Mit6 003f11 Lec10

6.
003: Signals and Systems

Feedback and Control
October 13, 2011

1
Courtesy of Jason Dorfman MIT / CSAIL. Used with permission.
Example: Perching
Can we make a xed-wing UAV land on a perch like a bird?
The Perching Problem
Courtesy of Leon van Dommelen and Szu-Chuan Wang. Used with permission.
Photo of a cardinal landing on a branch removed due to copyright restrictions.
Photo from Naval Historical Center Aircraft Data Series.
Dimensionless Analysis
Bird or plane...
with mass m, wing area S, operating in a uid with density which requires a distance x to slow from V0 to Vf
Distance-averaged drag coefcient, CD:
2m ln CD = Sx
V0 Vf
Photo of the Boeing 747-400ER removed due to copyright restrictions.
A few (very preliminary) reference points:

Vehicle Boeing 747 X-31 Average CD 0.16 0.3 0.25 10
Cornell Perching Plane Common Pigeon
Photos of Cornell perching plane and landing pigeon removed due to copyright restrictions.
U.S. Navy photo by James Darcy.
Image removed due to copyright restrictions. Please see SlowMoHighSpeed. "Photron SA2 Camera - Eagle Owl in Flight." October 27, 2008. YouTube. Accessed September 25, 2012. http://www.youtube.com/watch?v=LA6XSrM0V_0
Experiment Design
Glider (no propellor) Flat plate wings Dihedral (passive roll stability) Offboard sensing and control
System Identication
Nonlinear rigid-body vehicle model Linear actuator model (+ saturations, delay) Real ight data (no wind tunnel)
System Identication
Lift Coefcient
1.5 Glider Data Flat Plate Theory 1 2.5 0.5 2 0
Cd Cl
Drag Coefcient
3.5 3 Glider Data Flat Plate Theory
1.5 1
!0.5 0.5 !1 0 !1.5 !20 !0.5 !20
20
40
60 80 Angle of Attack
100
120
140
160
20
40
60 80 Angle of Attack
100
120
140
160
A Dynamic Model
Fw
lw (x, z) m, I
Fe le
l g
Planar dynamics Aerodynamics t from data State:
] x = [x, y, , , x, y,
Actuator: u =
10
Perching Results
Enters motion capture @ 6m/s Perch in < 3.5m away Entire trajectory < 1s
Requires separation!
11
Flow visualization
Courtesy of Jason Dorfman MIT / CSAIL. Used with permission.
12
Dimensionless Analysis
Vehicle Boeing 747 X-31 Cornell Perching Plane Common Pigeon Our glider Cobra maneuver (Mig)
Average CD 0.16 0.3 0.25 10 1.1 0.9
13
Feedback is essential...
to compensate for initial condition errors, disturbances, and imperfect model agile airplanes are open loop unstable
open loop
feedback
14
Todays goal
Use systems theory to gain insight into how to control a system.
15
Example: wallFinder System

Approach a wall, stopping a desired distance di in front of it.
di = desiredFront do = distanceFront do K = 0.5 do K = 2 do K = 1 do K = 8
What causes these dierent types of responses?

16
Structure of a Control Problem

(Simple) Control systems have three parts.
E controller S
C plant
sensor
The plant is the system to be controlled. The sensor measures the output of the plant. The controller species a command C to the plant based on the dierence between the input X and sensor output S .
17
Analysis of wallFinder System

Cast wallFinder problem into control structure.
X + E controller S sensor plant C Y
di = desiredFront do = distanceFront
proportional controller: v [n] = Ke[n] = K di [n] ds [n] locomotion: do [n] = do [n 1] T v [n 1] sensor with no delay: ds [n] = do [n]
18
Analysis of wallFinder System: Block Diagram

Visualize as block diagram.
proportional controller: v [n] = Ke[n] = K di [n] ds [n] locomotion: do [n] = do [n 1] T v [n 1] sensor with no delay: ds [n] = do [n]
Di
Do
19
Analysis of wallFinder System: System Function

