Available online at www.sciencedirect.

com

Journal of the Franklin Institute 353 (2016) 3021–3039

www.elsevier.com/locate/jfranklin

A continuous version of the LuGre friction model

applied to the adaptive control of a pneumatic
servo system
M.R. Sobczyka,n, V.I. Gervinib, E.A. Perondia, M.A.B. Cunhac
a
Departamento de Engenharia Mecânica, Universidade Federal do Rio Grande do Sul,
Rua Sarmento Leite, 425, Centro. Porto Alegre, RS, Brazil
b
Centro de Ciências Computacionais, Universidade Federal do Rio Grande,
Av. Italia, km. 8, Campus Carreiros, Rio Grande, RS, Brazil
c
Instituto Federal de Educação, Ciência e Tecnologia Sul-rio-grandense,
Praça Vinte de Setembro, 455, Pelotas, RS, Brazil

Received 30 October 2015; received in revised form 28 March 2016; accepted 1 June 2016
Available online 8 June 2016

Abstract

This work presents the application of a friction compensation scheme to the trajectory-tracking control of
a pneumatic servo actuator. Such scheme is based on a continuous approximation of the LuGre friction
model, developed so as to allow complete Lyapunov stability analyses without resorting to assumptions that
are difficult to satisfy in practice due to their physical meaning. By using adaptive estimation, extensive
identification procedures are also avoided for determining friction parameters. Experimental results illustrate
the most significant advantages and potential limitations of the proposed scheme in real applications.
& 2016 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.

1. Introduction

Pneumatic actuators are an interesting choice for many industrial applications because, while
presenting good ratios between size/weight and available forces/torques, these systems are

n
Corresponding author.
E-mail addresses: [email protected] (M.R. Sobczyk), [email protected] (V.I. Gervini),
[email protected] (E.A. Perondi), [email protected] (M.A.B. Cunha).

http://dx.doi.org/10.1016/j.jfranklin.2016.06.003
0016-0032/& 2016 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.
3022 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039

usually cheap and easy to assemble. Also, they are easy to integrate in industrial facilities (which
usually have compressed-air supply lines as a standard feature) and environment-friendly,
because air leakages do not pose serious pollution hazards. On the other hand, as their dynamic
behavior is highly nonlinear, they are poorly suited for performing high-precision tasks. As a
result, since their application advantages are significant, much effort has been spent in finding
ways to monitor these systems effectively [1–9].
Friction is one of the most difficult phenomena to deal with in high-precision applications
[10–12], because it is the source of strongly nonlinear effects such as pre-sliding displacement,
break-away force variations, and stick-slip motion [13]. Due to its relevance, this is a widely
addressed problem [2,6,14–23], with model-based compensation techniques appearing as a
frequent approach [2,6,15,18,20–23]. Several models for this purpose can be found and there is
no consensus regarding which is the most suitable one [24], but the LuGre model [13] seems to
be the most frequent choice (see [2,6,18,22,25–29], among many other examples), being even
considered as “the reference model” for representing dry-friction forces [16].
In spite of its usefulness in many applications, the LuGre model has been criticized for flaws
both in its theoretical properties and in its practical usage [9,16,24,30–32]. More specifically, there
are two points of interest in this work: (i) the experimental identification of the parameters
employed by the model is a difficult task [14,16,22,26,32] and (ii) the model depends on
discontinuous mathematical terms, which limit its application for certain classes of systems
[6,22,33,34]. In the first case, direct measurement of the necessary variables requires sophisticated,
high-resolution equipment [26,32,35], and measurements can be affected by variables that are not
accounted for in the structure of the model, such as temperature and loading conditions [19].
Therefore, many works rely on time-consuming, indirect identification approaches [9,36–38] and/
or combined simulation-experiment procedures, using parameter values ranging within the same
orders of magnitude that are usually found in works based on directly measured data [13,22,39].
Regarding the second problem, the discontinuous terms represent a major hindrance when it is
explicitly required to estimate the time derivative of the friction force, a situation that is common to
fluid-driven servo actuators [2,6,22,33,40,41] and may also occur in systems whose actuators'
internal dynamics are significant [17,18]. This difficulty affects directly the stability guarantees that
can be provided for the closed-loop systems when model-based controllers are employed, leading
many authors to provide analysis only for restricted cases [6,33,42,43] or to apply regularization
methods that alter basic properties of the LuGre friction model, such as boundedness and passivity
[12,22,33,44]. In some cases, these measures lead to assumptions that are difficult to satisfy in the
practical implementation of the proposed algorithms. In [6,33], for instance, the time derivative of
the friction force is neglected, which is not a valid hypothesis for low velocities. The control
method discussed in [42] uses a discontinuous and potentially unbounded term for representing
friction effects in its Lyapunov-based stability analysis. The analysis given in [22] relies on the
existence of an upper bound to the amplitude of the internal state that is employed for estimating
the friction force, which is not true for the smooth approximation of the LuGre model that is used
to that end. In [41], the terms related to viscous friction effects are taken to be null, which contrasts
with the nature of the hydraulic fluid that is usually employed in this type of actuator. Finally, the
analysis presented in [2] is built on the assumption that the friction force estimated by the algorithm
actually converges to its corresponding value in the experimental system, without presenting any
theoretical or experimental arguments to back up such claim.
In [40], we introduced a continuous approximation of the Lugre friction model. Such
approximation differs from other ones in two critical points, which were proven in that work:
(i) its internal state zðt Þ is bounded and (ii) with an appropriate choice of its parameter values the
 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039 3023

input–output mapping defined by such approximation is passive. Here, we extend that work by
employing such approximation with an adaptive control algorithm to be applied to trajectory-
tracking problems in a pneumatic servo positioner. The main contribution presented in this paper
lies in allowing the use of the LuGre model with fluid power actuators without relying on
questionable assumptions or affecting any aspect of a formal, complete stability analysis of the
closed-loop system. Also, since it is an adaptive controller, we avoid extensive efforts to identify
the parameters of the friction model. Finally, experimental results are used to evaluate the main
advantages and limitations of the proposed strategy in real applications. To the best of our
knowledge, this is the first work to address these three features simultaneously.
This work is organized as follows. Initially, in Section 2, the mathematical model of the
system is presented. The proposed control algorithm is discussed in Section 3, whereas the
stability analysis of the closed loop system is the focus of Section 4. Section 5 is dedicated to
presenting and discussing experimental results. Finally, the main conclusions are given in
Section 6.

2. System model

A typical pneumatic servo actuator is depicted schematically in Fig. 1. Its pneumatic source is
assumed to provide a constant pressure Ps to a symmetric actuating cylinder with utile stroke L
and piston area A. An electric servovalve regulates the flow rates of compressed air moving into
and out of the chambers of the cylinder. The air passage area of the valve is directly proportional
to its electric control voltage u. The dynamic model for this type of system is discussed at length
in many works (see, for instance, [6] or [9]), and the details of the specific model that is used in
this work can be found in [22,45].
The operation of this system can be understood as two interconnected subsystems:
(i) mechanical and (ii) pneumatic. The first one is regarded as a mass M subject to a friction

Fig. 1. Pneumatic positioning system.

