Open Architecture Robot Control System

Volume 2, Number 3, March 2012 (Serial Number 9)
Journal of Mechanics
and
Automation
Engineering
David
David Publishing Company
www.davi dpubl i shi ng.com
Publishing David

Publication Information:
Journal of Mechanics Engineering and Automation is published monthly in hard copy (ISSN 2159-5275) and online
(ISSN 2159-5283) by David Publishing Company located at 9460 Telstar Ave Suite 5, EL Monte, CA 91731, USA.

Aims and Scope:
Journal of Mechanics Engineering and Automation, a monthly professional academic journal, particularly emphasizes
practical application of up-to-date technology in realm of Mechanics, Automation and other relevant fields. And
articles interpreting successful policies, programs or cases are also welcome.

Editorial Board Members:
Konstantin Samsonovich Ivanov (Kazakhstan) Isak Karabegovic (Bosnia and Herzegovina)
Curtu Ioan (Romania) Adel Abdel-Rahman Megahed (Egypt) Zhumadil Baigunchekov (Kazakhstan)

Manuscripts and correspondence are invited for publication. You can submit your papers via web submission, or
E-mail to mechanics@davidpublishing.com. Submission guidelines and web submission system are available at
http://www.davidpublishing.com.

Editorial Office:
9460 Telstar Ave Suite 5, EL Monte, CA 91731, USA
Tel: 1-323-984-7526; Fax: 1-323-984-7374
E-mail: mechanics@davidpublishing.com

Copyright2012 by David Publishing Company and individual contributors. All rights reserved. David Publishing
Company holds the exclusive copyright of all the contents of this journal. In accordance with the international
convention, no part of this journal may be reproduced or transmitted by any media or publishing organs (including
various websites) without the written permission of the copyright holder. Otherwise, any conduct would be
considered as the violation of the copyright. The contents of this journal are available for any citation. However, all
the citations should be clearly indicated with the title of this journal, serial number and the name of the author.

Abstracted / Indexed in:
Database of EBSCO, Massachusetts, USA
Chinese Database of CEPS, Airiti Inc. & OCLC
CSA Technology Research Database
Ulrichs Periodicals Directory
Summon Serials Solutions
Norwegian Social Science Data Services (NSD), Norway
Chinese Scientific J ournals Database, VIP Corporation, Chongqing, China

Subscription Information:
Price (per year):
Print $450; Online $320; Print and Online $600

9460 Telstar Ave Suite 5, EL Monte, CA 91731, USA
Tel: 1-323-984-7526; Fax: 1-323-984-7374
E-mail: order@davidpublishing.com

www.davidpublishing.com
DAVID PUBLISHING
D

Jour nal of Mechani c s
Engi neer i ng and
Aut omat i on

Volume 2, Number 3, March 2012 (Serial Number 9)

Contents
Techniques and Methods
137 An Adaptive Dominant Type Hybrid Adaptive and Learning Controller for Geometrically
Constrained Robot Manipulators
Munadi, Tomohide Naniwa and Yoshiaki Taniai
149 Basic Aspects of Defining Mechanical-Technological Solutions for the Production of Biogas from
Liquid Manure
Nataa Soldat and Mirjana Radii
154 Block Design of Robust Control for a Class of Dynamic Systems by Direct Lyapunov Method
Vjacheslav Pshikhopov and Mikhail Medvedev
163 Cycle Analysis of Internally Reformed MCFC/SOFC-Gas Turbine Combined System
Abdullatif Musa and Abdussalam Elansari
169 Design, Development and Testing of a Wireless Multi-Sensors Network System
Chelakara Subramanian, Jean-Paul Pinelli, Ivica Kostanic and Gabriel Lapilli
184 Distributed Robot Control System Based on the Real-Time Linux Platform
Goran Ferenc, Zoran Dimi, Maja Lutovac, Vladimir Kvrgi and Vojkan Cvijanovi
190 Hybrid Algorithms for Multiobjective Optimization of Mechanical and Hydromechanical Systems
Valeriy D. Sulimov and Pavel M. Shkapov
Investigation and Analysis
197 Preventive Maintenance of Passengers Cars Driving in the Territory of the Republic of Kosovo
Xhemajl Mehmeti, Naser Lajqi, Bashkim Baxhaku, Shpetim Lajqi and Hajredin Tytyri
Journal of Mechanics Engineering and Automation 2 (2012) 137-148

An Adaptive Dominant Type Hybrid Adaptive and
Learning Controller for Geometrically Constrained
Robot Manipulators
Munadi
1
, Tomohide Naniwa
2
and Yoshiaki Taniai
2

1. Department of Mechanical Engineering, Diponegoro University, Semarang 50275, Indonesia
2. Department of Human and Artificial Intelligent Systems, University of Fukui, Fukui 910-8507, Japan

Received: December 20, 2011 / Accepted: January 04, 2012 / Published: March 25, 2012.

Abstract: In this paper, the problem of position/force tracking of a robot manipulator for the repetitive tasks in touch with a
constraint surface is considered. When an end-effector of robot manipulator is requested to move to touch an environment given as a
rigid smooth surface, the controller has to achieve both desired position and contact force trajectories. In this paper, an adaptive
dominant type hybrid adaptive and learning controller (HALC) is proposed for a geometrically constrained robot manipulator. The
proposed controller deals with a model-based adaptive control (MBAC) and a repetitive learning control (RLC). The proposed
controller is designed so that the robot manipulator can follow the response of desired position and contact force trajectories. The
Lyapunov-like method is utilized for the convergence of joint angle errors and contact force errors which is proven under an
appropriate initial conditions and the smoothness of the constraint surface. Finally, the effectiveness of the proposed controller by
using a dynamical model of two-link robot manipulator is verified by the computer simulation results.

Key words: Adaptive dominant, adaptive control, learning control, position/force control, geometrically constrained, robot
manipulator.

1. Introduction

In many industrial applications, robot manipulators
are employed to perform an assigned task repeatedly,
and each task is repeated over a given operation time.
The examples of above tasks include materials
transfer in assembly-line workplace, in which the
end-effector of robot manipulator is required to make
contact with an environment or other objects. In such
cases, the interaction force due to the contact between
the end-effector and an object at a point that is
expressed by a constraint surface has to be controlled
accurately, since otherwise the arising contact force

Tomohide Naniwa, professor, research fields: adaptive and
learning control, adaptive position/force control, and
coordinated control of multiple manipulators.
Corresponding author: Munadi, Dr. Eng., research fields:
hybrid adaptive and learning controller. E-mail:
munadi@undip.ac.id.
may damage the object or the robot structure.
Therefore, it is necessary to control the robot
manipulator not only for the desired position
trajectory but also for the contact force trajectory
when the end-effector moves and keeps in touch with
a constrained surface of the object. Many research
efforts have been investigated and developed
extensively to define and analyze several control laws
for force control methods of robot manipulator since
1950. Most of the approaches are possible to classify
into two main categories: impedance control [1] and
hybrid position/force control [2]. Impedance control
emphasizes at controlling position and force by
adjusting the mechanical impedance of the
end-effector to external force generated by contact
with the environment. Mechanical impedance is
roughly an extended concept of the stiffness of a
D
DAVID PUBLISHING
An Adapti ve Dominant Type Hybrid Adaptive and Learning Controller for Geometrically

138
mechanism against a force applied to it. Next, as the
name implies, the hybrid position/force control can be
used to track positional trajectories and force
trajectories in different subspaces simultaneously by
using two feedback loops for direct separate control of
position and force.
Refs. [3-4] have proposed the approach of
impedance control and hybrid position/force control,
in which they assumed that the dynamical parameters
of robot manipulator are known precisely such as
moment of inertia, mass of link, and joint friction. In
fact, it is difficult to measure the exact value of
dynamical parameters due to the uncertainty of
dynamical parameters of robot manipulator, hence, the
identification and estimation techniques [5-6] are
proposed. Furthermore, when the end-effector touches
with an object, the model-based adaptive control
methods [7-8] are proposed to track the desired
position trajectory and contact force trajectory by
estimating the unknown dynamical parameter of robot
manipulator accurately. It is also reinforced by the
experimental results in Ref. [9].
Meanwhile, the learning control is used to
compensate for the repeated errors for robot
manipulators to perform a given task repeatedly. The
learning control is a concept for controlling uncertain
dynamical system in an iterative manner or repetitive
manner. In Refs. [10-11], several learning control laws
have been discussed in which the learning control
does not require exact knowledge of the dynamics of
robot manipulator. So far, most researches on the
learning control have been focused on the problem of
periodic trajectory tracking. Furthermore, relating to
the implementation of the hybrid position/force
control, there were several approaches which have
been proposed to describe a constraint surface for
force control [12-15]. In Ref. [16], the principle of
orthogonalization for position and force control was
presented. It introduces a projection matrix that
projects error vectors to the tangent plane of the
constraint surface in joint space.
In this paper, we are motivated to extend a simpler
of hybrid adaptive and learning control (HALC) for
hybrid position/force control that can achieve both
desired position and contact force trajectories
accurately when the robot manipulator is limited in
motion for keeping in touch with the smooth
constraint surface. The hybrid position/force approach
is developed based on the HALC law in Ref. [17] that
consists of the model-based adaptive control to cope
the unknown dynamical parameters, the learning
control to handle an assigned task repeatedly and also
the proportional-derivative control to stabilize the
closed-loop system and ensure the error convergence,
in which the adaptive control input becomes dominant
than other inputs. Domination of adaptive control
input gives the advantage that the hybrid
position/force control could adjust the feed-forward
motion control input immediately. Whereas, if the
learning control input becomes dominant, the
proposed controller will need much time to relearn the
learning control input when the desired trajectory is
changed during motion. Furthermore, a Lyapunov-like
method is presented to prove the stability of the
proposed hybrid position/force control in which
asymptotic convergence of position and force errors to
zero is guaranteed. The effectiveness of the proposed
method is evaluated by computer simulation with a
model of two-link robot manipulator.
The paper is organized as follows: The dynamics of
robot manipulators and a constraint surface for the
end-effector are described in section 2. Section 3 then
presents the proposed controller for geometrically
constrained robot manipulator. The stability analysis is
explained in section 4. Section 5 reports the
simulations results and section 6 concludes the paper.
2. Description of Constrained Robotic
System
2.1 Dynamics of Robot Manipulators
In this section, we consider a constraint surface of
robot system in which an end-effector of n serial-link

139
robot manipulator is required to move on a smooth
constraint surface, as shown in Fig. 1. Furthermore, let
x = (x
1
, , x
n
)
1
e R
n
denote a position vector of a
contact point of end-effector in a fixed task space
coordinates such as Cartesian coordinates and let also
q e R
n
denote the joint angle position of robot
manipulator. The forward kinematic relation of x and
q is expressed by a vector function as follows:
x = b(q) (1)
where b(. ) e R
n
- R
n
is generally a nonlinear
transformation describing the relation between the
joint space and task space. Then, the velocity x is
given as follows:
x = [
x
(q) q (2)
where [
x
(q) e R
n x n
is the manipulators Jacobian
matrix of x with respect to q. Therefore, the
equation of motion for the constraint robot
manipulator is expressed in following form by
neglecting joint friction:
R(q)q (t) + _
1
2
R
(q) + S(q, q )_ q (t) + 0(q)

= -u(t) + [
x
1
(q)F(t) (3)
where q (t), q (t) e R
n
are the joint angle velocity
and acceleration vector, respectively. R(q) e R
nxn

represents inertia matrix, which is symmetric and
positive definite, S(q, q ) e R
nxn
represents a
skew-symmetric matrix came from Coriolis and
Centrifugal force that is expressed by
S(q, q )q =
1
2
R
(q)q -
1
2
]
q
q
1
R(q)q
1
(4)
And also 0(q) e R
n
represents the gravitational
force vector, u(t) e R
n
represents the control input
vector generated by independent torque sources at
each joint, and F(t) e R
n
represents the contact
force vector.
2.2 Constraint Surface
Further, we consider a constraint surface for the
end-effector of robot manipulator, which moves in
touch with an object, and it can be defined in the
algebraic term as follows:
(x(q)) = u (5)
where (x(q)): R
n
- R
1
is a given scalar function.

Fig. 1 The constrained robot manipulator.

