November 29, 2005
Auto-Calibration and Control Applied to Electro-Hydraulic Valves
A Ph.D. Thesis Proposal Presented to the Faculty of the
George Woodruff School of Mechanical Engineering at the Georgia Institute of Technology
By
PATRICK OPDENBOSCH
Committee Members:Nader Sadegh (Co-Chair, ME) Wayne Book (Co-Chair, ME)
Chris Paredis (ME) Bonnie Heck (ECE)
Roger Yang (HUSCO Intl.)
November 29, 2005 2
PRESENTATION OUTLINE
INTRODUCTION PROBLEM STATEMENT OBJECTIVES REVIEW OF MOST RELEVANT WORK PROPOSED RESEARCH PRELIMINARY WORK EXPECTED CONTRIBUTIONS CONCLUSION
November 29, 2005 3
INTRODUCTION
CURRENT APPROACH Electronic control Use of solenoid Valves Energy efficient operation New electrohydraulic
valves Conventional hydraulic
spool valves are being replaced by assemblies of 4 independent valves for metering control
Spool Valve
Spool piece
Piston
Low Pressure
High Pressure
Piston motion
Spool motion
November 29, 2005 4
INTRODUCTION
CURRENT APPROACH Electronic control Use of solenoid Valves Energy efficient operation New electrohydraulic
valves Conventional hydraulic
spool valves are being replaced by assemblies of 4 independent valves for metering control
Piston motion
Low Pressure
High Pressure
Valve motion
November 29, 2005 5
INTRODUCTION
ADVANTAGES Independent control More degrees of freedom More efficient operation Simple circuit Ease in maintenance Distributed system No need to customize
Piston motion
High Pressure
Valve motion
Low Pressure
November 29, 2005 6
INTRODUCTION
METERING MODES Standard Extend Standard Retract High Side Regeneration Low Side Regeneration
DISADVANTAGES Nonlinear system Complex control
Piston motion
High Pressure
Valve motion
Low Pressure
November 29, 2005 7
INTRODUCTION
POPPET ADVANTAGES Excellent sealing Less faulting High resistance to
contamination High flow to poppet
displacement ratios Low cost and low
maintenance
Pilot Pin
Main Poppet
Reverse (Nose) Flow
Forward (Side) Flow
ControlChamber
ModulatingSpring
Coil
Armature
ArmatureBias Spring
PressureCompensatingSpring
Coil CapAdjustmentScrew
Input Current
U.S
. P
ate
nts
(6
,32
8,2
75
) &
(6
,74
5,9
92
)
November 29, 2005 8
INTRODUCTION
Electro-Hydraulic Poppet Valve (EHPV) Poppet type valve Pilot driven Solenoid activated Internal pressure
compensation Virtually ‘zero’ leakage Bidirectional Low hysteresis Low gain initial metering PWM current input
Pilot Pin
Main Poppet
Reverse (Nose) Flow
Forward (Side) Flow
ControlChamber
ModulatingSpring
Coil
Armature
ArmatureBias Spring
PressureCompensatingSpring
Coil CapAdjustmentScrew
Input Current
U.S
. P
ate
nts
(6
,32
8,2
75
) &
(6
,74
5,9
92
)
November 29, 2005 9
INTRODUCTION
VALVE CHARACTERIZATION
Flow Conductance Kv
or
PKPPKQQ 2
V21
2
V
Kv
P2 P1
Q 2121V sgn PPPPKQ
November 29, 2005 10
INTRODUCTION
FORWARD MAPPING
REVERSE MAPPING
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.80
10
20
30
40
50
60
70
80
90
100
Pressure Differential [MPa]
Kv
[LP
M/s
qrt(
MP
a)]
EHPV Forward Flow Conductance Coefficient Measurement
1.5044 1.3565 1.2074 1.0584 1.4308 1.2818 1.13260.98395
0 0.2 0.4 0.6 0.8 1 1.2 1.40
20
40
60
80
100
120
Pressure Differential [MPa]
Kv
[LP
M/s
qrt(
MP
a)]
EHPV Reverse Flow Conductance Coefficient Measurement
1.5071.35871.20911.05941.43331.2838 1.1340.9845
Forward Kv at different input currents [A]
Reverse Kv at different input currents [A]
Side to nose
Nose to side
November 29, 2005 11
INTRODUCTION
MOTIVATION Need to control valve’s KV
Currently done by inversion of the steady-state input/output characteristics
Requires individual offline calibration
CHALLENGES Online learning of steady
state and transient characteristics
Online estimation of individual Kv.