3024 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039

force f and driven by a pneumatic force due to the differences in the pressures p1 and p2 in the
chambers of the cylinder. Thus, considering that M encompasses the mass of the entire piston–
load assembly, we apply Newton's Second Law to obtain
M y€ þ f ¼ Aðp1  p2 Þ ð1Þ
where y€ is the second derivative with respect to time of the piston position y, i.e., the
acceleration. Accordingly, the piston velocity is defined as y_ and the time derivative of the
…
acceleration, commonly referred to as jerk, is denoted as y . The friction force f will be discussed
separately in Section 2.1.
The second subsystem is composed of the equations that govern the dynamics of the chamber
pressures p1 and p2. These equations can be written with basis on energy conservation arguments
and are given by:
Ary_ rRT
p_1 ¼  p1 þ q ð2Þ
V 10 þ Ay V 10 þ Ay m1

Ary_ rRT
p_2 ¼ p þ q ð3Þ
V 20 þ AðL  yÞ 2 V 20 þ AðL  yÞ m2
where r is the ratio between the specific heat values of the air, R is the gas constant, T is the air
supply temperature, y is the piston position, V10 and V20 are the dead volumes of air at the
extremities of both chambers, and qm1 and qm2 are the mass flow rates of air into or out of each
chamber. These rates are nonlinear functions of the pressures in the chambers and of the voltage
signal u, whose effects are modeled by empirically determined polynomial approximations, as
described in Appendix A.
As the piston moves due to the differences of pressure between the two chambers, it is
convenient to rewrite Eqs. (1)–(3) in terms of a differential pressure pΔ ¼ p1  p2 . We accomplish
this by subtracting Eqs. (2) and (3) and rearranging terms, obtaining:
^ 1 ; p2 ; y; y_ Þ þ uðp
p_ Δ ¼ hðp ^ 1 ; p2 ; u; yÞ ð4Þ
^ 1 ; p2 ; y; y_ Þ and uðp
where hðp ^ 1 ; p2 ; u; yÞ are grouped terms that are functions only of the states of
the system and those that are affected by the input voltage u, respectively. Such terms are defined
as:
 

 qm1 qm2
u^ p1 ; p2 ; u; y ¼ RrT  ð5Þ
V 10 þ Ay V 20 þ AðL  yÞ
 

 p1 p2
h^ p1 ; p2 ; y; y_ ¼  rA_y  ð6Þ
V 10 þ Ay V 20 þ AðL  yÞ

Remark 1. From Eq. (4), it is clear that the control signal applied to the system ðuÞ does not
affect directly the force applied to the piston–load assembly. Instead, it affects the dynamics of
the pressures in the chambers, i.e., the time derivative of the applied pneumatic force. Therefore,
the proposed feedback-linearization control algorithm requires an explicit estimate of the time
derivative of the friction force, because this is the quantity that is directly related to the control
input. This topic is discussed in more detail in Section 2.1.
 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039 3025

2.1. Friction model

In this work, we use a continuous approximation of the LuGre model to account for the
friction forces in the system. The original LuGre model was proposed by Canudas de Wit et al.
[13], being well known and widely used in control applications because of its ability to represent
most friction effects with satisfactory accuracy. The model relies on estimating an average
deflection z of the microscopic asperities (bristles) of two contacting surfaces, which is
accomplished by means of the following nonlinear state observer:
_
jyj
z_ ¼ y_  z ð7Þ
gðy_ Þ
where gðy_ Þ is a positive function defined as [13]:
1h i
F C þ ðF S  F C Þe  ðy_ =_y S Þ
2
gðy_ Þ ¼ ð8Þ
s0
where FC is the Coulomb friction force, FS is the static friction force, and y_ S is the Stribeck
velocity. The meaning of each of these quantities is illustrated in Fig. 2, which depicts the
friction force as a static function of velocity.
Based on the state z, the friction force f acting between two bodies with relative velocity y_ is
represented as:
f ¼ s0 z þ s1 z_ þ s2 y_ ð9Þ
where s0 is a stiffness coefficient, s1 is a damping coefficient related to z_ , and s2 is a viscous
friction coefficient, which is also illustrated in Fig. 2.  
As discussed in further detail in [40], due to the presence of the term y_ ðt Þ, whose time
derivative is discontinuous when y_ ðt Þ ¼ 0, application of the LuGre model becomes seriously
hindered in control tasks involving actuators whose internal dynamics cannot be neglected, i.e.,
the control input u affects the time derivative of the force instead of the force itself, which is the
case for fluid-driven servo actuators. This occurs because, for such systems, it is necessary to
estimate the time derivative of the friction force, which cannot be done for all operation
conditions due to the  aforementioned
 discontinuity. In many works, this problem is avoided by
replacing the term y_ ðt Þ by “intuitive” continuous approximations, such as 2_y arctanðkv y_ Þ=π or
y_ tanhðk v y_ Þ [12,22,33], where kv is a positive constant in both cases. However, this procedure
causes the loss of two of the most important properties of the LuGre model, namely,
boundedness of its internal state z(t) and passivity. So, in order to avoid these problems in
applying the LuGre model to the control of a pneumatic actuator, we use the following

Fig. 2. Friction force f as a function of piston velocity v.

3026 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039

continuous approximation:
S2 ðy_ Þ
z_ ¼ y_ S1 ðy_ Þ z ð10Þ
gðy_ Þ
where S1 ðy_ Þ and S2 ðy_ Þ are respectively defined as:
(
S1 ðy_ Þ ¼ ½S0 ðy_ Þ2
ð11Þ
S2 ðy_ Þ ¼ y_ S0 ðy_ Þ

and S0 ðy_ Þ is any odd, continuous and monotonically increasing function whose absolute value
converges asymptotically to 1 as j_y j-1. Here, we use S0 ðy_ Þ ¼ 2arctanðkv y_ Þ=π. Even though this
version of the LuGre model differs from other ones in minor details, it presents an important
advantage since it can be proven to maintain the properties of boundedness and passivity. For the
complete proofs of these two properties, refer to [40].
The calculation of the friction force itself is not modified, i.e., Eq. (9) is still employed in its
original form. As for its corresponding time derivative f_ , it can be readily determined as:
f_ ¼ s0 z_ þ s1 z€ þ s2 y€ ð12Þ
where the term z€ can be calculated by taking the time derivative of Eq. (10):
z€ ¼ S_ 1 y_ þ S1 y€  ðΓz
_ þ Γ_z Þ ð13Þ
with Γ ðy_ Þ ¼ S2 ðy_ Þ=gðy_ Þ. The derivatives of the intermediate terms involved in this equation are
given by the following expressions:
S_ 1 ¼ 2S0 S_ 0 ð14Þ

2kv
S_ 0 ¼   y€ ð15Þ
π 1 þ ðkv y_ Þ2

S_ 2 g  S2 g_
Γ_ ¼ ð16Þ
g2 ð_y Þ

S_ 2 ¼ S_0 y_ þ S0 y€ ð17Þ

 2ð F s  F c Þ
y_ y€ e  ðy_ =_y s Þ
2
g_ ¼ ð18Þ
s0 y_ s
2

In sum, the friction forces acting on the piston of the actuating cylinder are calculated by
means of Eqs. (9)–(12), with the aid of their corresponding auxiliary terms. This completes the
open-loop model.

