NSF GRANT # 0600516
NSF PROGRAM NAME: Dr. Suhada Jayasuriya
Adaptive Robust Control – Theory and Application to Precision and Energy
Saving Control of Electro-Hydraulic Systems
Abstract: The salient features of the proposed adaptive
robust control (ARC) lie in the seamless integration of
(i) appropriate nonlinear robust feedback controls that
are physically realizable while having targeted local
high-gain feedback so that the effect of various model
uncertainties can be sufficiently attenuated to guarantee
certain consistent robust control performance and (ii)
controlled learning to reduce model uncertainties to
maximize the achievable control performance and
facilitate the construction of certain built-in
intelligences. Such a design philosophy is well in-line
with how human beings utilize feedback information
and makes it well suited for the integrated design of
intelligent and precision mechatronic systems, as
demonstrated through various applications including
the electrical motor driven mechanical systems with
control accuracy down to sub-micrometer levels and
piezo-electrical actuator driven nano-positioning
systems for nanotechnology applications. This paper is
to introduce researchers and practicing engineers to the
essences of such an advanced nonlinear control design
methodology. The energy-saving control of electro-
hydraulic systems using novel programmable valves are
used as an application example and comparative
experimental results are presented to illustrate the
effectiveness of the presented ARC approach.
1. Introduction: Modern mechanical systems such as
microelectronics manufacturing equipment, robot
manipulators, and automatic inspection machines are
often required to operate at high speeds to yield high
productivity. At the same time, as the society moves
into the era of micro and nano-technology, these
systems are also required to have motion accuracy in
the micro or nano-meter range. These industrial trends
put an urgent demand on performance oriented
advanced controls that can effectively handle (i)
nonlinearities due to the strong nonlinear coupling
effect of multiple-degree-of-freedom (MDOF)
mechanical systems during high-speed operations, and
(ii) the unavoidable uncertainties associated with any
modeling of physical systems. The modeling
Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii
Bin Yao
School of Mechanical Engineering
Purdue University
West Lafayette, IN 47907-1288
uncertainties may come from either (i) parameters of
the system model which may depend on operating
conditions and may not be precisely known in advance
(e.g., inertia of an object grasped by a robot), or (ii)
nonlinearities which cannot be modeled exactly (e.g.,
nonlinear friction characteristics) or terms which
change with time randomly (e.g., external
disturbances); the former is referred to as parametric
uncertainties and the latter as uncertain nonlinearities.
To meet these ever increasingly stringent control
performance requirements, control engineers have been
forced to look beyond traditional linear control theory
for more advanced nonlinear controllers which can
directly deal with various nonlinearities and model
uncertainties. There has been an exponential growth in
nonlinear control research during the past two decades,
with major advances and breakthroughs reported in
both the nonlinear deterministic robust control (DRC)
area [1-4] and adaptive or robust adaptive control
(RAC) area [5-12]. Systematic nonlinear control design
methodologies such as the backstepping technique [10]
have been proposed as well.
The design of reliable industrial systems involves more
than just system modeling and control synthesis that
meets the performance requirements. In addition to
good control performance, it is also desirable for the
system to have the ability to detect and react favorably
to unexpected changes in operating conditions due to
machine component failures and various faults [13]. As
revealed in [14], faults that occur in a dynamic system
are often reflected in the physical parameters of the
process. As such, accurate on-line parameter estimation
has become one of the critical elements in fault
detection and isolation. Furthermore, it is also one of
the key ingredients for the construction of certain
intelligent features for modern mechanical systems such
as machine component condition monitoring and
prognosis; for example, accurate on-line estimation of
the friction coefficients can be used to monitor bearing
health, thus maintenance is only performed when
needed to reduce costs.
Grant #0600516
During the past decade, with the support of a NSF
CAREER award and subsequent continuing grants, the
PI and his students have been focusing on the
development of adaptive robust control (ARC) theory
to lay a solid foundation for the design of a new
generation of controllers which will help industry build
modern machines of great performance and
intelligence. The developed ARC theory bridges the
gap between two of the main control research areas -
nonlinear robust adaptive controls (RAC) [5-12] and
deterministic robust controls (DRC) [1-4].
Traditionally, those two research areas have been
presented as competing control design approaches since
each has its own benefits and limitations. By
recognizing the fundamentally different working
mechanisms of the two approaches, the PI is able to
develop a mathematically rigorous theoretical
framework which preserves the theoretical performance
results of both design approaches while overcoming
their well-known practical performance limitations.
Those theoretical results have been well publicized
through a number of publications (e.g., [15-26]).
Using NSF supports as leverages, the PI has also been
able to obtain matching supports from Purdue
University and various industrial sponsors to apply the
developed ARC theory to the integrated design of
various intelligent and/or precision industrial
mechatronic systems, including electro-hydraulic
actuator driven large-scale mechanical systems [27,28],
energy-saving control via novel programmable valves
[29,30], electrical motor driven mechanical systems
with control accuracy down to sub-micrometer levels
(e.g., linear motor driven positioning stages [31-34] and
ultra-high density hard disk drives [35]), and piezo-
electrical actuator driven nano-positioning systems for
nanotechnology applications [36,37]. The results from
this comprehensive set of applied research projects
have shown the developed ARC to be a powerful
approach to the control of a wide variety of
mechatronic systems. Other researches have
demonstrated the effectiveness of adaptive robust
control in various applications as well (e.g., machine
tools [46], the control of pneumatic muscles driven
parallel manipulators [38-40], hard disk drives [41,42],
active suspension systems [43,44], and vehicles [45]).
The theoretical breakthrough and the significant
performance improvement of the proposed adaptive
robust control (ARC) in various implementations make
the approach an ideal choice for industrial applications
demanding stringent performance. In addition, the
recently developed integrated direct/indirect adaptive
robust controls (DIARC) [16] also enable accurate on-
line parameter estimations to be obtained in actual
implementation. Such a by-product of the approach –
Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii
accurate parameter and nonlinearity estimations –
makes adding intelligent features such as the automated
on-board modeling [30] and the fault detection and
prognosis [47] possible. It is thus beneficial for control
engineers to get exposed to such an advanced nonlinear
control design methodology and to master how the
method can be used to build intelligent and yet
precision mechatronic systems, which is the main
objective of the paper.
2. Adaptive Robust Control Theory: To avoid
getting bugged down to the technical design
complexity, in this section, the tracking control of a
simple first order nonlinear systems with uncertainties
will be used to illustrate the advantages and limitations
of different types of adaptive robust controls. The
system is described by
x = f ( x, t ) + u , f = ϕ ( x)T θ + Δ( x, t ) (1.1)
where x, u ∈ R and is an unknown nonlinear function.
In general, f can be approximated by a group of
known basis functions ϕ ( x) ∈ R p with unknown
weights θ ∈ R , and the approximation error is denoted
p
by the unknown nonlinear function Δ( x, t ) . The
objective is to let x track its desired trajectory xd (t ) as
closely as possible. The following reasonable and
practical assumption is made, which is satisfied by most
applications:
A1 . The extent of parametric. uncertainties and
uncertain nonlinearities is known, i.e.,
θ ∈ Ωθ = {θ : θ min < θ < θ max }
(1.2)
) ≤ δ ( x, t ) }
Δ ∈ Ω Δ = {Δ : ‖Δ( x, t ‖
where θ min , θ max and δ ( x, t ) are known.
2.1 Direct Adaptive Robust Control (DARC)
The simplest ARC design is the direct ARC (DARC)
designs presented in [48,49,22], in which learning laws
such as parameter adaptations are synthesized along
with the control law to achieve the sole purpose of
reducing output tracking error. Specifically, for (1.1),
the parameter estimate θˆ is updated through a
parameter adaptation law having the form given by