Solve.
di = desiredFront do = distanceFront Di +
KT R KT R KT R 1R = = KT R 1 R KT R 1 (1 + KT )R 1+ 1R
Do
Do = Di
20
Analysis of wallFinder System: Poles

The system function contains a single pole at z = 1 + KT . KT R Do = Di 1 (1 + KT )R Unit-sample response for KT = 0.2:
h[n] 0.2 n 0
Unit-step response s[n] for KT = 0.2:
n 0
What determines the speed of the response? Could it be faster?
21
Check Yourself
Find KT for fastest convergence of unit-sample response. KT R Do = Di 1 (1 + KT )R 1. 2. 3. 4. 5. 0. KT = 2 KT = 1 KT = 0 KT = 1 KT = 2 none of the above
22
Check Yourself
Find KT for fastest convergence of unit-sample response. Do KT R = 1 (1 + KT )R Di If KT = 1 then the pole is at z = 0. KT R Do = =R 1 (1 + KT )R Di Unit-sample response has a single non-zero output sample, at n = 1.
23
Analysis of wallFinder System: Poles

The poles of the system function provide insight for choosing K . Do KT R (1 po )R ; = = Di 1 (1 + KT )R 1 po R p0 = 1 + KT
Im z
Im z
Im z
Re z
Re z
Re z
0 < p0 < 1 1 < KT < 0 monotonic converging
1 < p0 < 0 2 < KT < 1 alternating converging
p0 < 1 KT < 2 alternating diverging
24
Check Yourself
Find KT for fastest convergence of unit-sample response. KT R Do = Di 1 (1 + KT )R 1. 2. 3. 4. 5. 0. KT = 2 KT = 1 KT = 0 KT = 1 KT = 2 none of the above
25
Analysis of wallFinder System

The optimum gain K moves robot to desired position in one step.
di = desiredFront=1 m do = distanceFront=2 m
KT = 1 1 1 K= = = 10 T 1/10 v [n] = K di [n] do [n] = 10 1 2 = 10 m/s exactly the right speed to get there in one step!
26
Analyzing wallFinder: Space-Time Diagram

di = desiredFront do = distanceFront position v = 10
time
27

di = desiredFront do = distanceFront position v = 10 v=0
time
28

di = desiredFront do = distanceFront position v = 10 v=0 v=0 v=0 v=0 v=0 v=0 time
29
Analysis of wallFinder System: Adding Sensor Delay

Adding delay tends to destabilize control systems.
proportional controller: v [n] = Ke[n] = K di [n] ds [n] locomotion: do [n] = do [n 1] T v [n 1] sensor with delay: ds [n] = do [n 1]
30

di = desiredFront do = distanceFront position v = 10
time
31

di = desiredFront do = distanceFront position v = 10 v=0
time
32

di = desiredFront do = distanceFront position v = 10 v=0 v = 10
time
33

di = desiredFront do = distanceFront position v v v v = 10 =0 = 10 = 10
time
34

di = desiredFront do = distanceFront position v v v v v time

35
= 10 =0 = 10 = 10 =0
Analysis of wallFinder System: Block Diagram

Incorporating sensor delay in block diagram.
proportional controller: v [n] = Ke[n] = K di [n] ds [n] locomotion: do [n] = do [n 1] T v [n 1]
sensor with sensor with no delay: ds [n] = do [n 1]

Di + R K V T + R Do
36
Check Yourself
Find the system function H =
Do . Di
Di
T R
Do
1. 3.
KT R 1R KT R KT R 1R
2. 4.
KT R 1 + R KT R2 KT R 1 R KT R2
5. none of the above
37
Check Yourself
Find the system function H = Do . Di
Di
T R
Do
Replace accumulator with equivalent block diagram.
Di
R 1R
Do
R KT R Do KT R 1R = = 2 Di 1 R KT R2 KT R 1+ 1R
38
Check Yourself
Find the system function H =
Do . Di
Di
T R
Do
1. 3.
KT R 1R KT R KT R 1R
2. 4.
KT R 1 + R KT R2 KT R 1 R KT R2
5. none of the above
39
Analyzing wallFinder: Poles