3. Control algorithm

The proposed controller is based on the cascade control methodology, which consists in
interpreting the whole system as two interconnected subsystems, mechanical and pneumatic, in
which the output of one subsystem (the differential pressure resulting from the pneumatic
dynamics) is regarded as an intermediate input that is applied to the other. Thus, regarding first
the mechanical subsystem, we develop a control law that generates a desired differential pressure
pΔd that leads the system to track a given desired trajectory yd . Then, we choose another control
 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039 3027

law to be applied to the pneumatic subsystem, which calculates the values of the servovalve
control voltage u that cause the actual pressure difference pΔ to track pΔd as closely as possible.
The advantage of this procedure lies in the fact that the two control laws are not directly tied to
each other, so that the control choices for one subsystem can be made without addressing the
other one explicitly.
The proposed control law to be applied to the mechanical subsystem is a version of the well
known one by Slotine and Weiping [46], which was augmented by Xie [21] to include an
adaptive friction-compensating term. In order to describe the proposed control algorithm, we
now define a number of auxiliary quantities. First, the trajectory tracking errors of the piston–
load assembly in terms of position, velocity, acceleration and jerk are given by, respectively:
… … …
y~ ¼ y  yd ; y~_ ¼ y_  y_ d ; y~€ ¼ y€  y€ d ; y~ ¼ y  y d ð19Þ
…
where yd, y_ d , y€ d , and y d are the desired values for their corresponding variables. Additionally,
the auxiliary error measure s is established as
s ¼ y~_ þ λ~y ð20Þ
With the aid of these terms, the control laws applied to the mechanical and pneumatic
subsystems are, respectively:

ApΔd ¼  K D s þ M y€ d  λy~_ þ f^ ð21Þ


u^ ¼  h^ p1 ; p2 ; y; y_  As  K P p~ Δ þ p_ Δd ð22Þ
where KD and KP are positive constants, p~ Δ ¼ pΔ  pΔd is the error in the differential pressure,
^ 1 ; p2 ; y; y_ Þ is the part of the pneumatic subsystem dynamics that is independent of the control
hðp
signal u(t) in Eq. (4), and f^ is the observed value of the friction force, i.e.:
f^ ¼ s^ 0 z^ þ s^ 1 z^_ þ s^ 2 y_ ð23Þ
In Eq. (23), all terms marked with a wedge ð 4 Þ are regarded as the estimated counterparts of
their corresponding quantities in the real system, as modeled in Section 2.1. The dynamics of the
estimated friction state z^ is:
S2 ð_y Þ
z^_ ¼ S1 ðy_ Þ_y  z^  μ0 s ð24Þ
f s ð_y Þ
Calculation of friction forces with the LuGre model requires six parameters: s0, s1, s2, FC, FS,
and vs. The first two are called dynamic parameters, and their experimental identification is a
rather difficult task [14,16,22,26,32]. The remainder are called static, because they can be easily
identified in a static friction map such as the one illustrated in Fig. 2, which can be drawn from
data obtained in relatively simple experiments as described in Section 5. Here, in order to avoid
the experimental identification of the dynamic friction parameters of the model, we employ
adaptive control, i.e., these parameters are updated on line as the task is performed. Adaptive
control is a widely addressed research topic, with numerous examples of different techniques and
applications (see, for instance, [47–51]). In our case, the adaptation scheme is based on the one
presented in [21], which estimates the dynamic parameters of the LuGre model in the case of a
DC servomotor. According to such scheme, the update rules for the estimated parameters are:
s^_ 0 ¼  γ^z s; s^_ 1 ¼  βz^_ s; s^_ 2 ¼  χ y_ s ð25Þ
where μ0, γ, β, and χ are positive constants.
3028 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039

Fig. 3. Proposed control scheme.

The complete procedure performed by the proposed algorithm is summarized in Fig. 3: based
on the measurement of the mechanical variables of the system (position, velocity and
acceleration), the control law given in Eq. (21) determines an appropriate value for the desired
differential pressure pΔd . Then, a second control law (Eq. (22)) is calculated, seeking to force the
actual pressure of the system to track such desired value. Finally, as such control law u^ does not
express the actual control voltage u applied to the servovalve, a nonlinear inversion procedure is
used to calculate this value. This inversion consists in solving Eq. (5) for u, using the desired
value of u^ and the polynomials given in Appendix A that represent the air mass flow rates going
through the servovalve ports.

4. Stability analysis

In this section, we analyze the convergence properties of the closed-loop tracking errors in the
system with basis on Lyapuonv's direct method. Prior to the analysis itself, in order to clarify its
development, we rearrange terms in some of the relations presented in the previous sections.
First, from the way that all tracking errors in the system are defined, we can write pΔ ¼ p~ Δ þ pΔd .
Thus, substitution of expressions (9), (21) and (23) into Eq. (1) yields:


M s_ ¼ K D s þ s^ 0 z^ þ s^ 1 z^_ þ s^ 2 y_  ðs0 z þ s1 z_ þ s2 y_ Þ þ Ap~
Δ ð26Þ
Also, as in the case of the previously defined variables, we denote the estimation error of the
friction state z as z~ ¼ z^  z. Thus, by subtracting Eq. (10) from Eq. (23), the time derivative of
such error can be shown to be:
S2 ð_y Þ
z~_ ¼  z~  μ0 s ð27Þ
f s ð_y Þ
Finally, we apply the control law given in Eq. (22) to the dynamics of the pneumatic
subsystem (Eq. (4)), obtaining
p~_ Δ ¼  As  K P pΔ ð28Þ
It is now possible to analyze the convergence properties of the trajectory-tracking errors of the
closed-loop system, which can be stated by means of Proposition 1.
Proposition 1. Consider the pneumatic positioning system described in Section 2. For this system,
h iT
we define ρt ¼ y~ ; y~_ ; p~ Δ as the trajectory-tracking error vector, ρe ¼ ½z~ ; s~ 0 ; s~ 1 ; s~ 2 T as the
 T
estimation error vector, and ρg ¼ ρTt ; ρTe as the global error vector. Assuming that the
desired values for the position, velocity, acceleration and jerk of the actuating piston are all
bounded, when this system is controlled in closed-loop form by means of the algorithm proposed in
 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039 3029

Section 3, it is possible to choose the parameter values for such controller so as to ensure that
ρg ¼ 0 is a stable equilibrium. Furthermore, such choice of parameters leads to J ρt J -0 as t-1,
i.e., the trajectory-tracking errors of the closed-loop system converge asymptotically to zero.