And taking the derivative of Eq. (5) with respect to
time gives the following expression:
d
dt
(x(q)) =
q
x
x
q
dq
dt
=
q
x
[
x
(q)q (t) = u (6)
It is assumed that the normal vector of the
constraint surface oox is not a zero vector, so we
obtain
[
q
(q) =
qx
[qx[
[
x
(q) (7)
where (oox)[oox[ e R
n
is a unit normal
vector in task space and [
q
(q) R
n
- R
1xn
is a
normal vector of (x(q)) = u in joint space such
that
[
q
(q)q = u (8)
It means we can obtain the derivative of Eq. (8)
with respect to time as follows:
d
dt
[
q
(q)q = [
q
(q)q +[
q
(q)q = u (9)
Hence, using Eq. (7), the dynamics of robot
manipulator in Eq. (3) can be represented as follows:
R(q)q (t) + _
1
2
R
(q) + S(q, q )_ q (t) + 0(t)

= u(t) +[
x
1
(q)(t) (10)
in which (t) represents the magnitude of contact
force, and
F(t) =
1
_
q
x
_
[
q
x
1
(t) (11)
3. Adaptive Dominant Type HALC
3.1 Definition of Error Signals and Regressor Matrix
A HALC is designed to make the robot
manipulators follow a periodic desired position
trajectory q
d
(t) and a desired contact force trajectory
d
(t)in accurately during an interval of finite duration
t e |u, I], in which I denotes a period of desired

140
trajectory. Furthermore, in order to define the
proposed controller, we make the standard assumption
with regard to Eq. (10) that all of the terms on
left-hand side of the dynamics of robot manipulators
are bounded if q
d
(t), q
d
(t) and q
d
(t) are bounded
and uniformly continuous.
Later, we can define the joint angular error q(t)
and integrated contact force error F(t) as follows:
q(t) = q(t) - q
d
(t), F(t) = ] ()J
t
0

(t) = (t) -
d
(t) (12)
where
d
(t) is also assumed uniformly continuous.
In this proposed controller, we define a reference
nominal velocity which is denoted as follows:
q
(t) = P
q
(q)(q
d
(t) - oq(t)) + y[
q
+
(q)F(t) (13)
where o and y is a positive scalar gain, and
P
q
(q) e R
n
denotes a projection matrix. The role of
the projection matrix is to project the joint space
vectors onto a tangent plane of the surface
(x(q)) = u at point q(t) . It is defined by the
following form:
P
q
(q) = I - [
q
+
(q)[
q
(q) (14)
in which [
q
+
(q) is the pseudo inverse matrix of
[
q
(q) defined as follows:
[
q
+
(q) = [
q
1
(q)|[
q
(q)[
q
1
(q)]
-1
(15)
As long as the end-effector of robot manipulator
touches a constraint surface, it holds (x(q)) = u at
point q(t), therefore we can confirm the following
properties:
P
q
(q) = P
q
1
(q), P
q
(q)q = q
P
q
[
q
1
(q) = ([
q
(q)P
q
(q))
1
= u (16)
Next, we consider the control input adopted from
Eq. (10). The dynamics of robot manipulators can be
rearranged in terms of unknown dynamical parameters
0 which will be rewritten in this following expression:
E(q, q , q , q )0 = u(t) + [
q
1
(q)(t) (17)
where E(q, q , q , q ) is the n x mnonlinear function
matrix known as the regressor matrix that consists of
known functions of joint position, velocity and
acceleration, while 0 e R
m
represents vector of
unknown dynamical parameters such as mass,
moments of inertia, and distance from joint to center
of mass of each link. Moreover, based on the periodic
desired trajectory, we can denote the desired regressor
matrix as follows:
E
d
(q
d
, q
d
, q
d
, q
d
)0 = R(q
d
)q
d
(t)
+]
1
2
R
(q
d
) + S(q
d
, q
d
) q
d
(t) + 0(q
d
) (18)
And also, we present another type of regressor
matrix which can be regarded as the residual regressor
matrix based on the reference nominal velocity in Eq.
(13) as the following expression:
E
(q, q , q
, q
)0 = R(q)q
(t)
+]
1
2
R
(q) + S(q, q ) q
(t) + 0(q) (19)

Based on Eq. (18) and Eq. (19), we can express a
correlation regressor matrix between E
and E
d
, and
it is defined by E
as follows:
E
= E
(q, q , q
, q
) - E
d
(q
d
, q
d
, q
d
, q
d
) (20)
Substituting Eq. (18), Eq. (20) into Eq. (10), in
which E
(q, q , q
, q
) is linear in q
(t) and q
(t),
we obtain the another form of dynamics of robot
manipulator that can be formulated by using s(t) in
the following expression:
R(q)s (t) + ]
1
2
R
(q) + S(q, q ) s(t) + E
0 + E
d
0 =
[
q
1
(q)(t) + u(t) (21)
s(t) e R
n
is another error signal called as a
filtering tracking error. It is defined as the difference
between current velocity q (t) and nominal reference
velocity q
(t) which is defined as follows:

s(t) = q (t) - q
(t) = P
q
(q)(q (t) + oq(t))
-y[
q
+
(q)F(t) (22)
where the right-hand side of the above equation
corresponds with following expression:
s
0
(t) = P
q
(q)(q (t) + oq(t))
s
1
(t) = -y[
q
+
(q)F(t) (23)
Note that this filtering tracking error s(t) plays
important role to design the proposed controller and to
prove the stability of the proposed controller.
3.2 Design of Proposed Controller
Certainly, the proposed controller is designed to be
capable of guaranteeing the convergence of position
and force tracking errors when time towards infinity.
Hence, we propose an adaptive dominant type HALC
for position/force control of the constrained robot
manipulator. To illustrate the detailed design of

141
proposed controller, we express the control input in
the right-hand side of Eq. (10) by following form:
u(t) = u
u
(t) +u
I
(t) +u
pd
(t)
-[
q
1
(q)(
d
(t) - F(t)) (24)
where u
u
(t) , u
I
(t) , and u
pd
(t) are the MBAC
input, the RLC input and the proportional-derivative
(PD) control input, respectively. Now let us to
describe the MBAC input based on the residual
regressor matrix in Eq. (19), which can be resulted
according to following law:
u
u
(t) = E
(q, q , q
, q
)0
`
(t) (25)
where 0
`
(t) e R
m
is an estimation of unknown
dynamical parameter vector 0(t) at time t, and
0(t) = 0
`
(t) - 0. 0
`
(t) is updated on-line according
to the following adaptive update rule:
0
`
(t) = 0
`
(u) -] K
u
E
1
(q, q , q
, q
)
t
0
s(t)Jt (26)
which implies
0
`
(t) = -K
u
E
1
(q, q , q
, q
)s(t) (27)
where K
u
is adaptive gain selected as a symmetric
positive definite matrix.
As the second component of control input, the RLC
input is resulted based on recalling the filtering
tracking error and storing error to update the control
input for next period. The RLC input is declared as an
original learning law by adding a forgetting factor in
the following expression:
u
I
(t +I) = [u
I
(t) - K
I
s(t) (28)
where [ is a forgetting factor selected as a positive
scalar value which satisfies u < [ < 1. We have to
notice for defining a forgetting factor in the RLC law,
because [ will make the RLC input approach to
zero and also make the MBAC input to be dominant.
This strategy will make the adaptive control input be
greater compared with other inputs when the control
input of the proposed controller achieves the actual
position and force trajectories converging to the
desired position and force trajectories. Next, K
I
is a
learning gain selected as a symmetric positive
definite matrix. In this approach, the RLC input is
initialized as u
I
(t) = u for e |u, I], also satisfies
u
I
(u) = u.
For the PD control input, it is resulted by the
filtering tracking error multiplied by its gain and is
defined as follows:
u
pd
(t) = -Ks(t)
= -K{P
q
(q)(q (t) + oq(t)) - y[
q
+
(q)F(t)] (29)
where K is a PD gain selected as a symmetric
positive definite matrix. Absolutely, utilization of this
PD feedback in the proposed controller is to stabilize
the closed-loop system and ensure the error
convergence.
Finally, we combine the dynamics of robot
manipulator in Eq. (21) with the proposed control
input law in Eq. (24), and obtain the closed-loop
system of robot manipulators expressed in following
compact form:
R(q)s (t) + ]
1
2
R
(q) + S(q, q ) s(t) =

(u
u
(t) - E
d
0) + (u
I
(t) - E
0) - Ks(t) +
[
q
1
(q)((t) + F(t)) (30)
4. Stability Analysis
Stability is a fundamental issue in analysis and
design of control system. In this section, we will prove
the stability of proposed controller, so the actual angle
position and contact force trajectories of the robot
manipulator converge to their desired position and
force trajectories as t - .
4.1 Lyapunov Function Candidate
We use the Lyapunov-like method to prove
asymptotic stability of the proposed controller. Now, a
Lyapunov function candidate I(t) e R
1
is defined in
the following lower bounded function:
I(t) =
1
2
{s
1
(t)R(q)s(t) +yF
2
(t)] +I
u
(t) +
I
I
(t) u (31)
where
I
u
(t) =
1
2
0
1
K
u
-1
0 (32)
I
I
(t) =
1
2
__ u
I
1
()
t+1
0
K
I
-1
[
-1
u
I
()J
- _ u
I
1
()K
I
-1
[u
I
()J
t
0
_
=
1
2
]] u
I
1
()
t+1
0
K
I
-1
[
-1
u
I
()J
] u
I
1
()K
I
-1
[
1-[
2
[
u
I
()J
t
0
(33) +

142
We show that the above equations satisfy I(t) u
since R(q), K
u
and K
I
are positive definite, and
also u < [ < 1.
Having determined the Lyapunov function
candidate to be positive definite, then we compute its
time derivative. By substituting Eq. (32) and Eq. (33)
into Eq. (31), and differentiating I(t) with respect to
time, we have
I
(t) = s
1
(t) ]R(q)s (t) +
1
2
R
(q)s(t)
+y(t)F(t) +I
u
(t) + I
I
(t) (34)
Utilizing the closed-loop system of robot
manipulators in Eq. (30) and differentiating I(t) by
considering the properties of robot manipulators in
which S(q, q ) is skew-symmetric, x
1
S(q, q )x =
u v x e R
n
, we can simplify I
(t) as follows:
I
(t) = -s
1
(t)Ks(t) + I
1
(t) +I
2
(t) +I
3
(t) (35)
where
I
1
(t) = (u
u
(t) -E
d
0)
1
s(t) +I
u
(t) (36)
I
2
(t) = (u
I
(t) - E
0)
1
s(t) +I
I
(t) (37)
I
3
(t) = s
1
[
q
1
(q)( (t) + F(t))
+y(t)F(t) (38)
Next, substituting Eq. (20), Eq. (25) and Eq. (27)
into Eq. (36) yield another form of I
1
(t) expressed
as follows:
I
1
(t) = (E
0
`
(t) - E
d
0)
1
s(t)
+0
1
K
u
-1
(-K
u
E
1
s(t)) = (E
0(t) +(E
-
E
d
)0)
1
s(t) -(E
0(t))
1
s(t) = s
1
(t)E
0 (39)
While based on the RLC law in Eq. (28), I
I
(t) in
Eq. (37) can be expressed in following result:
I
I
(t) =
1
2
{u
I
1
(t + I)K
I
-1
[
-1
u
I
(t + I)
- u
I
1
(t)K
I
-1
[u
I
(t)]
=
1
2
]([u
I
(t) -K
I
s(t))
1
K
I
-1
[
-1
([u
I
(t) - K
I
s(t))
- u
I
1
(t)K
I
-1
[u
I
(t)
|

=
1
2
{u
I
1
(t)K
I
-1
[u
I
(t) - u
I
1
(t)K
I
-1
K
I
s(t)
- s
1
(t)K
I
1
K
I
-1
u
I
(t)
+ s
1
(t)K
I
1
K
I
-1
[
-1
K
I
s(t)
- u
I
1
(t)K
I
-1
[u
I
(t)]
= -u
I
1
(t)s(t) +
1
2
s
1
(t)[
-1
K
I
s(t) (40)
And I
2
(t) in Eq. (37) can be asserted as follows:
I
2
(t) = (u
I
(t) - E
0)
1
s(t) -u
I
1
(t)s(t) +
1
2
s
1
(t)[
-1
K
I
s(t) =
1
2
s
1
(t)[
-1
K
I
s(t) - s
1
(t)E
0 (41)
Meanwhile, after substituting Eq. (22) and Eq. (16)
into Eq. (38), we have I
3
(t):
I
3
(t) = (q (t) + oq(t))
1
P
q
(q)[
q
(q)
1

((t) + F(t)) -yF(t)[
q
(q)
+1
[
q
(q)
1

((t) + F(t)) +y(t)F(t)
= -yF(t)(t) -yF
2
(t) + y(t)F(t)
= -yF
2
(t) (42)
Finally, by substituting I
1
(t), I
2
(t)and I
3
(t) into
I
(t)in Eq. (35), we can represent the derivative of the

Lyapunov function candidate as follows:
I
(t) = -s
1
(t)Ks(t) + s
1
(t)E
0 +
1
2
s
1
(t)[
-1
K
I
s(t)
- s
1
(t)E
0 -yF
2
(t)
= -s
1
(t) ]K -
1
2
[
-1
K
I
s(t) - yF
2
(t) (43)
Since K and K
I
are positive definite, and [ is
positive, we can select the control gains in above
based on the following sufficient condition:
K -
1
2
[
-1
K
I
> u, and y > u (44)
and it means that I
(t) u . This implies the

boundedness of I
2
of s(t) and F(t). In addition,
based on Eq. (32) and Eq. (33), 0(t) and u
I
(t) are
also bounded.
4.2 Uniform Boundedness of Joint Angular Error
In this section, we will prove the uniform
boundedness of q(t) when the constraint surface
(x(q)) = u is smooth enough. According to
definition of s
0
in Eq. (23) and P
q
(q in Eq. (14),
we can denote
q
1
(t)s
0
(t) =
1
2
J
Jt
q
1
(t)q(t) +oq
1
(t)q(t)
- oq
1
(t)I
q
(q)q(t)
-q
1
(t)I
q
(q)q (t) (45)
where
I
q
(q) = [
q
+
(q)[
q
(q) (46)
I
q
(q) is a projection matrix into the
complementary subspace of the contact force vector.
Next, we consider the fact of Eq. (8), we have
I
q
(q)q (t) = -I
q
(q)q
d
(t)
= -[I
q
(q) - I
q
(q
d
) q
d
(t)
(47)
and the inner product of q(t) and s
0
(t) in Eq. (45)

