reference Inputs r (with a constant friction L ) KL K m L yss = GYR ( 0 ) r + GYT ( 0 ) L = r
1 + Km Kc K K ess = r y ss = L m L 1 + Km Kc
ME 475 Session 2: Principle of Feedback Control Xinyan Deng Purdue University

decrease with increasing Kc

Slide 16

Closed-Loop Control Example

Steady-State Output Speed Error

K K ess = r y ss = L m L 1 + Km Kc
Q: When can we consistently have the desired steady-state speed regardless certain amount of Coulomb friction that may exist ?

Q: How is response speed of the closed-loop system compared with the response speed of original open-loop system ?

ME 475 Session 2: Principle of Feedback Control

Xinyan Deng Purdue University

Slide 17

Closed-Loop Experimental Results

1.5 x 10
4

Kc=0.01 Kc=0.06

Velocity

0.5

Kc=0.12 reference

-0.5

-1

-1.5

0.5

1.5

time (sec)

2.5

3.5

Closed Loop Step Response (with Feedforward)

ME 475 Session 2: Principle of Feedback Control Xinyan Deng Purdue University Slide 18

Closed-Loop without Feedforward

TL(s) Coloumb Friction
Torque

KL
+ R(s)

Kc
Feedback Controller Closed-loop Controller

U(s)

Km ms +1
Motor Model
(neglect electrical dynamics)

Y(s)

Steady-State Output Speed Error for Constant Desired Speed Reference Inputs (with a constant friction) without Feedforward Control K m = = Y U ( s ) K LTL ( s ) Gp ( s ) K c R ( s ) K cY ( s ) K LTL ( s ) (s) s +1 m
Gp ( s)

(1 + K G ( s ) ) Y ( s ) = G ( s ) K R ( s ) K T ( s )
c p p c L L
ME 475 Session 2: Principle of Feedback Control Xinyan Deng Purdue University Slide 19

Feedforward vs Feedback Control

K LG p ( s ) G ( s ) Kc Y (s) = p R (s) + TL ( s ) 1 + K cG p ( s ) 1 + K c G p ( s )

( s) R( s) Y ( s ) E = 1 1 + Km Kc

with Feedback Action Only

R (s) + Km KL 1 + Km Kc

Km Km Kc KL s +1 ms +1 =m R (s) TL ( s ) Km Km 1 + Kc 1 + Kc ms +1 ms +1

1 + Km Kc

s +1

1 + Km Kc

s +1

TL ( s )

Q: How is the performance of the closed-loop control with feedback control action only compared to that of the closed-loop control with both feedforward and feedback actions ?
Attenuation ability of the system to disturbances such as the Coulomb friction: Response speed: Steady-state output speed error for unit constant desired speed without 1 considering friction effect:
1 + Km Kc 0
ME 475 Session 2: Principle of Feedback Control Xinyan Deng Purdue University

when Kc is very large.

Slide 20

Closed-Loop Experimental Results

1.5 x 10
4

Kc=0.01
1

Kc=0.06 Kc=0.12

0.5

velocity

reference

-0.5

-1

-1.5

0.5

1.5

2 time (sec)

2.5

3.5

Closed Loop Step (w.o feedforward)

ME 475 Session 2: Principle of Feedback Control Xinyan Deng Purdue University Slide 21

Why Feedback ?
The previous speed control examples illustrate that, by using feedback, we can change the closed-loop systems dynamic behavior, as the Closed-Loop Transfer Function (CLTF) is different from the original systems (open-loop) transfer function. As such, through feedback, we have the ability to achieve the following objectives:

Stabilize Unstable Systems

For example, unstable plants such as inverted pendulum and the position control of DC motor can be stabilized using feedback.

Improve System Performance to meet stringent performance requirements

Steady State Performance -- for example, reduce steady state error due to disturbances ... Transient Performance -- for example, reduce rise time and settling time to speed up system response ,

Reduce (Attenuate) the effect of modeling uncertainty (error) and various disturbances through High-Gain Feedback

ME 475 Session 2: Principle of Feedback Control

Xinyan Deng Purdue University

Slide 22

Why Feedforward ?
The previous examples also illustrate that, feedforward control action makes the control input close to the desired control input that is needed to accomplish the task. As such, only small amount of control correction needs to be provided by feedback control action, which results in smaller tracking error (keep in mind that feedback control needs tracking error to generate the control action): Feedforward is very important for applications having stringent performance requirements such as control of precision electromechanical devices Usefulness of feedforward heavily depends on the accuracy of models used for physical plants, which normally have quite large variations of system parameters. As such, learning mechanisms such as on-line parameter adaptation in adaptive control are needed to build accurate model based on various information obtained by sensors including stored past information
ME 475 Session 2: Principle of Feedback Control Xinyan Deng Purdue University Slide 23

General Control Design Principles

Control design is nothing but an Inversion Process. The inversion can be achieved by two key mechanisms: (High-Gain) Feedback and (Model Compensation) Feedforward
R( s ) E (s)
K
Controller

U (s)

G (s)
Plant

Y (s)

U (s) = R( s)

K 1 + KG

large K

G 1

High-gain feedback gives approximate inversion in the presence of modeling errors and disturbances, which is the essence of control. However, in practice, the choice of feedback gain is part of a complex web of design trade-offs; high-gain leads to high sensitivity to measurement noises and makes the stability of closed-loop system sensitive to control input saturation and neglected high-frequency dynamics. Understanding and balancing these trade-offs is the essence of feedback control system design.

On-line learning is key to have a good model compensation or feedforward design.

ME 475 Session 2: Principle of Feedback Control Xinyan Deng Purdue University Slide 24

Use nonlinear feedback instead of linear feedback to achieve a better trade-off !

General Structure of Controller

In general, a controller is nothing but a strategy to determine a control action based on all available information; information not only comes from the measured output but also from the measured internal state variables, measured disturbance, reference trajectory, and plant model structure. It can have any form and is illustrated below:
(t ) d
d (t )

Sensor

Disturbances

Measured Disturbance Reference Trajectory r(t) Other Available Information

Information Synthesis

u (t )

Plant Dynamics

y (t )

Control Strategy
Measured State Variables Measured Output
ME 475 Session 2: Principle of Feedback Control Xinyan Deng Purdue University Slide 25

Sensors