Substitute R 1 in the system functional to nd the poles. z
1 KT z KT z KT R Do = = = 2 2 1 1 Di 1 R KT R z z KT 1 z KT 2 z
The poles are a then the roots of the denominator. 1 1 2 z= + KT 2 2
40
Feedback and Control: Poles

If KT is small, the poles are at z KT and z 1 + KT . z=
1 2
1 2 + KT 2
1 2
1 2
+ KT
Im z
= 1 + KT, KT
z-plane
KT 0
Re z
Pole near 0 generates fast response. Pole near 1 generates slow response. Slow mode (pole near 1) dominates the response.
41

As KT becomes more negative, the poles move toward each other 1 and collide at z = 1 2 when KT = 4 . 1 2 + KT = 1 1 2 1 = 1, 1 z=1 2 2 4 2 2 2 2
KT = 1 4 Im z z-plane
2 1 Re z
Persistent responses decay. The system is stable.
42

If KT < 1/4, the poles are complex. z=1 2
1 2 2
+ KT = 1 2 j
KT = 1
2 KT 1 2
Im z z-plane
Re z
Complex poles oscillations.
43
Same oscillation we saw earlier!

di = desiredFront do = distanceFront position v v v v v time

44
= 10 =0 = 10 = 10 =0
Check Yourself
KT = 1
Im z
z-plane
Re z
What is the period of the oscillation? 1. 1 4. 4 2. 2 5. 6 3. 3 0. none of above
45
Check Yourself
KT = 1 Im z
z-plane
Re z
1 3 p0 = j = ej/3 2 2
jn/3 pn 0 =e j 0/3 j/3 j 6/3 e , ej 2/3 , ej 3/3 , ej 4/3 , ej 5/3 , e e -v ", e e -v " 1 ej 2 =1
46
Check Yourself
KT = 1
Im z
z-plane
Re z
What is the period of the oscillation? 1. 1 4. 4 2. 2 5. 6 3. 3 0. none of above
47

The closed loop poles depend on the gain.
Im z z-plane
Re z
1 1 If KT : 0 : then z1 , z2 : 0, 1 1 2 , 2 2 j
48
Check Yourself
Find KT for fastest response.
Im z
z-plane closed-loop poles 1 2 1 2 + KT 2
Re z
1. 0 4. 1
2. 1 4 5.
3. 1 2 0. none of above
49
Check Yourself
z= 1 2 a 1 2 + KT 2
1 The dominant pole always has a magnitude that is . 2 1 It is smallest when there is a double pole at z = . 2 1 Therefore, KT = . 4
50
Check Yourself
Find KT for fastest response.
Im z
z-plane closed-loop poles 1 2 1 2 + KT 2
Re z
1. 0 4. 1
2. 1 4 5.
3. 1 2 0. none of above
51
Destabilizing Eect of Delay

Adding delay in the feedback loop makes it more dicult to stabilize. Ideal sensor: ds [n] = do [n] More realistic sensor (with delay): ds [n] = do [n 1]
Im z Im z
Re z
Re z
Fastest response without delay: single pole at z = 0. 1 Fastest response with delay: double pole at z = . much slower! 2
52
Destabilizing Eect of Delay

Adding more delay in the feedback loop is even worse. More realistic sensor (with delay): ds [n] = do [n 1] Even more delay: ds [n] = do [n 2]
Im z Im z
Re z
2 1
Re z
1 Fastest response with delay: double pole at z = . 2 Fastest response with more delay: double pole at z = 0.682. even slower
53
Feedback and Control: Summary

Feedback is an elegant way to design a control system. Stability of a feedback system is determined by its dominant pole. Delays tend to decrease the stability of a feedback system.
54
Photo from Naval Historical Center Aircraft Data Series.
55
Block diagram of the F-14 control system as modeled in Simulink removed due to copyright restrictions. Please see "F-14 Longitudinal Flight Control." The MathWorks, Inc.
56
MIT OpenCourseWare http://ocw.mit.edu
6.003 Signals and Systems

Fall 2011
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.