143
can be rewritten as follows:
1
2
J
Jt
[q(t)[
2
= -]o[q(t)[
2
- [I
q
(q)q(t)[
2
- q
1
(t)I
q
1
(q) [I
q
(q)
- I
q
(q
d
)
q
d
(t)
+q
1
(t)s
0
(t) (48)
since
I
q
1
(q)I
q
(q) = I
q
(q) (49)
Furthermore, the assumption of the constraint
surface is smooth, it will make q(t) becoming
nearly perpendicular to the unit normal I
q
(q) when
q(t) tracks approaching to q
d
(t) in same direction,
in which it satisfies [I
q
(q)q(t)[ [q[ . This
means that there is a constant scalar value p
0

satisfying [q(t)[ p
0
which implies
[I
q
q(t)[
1
2
[q(t)[ (50)
At the same time and in the same way, we can also
declare a constant scalar value c
0
which is used to
satisfy the following assumption:
[I
q
(q) -I
q
(q
d
)[ c
0
[q[ (51)
when [q(t)[ p
0
. To note in above equation that
c
0
depends on the radius of curvature of the
constraint surface at point q
d
(t) , in which the
constant c
0
will be smaller as long as the radius of
curvature increases.
Thus, we submit Eq. (50) and Eq. (51) into Eq. (48),
and the result is divided by [q(t)[, then we will
obtain following equation:
d
dt
[q(t)[ -[
3u+2c
0
c
1
4
[q(t)[ + [s
0
(t)[ (52)
where
c
1
= sup
t>0
[q
d
(t)[ (53)
According to Eq. (31), I(t) is non-increasing in t
and s(t) that consists of s
0
(t) and s
1
(t) are
orthogonal to each other is bounded, so we can
express I(t) more precisely by following equation:
1
2
z
mn
[s
0
(t)[
2
I(t) I(u)
1
2
]z
mux
[s
0
(u)[
2
+
0
1
(u)K
u
0(u) +] u
I
1
()K
I
-1
[
-1
u
I
()J +
1
0
] u
I
1
()K
I
-1
[
1-[
2
[
u
I
()J
1
0
(54)
where z
mn
and z
mux
denote the minimum and
maximum eigenvalues of matrix R(q) over all q(t).
Later, we can define
r
2
= _ u
I
1
()K
I
-1
[
-1
u
I
(t)J
1
0

+] u
I
1
()K
I
-1
[
1-[
2
[
u
I
()J
t
0
(55)
From Eq. (54), it follows that for any t u,
[s
0
(t)[ c[s
0
(u)[ +
1
z
mn
|[0(u)[
K
c
+r|
c{[q (u)[ + o[q(u)[] +
1
x
min
|[0(u)[
K
c
+
r] H (56)
where
c = _
x
mcx
x
min
; [0(u)[
K
c
2
= 0
1
(u)K
u
0(u)
Further, substituting Eq. (56) into Eq. (52), it yields
[q(t)[ c
-ut
_[q(u)[ + _ c
u:
HJ
t
0
_
M
u
-c
-ut
j
M
u
- [q(u)[[ (57)
provided that o 2c
0
c
1
. Based on Eq. (56) which
defines H, and the fact c 1, and referring Eq. (56),
Eq. (57) can be represented as follows:
[q(t)[
M
u
= c[q(u)[ +
1
u
_c[q (u)[ +
1
x
min
|[0(u)[
K
c
+ r|
_
(58)
Next, if o is large enough satisfying
1
u
_c[q (u)[ +
1
x
min
|[0(u)[
K
c
+ r|_ <
1
2
p
0

(59)
and q(u) satiesfies
[q(u)[
1
2
p
0
(60)
then it follows from Eq. (58) that
[q(t)[ p
0
(61)
Eq. (61) shows the uniformly boundedness of joint
angular error q(t).
4.3 Convergence of Filtering Tracking Error and
Contact Force Error
For showing convergence of s(t) and F(t), we
can assume that the second order differential of the
constraint surface (x(q)) = u is continuous and
bounded in q(t) and thus P
q
, oP
q
oq, [
q
, and
o[
q
oq is uniformly continuous in q(t). Next, the
dynamics of robot manipulator in Eq. (30) can be
rewritten as follows:

144
R(q)s (t) + ]
1
2
R
(q) + S(q, q ) s(t) = E
0 +
u
I
(t) + Ks(t) + [
q
1
(q)((t) + F(t)) (62)
Since E
0
`
(t) is used in the dynamics of robot
manipulator, it can be described in the following
expression:
E
0
`
(t) = -R
`
(q)q
(t)
-]
1
2
R
`
(q) + S
`
(q, ) q
(t) - 0
`
(q) (63)
Now, we can rewrite the dynamics of manipulator
as follows:
R(q)s (t) + R
(q)q
(t) + N = [
q
1
(q)((t) +
F(t)) (64)
where
N = _K +
1
2
R
(q) +S(q, q )_ s(t) +

_
1
2
R
(q) + S
(q, q )_ q
(t) +0
(q) - u
I
(t)
R
(q) = R(q) - R
`
(q)
S
(q, q ) = S(q, q ) - S
`
(q, q )
0
(q) = 0(q) - 0
`
(q) (65)
R
`
(q) indicates the inertia matrix calculated on the
basis of the estimation of unknown dynamical
parameter vector 0
`
(t), and N represents the
remaining bounded terms of s(t) and q
(t). Since
s(t) is bounded, based on Eq. (22), q (t) is also
bounded together with q(t) and F(t). Then, the
reference nominal velocity q
(t) in Eq. (13) can be

derived with respect to time, and it yields
q
(t) = q
0
(t) +[
q
+
(q)F(t) + [
q
+
(q)(t) (66)
where
q
0
(t) = P
q
(q)(q
d
(t) - oq(t))
+P
q
(q)(q
d
(t) - oq (t)) (67)
Hence, q
(t) -[
q
+
(q)(t) are also bounded.
Next, substituting Eq. (66) into Eq. (64) yields:
R(q)s (t) + R
(q)q
0
(t) + R
(q)|y[
q
+
(q)F(t) +
y[
q
+
(t)| + N = [
q
1
(q)((t) + F(t)) (68)
Multiplying Eq. (68) by [
q
(q)R
-1
(q) and then it
is continued by substituting
[
q
(q)s (t) = -[
q
(q)s(t) - y, we have
|[
q
(q)R
-1
(q)[
q
1
(q) - y[
q
1
(q)
- y[
q
(q)R
-1
(q)R
(q)[
q
+
(q) + y|(t)

= -[
q
(q)s(t) +[
q
(q)R
-1
(q)(R
(q)q
0
(t) + N)
+|y[
q
(q)R
-1
(q)R
(q)[
q
+
(q)
-[
q
(q)R
-1
(q)[
q
1
(q)|F(t) (69)
For the left-hand side of eq. (69), it can be rewritten
as follows:
[
q
(q)R
-1
(q)[
q
1
(q) - y[
q
(q)R
-1
(q)R
(q)[
q
+
(q) + y
= [
q
(q)R
-1
(q)[
q
1
(q) +
y
2
[
q
(q)|R
-1
(q)R
`
(q) +
R
`
(q)R
-1
(q)][
q
1
(q)|[
q
(q)[
q
1
(q)]
-1
(70)
Since y > u is small enough as a positive constant
value, R
`
(q) is bounded, and the inside of the square
bracket of Eq. (70) is also bounded and positive
definite. Whereas q
(t) in Eq. (66) shows the

boundedness of (t), and the boundedness of s (t)
follows next equation:
s (t) = -R
-1
(q)|R
(q)q
(t) +N -[
q
1
(q)((t) +
yF(t))| (71)
Further, both s(t) and F(t) are uniformly
continuous referring to Eq. (69), and belong to the I
2

since (t) is bounded, and certainly it implies
lim
t-
s(t) = u ; lim
t-
F(t) = u (72)
This condition is identical form to s
0
(t)in Eq. (23)
converging to zero:
lim
t-
s
0
(t) = lim
t-
P
q
(q)(q (t) +oq(t)) = u
(73)
5. Computer Simulation Results
In this section, the simulation results are presented
to illustrate the performance of the proposed controller.
We consider a model of two-link robot manipulator
with two revolute joints of joint variables q
1
, q
2
, link
lengths l
1
= l
2
= u.4 |m], masses m
1
= m
2
=
u.2 |kg], distance between the joint to the center of
mass l
c1
= l
c2
= u.2 |m] , and moment inertias
I
1
= I
2
= u.u1u7 |kg m
2
]. Fig. 2 shows the model of
two-link robot manipulator in which the end-effector
is required to move along on a constraint surface in
the Cartesian coordinates.
For the dynamics of robot manipulator, according
Eq. (10), it can be written in detail as follows:
R(q) = _
R
11
R
12
R
21
R
22
_ (74)
for
R
12
= I
1
+ I
2
+ m
1
l
c1
2
+ m
2
(l
1
2
+l
c2
2
+ 2l
1
l
c2
c
2
)
R
12
= R
21
= I
2
+m
2
(l
c2
2
+ l
1
l
c2
c
2
)
R
22
= l
2
+ m
2
l
c2
2


145

Fig. 2 A model of two-link constrained robot manipulator.

1
2
R
(q) +S(q, q ) =
_
-m
2
l
1
l
c2
q
2
s
2
-m
2
l
1
l
c2
(q
1
+q
2
)s
2
m
2
l
1
l
c2
q
1
s
2
u
_ (75)
and
0(q) = _
m
1
gl
c1
c
1
+ m
2
g(l
1
c
1
+ l
c2
c
12
)
m
2
gl
c2
c
12
_ (76)
where s
1
= sin(q
1
), s
2
= sin(q
2
), s
12
=
sin(q
1
+ q
2
) , c
1
= cos(q
1
) , c
2
= cos(q
2
),
c
12
= cos(q
1
+ q
2
) and g = 9.81 |ms
2
].
Furthermore, for the regressor matrix E(q, q , q, q
),
the unknown dynamical parameters 0 is defined by
l
l
l
l
l
0
1
0
2
0
3
0
4
0
5
1
1
1
1
1
=
l
l
l
l
l
I
1
+ m
1
l
c1
2
I
2
+ m
2
l
c2
2
m
1
l
c1
m
2
l
c2
m
2
1
1
1
1
1
(77)
Meanwhile, when the length of links is denoted by
l
1
and l
2
with reference to the geometry given in Fig.
2, the position of the end-point is given as follows:
j
x
y
[ = _
l
1
c
1
+ l
2
c
12
l
1
s
1
+ l
2
s
12
_ (78)
For the Jacobian matrix [
x
(q) which maps from
joint space to Cartesian space for two-link robot
manipulator is given by
[
x
(q) = _
-l
1
s
1
-l
2
s
12
-l
2
s
12
l
1
c
1
+ l
2
c
12
l
2
c
12
_ (79)
In this simulation, the geometric constraint of
end-point is described by following expression:
(x(q)) = y -b = l
1
s
1
+ l
2
s
12
- b (80)
in which we specify b = u.27S |m]. Furthermore, the
robot manipulator is requested to track the periodic
desired position trajectory q
d
(t) , in which by
considering the relation between the constraint surface
and q
d1
(t) for l
1
= l
2
= l , we have to define
q
d1
(t) and it yields q
d2
(t) as follows:
l
l
l
l
l
q
d1
(t)
q
d2
(t)
1
1
1
1
1
= _
2n
327
(2 sin(nt) - sin(2nt)) +
n
2
osin _
h
I
- sin(q
d1
(t))] - q
d1
(t)
_ (81)
Both desired position trajectories are shown in Fig.
3. For
d
(t), the desired force trajectory is given as a
constant value, 2 [N]. And according to Eq. (44), the
control gains K, K
I
, and [ are selected as:

(a)

(b)
Fig. 3 The periodic desired trajectory q
d
(t) for each
period at joint (a) 1 and (b) 2.

146
[ = u.S, K
u
= Jiog{u.u1, u.u1, u.u1, u.u1, u.u1],
K = Jiog{S.u, S.u], and K
I
= Jiog{2.9, 2.9]
respectively. For other gains, we define o =
Jiog{S.u, S.u] , y = 8.u, anu = S.u.
The simulation is started by initialing all initial
condition of estimation of unknown dynamical
parameters as 0, 0
`
(u) = u. Later, the simulation
results of the proposed controller are shown in Figs.
4-7. Fig. 4a shows position tracking error resulted at
joint 1 and Fig. 4b is error resulted at joint 2,
respectively. At joint 1, the position tracking error is
reduced each period, especially the error shrinks
drastically after the second repetition at 4 [s] from
0.045 to 0.006 [rad]. And also at joint 2, It is
decreased from 0.0330 to 0.0055 [s]. It can be seen
that the tracking performance was considerably
improved after the second repetition and it converges
asymptotically to 0.

(a)

(b)
Fig. 4 The position tracking error q(t) at joint (a) 1
and (b) 2.

(a)

(b)
Fig. 5 The velocity tracking error q (t) from repetition
of trajectory at joint (a) 1 and (b) 2.

Figs. 5a-5b show the velocity tracking performance
improvement for the two joint. At the initial repetition,
the maximum velocity errors were about 0.18 and
0.15 [rad/s], respectively. But after second repetition,
the maximum values were reduced to 0.050 and 0.008
[rad/s]. Meanwhile, Fig. 6a shows the force tracking
performance of the controller and the force tracking
error converges asymptotically, but it is very slowly.
And Fig. 6b shows the estimation of unknown
dynamical parameters during the execution of
prescribed desired trajectory. Based on Fig. 6b for
t = u |s], all estimated value of 0
`
is 0, then 0
`
1
, 0
`
3

and 0
`
5
increase leading up to relatively fixed value
of 0
1
, 0
3
and 0
5
, in which we obtain various
estimation of dynamical parameters 0
`
1
= u.uS ,
0
`
3
= u.u22 and 0
`
5
= u.u1S. Meanwhile, 0
`
2
and
0
`
4
decrease and will tend to -0.015 and -0.042,
respectively.

