Biomechatronics - Lecture 12. Artificial motor control part 2
-
Upload
tu-delft-opencourseware -
Category
Education
-
view
612 -
download
0
Transcript of Biomechatronics - Lecture 12. Artificial motor control part 2
BiomechatronicsDelft University of Technology
Course 2006-2007(Wb 2432)
Frans van der Helm
Lecture 13Artificial motion control
Contents• Introduction• Definitions from control engineering
– dynamical systems– stability– Phase-margin and gain margin
• Model of human controller– stability analyis– limitations and adaptation
• Supervisory control situations– supervisor over automated control loops
The human as controller
• Marine vessels• Vehicles• Airplanes• ….
Humancontroller vehicle
+
-
Closed-loop systeem
Visualsystem
����������� ���� �����������
Block diagram of a control loop
controller system+ -
sensor
controlorgan
disturbancereal
valuedesiredvalue
H1(ωωωω) H3(ωωωω)+ -
H4(ωωωω)
H2(ωωωω)
N(ωωωω)Y(ωωωω)U(ωωωω)
Block diagram of a control loop
H1(ωωωω) H3(ωωωω)+ -
H4(ωωωω)
H2(ωωωω)
N(ωωωω)Y(ωωωω)U(ωωωω)
YH H HH H H H
UH H H H
N=+
++
1 2 3
1 2 3 4 1 2 3 411
1
Examples in biomechatronics
humancontroller
Limb +prosthesis+ -
tactileinterface
muscles
disturbance
jointangle
setpointforce
skinstimulation
Examples in biomechatronics
humancontroller
Handprosthesis+ -
visualsystem
electro-motor
disturbancehand
openingsetpoint
EMG force
Stability of a control loop� �� ������� ��� ���� ���������
�������������� ���!
� ��"ω#�$������%�
������"ω#���� �&� �����"ω#����� $�
� '� $�(� ����
)��������"ω# ��� ���� �����"ω# *�( �+,�
� - �(� �����
)�� �����"ω# ��(�+,� �����������"ω# .��
-1.5 -1 -0.5 0 0.5 1 1.5-2.5
-2
-1.5
-1
-0.5
0
0.5
Re (Hr)
Im (H
r)
-1
Gain-margin
Gain = 1
Phase-margin
'� $�(� ����/�� �(� ���
0�1�$������
Cross-overfrequency
ωωωω = ωωωωc
frequency (rad)
'� $�(� ���
10^0 10^1 10^2 10^3 10^-2
10^-1
10^0 = 1
10^1
10^2
10^3
2���(����������
gain
10^-1 10^0 10^1 10^2 -315
-270
-225
-180
-135
-90
-45
0
2���(���� �����
frequency (rad)
phas
e (d
egre
es)
'� $�(� ���
ωωωω�Cross-over frequency
10^0 10^1 10^2 10^3 10^-2
10^-1
10^0 = 1
10^1
10^2
10^3
frequency (rad)
gain
10^-1 10^0 10^1 10^2 -315
-270
-225
-180
-135
-90
-45
0
frequency (rad)
phas
e (d
egre
es)
- �(� ���
2���(���������� 2���(���� �����
- �(� ���
Input- & output signals� )���������"�#���$�"ω�#� 3��������"�#�����$��"ω��4�ϕ#� - ��������� $��ϕϕϕϕ
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5-1.5
-1
-0.5
0
0.5
1
1.5
�
ϕϕϕϕ
Stability of a control loop
H1(ωωωω) H3(ωωωω)+ -
H4(ωωωω)
H2(ωωωω)
N(ωωωω)Y(ωωωω)U(ωωωω) E(ωωωω)
0 1 2 3 4 5-3
-2
-1
0
1
2
3
time (sec)
ampl
itude
'� $�( ���� ��(�+,�
� - ���� .��
� $� ��
� - ���� ���
� ���5���������$� ���
� - ���� *��
� ��$� ��
Stability of a control loop
H1(ωωωω) H3(ωωωω)+ -
H4(ωωωω)
H2(ωωωω)
N(ωωωω)Y(ωωωω)U(ωωωω) E(ωωωω)
Phase shift = -180 degreesgain = 0.75
Positive gain margin
0 1 2 3 4 5-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
Model human controller
human plant+ -
U(ωωωω) Y(ωωωω)E(ωωωω)Control
interface
sensor
human system+