Proof. Consider the following non-negative function:

 
1 1 1 1 s0
V ðt Þ ¼ Ms2 þ s~ 20 þ s~ 21 þ s~ 22 þ z~ 2 þ p~ 2Δ ð29Þ
2 γ β χ μ0
by differentiating Eq. (29) and using Eqs. (25)–(28), we can show that

 s0 S2 ð_y Þ 2 s1 S2 ð_y Þ
V_ ðtÞ ¼  K D þ s1 μ0 s2  K P p~ 2Δ  z~  z~ s ð30Þ
μ0 gð_y Þ gð_y Þ


Now, applying the inequality ab r 12 a2 þ b2 to the term that depends on the product z~ s and
rearranging terms, we obtain:
V_ ðt Þr  βs s2  K P p~ 2Δ  βz z~ 2 ð31Þ
where βs and βz are, respectively:
8
< βs ¼ K D þs1 μ0  s
> _ Þ=½2gðy_ Þ
1 S2 ð y
S2 ð_y Þ s0 s1 ð32Þ
: βz ¼ f ð_y Þ μ  2
>
s 0

Thus, V_ ðt Þ can be made non-positive if

8
< μ0 o2s0 =s1
>
S2 ðy_ Þ ð33Þ
: K D 4s1 2gðy_ Þ  μ0
>

Since S2 ð_y Þ and gð_y Þ are bounded, the aforementioned values of μ0 and KD exist. Thus,
_
V ðt Þr 0 and this equilibrium is stable, i.e., Jρt J and Jρe J are both limited. Moreover, as the
dependence on s and p~ Δ is negative for any nonzero combination of these quantities, both must
tend to zero as t-1. However, this is not true for z~ , as its multiplying factor in Eq. (30) depends
on y_ , nor the remainder elements of ρe , because they do not appear in V_ ðt Þ.
As s ¼ y~_ þ λ~y can be interpreted as a first order filter applied to y~ , and since s converges
asymptotically to zero, it follows that both y~_ and y~ must also converge asymptotically to zero,
completing the proof. □

Remark 2. The second condition stated in Eq. (33) can be difficult to attain in practice because,
even for moderate velocities (about 0.1 m/s), the lower bound required for KD is roughly equal to
s0 s1 =100, which can often reach orders of magnitude of 103 or higher [15,22,37,42,43],
implying very large feedback gains in the practical implementation of the controlled system.
Even if the control apparatus is able to apply such values in practice, it is undesirable to do so
because that can render the system to be very sensitive to measurement noise or unmodeled
dynamic effects. This particular topic will be discussed with more detail in Section 5.
3030 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039

5. Experimental evaluation

The experiments were performed on a pneumatic workbench comprising a rodless, double-

acting cylinder (DGPL-1000 FESTO) and a 5/3 servovalve (MPYE-5-1/8 FESTO). A variable-
resistance sensor (MLO-POT-1000-TLF FESTO) was used for measuring piston position, whereas
velocity and acceleration were estimated by numerical differentiation. Chamber pressures were
monitored by membrane-type transducers (510-Huba Control). All signals were fed to a real-time
control electronic board (DS1104-dSPACE GmbH), hosted by a standard PC, running the
algorithm described in Section 3. The processing time for each control loop was 1.0 ms.
According to its manufacturer's catalog, the nominal parameters values for this workbench are:
A ¼ 4:19  10  4 m2 , L ¼ 1 m, V 10 ¼ 1:96  10  6 m3 , and V 20 ¼ 4:91  10  6 m3 . The parameters
for compressed air were assumed as r ¼ 1:4, R ¼ 286:9 J kg=K, T ¼ 293:15 K, as commonly found
in other works on the subject [22,9,8]. The payload mass was M ¼ 3:66 kg. As for the admissible
trajectory requirements, the maximum stroke and velocity for this pneumatic cylinder are 0.9 m and
1.0 m/s, respectively. Also, due to the continuity conditions implied in the stability analysis
presented in Section 4 [52], all required position trajectories and its derivatives up to the 3rd order
must be bounded continuous functions of time. So, if a discontinuous trajectory (e.g., a step input) is
required, it is necessary to filter its corresponding signal before applying it to the system.
As already mentioned in Section 3, LuGre model parameters can be classified in two
categories: static and dynamic. Those in the first group can be experimentally identified by the
following procedure: using a simple algorithm such as P- or PI-controller, the actuator is led to
track a constant velocity trajectory while the pressure difference applied to the piston is
measured. Since the velocity is constant, this pressure difference multiplied by the piston area
gives the friction force. Repeating the test for several speed values leads to a static map such as
the one depicted in Fig. 2, from which the following parameters were determined:
s^ 2 ¼ 89:9 N s=m, F C ¼ 64:3 N, F S ¼ 75 N, vs ¼ 0:02 m=s. As the parameters in the second
group are very difficult to determine experimentally, their values were assumed as unknown,
being estimated on line as described in Section 3. The workbench used here is of the same
manufacturer and model as the one employed in [22], which deals with friction compensation in
a pneumatic actuator using the LuGre model with fixed parameters. Thus, the initial values for
our dynamic friction parameters were the same as those used in that work: s^ 0 ¼ 4500 N=m,
s^ 1 ¼ 93:1 Ns=m.
Due to its cascaded structure, the proposed controller allows a clear separation of the effects of
each control law: while Eq. (21) relates an intended position/velocity trajectory to its
corresponding desired pressure pΔd , Eq. (22) is concerned with causing the actual differential
pressure pΔ to match such desired value. Likewise, the controller gains KD and λ have a direct
effect on the tracking of the mechanical variables (y and y_ ) of the system, whereas KP affects
mostly the applied pressure pΔ . Since our focus is on the effect of the proposed friction
compensation scheme on the mechanical trajectory performed by the system, only the gains
referring to the mechanical subsystem were varied, whereas the pneumatic subsystem gain was
set as K P ¼ 150 s  1 . The tests were repeated for five different sets of controller gains, as
informed in Table 1. These sets were adjusted by means of approximate transfer function
considerations combined with direct experiments, and the chosen values reflect an increasing
degree of required control effort.
In each test, the system was required to track a sinusoidal trajectory with amplitude 0.4 m and
frequency 2.0 rad/s. Because the proposed friction compensating scheme is adaptive, in order to
highlight its specific effects, the time span for each test was 5 min (300 s). Initially, the friction
 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039 3031

compensating term of the control law (Eq. (23)) was turned off. Then, this term was turned on, with
fixed dynamic parameters. Finally, parameter values were allowed to adapt with time as determined
in Eq. (25). The results from the experiments are illustrated in Figs. 4–6. Due to the relatively long
test period when compared to that of each movement cycle, to facilitate the visualization of the
results, Fig. 4 depicts only a few cycles of interest at the beginning and at the end of one of the tests,
whereas the complete picture is given in the other two graphs. Also for the sake of facilitating their
understanding, the results in these figures are summarized in periods of 15 s.
As seen in Figs. 4 and 5, the compensation term with nominal (initial) parameters caused
overcompensation of the friction forces and the amplitudes of the tracking errors were increased.
However, as the adaptation scheme acted, such errors were reduced until reaching approximately

Table 1
Gain sets employed during the experimental tests.