147

(a)

(b)
Fig. 6 The force error trajectory (t) (a) and the
estimation of unknown dynamical parameter 0
(b).

Furthermore, the motion control inputs are shown
in Fig. 7, in which the MBAC input is shown by solid
line and the RLC input by dashed line. Based on Fig.
7a which shows the MBAC input and RLC input at
joint 1, the initial MBAC input is comparable with
RLC input in early period, but for next period, the
MBAC input increases and becomes dominant to
achieve desired trajectory be compared with other
inputs. This condition describes the meaning of
domination of adaptive input in the proposed
controller. Whereas, the RLC input has the highest
torque in the second period, and then it decreases
according to the learning updated law.
6. Conclusions
In this paper, we have studied the hybrid
position/force control problem with the geometric

(a)

(b)
Fig. 7 The required torque profile for MBAC input and
RLC input at joint (a) 1 and (b) 2.

constraint of end-effector of robot manipulator. This
control method is a simpler combination of
model-based adaptive control that estimates the
unknown dynamical parameters, a repetitive learning
control that uses the input torque profile obtained
from the previous repetition, and a traditional PD
control. The proposed controller incorporates both
adaptive and learning capabilities, therefore, it can
provide an incrementally improved tracking
performance of position error by increasing the
number of repetitive tasks. The position/force tracking
errors have been proven to converge asymptotically
via Lyapunov-like stability analysis. The numerical
simulations have validated the effectiveness of the
proposed controller by showing the position and
velocity tracking errors decrease with the increase of
the repetition number.

148
References
[1] N. Hogan, Impedance control: An approach to
manipulation: Part ITheory, Part IIImplementation,
and Part IIIApplication, J. Dyn. Systems, Meas.,
Control 107 (1985) 1-24.
[2] M.H. Raibert, J.J. Craig, Hybrid position/force control of
manipulators, J. Dyn. Systems, Meas., Control 102 (1981)
126-132.
[3] T. Yoshikawa, T. Sugie, M. Tanaka, Dynamic hybrid
position/force control of robot manipulators: Controller
design and experiment, IEEE J. Robot. Automat. 4 (1988)
699-705.
[4] D. Wang, N.H. McClamroch, Position/force control design
for constrained mechanical systems: Lyapunovs direct
method, IEEE Trans. Robot. Automat. 9 (1993) 308-313.
[5] W. Li, J.J.E. Slotine, A unified approach to compliant
motion control, in: Proceedings of Amer. Contr. Conf.,
1989, pp. 1944-1949.
[6] R. Carelli, R. Kelly, An adaptive impedance/force
controller for robot manipulators, IEEE Trans. Automat.
Contr. 36 (1991) 967-972.
[7] S. Arimoto, Y.H. Liu, T. Naniwa, Model-based adaptive
hybrid control for geometrically constrained robots, in:
Proceedings of the 1993 IEEE International Conference
on Robotics and Automation, 1993, pp. 618-623.
[8] T. Naniwa, S. Arimoto, Model-based adaptive control for
geometrically constrained robot manipulators, ISSUE 8
(9) (1995) 482-490.
[9] L.L. Whitcomb, S. Arimoto, T. Naniwa, F. Ozaki,
Adaptive model-based hybrid control of geometrically
constrained robot arms, IEEE Trans. Robot. Automat. 13
(1) (1997) 105-116.
[10] S. Arimoto, S. Kawamura, F. Miyazaki, Bettering
operation of robots by learning, J. Robot. Syst. 1 (2)
(1984) 440-447.
[11] M. Aicardi, G. Cannata, G. Gasalino, Hybrid learning
control for constrained manipulators, Advanced Robotics
6 (1992) 69-94.
[12] M. Vukobratovic, Y. Ekalo, New approach to control of
robotic manipulators interacting with dynamic
environment, Robotica 14 (1) (1996) 31-39.
[13] C.C. Cheah, S. Kawamura, S. Arimoto, Stability of
hybrid position and force control for periodic manipulator
with kinematics and dynamics uncertainties, Automatica
39 (2002) 847-855.
[14] C.S. Chiu, K.Y. Lian, T.C. Wu, Robust adaptive
motion/force tracking control design for uncertain
constrained robot manipulators, Automatica 40 (2004)
2111-2119.
[15] Y. Karayiannidis, G. Rovithakis, Z. Doulgeri,
Force/position tracking for a robotic manipulator in
compliant contact with a surface using neuro-adaptive
control, Automatica 43 (2007) 1281-1288.
[16] S. Arimoto, Y.H. Liu, T. Naniwa, Principle of
orthogonalization for hybrid control of robot arms, in:
Proceedings of 12th IFAC World Congress, 1993, pp.
507-512.
[17] Munadi, T. Naniwa, An adaptive controller dominant
type hybrid adaptive and learning controller for trajectory
tracking of robot manipulators, Advanced Robotics 26
(2012) 45-61.


Basic Aspects of Defining Mechanical-Technological
Solutions for the Production of Biogas from Liquid
Manure
Nataa Soldat
1
and Mirjana Radii
2

1. Department of Mechanical Engineering, Faculty of Mechanical Engineering, University of Belgrade, Belgrade 11000, Serbia
2. Department of Mechanical Engineering, Technical College of Applied Sciences in Zrenjanin, Zrenjanin 23000, Serbia

Received: January 03, 2012 / Accepted: January 30, 2012 / Published: March 25, 2012.

Abstract: Defining mechanical-technological solutions for the production of biogas requires defining of single devices and their
functional integration into a whole. This paper defines mechanical and technological solutions for the production of biogas from
liquid manure, the choice of material, the method, the possibility of adaptation of existing devices and the removal of hydrogen
sulphide. The research and the results thereof are based on years of research done on existing facilities for the production of biogas.
The results show that the volume of the digester increases proportionally to the daily influx of fertilizer. Besides that, it should be
noted that all the insulating materials must be coated with a watertight material (thin aluminum foil) in order to prevent the alteration
of thermal insulating material.

Key words: Liquid manure, biogas, facilities.

1. Introduction

The production of biogas from liquid manure is
carried out using anaerobic fermentation in a facility
called reactor, where this process is carried out in
multiple phases. For each phase a functional and
dedicated group of devices is used.
This paper shows the research that was carried out
with regard to finding mechanical-technological
solutions for the production of biogas fro liquid manure,
the selection of a facility for anaerobic fermentation,
the method for their construction, and also the
possibility of adjusting existing devices and purifying
biogas.
The objective of this paper is to identify the most
suitable reactors for the production of biogas
fromliquid manure by studying the mentioned

Mirjana Radii, Ph.D., research field: manufacturing of
biomaterial.
Corresponding author: Nataa Soldat, M.Sc., research field:
materials. E-mail: nsoldat2000@yahoo.com.
mechanical-technological solutions, along with the
possibility of adjusting devices and purifying biogas.
The paper consists of the following sections:
Section 2 proposes mechanical-technological solution
for the production of biogas; section 3 introduces
digester volume; section 4 is the insulation of digester;
section 5 is choosing an anaerobic fermentation
facility and section 6 talks about the possibility of
reactor adjustment and purification of biogas.
2. Mechanical-Technological Solution for the
Production of Biogas
Defining a mechanical and technological solution
for the production of biogas from liquid manure is
based on defining the single devices and putting them
together into one functional unit.
The production of biogas comprises a number of
phases [1]:
Preparation of the substrate;
Fermentation;
D
DAVID PUBLISHING
Basic Aspects of Defining Mechanical-Technological Solutions
for the Production of Biogas from Liquid Manure

150
Capturing the gas;
Using biogas.
To every single phase a functional and dedicated
group of devices is assigned.
Every biogas production plant consists of the
following:
Raw material storage and processing facility;
Digester;
Overfermented substrate disposal system;
Biogas storing system.
The basic-centre part of a biogas facility is the
digester, i.e., biogas reactor for anaerobic digestion, to
which other components of the biogas facility, i.e.,
biogas processing equipment is connected.
A group of devices for maceration and separation of
coarse particles is used for the process of collecting
and preparation of the substrate. Actually,
polypropylene meshes and centrifugal fecal sledge
pumps, usually with two buckets are used for this.
The main phase of the biogas production process is
the anaerobic fermentation which is a multistage
biochemical process used on a number of different
types of organic substances. The technological process
of anaerobic fermentation depends on a number of
conditions, which, when met, produce a result
showing a high degradation level for organic matter
together with an acceptable quality and amount of
biogas.
Some of them can be monitored and in that way the
production of biogas can be controlled. That refers
mostly to: pH level (acidity), temperature, retention
time, filling level and toxicity.
3. Digester Volume
In almost all types of digesters, all three phases of
anaerobic digestion are carried out simultaneously
within the same volume-digester. That also is the
reason why the substrate hydraulically is kept for such
a long time in the digester. This time ranges between
12 and 20 days (sometimes 30). The degradation level
of organic matter is from 45 to 65%. The volume of
the digester increases proportionally to the daily influx
of fertilizer. This function is shown in Tables 1-2.
4. Insulation of Digester
As most digesters for biogas production work in a
mesophile temperature range, it is necessary to
thermally insulate the digesters well, to minimize the
loss of heat. Insulating materials can be natural and
synthetic. Synthetic materials (polystyrol and
polyurethane) are used more frequently, because they
are easier to shape and process, and they also have
better thermal insulating characteristics and they are
also cheaper.

Table 1 Required digester volume depending on the content of dry material in liquid fertilizer.
Content of dry material (%) 8 7 6
Liquid fertilizer mass (kg) 76,240 87,129 101,650
Usable digester volume (m
3
) 1,530 1,742 2,033
Index 100 114 133
Organic load of digester (kg SM/m
3
/day) 3.98 3.50 3.00
Specific production of biogas (Nm/m
3
/day) 2.14 1.85 1.58

Table 2 Required digester volume depending on the content of dry material in liquid fertilizer.
Content of dry material (%) 5 4 3.5
Liquid fertilizer mass (kg) 121,980 152,475 174,254
Usable digester volume (m
3
) 2,433 3,044 3,485
Index 159 199 278
Organic load of digester (kg SM/m
3
/day) 2.50 2.00 1.75
Specific production of biogas (Nm/m
3
/day) 1.34 1.06 0.92


151

All insulating materials must be coated with
watertight material (thin aluminium foil) in order to
prevent the alteration of thermal insulating material.
The technical process of biogas production is
economical if the heat loss does not exceed 15 to 40%.
Tables 3-4 show the thickness of the digesters
thermal insulation.
5. Choosing an Anaerobic Fermentation
Facility
The biogas production facility consists of a
complex installation incorporating a large number of
parts. The appearance of the facility dependa on the
type and amount of raw material that will be used for
the production of biogas.
The facility where the anaerobic fermentation takes
place is called reactor. Its solution is a basic condition
for a good performance of the whole installation.
Globally, as well as in our country, there are a number
of reactors that are different in structure and the
material they are made of [2-3].
Reactors most often are discontinuous and usually
there are at least two or three in series. Continuous
reactors are rarer. Both reactor types are supplied
through a hydraulic sealing system, which, for now, is
a practical and simple solution.
As already mentioned, during the process of
anaerobic fermentation, the matter is biologically
degraded. Besides the absence of oxygen, this process
also needs constant temperature. The degradation of
matter is most efficient on a temperature of 15
C
(psychrophile), 35
C (mesophile) and 55
C
(termophile process). The mesophile process is used
most frequently, and in the summer season also the
termophile process [2].
Reactors are classified by shape, size, type, by the
material they are made of, the mixing system or
substrate heating used. Depending on the material they
are made of reactors are from: steel, concrete or
plastics. Very rarely they are made of stainless steel.
Fig. 1 shows a vertical reactor made of concrete,
one that is most widely used [3].

Table 3 Required thickness of the digesters thermal insulation.
Container diameter
Minimum production
of biogas in m
3
/day
Mineral wool Epoxide resin
1.6 7.1 125 109
1.8 10.1 111 97
2.0 13.8 100 88
2.2 18.5 91 80
2.4 24.0 83 73
2.8 36.0 73 64
3.0 47.0 67 59
3.5 75.0 58 51
Table 4 Required thickness of the digesters thermal insulation.
Container diameter
Minimum production
of biogas in m
3
/day
Polystyrene plates Dry chips
1.6 7.1 105 250
1.8 10.1 93 230
2.0 13.8 84 210
2.2 18.5 76 190
2.4 24.0 70 170
2.8 36.0 61 150
3.0 47.0 56 135
3.5 75.0 49 120

152

Fig. 1 Vertical reactor made of concrete.