-
U(ωωωω) Y(ωωωω)E(ωωωω)
display
Model human controller 1
human system+
-
U(ωωωω) Y(ωωωω)E(ωωωω)
Hm (ωωωω) Hs(ωωωω)+
-
U(ωωωω) Y(ωωωω)E(ωωωω)N(ωωωω)
6�7����
� ��� ��$��8$������������� 5����8���
� ������������������$�$���$
� ��� �����$����$�� ���$��5�(��� 5����
��"ωωωω# ����"ωωωω#��$"ωωωω# ���9:ωωωω
3��� �$��5����� 5�������"ωωωω# ���9:ωωωω
10^0 10^1 10^2 10^3 10^-2
10^-1
10^1
10^2
10^3
2���(����������
frequency (rad)
gain
10^-1 10^0 10^1 10^2
2���(���� �����
frequency (rad)
phas
e (d
egre
es)
-90
Slope -110^0
3��� �$��5����� 5�������"ωωωω# ���9:ωωωω
-1.5 -1 -0.5 0 0.5 1 1.5-2.5
-2
-1.5
-1
-0.5
0
0.5
Re (Hr)
Im (H
r)
-1 0�1�$������
�8 �$$� ��
10^0 10^1 10^2 10^3 10^-2
10^-1
10^1
10^2
10^3
gain
phas
e (d
egre
es)
;� �$�������(������
frequency(rad)10^-1 10^0 10^1 10^2
frequency (rad)
��"ωωωω#��9:ωωωω+
-
U(ωωωω) Y(ωωωω)E(ωωωω)H
YU
j
jj
uy( )( )( )
ωωω
ω
ωω
= =+
=+
1
11
11
Low frequencies: transfer =1Input is followed accurately
High frequencies: transfer almost 0Input is followed badly
0
-90
1
;� �$�������(�����
10^0 10^1 10^2 10^3 10^-2
10^-1
10^1
10^2
10^3
frequency (rad)
gain
10^-1 10^0 10^1 10^2
frequency (rad)
phas
e (d
egre
es)
��"ωωωω#��9:ωωωω+
-
U(ωωωω) Y(ωωωω)E(ωωωω)H
EU
j
jj
ue( )( )( )
ωωω
ω
ωω
= =+
=+
1
11 1
1
0
-90
Low frequencies: error almost 0Input is followed accurately
High frequencies: error = inputInput is followed badly
Model human controller 2
6�7����
� ��� �����$���� ����%� ������� �$��5�(
� ��� 5�������"ωωωω# ����"ωωωω#��$"ωωωω# ���9:ωωωω� 2���$������������<
≈≈≈≈
human system+
-
U(ωωωω) Y(ωωωω)E(ωωωω)
Hm (ωωωω) Hs(ωωωω)+
-
U(ωωωω) Y(ωωωω)E(ωωωω)N(ωωωω)
Limitations & adaptability human controller
� =�� ���$�
> ��� ������$������$$������
> )���� �������$�� ��$�$���� ���8������������� �
> ��� ��� $ ����� ���� ��
> ��� ��� ���$�� ���$������� $�� ��
����� ������$�� ��� ����� ���
� �� �� ����
> ��� ��� �� � ���8� ���$�$������� ��$�
> ��� �����$������5�������� ��������$�� ����
"�������5��$�$���$��#
Cross-over model
Hm (ωωωω) Hs(ωωωω)+
-
U(ωωωω) Y(ωωωω)E(ωωωω)N(ωωωω)
vjmm e
jjj
KH ωτ
ωτωτωτω −
+++= .
11
.11
)(32
1
� �� ��� �� ���$
?����8����1�����$�"ωωωω .� �� �9$�≈ ,�@���#�
ejmm e
jj
KH ωτ
ωτωτω −
++= .
11
)(2
1
Cross-over model
For frequencies around ωωωω = ωωωωc human tries to:
• Loop gain is then equal to 1• Loop gain function approximates integrator• ωωωωc roughly between 2 & 6 rad/s (≈ 0.3 & 1 Hz)• τe equal to 200 till 300 ms
ejcmm e
jHH ωτ
ωωωω −=)().( s
Hm (ωωωω) Hs(ωωωω)+
-
U(ωωωω) Y(ωωωω)E(ωωωω)N(ωωωω)
10^0 10^1 10^2 10^3 10^-2
10^-1
1
10^1
10^2
10^3
2���(����������
frequency (rad)
gain
10^-1 10^0 10^1 10^2
-540
-450
-360
-270
-180
-90
2���(���� �����
frequency (rad)
phas
e (d
egre
es)
Hr(ω)= Hm(ω).Hs(ω)= ωc / j(ω) . e-jωτe
-630
0
ωωωωc = 3 rad/s
'� $�(� ���
Hr(ω)= Hm(ω).Hs(ω)= ωc / j(ω) . e-jωτe
-1.5 -1 0 0.5 1 1.5-2.5
-2
-1.5
-1
-0.5
0
0.5
Re (Hr)
Im (H
r)
-1
Gain = 1
ωωωω = ωωωωc
Phase-margin
-0.5
��� �����$����
��� ��$� ������
Hm (ωωωω) Hs(ωωωω)+
-
U(ωωωω) Y(ωωωω)E(ωωωω)N(ωωωω)
ejmm e
jj
KH ωτ
ωτωτω −
++= .