Set 1 2 3 4 5

λ(s  1) 50 60 70 80 90
KD (N s/m) 10 15 20 25 30

Fig. 4. Trajectory tracking performance of the closed-loop system – general view: (a) initial stages and (b) final stages.
For 0rto15 s, without friction compensation, stiction always caused the piston to stop before reaching the end of its
desired trajectory. As the compensator was turned on with nominal parameters (t¼15 s), the piston was able to reach the
entire trajectory, but at the price of overcompensation. As the parameters were adapted, such overcompensation was
progressively reduced and the overall tracking performance was enhanced.
3032 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039

Fig. 5. Position tracking errors in the controlled system along 300 s (20 periods of 15 s), for the 5 sets of controller gains
given in Table 1: (a) peak errors and (b) rms errors. In the first period (0–15 s), no friction compensation was used. During
the second period (15–30 s), friction compensation was used with nominal parameters. Starting from t¼30 s, friction
parameters were adapted with time. After long periods of operation in adaptive mode, the peak errors obtained by using
the friction compensation term were just slightly smaller than those achieved without it, but the rms errors were clearly
reduced, thus improving the overall trajectory tracking performance.

constant amplitudes, the smallest of which was 2.9 cm, representing 7.3% of the amplitude of the
desired trajectory. Moreover, due to the adaptation of the friction parameters, the performed
trajectory became significantly closer to its reference, a fact confirmed by the expressive
reduction of the rms tracking errors. When compared to traditional control approaches, the
proposed control scheme led to a clear improvement in the trajectory tracking performance of the
controlled system. With linear state feedback control, for instance, the obtained tracking error
amplitude for the same trajectory was about 6 cm, or 13.3% of relative error. Direct comparison
with other works is more difficult because there is significant variation in experimental
apparatuses. However, in those examples where experimental setups and trajectory requirements
were similar to ours, the reported relative errors ranged from about 7.5% to 11% [6,22,53–55].
Therefore, although such comparison must be considered just in qualitative terms, it indicates
that the proposed controller can be regarded as a suitable approach.
Addressing specifically the largest observed errors, Fig. 4 clearly shows that they occurred
mostly at the moments of trajectory reversal, regardless of the use of friction compensation. This
feature can be explained by stiction: when compensation was not used, at near-zero velocities,
the sudden rise in friction force caused the piston to stop short of its desired position; with the aid
of the compensation scheme, the controller was able to track friction disturbances more closely
 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039 3033

Fig. 6. Average values of the estimated friction parameters during the tests: (a) s^ 0 ðt Þ, (b) s^ 1 ðtÞ, and (c) s^ 2 ðt Þ. The periods
depicted here are the same shown in Fig. 5.

as velocity was reduced. However, after the piston stopped, it took relatively long for it to move
once again. The main reason for such delay lies in the structure of the model (see Eqs. (10) and
̇
(11)): since y ¼ 0, the value of the estimated friction state z(t) remains fixed. Therefore, the
estimated friction force that is used in the mechanical subsystem control law (Eq. (21)) cannot
match the variations of the actual one unless the velocity is nonzero. This is a fundamental
limitation not only of our version of the LuGre model, but of the original one as well. To
overcome it, it is necessary to add an entirely new term to the structure of the model, allowing for
the update of the estimated friction force while the body remains still. Here, this necessity is
somewhat abated by estimating the time derivative of the friction force, which is part of the
3034 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039

control law applied to the pneumatic subsystem (Eq. (22)) and depends on piston acceleration,
but the problem clearly persists.
The values of the estimated friction parameters over time during the tests are depicted in
Fig. 6. In most cases, such estimates tended to stabilize with time, but two important exceptions
must be pointed out. First, for higher gain values, the estimates for s^ 0 ðt Þ and s^ 2 ðt Þ behaved much
differently from the cases with reduced gains. In particular, s^ 2 ðt Þ for gain set 5 showed signs of a
possibly unstable behavior, because its value decreased until it saturated at its inferior bound (all
three s^ i are lower-bounded by zero because of their physical meaning). Second, s^ 1 ðt Þ was
monotonically increasing in all tests (even for gain set 5, although much more slowly), which is
also an evidence of possible instability. From the analysis carried out in Section 4, this result may
seem unexpected, because it was proven that all tracking errors and estimated parameters should
be limited quantities. However, such proof relies on the existence of a lower bound for KD,
which, for the values of the parameters used in this work, should be of an order of magnitude of
103 (see Eq. (33) and Remark 2). In practice, however, this gain could be hardly set at values
greater than about 100 Ns/m, because of the appearance of intense oscillations in the response of
the controlled system. These oscillations may be related to unmodeled dynamics in the controlled
plant, but there is no doubt that they are largely affected by measurement noise, which, in the
case of the employed workbench, is seriously enhanced by the use of numeric differentiation for
obtaining the velocity and acceleration signals. Due to this problem, the gains of the controller
had to be limited to the relatively narrow range given in Table 1, which reflects a compromise
between avoiding excessive oscillations and obtaining acceptable tracking-error amplitudes.
Noise is ubiquitous to any experimental apparatus, and even more so in common industrial
environments, representing a problem that can only be mitigated at very large costs in terms of
equipment, time and effort. Thus, even in those uncommon cases when the restrictions imposed
by the control hardware can be neglected, limiting gains to avoid noise amplification is always
necessary. This is where a major drawback in using the LuGre model appears: since it emulates
the elastic behavior of microscopic contact elements, its dynamic parameters tend to present
rather large values: in the closely related case of a hydraulic actuator, for instance, the elastic
deformation of seals lead to pre-sliding displacements of a few tenths of millimeter [19], which
are consistent with an order of magnitude of 105 N/m or higher for s0. In other works, typically
found values are about 104 N/m or 105 N/m for s0 and 102 N s/m for s1 [15,22,37,42,56].
Therefore, since the stability analyses for nonlinear systems are often based on inequalities that
depend on the values of their parameters (see [9,21,22,57–60], among many other examples),
when the LuGre model is used, it can be very difficult to attain in practice the gain values
required by such analyses.
It must be stressed that Lyapunov stability methods are based on sufficient conditions, not on
necessary ones. Thus, even if the usual values associated with the LuGre model cause such
analyses to fail in guaranteeing the stability of the controlled system, it does not follow that the
system is unstable. Moreover, the experimental results presented here indicate that, with smaller
feedback gains and an appropriate set of friction parameters, the proposed friction compensation
scheme tends to improve the performance of the controlled system. As for the potential
instability, since the only indication for this possibility lies in the behavior of the estimated
parameters, this difficulty can be overcome by employing other adaptation rules, or setting strict
limits for the variation range of their values. Finally, as already pointed out in Section 1, even
though the stability condition found here could not be implemented in practice, we must
emphasize that the use of the proposed approximation allowed the corresponding analysis to be
carried out in a complete and formal manner, without need to assume invalid or excessively
 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039 3035

restricted operation hypotheses. In contrast, all of the aforementioned works with similar
experimental results [6,22,53–55] suffer from more severe restrictions regarding this topic: the
first two examples rely on stability analyses based on physically unrealistic hypotheses, as
already discussed in Section 1, whereas the analysis presented in the third one does not take dry-
friction effects explicitly into account, and the last two discuss stability only in the context of
linear approximations. This reinforces the notion that just the possibility of performing a more
complete stability analysis of the system is already an important advance by itself. Therefore,
although there are some important issues to overcome, the overall performance of the proposed
control approach can be classified as promising.