Due to the corrosion of steel and the porosity of
concrete, it is necessary to plastify the first two types.
Due to its chemical stability, easy use and low
plastification cost and unsaturated polyester is used as
a basic reinforcing material. In order to achieve
mechanical reinforcement, the polyester layer is
reinforced with glass fiber, and to enhance the shock
strength, a bitumen component is added to polyester in
form of a 50% bitumen stirene or toluene solution. An
ortho-or isonaphtal polyester resin is used for this.
Such reinforced protection containing approx. 30% of
alcohol-free glass fiber, draws us towards the idea that
most of the vital device parts, such as the ractor and
the bell, can entirely be made of polyester laminate [2].
Due to gas permeability and water absorption, and
also the inhibition of enzymate reactions, especially
saponification under the influence of different types of
estherases, the hydrophility of the laminate is reduced
by adding 3% of a five percent solution pf paraffin in
styrene. Gas permeability is reduced by foliar
multilayer lamination.
Mechanical and other properties of laminate are
shown in Table 5.
As for the overfermented substrate disposal system,
it is pumped out of the digester and through the piping it
reaches the tanks where it is stored. The tanks are located
near the digester where it is stored for a limited time (few
days). The digested material can be stored in concrete
facilities which are covered with natural or artificial
floating layers or membranes or even in lagoons.
Fig. 2 shows a storage tank covered by a
membrane.
It is possible that a part of the methane and
nutritious matter is lost during storing and handling of
overfermented substrate. It has been empirically
shown that up to 20% of total biogas production
occurs in the tanks. To avoid methane emissions and
to collect the additionally produced gas, tanks always
have to be covered with a gas-non-permeable
membrane in order to collect the gas.

Fig. 2 Storage tank covered by a membrane.

Table 5 Mechanical and other properties of laminate.
Tensile strength 100-140 N/mm
2

Bending strength 120-160 N/mm
2

E-module from bending 6000-8000 N/mm
2

Pressure strength 240-300 N/mm
2

Shock strength 70-90 N/mm
2

Max. Stretch till break 2%
Content of glass 30%
Density 1450 kg/m
3

Coefficient of thermic conductivity 0.20 W/mK
Resistance to temperature change -40-+120
C
Linear expansion coefficient 3 10
-5
K
-1

Absorption of water (24 h on 20
C) max 0.2 %

153

Biogas can also be stored within the gas space of
the digester itself (with a mobile or a fixed canopy)
and in biogas tanks which can be as follows:
Wet, low-pressure tanks (gas bell over substrate
or water;
Plastic foil gas cap;
Low-pressure dry tank;
Pressurized dry tank.
The pressure on low-pressure tanks is never greater
than 30 kPa. U Empty space is left in the digester (at
least 2-8% of the volume on large, and about a third
on small digesters) that is intended for the reception
and storage of biogas. If that space is small or if the
production and consumption of biogas on a farm are
not balanced, then it is necessary to build a special
biogas tank.
The storage of biogas in special tanks is expensive.
Building costs for tanks of up to 200 m
3
capacity are
20 to 30% of the digesters price.
6. Possibility of Reactor Adjustment and
Purification of Biogas
The adjustment of both steel and concrete reactors
is easy. On concrete facilities, non-permeability of gas
is also achieved using a penetration layer and foliar
lamination.
Purification of biogas, i.e., removing hydrogen
sulfate and water prevents the installations from
corroding, and the removal of CO
2
increases the
caloric capacity of biogas.
The level and method for the purification of biogas
depends on the method of use, purpose and other
factors. When biogas is compressed it has to be dried.
Usually it is dried through absorption, i.e., using
agents that are binding water, as are calcium
hydroxide or calcium chloride.
The deficit of this procedure is that also a part of
CO
2
, CaO and CaCl
2
is absorbed, increasing the
consumption of lime. When biogas passes through the
granular layer, the granules of CaO and CaCl
2
adhere
to each other, binding water. In that way the absorbent
cloddes and prevents the flow of biogas.
7. Conclusions
Studying mechanical-technological solutions for the
production of biogas from liquid manure, we can
conclude that due to their characteristics, concrete,
steel and plastic reactors are most frequently used.
Due to the corrosion of steel reactors and the
porosity of concrete reactors, they need to be
plastified. Polyester is used as a basic reinforcement
material, due to its chemical stability, easy use and
low plastification cost, in form of a polyester coating,
the reinforcement is achieved through glass fiber and
the shock strength is enhanced by adding a bitumen
component in shape of a 50% bitumen-styrene or
toluene solution.
The adjustment of steel or concrete reactors is
carried out in using a penetration layer and foliar
lamination for concrete facilities, achieving gas
non-permeability at the same time.
Purification of biogas, i.e., the removal of hydrogen
sulfate and water prevents the installations from
corroding, and the removal of CO
2
increases the
caloric capacity of biogas.
References
[1] M. ulbi, Biogas (dobijanje, korenje i gradnja
ureaja), Novinsko-izdavaka radna organizacija
Tehnika knjiga, Beograd, 1986.
[2] M. Radii, Proizvodnja i primena biogasa, Technical
College of Appplied Sciences in Zrenjanin, Zrenjanin.
2006.
[3] N. Soldat, Biogas-mogunosti proizvodnje i primene
(diplomski rad), Faculty of Mechanical Engineering,
University of Belgrade, Beograd, 2010.


Block Design of Robust Control for a Class of Dynamic
Systems by Direct Lyapunov Method
Vjacheslav Pshikhopov and Mikhail Medvedev
Department of Mechanical Engineering and Mechatronics, Technological Institute of Southern Federal University, Taganrog
347928, Russian Federation

Received: January 31, 2012 / Accepted: February 15, 2012 / Published: March 25, 2012.

Abstract: In this paper, a new method of a robust control design for a class of nonlinear multilinked dynamic systems is developed.
The method is advanced block control theory, which allow to provide compensation of object nonlinearity to closed-loop system with
sliding regime. This paper suggests approach, which allows to eliminate disadvantage of all sliding regime systems, that is unstability
in small. It is made by means of substitution of relay control by similar continuous control. The considered class of systems is called
block systems. Block form allows to describe a big class of controlled objects. In this work new controllability estimation approach
for nonlinear multilinked systems is suggested on the base of optimal control ability conditions check. The design procedure is based
on the direct Lyapunov method. It provides closed-loop system asymptotic stability. Control design for a single block system is based
on quadratic function of Lyapunov. For a general block system step-by-step design procedure is developed. Suggested synthesis
method provides closed-loop system stability characteristic robustness to right side equations of object in an area, defined by
limitations for control actions. Suggested approach also considers limitations for object variable states. The results of theoretical
analysis, solvability conditions of the control design equations, and robust control algorithms are presented. Theoretic results are
implemented on experimental robotic mini-airship and wheeled vehicle.

Key words: Nonlinear system, robust control, block system.

1. Introduction

Design of adaptive control systems is of significant
interest nowadays. This interest is attracted by unique
capability of adaptive control systems [1-5]: capability
to operate under condition of uncertain parameters,
uncertain mathematical model, and unmeasured
disturbances. Nowadays adaptation of systems are
based on searchless direct and indirect adaptive
control, robust control, search adaptive control,
invariant control, relay control, fuzzy logic control,
and neural network control.
In Refs. [6-7], a new method of the relay robust
control systems design was proposed. In this paper a
block method of the robust control by a class of
nonlinear control systems is developed. The method

Corresponding author: Mikhail Medvedev, professor,
D.Sc., research fields: automatic control, robotics, adaptive
control, estimation. E-mail: medvmihal@gmail.com.
takes into account state variables, control actions
limits. In addition the function of Lyapunov for the
designed control systems is constructed. Moreover the
matrix controllability conditions are presented via
scalar inequalities.
In Refs. [6-7], a relay control is designed. Therefore
the designed closed-loop systems are unstable in small
neighborhood of steady-state point. In this paper, a
continues approximation of relay control is applied.
Parameters of the approximation are defined by direct
Lyapunov method.
The paper is organized as follows: Section 2
contains description of synthesis method for object,
consisting of one block, and also stability conditions
of these objects. Suggested approach is generalized for
system class, composed of a few sequentially
connected blocks in section 3. Suggested approach is
advanced in section 4 according to limitations of
D
DAVID PUBLISHING
Block Design of Robust Control for a Class of Dynamic Systems by Direct Lyapunov Method

155
object variables states. Similarities of suggested
synthesis method and time optimal control are
discussed in section 5. Robust control systems
synthesis examples for test objects, for wheeled
mobile robot and for airship-based robotized complex
are described in section 6. Implementation of robots
using algorithms, synthesized with suggested in the
paper algorithm is described in section 7.
2. Robust Control Design
Suppose that a control object is
( ) ( )u x B x f x + =
(1)
where x a vector of state variables; u a vector of
control action; ( ) ( ) ( )
T
i
f x x f = , n i , 1 = a
functional vector of the state variables;
( ) ( ) ( )
T
i
x b x B = , n i , 1 = a m n functional
matrix.
Let the control action u be bounded:
max
j j
u u , m j , 1 = (2)
where
max
p
u
positive constants.
Let the control ( )
T
m
u u u
2 1
= u be a function
of the state variables x . We have to design the
control vector u such that the closed-loop system is
stable, and robust.
Now we introduce the following auxiliary vectors:
x = , ( )
T
m
u u u
max max
2
max
1
max
, , = u (3)
Using the vectors , and
max
u we introduce the
next theorem.
Theorem 1: Suppose the control vector for object
(1), (2) is
( ) ( ) x B u u
T
sign =
max
(4)
and the next inequalities
( ) ( ) x u x b
i i
f >
max
,
n i , 1 =
(5)
are satisfied; then function of Lyapunov for
closed-loop system (1)-(4) is

T
V
2
1
=
(6)
Theorem 1 can be proved by direct calculation of
function (6) time derivative. We have
( ) ( ) ( ) ( ) ( ) x x B u x B x f x
x x
T T
T T
sign
V

= = =
max

(7)
If (5) is satisfied, then the time derivative (7) is a
negative definite function. The theorem is proved.
From theorem 1, it follows that control (4) ensures
stability of the closed-loop system (1)-(4) if (5) is
satisfied. The assumptions of theorem 1 does not
include continuity of the ( ) x f . But the functional
vector ( ) x f is bounded. Note that control (4) does
not depend from the vector ( ) x f .
Inequalities (5) are equivalent to the controllability
condition presented in Ref. [8].
If n m then we have the following theorem.
Theorem 2: Suppose (5) is satisfied, and n m ,
then the state commonness condition [9] is satisfied:
( )
n
D G G G ,..., ,
2 1
= , n rankD = (8)
( ) x B G =
1

( ) ( ) ( )
( ) ( )
j 1
j
j 1
= +

G
G f x B x u
x
f x B x
u G
x x
,
n j , 2 =

Using (5) we get
( ) 0
max
> u x b
i
, n i , 1 = (9)
Since (9) is performed it follows that
( ) 0
max
u x b
i
, n i , 1 = (10)
Writing (10) in a vector form, we obtain
( ) 0 u x B
max
(11)
where 0 a zero vector.
From (11) it follows that
( ) ( ) n rank = x B (12)
If (12) is performed, then (8) is satisfied. Theorem 2
is proved.
Theorem 2 defines sufficient conditions of
controllability of the system (1)-(2). Thus the
sufficient condition of controllability (8) can be
substituted by scalar inequalities (5).
Moreover condition (12) is necessary condition of
the system (1), (2), (4) stability. Function (7) must be

156
continues. Therefore we have
( ) ( ) ( ) ( ) ( )
( )
( ) ( ) ( )
max
0
0
max
0
lim
lim
lim 0.
+
+
+

= +

=

T T
x
T
x
T T
x
sign
sign
x f x B x u B x x
x f x
x B x u B x x

( ) ( ) ( ) ( ) ( )
( )
( ) ( ) ( )
max
0
0
max
0
lim
lim
lim 0.

= +

=

T T
x
T
x
T T
x
sign
sign
x f x B x u B x x
x f x
x B x u B x x

But the last equations are satisfied if (12) is
performed.
3. Block Robust Control Design
In general case controls can directly impact just on
part of state space variables. Therefore the block
control design is developed in this section.
Now we introduce the next definition.
Definition 1: System (1) is a block system if it can
be presented in the next form:
( ) ( ) ( )u x B x f x x x f x + = =
+ k k i i i
, ,...,
1 1
(13)
1 , 1 = k i

where
( )
T
i
l
i i i
i
x x x
2 1
= x
,
( )
T
k
x x x x ..., , ,
2 1
=
;
( ) ( ) ( ) ( )
T
i i
l
i i i i
f f
1 1 1 1
1
1 1
,..., ,..., ,...,
i
+ + +
= x x x x x x f
,
( ) ( ) ( ) ( ) ( )
T
k k
x f x f x f x f ,..., ,
2 2 1 1
=
;
( ) ( ) ( ) x x B
ij
b =
n x m-matrix, k a positive integer.
A block system is a special case of a controllable
Jordan form [10].
If 2 k , then the vector u impacts to the vector
k
x . The vector
k
x impacts to the vector
1 k
x
,

etc.
In general case a vector
i
x is a fictitious control if
i
x impacts to
1 i
x .
If system (1) is a block system, then the control
design block procedure is used.
To design the control vector ( )
k
x x u u ,...,
1
= we
introduce the next auxiliary vectors:
k k
x = ,
k i i i
x H + = , 1 , 1 = k i (14)
where
i
H matrices of weighting factors.
We introduce the next theorem using (14).
Theorem 3: Suppose for block system (2), (13) the
next conditions are satisfied:
( )
=

=
n
i
i T
sign
1
max
x B u u
(15)
( )
k
i
i
i j
f f + >
max
u x b (16)
Then Lyapunovs function of system (2), (13)-(15)
is
( )
=
=
k
i
i
T
i
V
1
2
1

(17)
function (17) time derivative. We have
( ) ( ) ( ) ( )
( ) ( )
k k
i T max T i
i 1 i 1
k 1
iT i k kT k
i 1
V sign
= =
=

=

+ + +

B x u B x
f f f
(18)
Under condition (16) of the theorem we have
function (18), which is a negative definite function.
The theorem is proved.
Any arbitrary system can be presented by (13) via
corresponding denotations of state variables.
Expression (14) transforms system (13) to the single
block system (1). It is clear, that transformation (14) is
nonsingular if
i
H is nonsingular.
4. Block Robust Control Design under State
Variables Bounded
Let the vector ( )
T
i
l
i i i
i
x x x
2 1
= x , k i , 2 = be
bounded by constant. Then we have
max
j
i
j
x x ,
k i , 1 =
,
i
l j , 1 =
(19)
where
max
j
x are positive constants.
Assume system (13) is presented in the next form:
( )
( )
1 1
,...,
i i i i i
k k k
+
= +
= +
x f x x B x
x f x B u
(20)
1 , 1 = k i

Now we shall give the following theorem.