ADVANTAGES No individual offline
calibration Design need not be perfect
and ‘sufficiently fast’ Maintenance scheduling
can be implemented from monitoring and detecting the deviations from the normal pattern of behavior.
November 29, 2005 12
PRESENTATION OUTLINE
INTRODUCTION PROBLEM STATEMENT OBJECTIVES REVIEW OF MOST RELEVANT WORK PROPOSED RESEARCH PRELIMINARY WORK EXPECTED CONTRIBUTIONS CONCLUSION
November 29, 2005 13
PROBLEM STATEMENT
PURPOSE Develop a general theoretical framework for auto-calibration
and control of general nonlinear systems. It is intended to explore the feasibility of the online learning of the system’s characteristics while improving its transient and steady state performance without requiring much a priori knowledge of such system.
APPLICATION This framework is applied to a hydraulic system composed of
electro-hydraulic valves in an effort to study the applicability of having a self-calibrated system.
November 29, 2005 15
PRESENTATION OUTLINE
INTRODUCTION PROBLEM STATEMENT OBJECTIVES REVIEW OF MOST RELEVANT WORK PROPOSED RESEARCH PRELIMINARY WORK EXPECTED CONTRIBUTIONS CONCLUSION
November 29, 2005 16
OBJECTIVES
THEORETICAL Development of a general
formulation for control of nonlinear systems with parametric uncertainty and time-varying characteristics
Development of a formulation for auto-calibration of nonlinear systems
Study of learning dynamics online along with fault diagnosis
Improve Kv control of EHPV’s
EXPERIMENTAL Analysis and validation on
the effectiveness of the proposed method
Study of the accuracy of the auto-calibration and possible drift problems
Development of computationally efficient algorithms
Development of a nonlinear observer for state estimation for unmeasurable states
November 29, 2005 17
PRESENTATION OUTLINE
INTRODUCTION PROBLEM STATEMENT OBJECTIVES REVIEW OF MOST RELEVANT WORK PROPOSED RESEARCH PRELIMINARY WORK EXPECTED CONTRIBUTIONS CONCLUSION
November 29, 2005 19
RELEVANT WORK REVIEW
The plant is linearized about a desired trajectory A Nodal Link Perceptron Network (NLPN) is employed in the
feedforward loop and trained with feedback state error The control scheme needs the plant Jacobian and controllability
matrices – obtained offline Approximations of the Jacobian and controllability matrices can be
used without loosing closed loop stability.
Sadegh (1995)
November 29, 2005 20
RELEVANT WORK REVIEW
Nodal Link Perceptron Network (NLPN) Functional approximation is achieved by the scaling of basis
functions The class of basis functions are to be selected as well as their
‘weights’ are to be trained so that the functional approximation error is within prescribed bounds
1
x1
2
N
x2
x3
xn
y1
ym
Wij
ΦWxxy T
1
N
iiiwf
N
iiiwf
1
xx
Sadegh (1998)
November 29, 2005 22
RELEVANT WORK REVIEW
O'hara (1990), Book (1998) Concept of “Inferred Flow
Feedback” Requires a priori
knowledge of the flow characteristics of the valve via offline calibration
Squematic Diagram for Programmable Valve
November 29, 2005 23
RELEVANT WORK REVIEW
Garimella and Yao (2002) Velocity observer based on
cylinder cap and rod side pressures
Adaptive robust techniques Parametric uncertainty for
bulk modulus, load mass, friction, and load force
Nonlinear model based Discontinuous projection
mapping Adaptation is used when
PE conditions are satisfied