6. Conclusions

The primary objective of this work was to propose a friction compensation scheme that
allowed the application of the LuGre friction model to fluid power-based actuators and still retain
the ability to produce a complete, formal stability analysis for the controlled system without
resorting to questionable assumptions regarding its operation conditions. Additionally, it was
intended to develop this compensation scheme in an adaptive manner, to avoid the experimental
determination of parameters that are difficult to measure. From a theoretical perspective, these
objectives were fulfilled. However, with respect to practical usefulness, even if some results
show potentially significant improvements, there are still some difficulties to overcome before
this algorithm may be considered as a fully attractive solution. Future works will focus on
avoiding such difficulties. Besides altering the adaptation rule employed in the proposed control
scheme, such works may also include further modifications in the structure of the LuGre model,
or the use of alternative models that are not as prone to the same hindrances.

Appendix A. Modeling of air flow rates

In this work, the air mass flow functions are modeled directly from experimental data, using
polynomial functions to represent the effects of the chamber pressures, p1 and p2, and of the cylinder
control voltage u. Such functions are normalized, 3rd order polynomials, whose coefficients are
adjusted so as to present only one real root within the interval [0,1]. The objective is to facilitate the
calculation of the inverse function uðu^ Þ, because such root can be easily determined from well-known
analytic relations. The general structure of these functions is given by:
m ðpu ; pd ; uÞ ¼ ½qm max f p ð pu ; pd Þf u ðuÞ
qproc ðA:1Þ
where pu and pd are, respectively, the upstream and downstream pressures with respect to the valve
orifice, ½qm max is the maximum air mass flow value identified during the experiment, f p ðpu ; pd Þ is the
polynomial function that models the effect of the pressures, and fu(u) is the polynomial that represents
the influence of the control input u. The index proc refers to the process being carried out in the
chamber of interest: filling (fil) or exhausting (exh). For instance, for a positive control input, chamber
1 is filled while chamber 2 is exhausted. Under such conditions, for the valve orifice that regulates the
flow into chamber 1, the upstream pressure is equal to the supply pressure Ps, whereas the
downstream pressure is p1. Likewise, for chamber 2, the corresponding values for these pressures are
p2 and the atmospheric pressure P0. As Ps and P0 are constant, they do not need to be mentioned
explicitly in the equations. Thus, the expressions for qproc1
m1 and qproc2
m2 are, respectively:
qfil fil fil
m1 ðp1 ; uÞ ¼ ½qm1 max f p ð p1 Þf u1 ðuÞ
fil
ðA:2Þ
3036 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039

m2 ð p2 ; uÞ ¼ ½qm2 max f p ð p2 Þf u2 ðuÞ

qexh ðA:3Þ
exh exh exh

Following this notation, the complete list of experimentally determined polynomials is:

p1 ðp 1 Þ ¼  1:888p 1 þ 1:157p 1  0:191p 1 þ 1:007

f fil ðA:4Þ
3 2

p1 ðp 1 Þ ¼ 0:346p 1  1:747p 1 þ 2:312p 1 þ 0:09

f exh ðA:5Þ
3 2

p2 ðp 2 Þ ¼  2:395p 2 þ 1:851p 2  0:362p 2 þ 1:014

f fil ðA:6Þ
3 2

p2 ðp 2 Þ ¼  0:029p 2  0:857p 2 þ 1:88p 2 þ 0:109

f exh ðA:7Þ
3 2

u1 ðuÞ ¼ 1:2u  3:5u þ 3:58u  0:3

f fil ðA:8Þ
3 2

u1 ðuÞ ¼ 1:265u  3:282u þ 3:148u  0:112

f exh ðA:9Þ
3 2

u2 ðuÞ ¼ 1:198u  3:446u þ 3:42u  0:161

f fil ðA:10Þ
3 2

u2 ðuÞ ¼ 1:2u  3:5u þ 3:6u  0:3

f exh ðA:11Þ
3 2

All these functions employ normalized values for its corresponding variables, i.e., u ¼ u=umax ,
where umax is the maximum input voltage that can be applied to the servovalve and
p i ¼ ðpi  P0 Þ=ðPs  P0 Þ. This procedure facilitates the calculation of these functions for cases
with different values of supply pressure or servovalves with other supply voltage definitions.
With these functions, the entire inversion procedure is carried out as follows. First, with the
desired value for the control law u^ and using Eq. (5), we calculate a weighted sum of the air mass
flow rates qm1 and qm2. Then, we substitute the polynomial functions that correspond to the
process determined by the desired control law, which result in a new 3rd order polynomial of u.
Finally, the calculation of the real root within the range [0,1] yields the value of the desired
control input to be applied to the servovalve.

References

[1] F. Abry, X. Brun, M.D. Loreto, S. Sesmat, Éric Bideaux, Piston position estimation for an electro-pneumatic
actuator at standstill, Control Eng. Pract. 41 (2015) 176–185, http://dx.doi.org/10.1016/j.conengprac.2015.04.
016 (URL 〈http://www.sciencedirect.com/science/article/pii/S0967066115000908〉).
[2] R.A. Rahman, L. He, N. Sepehri, Design and experimental study of a dynamical adaptive backstepping-sliding
mode control scheme for position tracking and regulating of a low-cost pneumatic cylinder, Int. J. Robust. Nonlinear
Control 26 (2016) 853–875, http://dx.doi.org/10.1002/rnc.3341.
[3] A. Estrada, F. Plestan, Second order sliding mode output feedback control with switching application to the control
of a pneumatic actuator, J. Frankl. Inst. 351 (4) (2014) 2335–2355. http://dx.doi.org/10.1016/j.jfranklin.2013.07.
011. (Special Issue on 2010–2012 Advances in Variable Structure Systems and Sliding Mode Algorithms. URL
〈http://www.sciencedirect.com/science/article/pii/S0016003213002846〉).
[4] B. Taheri, D. Case, E. Richer, Force and stiffness backstepping-sliding mode controller for pneumatic cylinders,
IEEE–ASME Trans. Mechatron. 19 (6) (2014) 1799–1809.
[5] F. Plestan, Y. Shtessel, V. Bregeault, A. Poznyak, Sliding mode control with gain adaptation application to an
electropneumatic actuator, Control Eng. Pract. 21 (5) (2013) 679–688.
[6] R.A. Rahman, N. Sepehri, Position tracking of a pneumatic actuator using backstepping-sliding mode control with
adaptive friction observer, in: ASME/BATH 2013 Symposium on Fluid Power and Motion Control, Sarasota,
Florida, USA, American Society of Mechanical Engineers, New York, NY, USA, 2013, p. V001T01A028.
 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039 3037