157
Theorem 4: Suppose for block system (20), (2), (19)
the next conditions are satisfied:
i i i
x f <
,
1
max
+
<
i i
j
i
x b f

max
u b f <
k
j
k
, 1 , 1 = k i (21)
( )
k k k
B q u u = tanh
max
, (22)
1 1
x =
,
( )
1 max
tanh

+ =
i i i i i i
B q x x

k i , 2 = (23)
1 1 1 1
q B B >
T
,
1 1 1
+ >
i i
T
i i i i
T
i
q B B q B B
,
k i , 2 = (24)
Then Lyapunov function of system (20), (22) is
k
T
k
V 5 . 0 = (25)
function (25) time derivative. In small neighborhood
of the 0 =
i
we have
( )
i i i i i i
B q B q ~ tanh (26)
From (22), (25)-(26) we obtain
(
( ) ( ))
(
( ) ( ))
k kmax k 1T k 1
T
k 1 k 1max k 2T k 2 k 1
k k max kT k k
kmax k 1T k 1
k 1 k 1max k 2T k 2 k 2
V
...
...

= +
+ +
+

+ +
x x B q
x x B q x
f B u B q
x B q
x x B q x

(27)

Under the conditions of (23)-(24), we have time
derivative (27) is a negative definite function.
In a tail region of the 0 =
i
we have
( ) ( )
0
tanh
1

dt
d
i i i
B q
(28)
From (25) and (28) we get
( ) ( )
k
T
k k k
T
k
sign V B u B f =
max
(29)
Time derivative (29) is a negative definite function
under condition (21). The theorem is proved.
If theorem 4 is satisfied, then control algorithm (22)
ensures stability of the closed-loop system. According
to conditions (21) the functions
i
f shall be bounded
both by the sectors
i i
x , as well as by the
constants
1
max
+
i i
j
x b , and
max
u b
k
j
. According to
(24) the sector
i i
x encloses the sector
1 1
i i
x .
Thus control (22) in a tail region is close to
bang-bang control. In a small neighborhood of
steady-state mode the control (22) is close to linear
quadratic regulator.
5. Interconnection of the Proposed Method
and the Pontryagin Principle of Maximum
Let H be a function of Pontryagin. Then the
interconnection between the Lyapunov method and
the Pontryagin principle of maximum is
V H

= (30)
where V is determined by (6).
If a function of Pontryagin is (30), then control (4)
satisfies to the Pontryagin principle of maximum [6].
In Ref. [11], a method of time suboptimal controls
for nonlinear systems was proposed. Let consider
system (1)-(2). Suppose the ( ) x B be a symmetrical
positive definite matrix. Let the dimension of the x
is equal to the dimension of u : m n = . The purpose
of control is given by (3). According to Ref. [11], we
have to satisfy the next equation:
0 T = + (31)
where T a positive definite function.
Combining (1), (3), and (31) we get
( ) ( ) [ ] 0 x u x B x f T = + +
(32)
From (32) we obtain
( ) ( ) ( ) ( ) ( ) [ ] x f x B x x TB u
1 1
= (33)
According to Ref. [11] to get a time suboptimal
control we have to calculate
( ) ( ) ( ) ( ) ( ) [ ] x f x B x x TB u
T U
1 1
0
lim
max

= sat (34)

158
where
max
U
sat the saturation function.
Expression (34) is the time suboptimal control
algorithm. Calculating limit in (34), we get
( ) ( ) ( ) ( ) ( ) [ ]
( ) ( ) [ ] ( ) ( ) x x B u x x TB
x f x B x x TB u
T U
T U
sign sat
sat
=
= =

max
1
0
1 1
0
lim
lim
max
max
(35)
It is clear that control (35) is equal to control (4).
6. Examples
Example 1: The example is to demonstrate the
proposed design method as a time optimal systems
method design.
Suppose that a control object is
( ) ( )
u
dt
t dx
x
dt
t dx
= =
2
2
1
, (36)
max
U u
(37)
It is known that the time optimal closed-loop
control for system (36)-(37) is
( ) ( )
2
2
2 1 max
5 . 0 x sign x x sign U u + = (38)
Applying the method proposed in this paper we get
( ) ( ) ( )
1
1
max 2 2
2
max
tanh tanh x q x x q U u + = (39)
Multiplying the right side of (38) by the right side
of (39) we obtain that the region of controls (38), and
(39) coincidence is
max 2 2
x x <
(40)
2
2 1
5 . 0 x x > (41)
In Fig. 1, there are both phase-plane portraits of
system (36)-(38), as well as system (36)-(37) and (39).
It is clear that in the region (40)-(41) the phase path of
system (36)-(38) and the phase path of system
(36)-(37) and (39) are same.
Parameters of Fig. 1 modeling results are:
5 . 0 , 2
max 2 max
= = x U , 10
2 1
= = q q .
Tr1 is the time optimal control system path. Tr2 is
the robust control system path. It is clear path Tr2 is
close to path Tr1 in the area bounded by (40)-(41).
Example 2: Consider a wheeled vehicle that
described by the next system:

-1.5 -1 -0.5 0 0.5 1 1.5
-1.5
-1
-0.5
0
0.5
1
1.5
x
1
x
2
Tr
1
Tr
2

Fig. 1 Phase portraits of time optimal, and robust control
system.
( )
( ) ( )
( )
( ) ( )
( ),
2
,
,
22 1 2
2
12 11
1
l r
r l
r l
b
r
dt
d
dt
t dx
dt
t dx
=
+ =
+ =
(42)
( )
( )
,
,
2 22 1 21 2
2 12 1 11 1
u b u b d
dt
t d
u b u b d
dt
t d
r
r
l
l
+ + =
+ + =
(43)
where
2 1
, x x external coordinates of the vehicle;
angle of orientation of the vehicle;
r l
, the
wheel rotation speeds; r the wheel radius; a a
kinematic factor;
i
d ,
ij
b , 2 , 1 = i ,
2 , 1 = j constants;
2 1
, u u control actions.
Functions ( )
11
, ( )
2 1
, ( )
1 2
, ( )
22
are
( ) ( ) ( )
11
0.5 cos sin r a = +

( ) ( ) ( )
12
0.5 cos sin r a =
( ) ( ) ( )
2 1
0.5 sin cos r a =
( ) ( ) ( )
12
0.5 sin cos r a = +

Let bounds be given by
max max
,
r l
(44)
2 max 2 1 max 1
, u u u u (45)
Eqs. (42) describe a kinematics of the vehicle. Eqs.

159
(43) describe a dynamics of the vehicle.
The initial state ( ) ( ) 0 , 0
2 1
x x of the vehicle
belongs to some area . Let the purpose state of the
vehicle is given by 0 , 0
2 1
= = x x . The orientation
of platform is arbitrary.
From section 4 of this paper we get
( ) ( )
( ) ( ) tanh
, tanh
22 2 12 1
2
2 max 2
21 2 11 1
2
1 max 1
b b q u u
b b q u u
=
=
(46)
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
1 l
1
max 1 11 2 21
2 r
1
max 1 12 2 22
tanh q x x 0
tanh q x x 0

=
=
=
=

(47)
Let the Lyapunov function is given by
( )
2
2
2
1
5 . 0 + = V
(48)
Differentiating the Lyapunov function (48) we obtain
( ) ( )
( ) ( )
( ) ( )
( ) ( )
2
11 1max 1 11 2 21
2
12 2max 1 12 2 22 1 l 1
2
21 1max 1 11 2 21
2
22 2max 1 11 2 21 2 r 2
V b u tanh q b b
b u tanh q b b d
b u tanh q b b
b u tanh q b b d

+ +
(49)
Function (49) is a negative definite function if
,
,
2 max 2 22 max 1 21
1 max 2 12 max 1 11
r
l
d u b u b
d u b u b
> +
> +
(50)
( ) ( )
( ) ( ) . 0
, 0
max 22 1 2
max 2 1 11
> +
> +
(51)
Necessary conditions for solution existence of
(50)-(51) are
2
22 21
12 11
=
b b
b b
rang
(52)
( ) ( )
( ) ( )
2
22 21
12 11
=

rang
(53)
Easy to prove that (52)-(53) are sufficient
conditions of the vehicle controllability.
There are modeling results of system (42)-(47) in
Figs. 2-4. Parameters of Figs. 2-4 modeling results are
10
2 max 1 max
= = u u , 10
max
= , 2 . 0 = r , 1 = a ,
( ) 15 . 0 05 . 0 J , ( ) 3 5 . 0
1
d , ( ) 3 5 . 0
2
d ,
1
22 11
= = b b , 0
21 12
= = b b , 10
2 1
= = q q .

Fig. 2 Path of the vehicle.

0 5 10 15 20
-15
-10
-5
0
5
10
15

omegar
omegal

Fig. 3 Wheels speeds of the vehicle.

0 5 10 15 20
-10
-8
-6
-4
-2
0
2
4
6
8
10
t,c
u1
u2

Fig. 4 The control action of the vehicle.

Example 3: Consider an airship that described by
the next system:

160
,
,
d u
F F x M
x R y
+ =
=
(54)
where x is a speed vector of the airship; y is a
coordinate vector of the airship;
u
F is a control
action vector;
d
F is a nonlinear uncertain functional
vector; M is matrix of masses as well as moments
of inertia; R is functional matrix of kinematical
connections.
Model (1) is described detailed in Ref. [1].
The aim of this example is to design the vector
u
F

such that airship (54) is stable in an area of the
undisturbed motion
0
x ,
0
y .
According to (14) we have
Ey x x x x + = =
2 1
,
(55)
where E is a nonsingular matrix.

Let the function of Lyapunov be given by
( ) ( )
2 2
5 . 0 x x
T
V = (56)
Differentiating (56) in time we get
( ) ( ) ( ) ( )
1 1 2 2 2
x R E F F M x x x + + = =

d u
T T
V
(57)
According to section 3 of this paper, and (57) the
robust control is
( ) ( ) Ey x M F F + =
T
u u
sign
1 max
(58)
where
max
u
F
vector of the control action bounds.
It is clear that (57) is a negative definite function if
1 1 max 1
ERx F M F M + >

d u
(59)
There are modeling results of system (54), (58) in
Figs. 5-6. The purpose of the control system is
movement along the straight line with speed about
5 m/s.
7. Hardware Implementation of the Control
System
Results of research are implemented in prototype of
airship-based autonomous mobile robot Sterkh,
shown in Fig. 7. These results also are implemented in
the control system of medium airship.
0 0.5 1 1.5 2 2.5 3
2
2.2
2.4
2.6
2.8
3
3.2
3.4
3.6
3.8
t,c
V
x
V
y

Fig. 5 The airship speed.

7 8 9 10 11 12 13
20
30
8
0
2
4
6
8
1
2
4
Z
g
X
g
Y
g

Fig. 6 The airship path.

Fig. 7 Autonomous mobile robot Sterkh based on
mini-airship.

Volume of the medium airship is 2 000 m
3
. Length
of the medium airship is 40 m. The movement of
airship is controlled by two engine installed in pylons.

161
-3
-2
-1
0 1
2
x 10
4
-2
0
2
4
x 10
4
0
200
400
600
800

Fig. 8 Path of the mobile robot on base of medium airship.

1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 11000
0
5
10
15
20
25
15 15 25 15 25 20 15 15 15 15 15 7 15 15 15 15 15 15

Fig. 9 Speed of the airship.

Fig. 10 Mobile robot Skif.

If speed of the airship exceeds 10 m/s (the inverse
speed), then aerodynamic control surfaces are used.
The airship control system is implemented as two
separate units. The first unit is a calculation block.
The second unit is a power electronic block. The
calculation block is based on computer PC/104
CMD58886PX1400HR-BRG512 (Rtd Embedded
Technologies). Computer allows to calculate
non-linear dynamic complex controls for the airship.
Navigation system of the airship consists of satellite
navigation system, inertial navigation system, laser
rangefinder, and video camera. In addition the airship
is equipped by sensor of air temperature, humidity,
sensor of atmospheric pressure, and wind sensor.
Actuators are servo-drives with local control systems.
Experimental results of the robust control systems
of the autonomous medium airship are shown in Figs.
8-9.
There are both paths of the autonomous airship as
well as the path speed. The deviation closed-loop
system is about deviation is 28 m
2
for path and 1.5 m
2

for speed. Causes of the deviation are control errors,
navigation errors, and actuators errors.
In addition results of research are implemented in
the wheeled mobile robot Skif, shown in Fig. 10.
8. Conclusions
New design methods of robust control systems
based on the direct Lyapunov method are developed in
this report. Function of Lyapunov is defined for the
block systems. The minimum of the Lyapunov
function is ensured by the robust relay control.
It was proved that the state commonness condition
is satisfied if the control actions are above or equal to
the disturbances.
Further the transformation of the block system to
the single block system was found.
References
[1] V.K. Pshikhopov, M.Y. Medvedev, M.Y. Sirotenko, V.A.
Kostjukov, Control system design for robotic airship, in:
Proceedings of the 9-th IFAC Symposium on Robot
Control, Gifu, Japan, September 9-12, 2009, pp. 123-128.
[2] J.D. Landan, Adaptive Control: the Model Reference
Adaptive Control, New York, Dekker, 1980.
[3] K.J. Astrm, V. Borrison, L. Ljung, B. Wittenmark,
Theory and applications of self-tuning regulators,
Automatica 13 (1977) 457-476.
[4] S.D. Zemlyakov, Some problems of analytical synthesis
in model reference control systems by the direct method
of Lyapunov: Theory of self tuning adaptive control
systems, in: Proc. of 1965 IFAC Symposium on Adaptive
Control, Teddington, England, 1965, pp. 145-152.
[5] V.Y. Rutkouvsky, V.M. Sukhanov, V.M. Glumov, S.J.