Mit6 003f11 Lec10

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mit6 003f11 Lec10

Uploaded by

Copyright:

Available Formats

6.

003: Signals and Systems

October 13, 2011

Courtesy of Jason Dorfman MIT / CSAIL. Used with permission.

Can we make a xed-wing UAV land on a perch like a bird?

The Perching Problem

Photo of a cardinal landing on a branch removed due to copyright restrictions.

Photo from Naval Historical Center Aircraft Data Series.

Distance-averaged drag coefcient, CD:

Photo of the Boeing 747-400ER removed due to copyright restrictions.

A few (very preliminary) reference points:

Cornell Perching Plane Common Pigeon

U.S. Navy photo by James Darcy.

!0.5 0.5 !1 0 !1.5 !20 !0.5 !20

Planar dynamics Aerodynamics t from data State:

Courtesy of Jason Dorfman MIT / CSAIL. Used with permission.

Average CD 0.16 0.3 0.25 10 1.1 0.9

Example: wallFinder System

di = desiredFront do = distanceFront do K = 0.5 do K = 2 do K = 1 do K = 8

What causes these dierent types of responses?

Structure of a Control Problem

Analysis of wallFinder System

Analysis of wallFinder System: Block Diagram

Analysis of wallFinder System: System Function

Analysis of wallFinder System: Poles

Find KT for fastest convergence of unit-sample response. KT R Do = Di 1 (1 + KT )R 1. 2. 3. 4. 5. 0. KT = 2 KT = 1 KT = 0 KT = 1 KT = 2 none of the above

Analysis of wallFinder System: Poles

0 < p0 < 1 1 < KT < 0 monotonic converging

1 < p0 < 0 2 < KT < 1 alternating converging

p0 < 1 KT < 2 alternating diverging

Find KT for fastest convergence of unit-sample response. KT R Do = Di 1 (1 + KT )R 1. 2. 3. 4. 5. 0. KT = 2 KT = 1 KT = 0 KT = 1 KT = 2 none of the above

Analysis of wallFinder System

Analyzing wallFinder: Space-Time Diagram

di = desiredFront do = distanceFront position v = 10

Analyzing wallFinder: Space-Time Diagram

di = desiredFront do = distanceFront position v = 10 v=0

Analyzing wallFinder: Space-Time Diagram

Analysis of wallFinder System: Adding Sensor Delay

Analysis of wallFinder System: Adding Sensor Delay

di = desiredFront do = distanceFront position v = 10

Analysis of wallFinder System: Adding Sensor Delay

di = desiredFront do = distanceFront position v = 10 v=0

Analysis of wallFinder System: Adding Sensor Delay

di = desiredFront do = distanceFront position v = 10 v=0 v = 10

Analysis of wallFinder System: Adding Sensor Delay

di = desiredFront do = distanceFront position v v v v = 10 =0 = 10 = 10

Analysis of wallFinder System: Adding Sensor Delay

di = desiredFront do = distanceFront position v v v v v time

Analysis of wallFinder System: Block Diagram

sensor with sensor with no delay: ds [n] = do [n 1]

Find the system function H =

5. none of the above

Replace accumulator with equivalent block diagram.

Find the system function H =

5. none of the above

Analyzing wallFinder: Poles

The poles are a then the roots of the denominator. 1 1 2 z= + KT 2 2

Feedback and Control: Poles

Feedback and Control: Poles

Persistent responses decay. The system is stable.

Feedback and Control: Poles

Complex poles oscillations.

Same oscillation we saw earlier!