θˆ = Proj (Γτ )
θˆ
(1.3)
where Γ is any symmetric positive definite adaptation
rate matrix, τ is an adaptation function to be specified
later, and Projθˆ (D) is the standard projection mapping
used in the adaptive control area [49]. Specifically,
when a diagonal Γ is used, with the discontinuous
projection, Projθˆ (•) = [ Projθˆ (•1 ), ⋅⋅⋅, Projθˆ (•n )]T , in
1 n
which
Grant #0600516
 ⎧0, if θˆi = θimax and •i > 0
⎪⎪
Projθˆ (•i ) = ⎨0, if θˆi = θimin and •i < 0 (1.4)
⎪• ,
i
otherwise
⎪⎩ i
The ARC control law consists of two parts given by
u = u f + us , u f = xd (t ) − ϕ T θˆ
(1.5)
us = us1 + us 2 , us1 = −kz
where z = x - xd is the tracking error. In (1.5), u f is
the adjustable model compensation needed for
achieving perfect tracking, and us is the robust control
law consisting of two parts: us1 is used to stabilize the
nominal system, which is a simple proportional
feedback in this case; and us 2 is a nonlinear robust
feedback used to attenuate the effect of model
uncertainties, which is synthesized to satisfy the
following two constraints
i. z[−ϕ T θ + Δ ( x, t ) + us 2 ] ≤ ε
(1.6)
ii. zus 2 ≤ 0
where ε is a positive design parameter representing the
attenuation level of the model uncertainties that one
would like to have. In (1.6), condition i is used to
represent the fact that us 2 is synthesized to dominate
the model uncertainties coming from both the
parametric uncertainties and uncertain nonlinearities to
achieve a guaranteed level of attenuation specified by
ε , and the passive-like constraint ii is imposed to make
sure that introducing us 2 does not interfere with the
nominal identification process of parameter adaptation.
The specific forms of us 2 satisfying constraints like
(1.6) can be found in ARC designs in [48,49,22].
Theorem 1: [48] If the adaptation function is chosen as
τ = ϕ ( x) z (1.7)
then, the ARC law (1.5) with the parameter adaptation
law (1.3) guarantees that
A. In general, all signals are bounded and the tracking
error is bounded by
ε In the DARC designs in the previous section, the
| z (t ) |2 ≤ exp (−2kt ) | z (0) |2 + [1 − exp (−2kt )] (1.8)
k
i.e., the tracking error exponentially decays to a ball.
The exponential converging rate k and the size of
ε which is not a problem for the direct ARC designs that
the final tracking error | z (∞) |≤ can be freely
k use gradient type adaptation laws only. For the indirect
adjusted by the controller parameters ε and k in a
known form.
B. If after a finite time, there exist parametric
uncertainties only (i.e., Δ( x, t ) = 0, ∀t ≥ t0 ), then, in
addition to the results in A, asymptotic tracking or
zero steady-state tracking error is achieved, i.e,
Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii
z → 0 as t → ∞ . Furthermore, if the desired
trajectory satisfies the following persistent excitation
(PE) condition
t +T
∫t
ϕ ( xd (ν ))ϕ T ( xd (ν ))dν ≥ ε p I p , ∀t ≥ t0 (1.9)
where T , t0 and ε p are some positive scalars, then,
the parameter estimates θˆ converge to their true
values as well (i.e., θ → 0 when t → ∞ ).
Remark 1: In the absence of parameter adaptation (i.e.,
Γ = 0 ), the proposed ARC law reduces to a
deterministic robust control (DRC) law and Result A of
Theorem 1 still holds. Therefore, the adaptation loop
can be switched off at any time without affecting the
stability and the guaranteed output tracking transient
performance. However, such a control law does not
discriminate the difference between parametric
uncertainties and uncertain nonlinearities and results in
a conservative design since Result B of Theorem 1 is
lost. As for adaptive control [10], the proposed ARC
uses certain coordination mechanisms (e.g., the
discontinuous projection mapping used in (1.3)) and
nonlinear robust feedback control us 2 to achieve a
guaranteed output tracking transient performance even
in the presence of uncertain nonlinearities (A of
Theorem 1) while without losing its nominal
performance (B of Theorem 1).
2.2 Indirect Adaptive Robust Control (IARC)
For the applications that need accurate parameter
estimates for other secondary purposes in addition to
the good output tracking performance, the indirect
adaptive robust control design (IARC) presented in [50]
can be used, which completely separates the
construction of parameter estimation law from the
design of underline robust control law as illustrated as
follows.
One of the key elements of the ARC design [48] is to
use the practical available prior process information to
construct projection type adaptation law for a controlled
learning process even in the presence of disturbances.
discontinuous projection mapping (1.4) is used for its
simplicity to ease implementation. However,
theoretically, such a discontinuous projection mapping
is valid only for diagonal adaptation rate matrix Γ ,
ARC introduced below, as the least square type
adaptation law will be used to achieve better
convergence of parameter estimations, the adaptation
rate matrix will be time-varying and non-diagonal. As
such, the standard projection mapping in the adaptive
control [10] should be used to keep the parameter
Grant #0600516
estimates within the known bounded set Ωθ , the
closure of the set Ωθ . The expression of Projθˆ (Γτ ) is
⎧ o
⎪⎪Γτ , if θˆ ∈ Ωθ or nθTˆ Γτ ≤ 0
⎨⎛ nθˆ nθTˆ ⎞
⎪ ⎜I −Γ T ⎟ Γτ , θˆ ∈ ∂Ωθ and nθTˆ Γτ > 0
⎜ n Γ n ⎟
⎩⎪ ⎝ θˆ θˆ ⎠
where Γ(t ) can be any time-varying positive definite
o ⎩0
symmetric matrix, Ωθ and ∂Ωθ denote the interior and
the boundary of Ωθ respectively, and nθˆ represents the
outward unit normal vector at θˆ ∈ ∂Ωθ . Such a
projection mapping has the same nice properties as
(1.4).
With the use of the projection type adaptation law
structure (1.3), the parameter estimates are bounded
within known bounds, regardless of the estimation
function τ to be used. As a result, the same adaptive
robust control law as in the direct ARC designs (i.e.,
(1.5) and (1.6)) can be used to achieve a guaranteed
output tracking transient and steady-state performance,
independent of the specific identifier to be used later.
Thus, the reminder of the IARC design is to construct
suitable estimation functions τ so that an improved
steady-state tracking accuracy– zero final tracking error
in the presence of parametric uncertainties only–can be
obtained with an emphasis on good parameter
estimation process as well. For this reason, it is
assumed that the system is absence of uncertain
nonlinearities, i.e., Δ = 0 in (1). Using a filter with a
stable transfer function H f ( s ) having a relative degree
of 1, the filtered system dynamics is obtained as
˙ parameters, but also introduces dynamic compensation
x f = ϕ Tf θ + u f
where x f = H f [ x], ϕ f = H f [ϕ ( x)] and u f = H f [u ]
are the filter output, regressor, and input respectively.
Define the estimation output and its estimate as
˙ u = ua + us , ua = ua1 + ua 2 , us = us1 + us 2
y = xf −uf , yˆ = ϕ Tf θˆ
With the calculable prediction error defined as
ε = ŷ − y , the resulting static prediction error model is
linearly parameterized in terms of parameter estimation
error θ as
ε = ϕ Tf θ (1.12)
Various estimation algorithms can then be used to
identify unknown parameters [50]. For example, when
the least squares type estimation algorithm with
exponential forgetting and co-variance limiting is used,
the resulting adaptation law is given by (1.3), in which
Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii
ϕf ε
τ =− (1.13)
1 + νϕ Tf Γϕ f
and Γ(t ) is updated by
ϕ f ϕ Tf
DΓ = αΓ − Γ Γ, Γ(0) = ρ0 I , ν ≥ 0
1 + νϕ Tf Γϕ f
GT G (1.14)
Γ(t ) = ⎨⎧ DΓ , if λM ( Γ ) < ρ M or vM DΓ vM < 0
else
where ν = 0 leads to the un-normalized algorithm, and
G
vM represents the eigenvector corresponding to
G
( )
λM ( Γ(t ) ) when λM Γ(t ) = ρ M , and the covariance is
initialized to a value of Γ(0) < ρ M I p . With the above
estimator and the adaptive robust control law, it is
shown in [50] that the same theoretical output tracing
performance results as in DARC in Theorem 1 are
achieved.
2.3 Integrated Direct/Indirect ARC
As shown in the comparative experimental results
[16,51], though the proposed IARC design has a much
better accuracy of parameter estimates than the direct
ARC, the output tracking performances of IARC are
not as good as those of DARC, especially during the
transient periods. A more detailed thorough analysis
reveals that the poorer tracking performance of IARC is
caused by the loss of dynamic compensation type fast
adaptation that is inherited in the DARC designs. To
overcome this loss of tracking performance problem of
IARC, an integrated direct/indirect ARC (DIARC)
design framework is developed in [16]. The design not
only uses the same adaptation process as in the IARC
design [14] for accurate estimation of physical
(1.10) type fast adaptation to achieve a better transient
performance as illustrated below.
For (1), the resulting DIARC law is:
(1.11)
ua1 = −ϕ T θˆ + xd (t ), ua 2 = −dˆc (1.15)
us1 = −k s1 z , us 2 = − ks 2 ( x, t ) z
In (1.15), ua1 represents the usual model compensation
with the physical parameter estimates θˆ(t ) updated
using the same indirect parameter estimator as in the
above IARC design, ua 2 is a model compensation term
similar to the fast dynamic compensation type model
compensation used in the DARC design, in which
dˆc can be thought as the estimate of the low frequency
component of the lumped model uncertainties defined
later. From (1.1) and (1.15), the error equation is
obtained as
Grant #0600516
 z = us + ua 2 − ϕ T θ + Δ (1.16)
Define a constant d c and time varying Δ (t ) such that
*
d c + Δ* (t ) = −ϕ T θ + Δ (1.17)
Conceptually, (1.17) lumps the disturbances and the
model uncertainties due to physical parameter