162
Dodds, Nonexciting control by orientation of flexible
space vehicle, in: Proc. of the 14-th IFAC Symposium on
Automatic Control in Aerospace, Seoul, Seoul National
University, 1998, pp. 339-344.
[6] M.Y. Medvedev, Design of suboptimal controls for
nonlinear multi-linked dynamical systems, Mechatronics,
Automatics, and Control 12 (2009) 2-8.
[7] V.K. Pshikhopov, M.Y. Medvedev, Block design of
robust systems with bounded controls and state variables,
Mechatronics, Automatics, and Control 1 (2011) 2-8.
[8] E.S. Pyatnickiy, Controllability of Lagrange systems with
limited controls, Automation and Remote Control 12
(1996) 29-37.
[9] V.A. Oleinikov, N.C. Zotov, A.N. Prishvin, The basis of
optimal and extremal control, oscow, High School,
1969.
[10] A.R. Gaiduk, Design of nonlinear systems on base of
controllable Jordan form, Automation and Remote
Control 7 (2006) 3-13.
[11] V.K. Pshikhopov, Time optimal path control of
electromechanical robot manipulator, Electromechanics 1
(2007) 51-57.


Cycle Analysis of Internally Reformed MCFC/SOFC-Gas
Turbine Combined System
Abdullatif Musa
1
and Abdussalam Elansari
2

1. Marine Engineering Department, Faculty of Engineering, Tripoli University, Tripoli 21821, Libya
2. Renewable Energy Authority of Libya, Tripoli 21821, Libya

Received: January 25, 2012 / Accepted: February 06, 2012 / Published: March 25, 2012.

Abstract: The use of high temperature fuel cells for distributed power generation presents advantages related to high electrical
efficiency that can be achieved through hybrid systems and low emissions. The intermediate temperature solid oxide fuel cell
(IT-SOFC) and molten carbonate fuel cell (MCFC) performances are calculated using numerical models which are built in Aspen
customer modeler for the internally reformed (IR) fuel cells. These models are integrated in Aspen Plus
TM
. In this paper, a new
combined cycle is proposed. This combined cycle consists of two-staged of MCFC and IT-SOFC. The combined and single-staged
MCFC cycles are simulated in order to evaluate and compare their performances. Moreover, the effects of important parameters such
as operating temperature, and cell pressure on the system performance are evaluated. The simulations results indicate that the net
efficiency of MCFC/IT-SOFC combined cycle is 64.6% under standard operation conditions. On the other hand, the net efficiency of
single-staged MCFC cycle is 51.6%. In other words, the cycle with two-staged MCFC and IT-SOFC gives much better net efficiency
than the cycle with single-staged MCFC.

Key words: Fuel cells, SOFC (solid oxide fuel cell), MCFC (molten carbonate fuel cell), gas turbine, cycle analysis.

1. Introduction

Currently, the annual global population growth rate
is about 2% while rising even more sharply in many
countries. Consequently, the global energy services
demand is expected to increase dramatically, with
primary energy doubling or tripling over the next five
decades. Therefore, there is a need for renewable and
environmentally benign power production in order to
resolve the seemingly inevitable energy crisis [1].
Fuel cells are electrochemical energy conversion
devices which typically run on hydrogen or methane
or methanol and produce electricity, heat and benign
emissions (water and, in the case of methane and
methanol, CO
2
). The fuel cells used for stationary
energy production are typically high temperature fuel
cells (HTFCs) such as solid oxide fuel cell (SOFC)
and molten carbonate fuel cell (MCFC).

Corresponding author: Abdullatif Musa, Ph.D., research
field: fuel cells. E-mail: Musa@tripoliuniv.edu.ly.
MCFC technology was established about 30 years
ago and has been developed considerably fast in the
USA, Korea, Japan and Europe during the last 10
years. In the literature there have been several studies
published on MCFC and SOFC systems and their
analyses [1-8].
Using serially connect fuel cells has emerged as a
new and highly efficient source of power source. In
Ref. [2] Araki and co-authors analysed a power
generation system consisting of two-stages externally
reformed SOFCs with serial connection of low and
high temperature SOFCs. They showed that the power
generation efficiency of the two-staged SOFCs is 50.3%
and the total efficiency of power generation with gas
turbine is 56.1% under standard operating conditions.
In previously paper [3], Two types of combined cycles
is investigated: a combined cycle consisting of a
two-staged combination of IT-SOFC and HT-SOFC
and another consisting of two stages of IT-SOFC. The
D
DAVID PUBLISHING
Cycle Analysis of Internally Reformed MCFC/SOFC-Gas Turbine Combined System

164
simulation results show that a combined cycle of
two-staged IT-SOFC can give 65.5% under standard
operational conditions. Furthermore, by optimizing
the heat recovery and the gas turbine use, the
efficiency can go up to 68.3%.
In this paper, a new combined cycle is proposed and
investigated. Thermodynamic models for internally
reformed IT-SOFC and MCFC are developed. These
fuel cells are combined in different ways in order
construct single-staged and two-staged fuel cell system
combining different cell types. The aim of the paper is
to find the best configuration for a single-staged or a
two-staged combined system. Therefore, the
performance of two types of cycles is analysed:
combined cycle consisting of two-staged of MCFC and
IT-SOFC and single-staged cycle with MCFC.
2. Cycles Description
The SOFC and MCFC cells currently in operation
are fuelled with natural gas. The high temperature
inside the cells stack allows for reforming the methane
directly inside the cell if the steam is provided at the
inlet. The heat necessary for this reforming reaction is
delivered by the electrochemical reaction in the cell.
Fuel is provided at atmospheric conditions. The fuel is
pure methane (CH
4
). In the cycles part of the anode
gasses is recycled, as the anode gasses contain steam
needed in the reforming reaction. This is a way of
avoiding a steam generator in the cycles. The
characteristics of the systems are given in Table 1.
2.1 Combined Cycle Configuration
Fig.1 shows a cycle diagram of the combined cycle
consisting of an MCFC and IT-SOFC. In this cycle,
the anode flow of the MCFC and IT-SOFC is in
parallel connected. Methane is admitted into the heat
exchanger H/E2 to preheat the methane. The
preheated methane is split into two parts. Part of the
preheated methane is mixed with the recycling anode
gases; the mixture is supplied to the anode side of
MCFC stack. The remaining part of the preheated
methane is mixed with the part of recycling anode
gases and then enters into the anode side in the
IT-SOFC stack. The compressors (C1 and C2) are
used to compensate the pressure drop through the
stacks. In both stacks the remaining part of anode
gases is recycled to the combustor. The combustor exit
gas which contains a major part of air, and CO
2
is split
into two parts. The first part is the cathode inlet gas of
the MCFC stack. The remaining part of the combustor
exit gas and cathode outlet gas of MCFC are mixed,
the mixture is sent to a gas turbine and heat exchanger
(H/E2) respectively. The cathode gases of the
IT-SOFC stack is recycled to the combustor. The
compressed air from the compressor (AC) is supplied
to the heat exchanger (H/E1) and then enters into the
cathode side of the IT-SOFC stack.
2.2 Single-Staged Cycle Configuration
The single-staged MCFC cycle is similar to the
combined cycle (Fig. 1), except that there is no
IT-SOFC stack. The combustor exit gas is split into
two parts. The first part of the combustor exit gas and
the compressed air from the compressor (AC) are
mixed. This mixture is sent to the heat exchanger
(H/E1) and then enters into the cathode side of the
MCFC stack. The remaining part of the combustor
exit gas is mixed with part of the cathode outlet gas,

Table 1 Setting parameters of the cycles.
Setting parameter Value Setting parameter Value
Current density 0.250 Acm
-2
Steam-to-carbon ratio 2
Active cell area 250 m Pressure drop in combustor 0.2 bar
Total fuel utilization rate 85% Pressure drop in SOFC 0.01 bar
Compressors isentropic efficiency 80% Pressure drop in heat exchangers 0.02 bar
Gas turbine and pump isentropic efficiencies 85% Gas turbine and compressor mechanical efficiencies 98%
Fuel recirculation rate 55% Pressure drop in MCFC 0.05 bar

165

Fig. 1 Configuration of the MCFC/IT-SOFC combined cycle (AC: air compressor; FC: fuel compressor; GT: gas turbine;
H/E: heat exchanger).

the mixture is sent to a gas turbine and heat exchanger
(H/E2) for recovering energy. The remaining part of
the cathode outlet gas is recycled to the combustor. In
this cycle part of the preheated methane is bypassed to
the combustor.
3. Water Gas Shift and Methane Reforming
Reactions
In the models, the chemical reactions are assumed
to be in equilibrium. This means that the reactions
occur instantaneously and reach the equilibrium
condition spontaneously at each position.
For SOFC model the electrochemical reaction is
implemented:

+
2
2 2
1
2 O e O

cathode (1)

+ + e O H O H 2
2
2
2

anode (2)
O H O H
2 2 2
1
2
+

overallreaction (3)
For MCFC model the electrochemical reaction is
implemented:

+ +
2
3 2 2
1
2
2 CO e O CO

cathode (4)

+ + + e CO O H CO H 2
2 2
2
3 2

anode (5)
O H O H
2 2 2
1
2
+

overallreaction (6)
The IT-SOFC and MCFC operate at a temperature
high enough to reform the steam reforming inside the
anode. The electrochemical and water gas shift
reactions are exothermic and, on the other hand, fuel
reforming is a strongly endothermic reaction. The
chemical reactions in the steam reforming and shifting
of methane are shown below:
2 2 4
3H CO O H CH + +
(7)
2 2 2
H CO O H CO + +
(8)
4. High Temperature Fuel Cells Models
Description
The fuel cell is treated as a single control volume in
which the steady state flow energy equation is applied
with the assumption of negligible change of kinetic
energy and potential energy. To determine actual cell
performance, the overpotential must be deducted from
the Nernst potential ( E ) which represents the ideal
performance.
loss cell
V E V =
(9)
4.1 Evaluation of Voltage Drops in IT-SOFC
The model of the IT-SOFC stack used in this paper
is based on an existing IT-SOFC model [4]. In the
651 C
Exhaust 353 C
Fuel
H/E2
H/E1
Combustor
1045 C
800 C
368 C
Air
15 C
CH
4

15 C
FC GT AC
P
MCFC

Anode
Cathode
MCFC
P
IT-SOFC

Anode
Cathode
IT-SOFC
C1
C2
634 C
656 C
750 C
3500
kg/h

166
model the cell will utilize an electrolyte with the
thickness of 25 m, a 50 m thick cathode and 250
m thick anode.
The cells are assumed to be stacked between bipolar
interconnect plates. The cell resistance is assumed to
have the following temperature dependence [4].
=
T
R
tot
exp
(10)
where T represents the cell operating temperature and
is the Boltzmann constant. The activation energy
(
19
10 01 . 1

= J) and the pre-exponential factor
(
4
10 98 . 2

= ) are chosen to fit the experimental
data reported by Ref. [4].
The cell voltage of the IT-SOFC can be calculated by
( )
tot
O H
O H
cell
iR
p
p p
F
RT
K
F
RT
V
+ =
2
2
1
2 2
ln
2
ln
2
(11)
4.2 Evaluation of Voltage Drops in MCFC
In this article, the internal resistances of the MCFC
stack can be calculated as in Ref. [5]. The ohmic cell
resistance (
ohm
r ), anodic reaction resistance (
an pol
r
,
)
and cathodic reaction resistance (
ca pol
r
,
) are reflected
in the functions of the reaction temperatures and
partial pressure of gas constituents. The ohmic cell
resistance included the ionic and electric resistance; it
is calculated using an Arrhenius equation as function
of the operating temperature (Eq. (12)).
=
923
1 1
3016 exp 5 . 0
cell
ohm
T
r
(12)
1 17 . 0 42 . 0 5
,
2 2 2
6435
exp 10 27 . 2

=
O H CO H
cell
an pol
P P P
T
r
(13)
09 . 0
2
34 . 0 6
,
2
9298
exp 10 505 . 7

=
CO O
cell
ca pol
P P
T
r
(14)
The cell voltage of the MCFC can be calculated by
( )
ca pol an pol ohm
O ca CO H
an CO O H
cell
r r r i
p p p
p p
F
RT
K
F
RT
V
, ,
2 2 2
2 2
ln
2
ln
2
+ +
=
(15)
5. Results and Discussion
The cycles layout under investigation is simulated
under the standard operating conditions shown in
Table 1. The aim of this analysis is to evaluate and
compared the performance of the cycles.
5.1 Performance of the Combined Cycle
In the simulations, the operating temperature of
MCFC and the cell pressure of stacks were changed to
see their effect on the MCFC/IT-SOFC combined
cycle performance. The total fuel utilization rate of the
combined cycle is 85% (42.5% in the MCFC, 42.5%
in the IT-SOFC). In the following analyses, the
triangles () indicate the net efficiency of the
MCFC/IT-SOFC combined cycle and the solid squares
() indicate the efficiency added by both MCFC and
IT-SOFC.
5.1.1 Operating Temperature of MCFC
Fig. 2 shows the changes in the net efficiency of the
combined cycle with the operating temperature of the
MCFC and gives an idea about the contribution of the
cells and the gas turbine to the net efficiency in the
combined cycle. Increasing the operating temperature
of the MCFC decreases the electrolyte resistance in
MCFC stack, the efficiency of MCFC increase. In the
bottoming gas turbine cycle, increase the operating
temperature of the MCFC causes an increase in the
outlet temperature of the MCFC. This means that the
turbine inlet temperature increases with increasing
turbine output. Since the increase in the MCFC and
gas turbine efficiencies are larger than the decrease in
the IT-SOFC efficiency, the net efficiency of the
combined cycle goes up.
5.1.2 Operating Pressure
combined cycle with cell pressure. Raising the cell
pressure causes an increase in the Nernst potential. Also
increasing the operating pressure leads to some
decrease in the overpotential, resulting in an increase
in the net efficiency in the combined cycle. An
increase in the compressors outlet temperature due to

167

Fig. 2 Change of net efficiency of combined cycle with
operating temperature of MCFC.