[7] R. Moreau, M.T. Pham, M. Tavakoli, M. Le, T. Redarce, Sliding-mode bilateral teleoperation control design for
master–slave pneumatic servo systems, Control Eng. Pract. 20 (6) (2012) 584–597.
[8] J. Weist, M.A. Arteaga, L.R. De la Cruz, H. Hebisch, Model free control for differential pneumatic pistons:
experimental comparison, Int. J. Control 84 (1) (2011) 138–164.
[9] K. Khayati, P. Bigras, L.-A. Dessaint, LuGre model-based friction compensation and positioning control for a pneumatic
actuator using multi-objective output-feedback control via LMI optimization, Mechatronics 19 (2009) 535–547.
[10] W. Yu, M.A. Moreno-Armendariz, Robust visual servoing of robot manipulators with neuro compensation, J.
Frankl. Inst. 342 (7) (2005) 824–838, http://dx.doi.org/10.1016/j.jfranklin.2005.06.003 (URL 〈http://www.
sciencedirect.com/science/article/pii/S001600320500058X〉).
[11] V. Lampaert, F. Al-Bender, J. Swevers, Comparison of model and non-model based friction compensation
techniques in the neighborhood of pre-sliding friction, in: Proceedings of the 2004 American Control Conference,
Boston, MA, 2004, pp. 1121–1126.
[12] J. Koopman, D. Jeltsema, M. Verhaegen, Port-hamiltonian formulation and analysis of the LuGre friction model, in:
47th IEEE Conference on Decision and Control, 2008, CDC 2008, 2008, pp. 3181–3186. http://dx.doi.org/10.1109/
CDC.2008.4739351.
[13] C. Canudas de Wit, H. Olsson, K.J. Astrom, P. Lischinsky, A new model for control of systems with friction, IEEE
Trans. Autom. Control 40 (3) (1995) 419–425.
[14] X. Wang, S. Wang, Output torque tracking control of direct-drive rotary torque motor with dynamic friction
compensation, J. Frankl. Inst. 352 (11) (2015) 5361–5379. (ISSN 0016-0032) (URL /http://www.sciencedirect.com/
science/article/pii/S0016003215003427S).
[15] C. Garcia, Comparison of friction models applied to a control valve, Control Eng. Pract. 16 (10) (2008) 1231–1243.
[16] R. Nouailletas, E. Mendes, D. Koenig, Hybrid modeling and identification of dry friction systems, application to a
clutch actuator, Control Eng. Pract. 18 (8) (2010) 904–917.
[17] F.-S. Ahmed, Modeling, simulation and control of the air-path of an internal combustion engine (Ph.D. thesis),
Université de Technologie de Belfort-Montbeliard, 2013.
[18] F.S. Ahmed, S. Laghrouche, M. Harmouche, Adaptive backstepping output feedback control of DC motor actuator
with friction and load uncertainty compensation, Int. J. Robust Nonlinear Control 25 (2015)
1967–1992, http://dx.doi.org/10.1002/rnc.3184.
[19] J.-C. Maré, Friction modelling and simulation at system level – considerations to load and temperature effects, Proc.
Inst. Mech. Eng. Part I: J. Syst. Control Eng. (2014) 22.
[20] L. Freidovich, A. Robertsson, A. Shiriaev, R. Johansson, Lugre-model-based friction compensation, IEEE Trans.
Control Syst. Technol. 18 (1) (2010) 194–200, http://dx.doi.org/10.1109/TCST.2008.2010501.
[21] W.-F. Xie, Sliding-mode-observer-based adaptive control for servo actuator with friction, IEEE Trans. Ind. Electron.
4 (3) (2007) 1517–1527.
[22] R. Guenther, E.A. Perondi, E.R. de Pieri, A.C. Valdiero, Cascade controlled pneumatic positioning system with
LuGre model based friction compensation, J. Braz. Soc. Mech. Sci. Eng. 28 (1) (2006) 48–57.
[23] V. Lampaert, F. Al-Bender, J. Swevers, A generalized Maxwell-slip friction model appropriate for control purposes,
in: Proceedings of the International Conference on Physics and Control, 2003, vol. 4, 2003, pp. 1170–1177. http://
dx.doi.org/10.1109/PHYCON.2003.1237071.
[24] J.F. O'Brien, D.J. Carruthers, Nonlinear dynamic compensation for large-feedback control of a servomechanism
with multiple nonlinearities, Control Eng. Pract. 21 (11) (2013) 1531–1541.
[25] H. Zeng, N. Sepehri, Dynamic surface control of cooperating hydraulic manipulators in the presence of friction, in:
American Control Conference, 2007, ACC'07, 2007, pp. 94–99. http://dx.doi.org/10.1109/ACC.2007.4282713.
[26] L. Lu, B. Yao, Q. Wang, Z. Chen, Adaptive robust control of linear motors with dynamic friction compensation
using modified LuGre model, Automatica 45 (12) (2009) 2890–2896, http://dx.doi.org/10.1016/j.automatica.
2009.09.007 (URL 〈http://www.sciencedirect.com/science/article/pii/S0005109809004178〉).
[27] D. Hoshino, N. Kamamichi, J. Ishikawa, Friction compensation using time variant disturbance observer based on the
LuGre model, in: 2012 12th IEEE International Workshop on Advanced Motion Control, AMC, 2012, pp. 1–6.
http://dx.doi.org/10.1109/AMC.2012.6197030.
[28] P. Green, K. Worden, N. Sims, On the identification and modelling of friction in a randomly excited energy
harvester, J. Sound Vib. 332 (19) (2013) 4696–4708, http://dx.doi.org/10.1016/j.jsv.2013.04.024 (URL
〈http://www.sciencedirect.com/science/article/pii/S0022460X13003635〉).
[29] X. Wang, H. Li, Y. Wang, J. Hua, Adaptive backstepping control of flexible joint robots with friction compensation
based on LuGre model, in: The 26th Chinese Control and Decision Conference, 2014, CCDC, 2014, pp. 1484–1489.
http://dx.doi.org/10.1109/CCDC.2014.6852401.
3038 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039