November 29, 2005 24
RELEVANT WORK REVIEW
Liu and Yao (2005) Modeling of valve’s flow
mapping Online approach without
removal from overall system
Combination of model based approach, identification, and NN approximation
Comparison among automated modeling, offline calibration, and manufacturer’s calibration
November 29, 2005 27
PRESENTATION OUTLINE
INTRODUCTION PROBLEM STATEMENT OBJECTIVES REVIEW OF MOST RELEVANT WORK PROPOSED RESEARCH PRELIMINARY WORK EXPECTED CONTRIBUTIONS CONCLUSION
November 29, 2005 28
PROPOSED RESEARCH
AUTO-CALIBRATION AND CONTROL
k = 0,1,2… (discrete-time index) 0 ≤ ui ≤ iUMAX, i = {1,2,…,m}
Set of admissible states
Set of admissible inputs
n
k x
kkk
kkkk
vxgy
ωuxfx
,
,,1
m
k um
k ω
p
k yp
k v
,0,: rrn xx
zxuxFzu ,,,:nU
November 29, 2005 29
PROPOSED RESEARCH
AUTO-CALIBRATION AND CONTROL
k = 0,1,2… (discrete-time index) 0 ≤ ui ≤ iUMAX, i = {1,2,…,m}
The control purpose is to learn the input sequence {uk} that forces the states of the system xk to follow a desired state trajectory dxk as k→∞
PROPOSED: Adaptive approach without requiring detailed knowledge about the system’s model
n
k x
kkk
kkkk
vxgy
ωuxfx
,
,,1
m
k um
k ω
p
k yp
k v
November 29, 2005 30
PROPOSED RESEARCH
SQUARE NONLINEAR SYSTEM
ASSUMPTIONS The system is strongly controllable:
The system is strongly observable:
The functions F and H are continuously differentiable
kkk
kkk
uxHy
uxFx
,
,
n
k xn
k un
k y
zuxFuzx ,thatsuchuniquea,, n
2121 ,,, xxuxuxu HHU
November 29, 2005 32
PROPOSED RESEARCH
SQUARE NONLINEAR SYSTEM
CONTROL DESIGN Tracking Error: Error Dynamics:
kkk
kkk
uxHy
uxFx
,
,
n
k xn
k un
k y
k
d
kkk
d
k
k
k
k
k ok
dk
dk
dk
d
uueuuu
Fe
x
Fe
uxux
,,,
k
d
kk xxe
k
d
kk
d
kk
d
k uuQeJe
November 29, 2005 33
PROPOSED RESEARCH
SQUARE NONLINEAR SYSTEM
CONTROL DESIGN Error Dynamics:
Deadbeat Control Law:
kkk
kkk
uxHy
uxFx
,
,
n
k xn
k un
k y
k
d
kk
d
kk
d
k uuQeJe
kk
d
k
d
k
d
k eJQuu 1
November 29, 2005 34
PROPOSED RESEARCH
SQUARE NONLINEAR SYSTEM
CONTROL DESIGN Deadbeat Control Law:
Proposed Control Law:
kkk
kkk
uxHy
uxFx
,
,
n
k xn
k un
k y
kkkkkk eJQuuu ~~~ 1
kk
d
k
d
k
d
k eJQuu 1
k
d
k xu
k
dT
kk xΦWu~~
November 29, 2005 36
PROPOSED RESEARCHNominal inverse
mapping
Inverse Mapping
Correction
Adaptive Proportional Feedback
NLPN PLANT
Jacobian Controllability
Estimation
xk
dxk
uk
kkkkkk uueJQu ~~~ 1
November 29, 2005 37
PROPOSED RESEARCH
ESTIMATION APPROACHES
Modified Broyden
kkk VMB
k
d
kk
d
kk
d
k uuQeJe
kk
kkkkkk VV
VVMBMM
T2
T
1
November 29, 2005 38
PROPOSED RESEARCH
ESTIMATION APPROACHES
Recursive Least Squares
k
d
kk
d
kk
d
k uuQeJe
k
T
kkb vm
kk
T
kk
k
T
kkkkkk
b
mPm
vmmPvv
11
111
kk
T
kk
k
T
kkkk
k
k mPm
PmmPPP
11
111
1
1
11 ekek
November 29, 2005 39
PROPOSED RESEARCH
APPLICATION Kv Observer
TBA PPxx
BLBB
BB
e
ALAA
AA
e
fLBA
AQQQAV
AQQQAV
fFAAm
2
10
2
10
2143
2
4
3
2
1 ,1
43
33
44
44
33
sgn
sgn
sgn
sgn
LL
RRVAA
SSVBB
RRVBB
SSVAA
KQ
PPKQ
PPKQ
PPKQ
PPKQ
kikVi
kikiki
ZhK
tuZfZ
,,
,,1, ,,
For each valve:
FL
x
PA PB
QB-
x
PR
PS
PB PA
QA+
QL QA
QB+
QB
QA-
AA AB
KvB- KvA+
KvB+ KvA-
VA0 VB0
m
Pump
Tank
KvP
KvT
M
November 29, 2005 40
PROPOSED RESEARCH
APPLICATION Health Monitoring
Failures: sensor fault, wear of the mating parts, contamination, break of a component, or component stiction
Assess valve’s behavior with respect to the nominal behavior.