estimation error together and divides it into the static
component d c (or low frequency component in reality)
and the high frequency components Δ* (t ) , so that the
low frequency component d c can be compensated
through fast adaptation similar to those in the above
direct ARC design as follows.
Let d cM be any pre-set bound and use this bound to
construct the following projection type adaptation law
for
 ⎧⎪ 0 if | dˆc |= d cM and dˆc z > 0
dˆc = ⎨ (1.18)
⎪⎩γ d z else
with γ > 0 and | dˆ (0) |≤ d . Such an adaptation law
d c cM
guarantees that | dˆc (t ) |≤ d cM , ∀t . Substituting (1.17)
into (1.16) and noting (1.15),
z = us + ua 2 + d c + Δ* (t )
(1.19)
= us1 + ⎡⎣us 2 − dc + Δ* (t ) ⎤⎦
Due to the use of projection type adaptation law, all
estimation errors are bounded within known bounds. As
such, the same as in the DARC, it can be shown that, as
long as the nonlinear feedback gain ks 2 is chosen large
enough, the same robust performance condition as (1.6)
can be satisfied
z ⎡⎣us 2 − dc + Δ* ⎤⎦ ≤ ε c (1.20)
With the above estimator and the adaptive robust
control law, it can be shown [16] that theoretically the
same output tracing performance results as in DARC in
Theorem 1 are achieved.
3. Essences of Adaptive Robust Control: Aside from
the mathematically rigorous designs presented in the
previous section, some intuitive explanations and the
link to the traditional fundamental control design
philosophy for linear systems will be presented in this
section, to reveal the essences of the proposed ARC
approach -- knowing the essences will significantly
help a control engineer in correctly applying the
proposed ARC approach in practice without getting
bugged down to the technical design complexity
necessary for theoretical rigorousness.
The salient features of the proposed adaptive robust
control (ARC) lie in the seamless integration of (i)
appropriate nonlinear robust feedback controls that are
Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii
physically realizable while having targeted local high-
gain feedback so that the effect of various model
uncertainties can be sufficiently attenuated to guarantee
certain consistent robust control performance (e.g., the
underlying robust control law structure allowing
adjustable but controlled model compensations) and (ii)
controlled learning (e.g., the proposed projection type
adaptation law with explicit on-line condition
monitoring) to reduce model uncertainties to maximize
the achievable control performance and facilitate the
construction of certain built-in intelligences. Such a
design philosophy is well in-line with how human
beings utilize feedback information: (i) routine
workouts and training combined with good physical
condition allow faster immediate reaction capability –
essential for survival and consistent control
performance in various uncertain situations, and (ii)
reasoning through large amounts of stored past
feedback information to understand the underlying
dynamics of the particular environment encountered
(i.e., slow learning) to gain prediction capability for
even faster response time or smoother actions. Through
this analogy, though a thorough understanding and
mastering of the proposed research needs advanced
control backgrounds, the underlying design philosophy
can be understood rather easily. This makes it possible
to incorporate the ARC design philosophy into the
current undergraduate control curriculum to re-enforce
their understanding of the general control design
principles. As pointed out in [52], “the core idea in
control is that of inversion; the inversion can be
conveniently achieved by the use of two key
mechanisms: feedback and feedforward” and “high
loop gain gives approximate inversion, which is the
essence of control” (page 36). These fundamental
control design principles are well reflected in the
proposed adaptive robust control theory – the adjustable
nonlinear model compensation with on-line parameter
adaptation or other learning tools in ARC can be
thought of as a modern version of traditional
feedforward design, and the emphasis of ARC design
on the use of nonlinear robust control law having
targeted local high-gain feedback not only touches the
essence of feedback control but also actually makes it
possible; it allows one to overcome typical
implementation problems of traditional linear robust
control law having high-gain feedback relatively
effortlessly as done in [18, 32]. It is strongly advised to
keep these fundamental design philosophies in mind
when going through various specific design techniques
presented to avoid getting bugged down in the
complicated mathematics.
4. Applications to Integrated Design of Intelligent
and Precision Mechatronic Systems: The seamless
integration of fast robust feedback and controlled slow
Grant #0600516
learning of the proposed ARC approach makes it well
suited for the integration design of intelligent and
precision mechatronic systems, as demonstrated
through various applications in [27-46]. Some general
aspects of this integrated design framework are outlined
below.
The first step in the integrated design is to figure out
various practical ways in achieving fast feedback,
which should put equal emphasis on both software and
hardware design optimizations. Software side, local-
high-gain-but-global-low-gain nonlinear robust
feedback structure instead of the traditional linear high
gain feedback should be used which enables the explicit
consideration of the control input saturation while
having enough flexibility in achieving a better trade-off
in meeting various conflicting design objectives as
illustrated in Fig.1. Hardware side, innovative
mechanical designs with product functionality in mind
should be sought to make the resulting physical system
capable of having fast response or conducive to high-
gain feedback. Two such practical examples are the
positioning systems driven by linear motors studied in
[31-34] and the novel energy-saving valves in [29,30].
Specifically, for electro-mechanical positioning
systems, the use of linear motor drive systems instead
of rotary motors provides the hardware possibility of
having fast reaction due to the rigid construction of
linear motor based positioning systems. For electro
hydraulic systems, when controlled properly, the use of
a unique combination of five independently controlled
poppet type cheap cartridge valves completely bypasses
the sandwiched deadband problem of traditional
expensive proportional directional control valves [53],
which in return provides the hardware possibility of
having fast reaction – sandwiched deadband physically
limits the response speed of a control valve, regardless
the type of advanced controls to be used.
Robust feedback us
High gain feedback for a better
disturbance rejection to improve
transient performance
tracking error e
Moderate feedback gain for a better
trade-
trade-off between noise attenuation
and disturbance rejection
Small feedback gain to avoid control
saturation and instability
Figure 1. Nonlinear Feedback for Improved Transient
Performance and Better Trade-off in Solving Conflicts
The next step in the integrated design is to look for
practical ways to have good and robust learning
Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii
capability. These include (i) separation of estimation
model from the controller design model [16], (ii)
parameter estimation algorithms with better
convergence properties, (iii) explicit online monitoring
of persistence excitation level for accurate parameter
estimates [47], and (iv) the use of available prior
process knowledge such as the physical bounds of
parameter variations to achieve a controlled learning
process; this helps get rid off the destabilizing effect of
on-line learning and enable a fast adaptation loop to be
used in practice for better control performance.
Finally, a system design perspective should be
emphasized through out the integrated design process.
An impossible-to-solve hard control problem (e.g., the
precision control of conventional rotary motor driven
high-speed/acceleration positioning systems that may
involve flexibility and backlash of drive mechanisms
along with the unavoidable highly nonlinear Coulomb
frictions) could be easily overcome through the
seamless integration of software based complex control
architectures (e.g., the presented adaptive robust control
which is very effective in dealing with the highly-
nonlinear Coulomb frictions) and the use of innovative
hardware redesign (e.g., the rigid construction of linear
motor driven positioning systems).
5. Application to Precision and Energy Saving
Control of Electro-Hydraulic Systems: Lower cost,
higher control performance and significantly less
energy consumption are the goals for the design of any
industrial systems including electro-hydraulics. This
section uses the precision and energy saving control of
electro-hydraulic systems as a specific example to
illustrate how to effectively integrate the proposed ARC
theory with novel hardware reconfiguration/re-design
as one of the tools to achieve these lofty goals
simultaneously.
5.1 Introduction
Electro-hydraulic systems are traditionally controlled
by four-way spool-type valves, either servo valves or
proportional directional control (PDC) valves.
Significantly large amount of research have been done
recently to improve the electro-hydraulic system
performance by applying advanced nonlinear adaptive
and robust control techniques to deal with the hydraulic
and mechanical nonlinearities (see references in [28]).
All the works presented were rooted on electro-
hydraulic systems controlled by the traditionally four-