[30] P. Dupont, B. Armstrong, V. Hayward, Elasto-plastic friction model: contact compliance and stiction, in: Proceedings of
the 2000 American Control Conference, vol. 2, 2000, pp. 1072–1077. http://dx.doi.org/10.1109/ACC.2000.876665.
[31] F. Al-Bender, V. Lampaert, J. Swevers, The generalized Maxwell-slip model: a novel model for friction simulation and
compensation, IEEE Trans. Autom. Control 50 (11) (2005) 1883–1887, http://dx.doi.org/10.1109/TAC.2005.858676.
[32] Z. Wei, B.L. Xiang, R.X. Ting, Online parameter identification of the asymmetrical Bouc–Wen model for
piezoelectric actuators, Precis. Eng. 38 (4) (2014) 921–927.
[33] L. He, Dynamical adaptive backstepping-sliding mode control of pneumatic actuator (Master's thesis), University of
Manitoba, Winnipeg, Canada, 2010.
[34] J.M. Lee, H.M. Kim, S.H. Park, J.S. Kim, A position control of electro-hydraulic actuator systems using the
adaptive control scheme, in: 7th Asian Control Conference, 2009, ASCC2009, 2009, pp. 21–26.
[35] V. Hayward, B. Armstrong, F. Altpeter, P. Dupont, Discrete-time elasto-plastic friction estimation, IEEE Trans.
Control Syst. Technol. 17 (3) (2009) 688–696, http://dx.doi.org/10.1109/TCST.2008.2001710.
[36] M.S. Madi, K. Khayati, P. Bigras, Parameter estimation for the LuGre friction model using interval analysis and set
inversion, in: 2004 IEEE International Conference on Systems, Man and Cybernetics, vol. 1, 2004, pp. 428–433.
http://dx.doi.org/10.1109/ICSMC.2004.1398335.
[37] T. Piatkowski, Dahl and LuGre dynamic friction models – the analysis of selected properties, Mech. Mach. Theory
73 (2014) 91–100.
[38] G. Palli, C. Melchiorri, Friction compensation techniques for tendon-driven robotic hands, Mechatronics 24 (2)
(2014) 108–117.
[39] P. Lischinsky, C. Canudas de Wit, G. Morel, Friction compensation for an industrial hydraulic robot, IEEE Control
Syst. Mag. 19 (1) (1999) 25–32.
[40] M. Sobczyk, E. Perondi, M. Cunha, A continuous extension of the LuGre friction model with application to the
control of a pneumatic servo positioner, in: 2012 IEEE 51st Annual Conference on Decision and Control, CDC,
2012, pp. 3544–3550. http://dx.doi.org/10.1109/CDC.2012.6426406.
[41] S.-H. Chen, L.-C. Fu, Observer-based backstepping control of a 6-dof parallel hydraulic manipulator, Control Eng.
Pract. 36 (2015) 100–112, http://dx.doi.org/10.1016/j.conengprac.2014.11.011 (URL 〈http://www.sciencedirect.
com/science/article/pii/S0967066114002731〉).
[42] M. Rahmat, Zulfatman, A. Husain, R. Ghazali, S. Rozali, Modeling and controller design of an industrial hydraulic
actuator system in the presence of friction and internal leakage, Int. J. Phys. Sci. 6 (14) (2011) 3502–3517.
[43] Zulfatman, M. Rahmat, A. Husain, R. Ghazali, S. Rozali, Smooth control action of sliding mode for a class of
electro-hydraulic actuator, in: 2011 4th International Conference on Mechatronics, ICOM, 2011, pp. 1–7. http://dx.
doi.org/10.1109/ICOM.2011.5937189.
[44] H. Mintsa, J.-P. Kenne, R. Venugopal, Adaptive control of an electrohydraulic position servo system, in:
AFRICON, 2009, AFRICON'09, 2009, pp. 1–6. http://dx.doi.org/10.1109/AFRCON.2009.5308385.
[45] M.R. Sobczyk, Variable structure cascade control with parameter adaptation and friction compensation applied to a
pneumatic positioning system (Ph.D. thesis), Federal University of Rio grande do Sul, Porto Alegre, RS, Brazil,
2009 (in Portuguese).
[46] J.-J. Slotine, L. Weiping, Adaptive manipulator control: a case study, IEEE Trans. Autom. Control 33 (11) (1988)
995–1003, http://dx.doi.org/10.1109/9.14411.
[47] F. Chen, R. Jiang, C. Wen, R. Su, Self-repairing control of a helicopter with input time delay via adaptive global
sliding mode control and quantum logic, Inf. Sci. 316 (2015) 123–131, http://dx.doi.org/10.
1016/j.ins.2015.04.023 (Nature-Inspired Algorithms for Large Scale Global Optimization. URL 〈http://www.
sciencedirect.com/science/article/pii/S0020025515002911〉).
[48] F. Chen, F. Lu, B. Jiang, G. Tao, Adaptive compensation control of the quadrotor helicopter using quantum
information technology and disturbance observer, J. Frankl. Inst. 351 (1) (2014) 442–455, http://dx.doi.org/10.
1016/j.jfranklin.2013.09.009 (URL 〈http://www.sciencedirect.com/science/article/pii/S0016003213003463〉).
[49] F. Chen, Q. Wu, B. Jiang, G. Tao, A reconfiguration scheme for quadrotor helicopter via simple adaptive control and
quantum logic, IEEE Trans. Ind. Electron. 62 (7) (2015) 4328–4335, http://dx.doi.org/10.1109/TIE.2015.2389760.
[50] W. Deng, J. Yao, D. Ma, Robust adaptive asymptotic tracking control of a class of nonlinear systems with unknown
input dead-zone, J. Frankl. Inst. 352 (12) (2015) 5686–5707, http://dx.doi.org/10.1016/j.jfranklin.
2015.09.013 (URL 〈http://www.sciencedirect.com/science/article/pii/S0016003215003609〉).
[51] M. Zhang, X. Liu, B. Yin, W. Liu, Adaptive terminal sliding mode based thruster fault tolerant control for
underwater vehicle in time-varying ocean currents, J. Frankl. Inst. 352 (11) (2015) 4935–4961, http://dx.doi.org/
10.1016/j.jfranklin.2015.08.009 (URL 〈http://www.sciencedirect.com/science/article/pii/S001600321500321X〉).
[52] H. Khalil, Nonlinear Systems, 3rd ed., Prentice Hall, Upper Saddle River, NJ, USA, 2002.
 M.R. Sobczyk et al. / Journal of the Franklin Institute 353 (2016) 3021–3039 3039

[53] R.M. Suzuki, Feedback linearization-based control applied to a pneumatic servo positioning system (Master's
thesis), Federal University of Rio Grande do Sul, Porto Alegre, RS, Brazil, 2010 (in Portuguese).
[54] S. Kaitwanidvilai, P. Olranthichachat, Robust loop shaping-fuzzy gain scheduling control of a servo-pneumatic
system using particle swarm optimization approach, Mechatronics 21 (1) (2011) 11–21, http://dx.doi.org/10.
1016/j.mechatronics.2010.07.010 (URL 〈http://www.sciencedirect.com/science/article/pii/S0957415810001340〉).
[55] B. Najjari, S.M. Barakati, A. Mohammadi, M.J. Futohi, M. Bostanian, Position control of an electro-pneumatic
system based on {PWM} technique and {FLC}, {ISA} Trans. 53 (2) (2014) 647–657, http://dx.doi.org/10.
1016/j.isatra.2013.12.023 (URL 〈http://www.sciencedirect.com/science/article/pii/S0019057813002334〉).
[56] H. Yanada, Y. Sekikawa, Modeling of dynamic behaviors of friction, Mechatronics 18 (7) (2008) 330–339, http://dx.
doi.org/10.1016/j.mechatronics.2008.02.002. (Special Section of Revised Papers from the 8th International {IFAC}
Symposium on Robot Control. URL 〈http://www.sciencedirect.com/science/article/pii/S0957415808000330〉).
[57] H. Zeng, N. Sepehri, Adaptive backstepping control of hydraulic manipulators with friction compensation using
LuGre model, in: American Control Conference 2006, 2006, p. 6. http://dx.doi.org/10.1109/ACC.2006.1657204.
[58] W.-H. Zhu, Fpga logic devices for precision control: an application to large friction actuators with payloads, IEEE
Control Syst. 34 (3) (2014) 54–75, http://dx.doi.org/10.1109/MCS.2014.2308691.
[59] D. Meng, G. Tao, H. Liu, X. Zhu, Adaptive robust motion trajectory tracking control of pneumatic cylinders with
LuGre model-based friction compensation, Chin. J. Mech. Eng. 27 (4) (2014) 802–815.
[60] J. Yao, W. Deng, Z. Jiao, Adaptive control of hydraulic actuators with LuGre model-based friction compensation,
IEEE Trans. Ind. Electron. 62 (10) (2015) 6469–6477, http://dx.doi.org/10.1109/TIE.2015.2423660.