11
)(2
1
ejcmm e
jHH ωτ
ωωωω −=)().( s
Examples of adaptation
•
•
•
•
H Hj
e Hj H
em sc j
mc
s
j( ). ( ) . ( ) .( )
.ω ωωω
ωωω ω
ωτ ωτ= → =− −1
H Kjj
em mj e( ) . .ω
ωτωτ
ωτ=++
−11
1
2
HKj
Kjj j
jK K
ss
mc
s
c
s( ) . .ω
ωωτωτ
ωω
ωτ τ
ω= →
++
= → = =11
1
22 1 m0, 0, K =
H K Kjj j K K
s s mc
s
c
s( ) . .
.ω
ωτωτ
ωω
τ τω τ
= →++
= → =11
11
22 1 m
2>> 1, 0, K =
( )( )
HK
jK
jj j
jK K
ss
mc
s
c
s( ) . .
.ω
ω
ωτωτ
ωω
ωτ τ
ωτ
= →++
= → =2
1
2
211
2 1 m1
0, >> 1, K =
( )( )
HK
jK
jj j
jK
ss
mc
s( ) . .ω
ω
ωτωτ
ωω
ω= →
++
= →3
1
2
311
not controllable
10^0 10^1 10^2 10^3 10^-2
10^-1
1
10^1
10^2
10^3
gain
10^-1 10^0 10^1 10^2 -1400
-1200
-1000
-800
-600
-400
-200
0
frequency (rad)
phas
e (d
egre
es)
;� �$�������(������
e + j
e = e
j + 1
e j =
UY
)(He
e
e
e
j-c
j-c
j-c
j-c
uy τω
τω
τω
τω
ωωω
ωω
ωω
ωω=ω
)()(
ωωωωc = 3 rad/s
10^0 10^1 10^2 10^3 10^-2
10^-1
1
10^1
10^2
10^3
10^-1 10^0 10^1 10^2 -20
0
20
40
60
80
100
frequency (rad)
phas
e (d
egre
es)
;� �$�������(�����
e + j
j =
e j
+ 1
1 =
)U()E(
= )(He
e
j-cj-c
ue τωτω ωω
ω
ωωω
ωω
ωωωωc = 3 rad/s
gain
10^-1 10^0 10^1 10^2 10^3 10^-2
10^-1
10^0
10^1
10^2
10^3
frequency (rad)
gain
10^-1 10^0 10^1 10^2 -40
-20
0
20
40
60
80
100
frequency (rad)
phas
e (d
egre
es)
ωωωωc = 1 rad/sωωωωc = 3 rad/sωωωωc = 5 rad/s
ωωωω� /�ττττ� ������������ ��/�$�� ���
A������� ��8���������
/�������$�$�������ωωωω� /�ττττ�
Higher bandwidth� More effort � higher ωωωωc and lower ττττe
Higher order system� More difficult � lower ωωωωc and higher ττττe
Demonstration
Conclusions
• Humans can adapt to the system to be controlled:– Km, τ1 and τ2: Adaptation parameters– ωc: Determines bandwidth of controlled system
• Still some limitations apply: – Only zero, first order and second order systems can be
controlled!!– Human controller properties known for systems with visual
and vestibulary feedback, not for tactile feedback, etc.
• Design of biomechatronic systems:– Be aware that humans must be able to control the system
Supervisory control situations
• System with automated control loops– FES system with artificial sensors, controller and stimulator– Above-knee prosthesis with automatic knee locking
• Humans should be in control of whole system– Monitoring:
• Feedback of result• Feedback of performance controller
– Actions:• Setpoint generation for controller• Overruling of controller
Manual control
human plant+ -
U(ωωωω) Y(ωωωω)E(ωωωω)Control
interface
sensor
display
Supervisory control
Automaticcontrol
loop
goal display
performance
Control actions
overrule
human
Setpoint
sensor
plantController
Actuator
Model Rasmussen
Skill-based
Rule-based
Knowledge-based
Model Rasmussen• Knowledge-based behaviour:
– Difficult tasks– performance based on reasoning– Requires much attention: High mental loading– Examples: control room, FES walking?
• Rule-based behaviour:– Tasks with intermediate difficulty– Performance based on applying standard rules– Examples: Control of helping robots
• Skill-based behaviour:– Easy tasks, incorporated in normal behaviour– Performance based on well-trained skills– Requires little mental effort– Examples: Body-powered prosthesis, Bladder control,
Cochlear implants
Model Rasmussen
• Decrease mental load: From knowledge-based behaviour to skill-based behaviour
• Humans generate internal model of system• ‘Ecological’ interface displays system output
resembling internal model• ‘Ecological’ interfaces allow for an intuitive control