way spool type valves and purely from control points of
view. Though the performances of those electro-
hydraulic systems were significantly improved, no
considerations were given to other practically important
objectives such as reducing energy consumptions.
Without some fundamental changes to the overall
Grant #0600516
electro-hydraulic system design, it is even impossible to
make a good tradeoff between the performance and the
cost; better control performance has to be gained
through the more expensive servo valves.
On the other hand, most fluid power engineers tend to
solve problems with novel fluid power hardware
designs only. It is thus no surprise that thousands types
of valves and hydraulic components were developed
and most of them have very narrow and specific
applications. As a result, the design of industrial
hydraulic systems becomes much more complex than
the electro-mechanical systems, which has posted a
significant challenge to the survival of fluid power
industry. The situation becomes even worse when
performance is concerned. Instead of relying on only
hardware or software via advanced controls alone, this
paper re-approaches various issues from an integrated
mechatronic design perspective, i.e., a balanced
emphasis on both the hardware and software control
designs in solving problems and in making a better
tradeoff among performance, cost and energy usage.
Specifically, by knowing what advanced controls can
do and cannot do, certain hardware re-design can be
carried out to remove the performance limiting
hardware barriers and/or to have a much simplified
mechanical design for cost reduction even though the
complexity of controlling the resulting system may
increases. As has been verified through various
application results, the proposed adaptive robust control
can effectively handle the effect of various smooth
nonlinearities and uncertainties. As such, hardware
designs do not have to focus on achieving a linear or a
perfect known nonlinear input-output relationship.
Instead, hardware designs should focus on solving the
performance limiting hard nonlinearities such as
deadband and backlash. In this regard, as will be
detailed later in this section, the use of cheap poppet
type cartridge valves to replace the traditional four-way
spool-type PDC valves when controlling an electro-
hydraulic system serves as a perfect example of such a
hardware redesign -- though the flow characteristics of
cartridge valves cannot be modeled as precisely and
simply as those of traditional PDC valves, the use of
cartridge valves enables one to overcome the
performance limiting deadband problems associated
with the traditional PDC valves. In addition, such a
hardware redesign significantly reduces the
manufacturing cost as cartridge valves are known as
economical alternatives to large PDC valves [54].
The hardware redesign often leads to a more capable
physical system of increased flexibility and
controllability, providing the hardware capability of
achieving other practically meaningful goals in addition
to better control performance (e.g., significantly
Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii
reducing system energy usage). It is thus imperative to
develop suitable advanced control strategies to make
full use of those capabilities gained by hardware
redesign. For example, the programmable valves
consisting of five cartridge valves proposed in this
paper eliminate the mechanical linkage between meter-
in and meter-out flows and enable the precise control of
regeneration flows. However, to make full use of these
hardware capabilities, the recently developed advanced
adaptive robust controls have to be used since the
simple linear control techniques used in the hydraulic
industry cannot handle the large extent of uncertainties
and the lack of suitable mathematical models of
cartridge valves, the main reason why cartridge valves
have traditionally not been used in the precision control
of electro-hydraulic systems.
5.2 Energy Saving Programmable Valves
5.2.1 Ways to Reduce Energy Consumptions: The
energy used by a hydraulic system to perform a task can
be calculated as:
t1
E = ∫ Ps (τ )Qs (τ )dτ (1.21)
t0
where t0 and t1 are the starting and the ending time of
the task, and Ps and Qs represent the supply pressure
and the flow rate from the hydraulic pump. It is obvious
that there exist two ways to reduce the energy usage:
1. reduce the pump supply pressure Ps (t )
2. reduce the flow rate Qs (t ) needed from a pump
Without considering the fluid compressibility, the flow
rate needed from a pump to perform a given task
depends on the cylinder piston movement only and
would be pretty much fixed unless additional flow paths
such as the regeneration flows between the two
chambers of the cylinder are created. To reduce the
pump supply pressure, working pressures at the two
cylinder chambers are desired to be as low as possible
so long their pressure difference can be maintained at
the level needed to perform the required motion.
Therefore, independently controlling flows to two
chambers of a cylinder with additional flow paths such
as the cross-port flows and having low working
pressures for the two chambers without losing
controllability and performance are the keys to reducing
energy consumption for energy saving.
Traditionally, a four-way valve, either servo or PDC, is
used to control each hydraulic cylinder as done in
almost all existing publications. With such a valve
configuration the meter-in and meter-out orifices are
mechanically linked and the flows to the two chambers
of the cylinder cannot be independently regulated.
Consequently only one of the two cylinder states (i.e.,
Grant #0600516
pressures of the two chambers) is completely because of these difficulties, there have not been any
controllable. There exists a one-dimensional internal published results on the simultaneous precise motion
dynamics [28] which cannot be modified by any input- tracking and significant energy saving.
output type control strategies. The control input to the
valve in those strategies is uniquely determined once Load
the desired motion is specified, which makes the
regulation of individual cylinder chamber pressures Cylinder
impossible. As a result, though high system
performance can be achieved with the use of advanced
controls, significant reduction of energy consumption is
unattainable.
Load
The need to eliminate the mechanical linkage between
the meter-in and meter-out orifices for independent
regulation of two cylinder chamber flows is well known
and has been used in heavy industrial applications. For
example, the spool type valve is replaced by four
poppet type valves in [55], and a number of slight
variations exist throughout the mobile hydraulics PS PT
industry [60]. Patents by various companies [57-59]
attest to the potential of this technique. Specialized Figure 2: (a) Four valve IMVs (b) Regeneration flows
hardware such as the independent metering valves
(IMV) shown in (a) of Fig. 2 by Caterpillar Inc. has Load
also been developed, in which each metering valve is Q2
independently controlled, allowing completely Cylinder
Q1
separated controls of flows to two cylinder chambers.
P1 P2
As illustrated in (b) of Fig.2, another well-known
Valve #3
technique to reduce energy usage is to make full use of
the regeneration flows [56], i.e., the flow pumped from
Qv3
one chamber of the cylinder to the other chamber using
Qv1 Qv5
energy of the external load (e.g., the potential energy of Valve #1 Valve #5
a load during downward motions and the kinematic
energy of a load during deceleration motions). Flow Qv2 Qv4
regeneration is a highly efficient process in which little Valve #2 Valve #4
or zero pump flow or energy is needed. Ideally, PS PT
regeneration should be used whenever the external
force is in the same direction as the desired motion for
attaining maximum efficiency. The four-valve IMVs Figure 3: Programmable valves configuration
also enables the use of regeneration flow but not to the
fullest extent possible. The reason is that, unlike in (b) The PI was exposed to the IMVs concept when he
where a dedicated valve is used, a completely started his fluid power control research with Caterpillar
controllable cross port flow path is not available in Inc. The hardware advantages of IMVs over traditional
IMVs since such a flow path has to be created by spool type valves impressed him very much and he was
opening both valves connecting to either the return line determined to reap the hardware benefits of IMVs with
to the tank or the pressurized line to the pump. Either the advanced adaptive robust controls he have
scenario can happen only at certain extreme loading developed over the years. To make full use of the
conditions. For example, to use the two valves regeneration flows for maximum energy-saving while
connecting to the pressurized line for the cross port having a controlled motion, a fifth valve is added to the
flow, the pressures of cylinder chambers should be IMVs configuration to create the programmable valves
higher than the pump supply pressure. Because of these, shown in Fig.3 and studied in [29,30]. It should be
though significant energy saving could be achieved for noted that such a configuration is used simply because
operations such as free fall or dump, precise control of it enables a completely controllable cross-port flow and
regeneration flows is normally unattainable and can be built with off-the-shelf one-way valves such as
simultaneous good motion control performance cannot the poppet type cartridge valves. As long as a
be accomplished with this set of hardware. Partly completely controllable cross-port flow path and

Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii Grant #0600516
independent control of flows to the two chambers of a
cylinder can be realized, other specialized hardware
with possibly even less number of valves through well
thought-out hardware integration can also be used.
5.2.2 Hardware Benefits of Programmable Valves:
The five independently controlled one-way proportional
valves in the proposed programmable valves provide
the hardware capability of fully decoupled control of
meter-in and meter-out flows and precise control of
cross-port flows at any time. To further reduce
manufacturing costs, poppet type proportional cartridge
valves, which are known as economical alternatives to
large proportional valves and have low maintenance
costs are used. In addition, cartridge valves are also
known for their low leakage characteristics. For
example, Vickers EPV10 series has a typical leakage of
less than 3cc/min. Thus, aside from a higher degree of
energy efficiency, the programmable valves can also be
used as a load holding device when they are not in the
energized modes.
Aside from the cost benefits, the proposed
programmable valves also have certain hardware
advantages over the traditional spool type valves from a
control’s perspective. Specifically, the much smaller
inertia of the cartridge valves’ moving parts leads to the
much faster valve dynamic response; in fact, cartridge
valves have been mainly used in hydraulic industry for
various safety precaution measures due to their fast
responses. This hardware benefit will significantly
simplify the advanced controller designs and put much
less restrictive limitations on the achievable closed-loop
bandwidth – all existing advanced controllers with
experimental verifications have neglected the valve
dynamics for simplicity, and consequently the
achievable closed-loop bandwidth in implementation is
limited by the bandwidth of the actual valve responses.
5.2.3 Solution to Performance Limiting
Sandwiched Deadband Control Problems: Closed-
center valves such as the proportional directional
control (PDC) valves are widely used in industry for
position or velocity control applications. An over-
lapped spool shown in Fig.4 is intentionally used in
these kinds of valves to prevent internal leakages during
non-energized modes so that the closed-loop system
can hold a position even when the power is off.
Unfortunately, such an over-lapped spool also
introduces a deadband between the spool displacement
and the actual valve orifice opening and creates the so-
called sandwiched deadband control problems shown in
Fig.5–the deadband is between the valve spool
dynamics and the plant dynamics.
Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii
dv , Deadand xvr= xv - dv, Actual valve
opening
xv , Spool x
displacement
v
Valve Body
Overlapped Spool Overlapped Spool
Valve Body
Spool in Neutral Position Spool in Neutral Position
Figure 4: Over-lapped closed-center PDC valves
Controller
u Valve xv xvr Orifice System
Dynamics Flows Dynamics
DBI DB
Figure 5: Hydraulic systems controlled by PDC valves
It is well-known that the sandwiched deadband control
problem is both theoretically and practically difficult to
solve [53]. Even though the destabilizing effect of the
deadband could be accounted for by suitable control
designs, there will always exist a hard limitation on the
achievable closed-loop control performance no matter
what type of advanced controls to be used. For
example, the traditional method is to use the direct
inverse of the deadband nonlinearity shown in Fig. 5 to
cancel the deadband effect. Successful implementation
of this strategy without closed-loop instability issues
relies on two conditions: a) the deadband is known or
accurately estimated, and b) the valve dynamics is fast
enough to be neglected. The first condition may be
achieved though off-line system identification, but the
second condition can be satisfied only by limiting the
achievable closed-loop system bandwidth low enough
so that the valve dynamics (usually pretty slow for PDC
valves) looks “faster enough” when compared with the
closed-loop bandwidth. Though both feed-forward
controller and local high-gain feedback control have
been proposed to boost the valve dynamic response, in
practice, neither method was able to improve the
overall control performance significantly due to several
implementation constraints as revealed by the
experimental results in [53,60]. Specifically, the
success of using the feed-forward controller heavily
depends on the accuracy of the valve dynamics, while
the local high-gain feedback controller needs the
measurement of valve spool position, which tends to be
too noisy to be of much usefulness, aside from the
much increased system cost of having the spool
position sensor. Because of these issues, PDC valves
have not been used in hydraulic industry for precision
motion controls. Instead, the critically centered or
under-lapped servo valves are used at the expense of
increased energy usage and the loss of the ability of
holding positions when the valve is shut off due to
significant leakage flows.
Grant #0600516
 The poppet type cartridge valve, as shown in Fig. 6,
also has deadband between the control input voltage to
the valve and the resulting valve flow due to the fact
that the input voltage has to be large enough to
overcome the pre-load spring force and the static
friction for the valve poppet to move to open the
orifice. However, such a deadband is fundamentally
different from the sandwiched deadband problem of
PDC valves and can be easily overcome by suitable
controller designs. Specifically, the electrical dynamics
between the commanded input voltage to the valve and
the resulting electro-magnetic force are usually in kilo-
Hz range. They are much higher than the PDC valve
dynamics relating the valve input voltage to the spool
displacement and can be safely neglected as in the
precision control of electrical motors. Therefore, as
illustrated in Fig. 7 where the valve dynamics refers to
the poppet dynamics between the net force acting on
the poppet to its displacement, the deadband is of the
input deadband type and can be easily canceled via the
well-known deadband inverse. Furthermore, as the
poppet only moves uni-directionally, the deadband can
be essentially removed by setting the right amount of
input off-set. All these make it a hardware reality of
bypassing the performance limiting sandwiched
deadband problem of traditional spool valves, paving
the way to achieve a higher control performance.
Figure 6: Poppet type cartridge valves
Controller
u Valve xvr Orifice
Dynamics Flows Dynamics
DBI DB
Figure 7: Cartridge valve controlled hydraulic systems
5.3 Coordinated Control Strategy
As shown in Fig.3, the use of programmable valves
provides the hardware capability of independently
controlling the pressures of both chambers, P1 and P2 ,
and the control of the cross-port flow between the two
chambers. However, it also increases the complexity of
controlling the resulting system significantly. The
difficulties of controlling such a system not only come
from the well-known issues of highly nonlinear
hydraulic dynamics and large extents of parameter
Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii
variations and uncertain nonlinearities but also from the
lack of accurate mathematical model of cartridge valve
flow characteristics and the need to coordinate the
control of the five cartridge valves.
How to effectively deal with nonlinearities and the
large extents of parametric uncertainties and uncertain
nonlinearities have been solved by applying the
recently developed adaptive robust control (ARC)
strategy in [28]. The problem of lacking accurate
mathematical model of the nonlinear cartridge valve
flow mappings can be solved by using look-up tables
that relate the valve flow rates to the valve input
voltages and the pressure drops across the valve
orifices. Such flow mapping tables can be obtained
either by off-line experiments or through automated on-
line modeling techniques in [30]. The difficulties in the
coordinated control of the five cartridge valves have
been dealt with through a two-level controller structure
shown in Fig. 8 in [29]. Specifically, given the current
system states and desired motion trajectory, the task
level controller determines the specific subsets of the
programmable valves to be used which will enable
maximal energy saving while without losing the
controllability for precise motion tracking. This process
is normally referred to as the working mode selection in
hydraulic industry. The valve level controllers use the
adaptive robust control strategy to precisely regulate the
pressures of both cylinder chambers under the chosen
working mode to achieve the dual objectives of precise
motion tracking and significant energy saving
Working Mode
Reference Selection
Motion
Valve Commands
State Motion/Pressure Flow Programmable EH
Feedback ARC Controllers Distribution Valves Systems
Meter In/Out
Flow Commands
System Figure 8: Two-level control structure
5.4 Comparative Experiments
Experiments have been done to test the proposed two-
level control system. The programmable valves shown
in Fig.10 were used to control the boom motion of an
electro-hydraulic arm shown in Fig. 9 to track a desired
motion trajectory for up and down motions shown in
Fig. 11. The programmable valves were compared in
terms of motion tracking performance as well as energy
usage with a critically centred servo valve and a close-
centred PDC valve with the direct deadband inverse
compensation using the same baseline ARC motion
controller [28].
Grant #0600516