Establish the criteria to declare faulting on the valves by studying the deviations from the nominal pattern.
-200 0 200 400 600 800 1000 1200 1400 16000
1000
2000
3000
4000
5000
6000
7000
8000
isol [mA]
KV
[LP
H/s
qrtM
Pa] ERR2
ERR1
ORIGINAL CURVE
TRUE CURVE
Kv as a Function of Input Current: Deviations from Nominal Patterns
November 29, 2005 41
PROPOSED RESEARCH
THEORETICAL TASKS Work on the convergence
properties of the estimated matrices
Perform analysis about the closed loop stability of the overall system.
Work on a nonlinear observer for the valves’ flow conductances.
EXPERIMENTAL TASKS Hydraulic testbed setup Sensor integration,
calibration, and filtering design
Data acquisition and analysis
Validation of theory Compare the performance
under learning to that of fixed input/output mapping
November 29, 2005 42
PRESENTATION OUTLINE
INTRODUCTION PROBLEM STATEMENT OBJECTIVES REVIEW OF MOST RELEVANT WORK PROPOSED RESEARCH PRELIMINARY WORK EXPECTED CONTRIBUTIONS CONCLUSION
November 29, 2005 43
PRELIMINARY WORK
NONLINEAR 1ST ORDER DISCRETE TIME SYSTEM
Desired State kd x Nominal Input ku
0 0 0.9 0.5 1.8 5.5 2.7 6.8 3.6 8.2 4.5 13.9 5 14.0 6 15.0 8 16.0
0
10
20
30
40
50
60
70
80
0 2 4 6 8
Desired State
Co
ntr
ol I
np
ut
u_nom u_ss
Comparison: implemented and true steady state mapping
Implemented Nominal Mapping
5.05.0
5.00
93.01567.021.0 45.0
1
kk
k
k
kkkk
uu
uu
xxux
November 29, 2005 44
PRELIMINARY WORK
xdkek
uk
FIRST ORDER DISCRETE SYSTEM TRAJ ECTORY CONTROL SIMULATION
Closed Loop
Open Loop
xkcl vs xdk vs xkol
xdk vs. xk(CLOSED LOOP)
uk
1-D T(u)
ubar = gamma(xdk).
1-D T(u)
ubar = gamma(xdk)
Q Q+delta
Zero Rejector
Q
[Q]
[J]
[E]
[Q]
[J]
[E]
[Q]
[Q]
[J]
[J]
uk xk
Nonlinear First OrderDiscrete System.
uk xk
Nonlinear First OrderDiscrete System
xdk
Ek
dukNL vs. LN
1
u
uk
Jk-1
Qk-1
xkLN
Linear Approx System
J
-1
xk
uk
Jk-1
Qk-1
ESTIMATION
(DSG)DISCRETE
SIGNALGENERATOR.