Figure 9: A scaled down backhoe loader arm
Figure 10 Programmable Valves
The tracking performances of the three systems were
plotted in Fig. 12 and quantified in Table 1 using the
following three indexes: (1) & e &1 defined as
20 20
∫0
e(t ) d (t ) ; (2) & e &2 defined as ∫
0
e2 (t )d (t ) ; and
(3) & e &∞ defined as max e(t ) . As seen from these
results, all three systems exhibit good tracking
performance as their maximum tracking errors are all
less than 0.015 rad. As expected, the servo valve has a
smaller tracking error when compared to the PDC
valve; the PDC valve has significantly larger transient
errors at the beginning and the end of each travel when
the deadband effects are most severe. The proposed
programmable valves do not have this deadband issue
and in fact achieve an even better tracking performance
when compared to the very expensive servo valve; such
a performance gain may come from the fact that the
cartridge valves have really fast response due to the
smaller inertia of the poppet.
A comparison of cylinder working pressures of each
system is shown in Fig. 13. The cylinder pressures in
the system controlled by the programmable valves are
much lower than the pressures in the other two systems,
Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii
with one chamber pressure always close to the tank
pressure, a preset pressure of 200KPa to compensate for
the line loss and to prevent cavitations. This is due to
the result of the decoupled control of cylinder pressures
which are made possible by the proposed
programmable valves.
Table 1: Comparison of Tracking Errors
Value & e &1 & e &2 & e &∞
PDC 0.0030 0.0046 0.0146
Servo 0.0014 0.0018 0.0051
Programmable 0.0007 0.0010 0.0036
The power usages of the three systems are shown in
Fig.14. The experiment setup used a pump with a
constant supplied pressure of 6900KPa (1000PSI). As
seen, during the upward motion periods (roughly, 1
through 3.5 seconds and 11 through 13.5 seconds in the
figures), the three systems use almost the same amount
of energy as pump flow is needed to lift the payload.
However, during the downward motion periods
(roughly, 6 through 8.5 seconds and 16 through 18.5
seconds in the figures), the programmable valves
controlled system did not use any pump energy at all
due to the use of regeneration flow for precise cylinder
motion controls, while the other two systems still need
pump flows for a controlled downward motion. It is
worth noting that the proposed programmable valves
not only enable the use of regeneration flow to reduce
energy consumption but also the precise control of
regeneration flow for a well controlled motion – the
resulting motion tracking error shown in Fig. 12 is even
better than that is achieved by a servo valve. Such a use
of regeneration flow is much different from existing
energy saving systems where the regeneration flows are
not precisely controlled and only coarse motions such
as the free falls can be performed. In addition, the
programmable valves controlled system uses about 2/3
of energy consumed by the other systems; the total
pump energy used by three systems for this specific
task were 32.4KJ and 32.7KJ for PDC and servo
systems, respectively, and only 21.3KJ for the
programmable valves system.
In mobile hydraulic industry, instead of a constant
supplied pressure pump, a load sensing pump (i.e., the
pump supplied pressure is varied according to the
required cylinder chamber pressures) is normally used
for further energy saving [1]. When a load sensing
pump is available, much more energy saving can be
obtained with the proposed programmable valves due to
the significantly reduced cylinder working pressures
shown in Fig.13. Unfortunately, our experimental setup
does not have a load sensing pump and it is impossible
to experimentally test the three valves with a load
Grant #0600516
sensing pump. Instead, a virtual comparison is done to
mimic a load sensing pump by adding 500KPa to the
cylinder working pressure connected to the pump as the
pump supplied pressure. With this assumption, the
comparative results for pump power usage of three
systems are shown in Fig.15, in which the
programmable valve controlled system uses about only
1/3 of the energy consumed by the other two systems.
6. Conclusions: The paper provides an overview on the
recently developed adaptive robust control approach. It
is shown that through a seamless integration of the new
physical capabilities gained by novel hardware redesign
(e.g., the proposed programmable valves) and the
proposed advanced adaptive robust controls,
traditionally conflicting requirements of lower cost,
higher performance, and less energy consumptions
could be simultaneously achieved. Such an integrated
mechatronic design philosophy has also been
successfully tested through the precision and energy
saving control of electro-hydraulic systems.
7. Acknowledgements: The work is supported in part
by the National Science Foundation (Grant No. CMS-
0600516).
8. References:
[1] V. I. Utkin, Sliding modes in control optimization.
Springer Verlag, 1992.
[2] B. Yao, W.B. Gao, S.P.Chan, and M. Cheng, “Vsc
coordinated control of two robot manipulators in the
presence of environmental constraints,” IEEE Trans. on
Automatic Control, vol.37, no.11, pp 1806–1812,1992.
[3] B. Yao, S. P. Chan, and D. Wang, “Unified
formulation of variable structure control schemes to
robot manipulators,” IEEE Trans. on Automatic
Control, vol. 39, no. 2, pp. 371–376, 1994.
[4] Z. Qu, Robust control of nonlinear uncertain
systems. John Wiley and Sons, Inc., 1998.
[5] K. J. Astrom, “Tuning and adaptation,” in IFAC
World Congress, Plenary Volume, pp. 1–18, 1996.
[6] I. D. Landau, R. Lozano, and M. M’Saad, Adaptive
control. New York: Springer, 1998.
[7] B. Yao, S. P. Chan, and D. Wang, “Variable
structure adaptive motion and force control of robot
manipulators,” Automatica, vol. 30, no. 9, pp. 1473–
1477, 1994.
[8] B. Yao and M. Tomizuka, “Adaptive control of
robot manipulators in constrained motion - controller
design,” ASME J. of Dynamic Systems, Measurement
and Control, vol.117, no.3, pp.320–328, 1995.
[9] P. A. Ioannou and J. Sun, Robust adaptive control.
New Jersey: Prentice-Hall, 1996.
[10] M. Krstic, I. Kanellakopoulos, and P. V.
Kokotovic, Nonlinear and adaptive control design.
New York: Wiley, 1995.
Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii
[11] R. A. Freeman, M. Krstic, and P. V. Kokotovic,
“Robustness of adaptive nonlinear control to bounded
uncertainties,” Automatica, vol. 34, no. 10, pp. 1227–
1230, 1998.
[12] F. Ikhouane and M. Krstic, “Adaptive
backstepping with parameter projection: robustness and
asymptotic performance,” Automatica, vol. 34, no. 4,
pp. 429–435, 1998.
[13] R. J. Patton, P. M. Frank, and R. N. Clark, Fault
Diagnosis in Dynamic Systems - Theory and
Applications. Prentice Hall, 1989.
[14] R. Isermann, “On the applicability of model-based
fault detection for technical processes.,” Control
Engineering Practice, vol.2, no.3, pp.439–450,1994.
[15] B. Yao, “Desired compensation adaptive robust
control,” ASME Journal of Dynamic Systems,
Measurement, and Control, 2009 (in press).
[16] B. Yao, “Integrated direct/indirect adaptive robust
control of SISO nonlinear systems transformable to
semi-strict feedback forms,” in American Control
Conference, pp. 3020–3025, 2003. The O. Hugo
Schuck Best Paper (Theory) Award from the American
Automatic Control Council in 2004.
[17] L. Xu and B. Yao, “Adaptive robust control of
mechanical systems with nonlinear dynamic friction
compensation,” International Journal of Control, vol.
81, no. 2, pp. 167–176, 2008.
[18] Y. Hong and B. Yao, “A globally stable saturated
desired compensation adaptive robust control for linear
motor systems with comparative experiments,”
Automatica, vol. 43, no. 10, pp. 1840–1848, 2007.
[19] J. Gong and B. Yao, “Output feedback neural
network adaptive robust control of a class of SISO
nonlinear systems,” ASME J. of Dynamic Systems,
Measurement, and Control, vol. 128, no. 2, pp. 227–
235, 2006.
[20] B. Yao and L. Xu, “Output feedback adaptive
robust control of uncertain linear systems with
disturbances,” ASME Journal of Dynamic Systems,
Measurement, and Control, vol. 128, no.4, pp. 1–9,
2006.
[21] B. Yao and L. Xu, “Adaptive robust control of
linear motors for precision manufacturing,”
International Journal of Mechatronics, vol. 12, no. 4,
pp. 595–616, 2002.
[22] B. Yao and M. Tomizuka, “Adaptive robust
control of MIMO nonlinear systems in semi-strict
feedback forms,” Automatica, vol. 37, no. 9, pp. 1305–
1321, 2001.
[23] B. Yao and L. Xu, “Observer based adaptive
robust control of a class of nonlinear systems with
dynamic uncertainties,” International Journal of Robust
and Nonlinear Control, no. 11, pp. 335–356, 2001.
[24] J. Gong and B. Yao, “Neural network adaptive
robust control of nonlinear systems in normal form,”
Asian Journal of Control, vol. 3, no. 2, pp. 96–110,
Grant #0600516
2001. (the Special Issue on Trend and Advancement in
Neural Networks Based Control Designs).
[25] J. Q. Gong and B. Yao, “Neural network adaptive
robust control of nonlinear systems in semi-strict
feedback form,” Automatica, vol. 37, no. 8, pp. 1149–
1160, 2001. (the Special Issue on Neural Networks for
Feedback Control).
[26] J. Q. Gong and B. Yao, “Neural network adaptive
robust control with application to precision motion
control of linear motors,” International Journal of
Adaptive Control and Signal Processing, vol. 15, no. 8,
pp. 837–864, 2001. (the Special Issue on Developments
in Intelligent Control for Industrial Applications).
[27] B. Yao, F. Bu, and G. T. C. Chiu, “Nonlinear
adaptive robust control of electro-hydraulic systems
driven by double-rod actuators,” International Journal
of Control, vol. 74, no. 8, pp. 761–775, 2000.
[28] B. Yao, F. Bu, J. Reedy, and G. Chiu, “Adaptive
robust control of single-rod hydraulic actuators: theory
and experiments,” IEEE/ASME Trans. on
Mechatronics, vol. 5, no. 1, pp. 79–91, 2000.
[29] S. Liu and B. Yao, “Coordinate control of energy-
saving programmable valves,” IEEE Transactions on
Control System Technology, vol. 16, no. 1, pp. 34–45,
2008.
[30] S. Liu and B. Yao, “Automated modeling of
cartridge valve flow mapping,” IEEE/ASME
Transactions on Mechatronics, vol. 11, no. 4, pp. 381–
388, 2006.
[31] L. Lu, Z. Chen, B. Yao, and Q.Wang, “Desired
compensation adaptive robust control of a linear motor
driven precision industrial gantry with improved
cogging force compensation,” IEEE/ASME
Transactions on Mechatronics, vol. 13, no. 6, pp. 617–
624, 2008.
[32] Y. Hong and B. Yao, “A globally stable high
performance adaptive robust control algorithm with
input saturation for precision motion control of linear
motor drive system,” IEEE/ASME Transactions on
Mechatronics, vol. 12, no. 2, pp. 198–207, 2007.
[33] L. Xu and B. Yao, “Output feedback adaptive
robust precision motion control of linear motors,”
Automatica, vol. 37, no. 7, pp. 1029–1039, 2001.
[34] L. Xu and B. Yao, “Adaptive robust precision
motion control of linear motors with negligible
electrical dynamics: theory and experiments,”
IEEE/ASME Transactions on Mechatronics, vol. 6, no.
4, pp. 444–452, 2001.
[35] J. Q. Gong, L. Guo, H.-S. Lee, and B. Yao,
“Modeling and cancellation of pivot nonlinearity in
hard disk drive,” IEEE Transaction on Magnetics, vol.
38, no. 5, pp. 3560–3565, 2002.
[36] J. Zhong and B. Yao, “Adaptive robust precision
motion control of a piezoelectric positioning stage,”
IEEE Transactions on Control System Technology, vol.
16, no. 5, pp. 1039–1046, 2008.
Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii
[37] J. Zhong and B. Yao, “Adaptive robust repetitive
control of piezoelectric actuators,” in the ASME
International Mechanical Engineers Congress and
Exposition, IMECE2005-81967, pp. 1–8, 2005. Finalist
of the Best Student Paper Competition of the ASME
Dynamic Systems and Control Division (DSCD).
[38] X. Zhu, G. Tao, B. Yao, and J. Cao, “Adaptive
robust posture control of a pneumatic muscle driven
parallel manipulator,” Automatica, vol. 44, no. 9, pp.
2248–2257, 2008.
[39] X. Zhu, G. Tao, B. Yao, and J. Cao, “Adaptive
robust posture control of a pneumatic muscles driven
parallel manipulator with redundancy,” IEEE/ASME
Transactions on Mechatronics, vol. 13, no. 4, pp. 441–
450, 2008.
[40] X. Zhu, G. Tao, B. Yao, and J. Cao, “Integrated
direct/indirect adaptive robust posture trajectory control
of a parallel manipulator driven by pneumatic muscles,”
IEEE Transaction on Control System Technology, 2009
(in press).
[41] L. Yi and M. Tomizuka, “Two degree-of-freedom
control with adaptive robust control for hard disk servo
systems,” IEEE/ASME Transactions on Mechatronics,
vol. 4, no. 1, pp. 17–24, 1999.
[42] H. D. Taghirad and E. Jamei, “Adaptive robust
controller synthesis for hard disk servo systems,” in
Proceedings of IEEE/RSJ International Conference on
Intelligent Robots and Systems, pp. 1154–1159, 2004.
[43] S. Chantranuwathana and H. Peng, “Adaptive
robust control for active suspensions,” in Proc. of the
American Control Conference, p. 1702C1706, 1999.
[44] S. Chantranuwathana and H. Peng, “Modular
adaptive robust control of SISO nonlinear systems in a
semi-strict feedback form,” Int. J. Robust Nonlinear
Control, vol. 14, p. 581C601, 2004.
[45] F. Bu and H.-S. Tan, “Pneumatic brake control for
precision stopping of heave-duty vehicles,” IEEE
Transactions on Control System Technology, vol. 15,
no. 1, pp. 53–64, 2007.
[46] B. Yao, M. Al-Majed, and M. Tomizuka, “High
performance robust motion control of machine tools:
An adaptive robust control approach and comparative
experiments,” IEEE/ASME Trans. on Mechatronics,
vol. 2, no. 2, pp. 63–76, 1997.
[47] P. Garimella and B. Yao, “An adaptive robust
framework for model-based state fault detection,” in
American Control Conference, pp. 5692–5697, 2006.
[48] B. Yao, “High performance adaptive robust control
of nonlinear systems: a general framework and new
schemes,” in Proc. of IEEE Conference on Decision
and Control, pp. 2489–2494, 1997.
[49] B. Yao and M. Tomizuka, “Smooth robust
adaptive sliding mode control of robot manipulators
with guaranteed transient performance,” ASME Journal
of Dynamic Systems, Measurement and Control, vol.
118, no. 4, pp. 764–775, 1996.
Grant #0600516
[50] B. Yao and A. Palmer, “Indirect adaptive robust Reference Trajectory
control of SISO nonlinear systems in semi-strict 1