(DSG)DISCRETE
SIGNALGENERATOR
November 29, 2005 45
PRELIMINARY WORK
0 0.5 1 1.5 2 2.5 3 3.50
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Time [sec]
xkClose-loopDesiredOpen-loop
Closed-loop and open-loop performance
November 29, 2005 46
PRELIMINARY WORK
0 0.5 1 1.5 2 2.5 3 3.50
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Time [sec]
Estim
ate
d V
alu
e
JQ
Estimated Jacobian and Controllability
November 29, 2005 48
PRELIMINARY WORK
Single EHPV learning control being investigated at Georgia Tech
Controller employs Neural Network in the feedforward loop with adaptive proportional feedback
Satisfactory results for single EHPV used for pressure control
November 29, 2005 49
PRELIMINARY WORKGeorgia Institute of TechnologyGeorge W. Woodruff School of Mechanical Engineering
Atlanta, GA 30332
EHPV TECHNOLOGY PROJECTDeveloped by: Patrick Opdenbosch
Date: April 6, 2005ROBOT-EHPV CONTROL
[NLPN_RLS/RLS-NOMD APPROACH]
Sampling Rate: 1kHz
1
x
1
ek
1
dP
1
Vsol
1
T
Target ScopeId: 3
Target ScopeId: 2
Target ScopeId: 1
Saturation
1
Qtot1
Qnet
1
Q
1
Pp
Kv k
Vsolk
J
Q
Kv a
Linear Approx System
KvdGENERATOR
1
Kvd1
Kva
1
Kv
1
J
1
Isol
x+noise x
Filter
x+noise x
Filter
Vsol
dP
Kv
Pp
T
isol
x
Qtot
Qnet
EHPV
dP
error
Kv d
J
Q
Vsol
CONTROLLER
Kv k
Vsolk
Vsol
J
Q
CONST. ESTIMATION
November 29, 2005 50
PRELIMINARY WORK
0 0.5 1 1.5 2 2.5 3 3.510
20
30
40
50
60
Time [sec]
Flo
w C
ond
ucta
nce
[LP
M/s
qrt(
MP
a)]
KvdKvKvappx
0 0.5 1 1.5 2 2.5 3 3.50.8
0.85
0.9
0.95
1
J [
]
Time [sec]0 0.5 1 1.5 2 2.5 3 3.5
0
2.5
5
7.5
10
12.5
Q [
LPM
/V-s
qrt(
MP
a)]
J
Q
0 0.5 1 1.5 2 2.5 3 3.510
20
30
40
50
60
Time [sec]
Flo
w C
ondu
ctan
ce [
LPM
/sqr
t(M
Pa)
]
KvdKvKvappx
0 0.5 1 1.5 2 2.5 3 3.50.8
0.85
0.9
0.95
1
J [
]
Time [sec]0 0.5 1 1.5 2 2.5 3 3.5
0
2.5
5
7.5
10
12.5
Q [
LPM
/V-s
qrt(
MP
a)]
J
Q
Estimated Jacobian and Controllability
Flow Conductance
Initial test response, no NLPN learning
November 29, 2005 51
PRELIMINARY WORK
Estimated Jacobian and Controllability
Flow Conductance
EHPV response with NLPN learning
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50
10
20
30
40
50
60
70
80
90
100
Kv
[LP
H/s
qrt(
MP
a)]
Time [sec]0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0
1
2
3
4
5
6
7
8
9
10
Inpu
t V
olta
ge [
V]
KvKvdVsol
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 520
25
30
35
40
45
50
Tem
pera
ture
[C
]
Time [sec]0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0
0.3
0.6
0.9
1.2
1.5
1.8
Pre
ssur
e D
iff [
MP
a]
Temp
dP
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 510
20
30
40
50
60
70
Time [sec]
Flo
w C
ond
ucta
nce
[LP
M/s
qrt(
MP
a)]
KvdKvKvappx
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50.9
0.92
0.94
0.96
0.98
1
1.02
J [
]
Time [sec]0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0
0.125
0.25
0.375
0.5
0.625
0.75
Q [
LPM
/V-s
qrt(
MP
a)]
J
Q
November 29, 2005 53
PRESENTATION OUTLINE
INTRODUCTION PROBLEM STATEMENT OBJECTIVES REVIEW OF MOST RELEVANT WORK PROPOSED RESEARCH PRELIMINARY WORK EXPECTED CONTRIBUTIONS CONCLUSION
November 29, 2005 54
EXPECTED CONTRIBUTIONS
An alternative methodology for control system design of nonlinear systems with time-varying characteristics and parametric uncertainty.
A method to estimate and learn the flow conductance of the valve online.
Guidelines to experimentally use this control methodology and health monitoring efficiently in the area of electro-hydraulic control.
November 29, 2005 55
PRESENTATION OUTLINE
INTRODUCTION PROBLEM STATEMENT OBJECTIVES REVIEW OF MOST RELEVANT WORK PROPOSED RESEARCH PRELIMINARY WORK EXPECTED CONTRIBUTIONS CONCLUSION
November 29, 2005 56
CONCLUSIONS
The proposed control methodology combines adaptive proportional feedback control with online corrected feedforward compensation
The input/output mapping of the system can be easily extracted via a functional approximator on the feedforward compensation
Extensive knowledge about the dynamics of the system are not needed a priori for satisfactory performance
The proposed method is to be employed in a Wheatstone bridge arrangement of novel Electro-Hydraulic Poppet Valves seeking a self-calibrated system