Boom Angle (rad)

feedback forms,” in IFAC World Congress, T-Tu-A03- 0.8

2, pp. 1–6, 2002. 0.6

[51] B. Yao and R. Dontha, “Integrated direct/indirect 0.4

adaptive robust precision control of linear motor drive 0.2

systems with accurate parameter estimations,” in the 0

2nd IFAC Conference on Mechatronics Systems, pp. 0 2 4 6 8 10 12 14 16 18 20

633–638, 2002.
[52] G. C. Goodwin, S. F. Graebe, and M. E. Salgado,

Angluar Velocity (rad/sec)

0.6

Control System Design. Prentice-Hall: Englewood 0.4

Cliffs, New Jersey, 2001. 0.2

[53] S. Liu and B. Yao, “Characterization and 0

attenuation of sandwiched deadband problem using −0.2

describing function analysis and its application to −0.4

electro-hydraulic systems controlled by closed-center 0 2 4 6 8 10 12 14 16 18 20

valves,” ASME Journal of Dynamic Systems, Time (sec)

Measurement, and Control, 2009 (in press). Figure 11: Desired trajectory for up and down motions
[54] Thomas J. Ulery, Proportional cartridge valves -
Performance (Tracking Error) Comparison
economical alternative to large valves, Agricultural 0.02

Engineering 71 (1990), no. 4, 11. PDC Valve

Error (rad)
0.01

[55] A. Jansson and J.-O. Palmberg, Separate control of 0

meter-in and meter-out orifices in mobile hydraulic −0.01

systems, SAE Transactions 99 (1990), no. 2, 337–383. −0.02

0 2 4 6 8 10 12 14 16 18 20
[56] K.D. Garnjost, Energy-conserving regenerative- 0.01
flow valves for hydraulic servomotors, United States Servo Valve
Error (rad)

0.005
Patent, 4,840,111, 1989. 0
[57] J.A. Aardema and D.W. Koehler, System and −0.005
method for controlling an independent metering valve, −0.01
United States Patent, 5,947,140, 1999. 0 2 4 6 8 10 12 14 16 18 20

[58] K. D. Kramer and E.H. Fletcher, Electrohydraulic 0.01

Programmable Valves
Error (rad)

valve system, United States Patent, Re. 33,846, 1990. 0.005

[59] R. Book and C.E. Goering, Programmable 0

electrohydraulic valve, SAE Transactions, vol.108, no. −0.005

2, 346–352, 1999. −0.01

0 2 4 6 8 10 12 14 16 18 20
Time (sec)
[60] Fanping Bu and Bin Yao, Nonlinear adaptive
robust control of hydraulic actuators regulated by Figure 12: Tracking performance comparison
proportional directional control valves with deadband Cylinder Pressure Comparison
and nonlinear flow gain coefficients, Proc. of American 8000
Pressure (KPa)

PDC Valve

Control Conference, pp. 4129–4133, Chicago, IL, 2000. 6000

4000

2000

0
0 2 4 6 8 10 12 14 16 18 20

8000
Pressure (KPa)

Servo Valve
6000

4000

2000

0
0 2 4 6 8 10 12 14 16 18 20

8000
Pressure (KPa)

Programmable Valves
6000

4000

2000

0
0 2 4 6 8 10 12 14 16 18 20
Time (sec)
Figure 13: Cylinder working pressures comparison

Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii Grant #0600516
 Comparison of Energy Usage Comparison of Energy Usage with Load Sensing Pump
8 8
PDC Valve PDC Valve
Power (KW)

Power (KW)
6 6

4 4

2 2

0 0
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20

8 8
Servo Valve Servo Valve
Power (KW)

Power (KW)
6 6

4 4

2 2

0 0
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20

8 8
Programmable Valve Programmable Valve
Power (KW)

Power (KW)
6 6

4 4

2 2

0 0
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (sec) Time (sec)
Figure 14: Energy usages (fixed supplied pressure Figure 15: Energy usages (load sensing pump)
pump)

Proceedings of 2009 NSF Engineering Research and Innovation Conference, Honolulu, Hawaii Grant #0600516