Fig. 3 Change of net efficiency of combined cycle with cell
pressure.

the increase of the operating pressure causes an
increase in the turbine inlet temperature, resulting in
an increase of gas turbine output. In particular, the
higher the operating pressure, the higher the cell
voltage determining a remarkable improvement of the
electrical of the system. It is also obvious that when
the value of the operating pressure is increased, higher
compressors and turbine costs must be considered [9].
5.2 Comparison of Performances between the
Single-Staged MCFC and Combined Cycles
Fig. 4 shows the changes in the total power of the
cycles with mean current density of MCFC. The total
power of the cycle is the power produced by the fuel
cells and turbine, minus the compressors power.
Increasing the current density increases the fuel cell
stacks overpotential, resulting a decrease of the cell
voltage. Since the increase in the current density is
larger than the decrease in the cell voltage, the electrical
outputs of the fuel cell stacks (Fig. 5) goes up.
The increase in the total fuel consumption in the

Fig. 4 Total powers of the cycles as a function of current
density of MCFC.

Fig. 5 Fuel cell stacks power of the cycles as a function of
current density of MCFC.

fuel cell stacks and the combustor in the cycles with
mean current density (Fig. 6) causes an increase in the
amount of inlet gas to the gas turbine; this means the
electrical output of gas turbine also goes up.
cycles with mean current density. Increasing the
current density increases the overpotential of stacks,
resulting in decrease of the efficiency of the stacks.
When the current density increases the MCFC outlet
temperature and the exhaust fuel, which is sent to the
gas turbine increase together with the increasing
turbine output. Since the decrease in the efficiency of
both stacks is larger than the increase in the efficiency
of gas turbine, the net efficiency of the cycles goes
down. As the total power of the cycles goes up (Fig. 4)
and total fuel consumption in the fuel cell stacks and
the combustor in the cycles changes greatly, net
efficiency of the cycles goes down. Though the fuel
consumption is lower in the single-staged MCFC
cycle, the total power is also lower, causing the
reduction in efficiency.
50
52
54
56
58
60
62
64
66
68
70
600 625 650 675 700
E
f
f
i
c
i
e
n
c
y

(
%
)

Operating temperature of MCFC (C)
MCFC/IT-SOFC cycle
MCFC+IT-SOFC
25
30
35
40
45
50
55
60
65
70
75
1 3 5 7 9 11 13 15 17
E
f
f
i
c
i
e
n
c
y

(
%
)

Cell Pressure (bar)
MCFC/IT-SOFC cycle
MCFC+IT-SOFC
200
400
600
800
1000
1200
1400
1600
0.09 0.11 0.13 0.15 0.17 0.19
T
o
t
a
l

p
o
w
e
r

(
k
W
)

Current density of MCFC (A/cm)
single-staged
MCFC cycle
MCFC/IT-SOFC
combined cycle
100
200
300
400
500
600
700
800
900
0.09 0.11 0.13 0.15 0.17 0.19
P
o
w
e
r

(
k
W
)

MCFC
IT-SOFC

168

Fig. 6 Total fuel consumption on the cycles as a function
of current density of MCFC.

Fig. 7 Change of the net efficiency with mean current
density of MCFC.
6. Conclusions
In this article, a new combined cycle is proposed
and investigated: a combined cycle consisting of
two-staged of MCFC and IT-SOFC. Moreover, the
combined and single-staged cycles are simulated in
order to evaluate and compare their performances.
Also, the effects of changing operating temperature of
MCFC and the operating pressure on the efficiency of
MCFC/IT-SOFC combined cycle are evaluated.
The simulation results show that the cycle
efficiency increased when the operating temperature
of MCFC and operating pressure increase.
Furthermore, the net efficiency of the combined and
single-staged MCFC cycles are 64.6% and 51.6%
respectively under standard operating conditions of
the cycles were mean current density of 0.150 Acm
-2

for MCFC, operating temperature of 650 C for the
MCFC, cell pressure of 6 bar, operating temperature
of 750 C for the IT-SOFC, and fuel recirculation rate
of 55%. In other words, the cycle with two-staged
MCFC and IT-SOFC gives much better net efficiency
than the cycle with single-staged MCFC.
Acknowledgment
The authors acknowledge University of Tripoli,
Tripoli, Libya for the support of this work.
References
[1] R. Rashidi, P. Berg, I. Dincer, Performance investigation
of a combined MCFC, International Journal of Hydrogen
Energy 34 (2009) 4395-4405.
[2] T. Araki, T. Ohba, S. Takezawa, K. Onda, Y. Sakaki,
Cycle analysis of planar SOFC power generation with
serial connection of low and high temperature SOFCs, J.
Power Sources 158 (2006) 52-59.
[3] A. Musa, M.D. Paepe, Performance of combined
internally reformed intermediate/high temperature SOFC
cycle compared to internally reformed two-staged
intermediate temperature SOFC cycle, International
Journal of Hydrogen Energy 33 (2008) 4665-4672.
[4] J. Plasson, A. Selimovic, P. Hendriksen, Intermediate
temperature SOFC in gas turbine cycles, in: ASME Turbo
Expo Conference, USA, 2001.
[5] S. Kimijima, N. Kasagi, Cycle analysis of micro gas
turbine-molten carbonate fuel cell hybrid system, JSME
Inter. J. Series B 48 (2005) 65-74.
[6] A. Musa, H. Steeman, M.D. Paepe, Performance of
internal and external reforming molten carbonate fuel cell
systems, Journal of Fuel Cell Science and Technology 4
(2007) 65-71.
[7] M. Baranak, H.A. Atakl, Basic model for analysis of
molten carbonate fuel cell behaviour, J. Power Sources
172 (2007) 831-839.
[8] V. Verda, F. Nicolin, Thermodynamic and economic
optimization of a MCFC-based hybrid system for the
combined production of electricity and hydrogen, I. J.
Hydrogen Energy 35 (2010) 794-806.
[9] F. Calise, M.D. dAccadia, A. Palombo, L. Vanoli,
Simulation and exergy analysis of a hybrid solid oxide
fuel cell (SOFC)-gas turbine system, J. Energy 31 (2006)
3278-3299.

25
50
75
100
125
150
175
200
0.09 0.11 0.13 0.15 0.17 0.19
F
u
e
l

(
k
g
/
h
)

single-staged
MCFC cycle
MCFC/IT-SOFC
combined cycle
45
50
55
60
65
70
75
0.09 0.11 0.13 0.15 0.17 0.19
E
f
f
i
c
i
e
n
c
y

(
%
)

single-staged MCFC cycle
MCFC/IT-SOFC combined cycle

Design, Development and Testing of a Wireless
Multi-Sensors Network System
Chelakara Subramanian
1
, Jean-Paul Pinelli
2
, Ivica Kostanic
3
and Gabriel Lapilli
1
1. Mechanical and Aerospace Engineering, Florida Institute of Technology, Melbourne 32901, USA
2. Civil Engineering, Florida Institute of Technology, Melbourne 32901, USA
3. Electrical and Computer Engineering, Florida Institute of Technology, Melbourne 32901, USA

Received: December 21, 2011 / Accepted: January 04, 2012 / Published: March 25, 2012.

Abstract: Full-scale field testing is necessary to understand and model the effects of natural hurricane winds on residential buildings.
Towards this goal, a group of researchers at Florida Tech. has developed a wireless multi-sensors network system to specifically
study wind induced roof pressure variations. The system has the capability of measuring pressure and temperature on a roof,
collecting data and sending it to a server to process and publish on the web in nearly real-time. Also wind speed and direction are
measured by the system with the use of an anemometer. The latest third generation system is suitable for 3 house installations, each
comprising of 30 remote pressure and temperature sensors, and 1 remote anemometers. Sensors are placed inside individual
custom-made plastic weatherproof cases. Small size of all components allows an aerodynamic shape, reducing the shape effect of the
sensor on the flow. Low power consumption combined with Li-ion batteries provide several days of continuous data collection. The
platform created allows using almost any type of sensor via a planned expansion port. This system has important wireless capabilities,
including individual sensor control, self-negotiating communication, and excellent data throughput. The software development effort
for the system included 2 major components: the embedded firmware written in PIC Basic that is flashed to the PIC processor on
each portable circuit board, and the end-user control and real-time display client written in C# that runs on a host computer with a
base unit connected through the USB port. The data processed from a three-house network deployment near the Florida east coast
confirmed the robustness of the network system by synchronously and continuously collecting data as well as uploading it to a
central server. The pressure sensors showed excellent precision, +/-0.68 mbar, with an accuracy of 0.0404 mbar, and sampling rate of
over 30 samples per second for a house deployment of 30 sensors.

Key words: Wireless sensors, hurricane monitoring, pressure sensors, anemometer, sensor network.

1. Introduction

Hurricanes are one of the main causes of natural
disasters. Through history, billions of dollars have
been lost as a result of these meteorological disasters.
The state of Florida bears the record of being hit by
the highest number of tropical or subtropical cyclones.
Since 1851 only eighteen hurricane seasons have
passed without a known storm impacting the State. In
total, over ten thousand human deaths have occurred
in Florida only.

Corresponding author: Chelakara Subramanian, professor,
program chair, Ph.D., research fields: fluid dynamics, wind
engineering, instrumentation, wind tunnel testing. E-mail:
subraman@fit.edu.
For decades, engineers and scientists have
attempted to understand the effect of a storm on many
particular type of structure in order to design
structures capable to withstand the hurricane wind
forces. For this, most of the wind engineering tests
have been conducted on scale models. But the
ultimate goal of predicting the real storms effects from
these model test results has not been very successful.
Nowadays, with the advanced computational
capabilities, better mathematical/physical models are
being developed to resolve the problem. For
establishing the reliability of prediction models we
need real storm measurement data. Joint research
efforts, such as the Florida Coastal Monitoring
D
DAVID PUBLISHING
Design, Development and Testing of a Wireless Multi-Sensors Network System

170
Program aim to create a reliable real storm database to
allow the scientific community to validate their
models.
1.1 Related Research
The Florida Coastal Monitoring Program (FCMP),
a joint-venture with University of Florida, Clemson
University, Florida International University, the
Institute for Business and Home Safety and Florida
Institute of Technology was started in year 1999, and
has a repository of more than 20 hurricanes/tropical
storms data [1]. The Programs focus is on developing
and using full-scale experimental techniques to
characterize and quantify near-surface hurricane wind
velocity and uplift loads on residential structures. The
Program instrumentation comprises of several
portable towers to measure wind velocity at two
different heights (as well as other quantities like
pressure and temperature) and roof pressure sensor
systems such as the one presented here. The previous
generation pressure sensors used wired connections,
making their deployment cumbersome. The new
wireless system described here eliminates the need of
wiring, thus enabling better portability and handling.
1.1.1 Wind Tunnel Testing
Although relatively good agreement was observed
in the general trend between models and full-scale
wind tunnel tests, the peak pressure coefficients have
shown significant differences [2-3]. This is due to
inadequate resolution of time scales and length scales
of real motions. Studies performed in Ref. [4] show
that peak pressure coefficients are affected by
small-scale turbulence, thus simulating only average
parameters of the flow does not necessarily give
accurate peak pressure results.
Modern research [5] is concentrated on testing on
real-size buildings by using multiple fan wind tunnels
controlled automatically by a computerized system.
This is able to reproduce entire time histories of wind
speed, occurring during real storms. The ultimate
facility built under this concept is the Insurance Center
for Building Safety Research (ICBSR) created by the
Institute for Business & Home Safety (IBHS),
currently in its calibration stage [6], and the Wall of
Wind at Florida International University [7].
1.1.2 Full-Scale Field Testing
The first attempts to measure pressure in full scale
buildings were performed during the 1970s in a
house-like structure with artifacts to modify the roof
slope [8]. Pressure was measured using differential
pressure transducers at different locations on the roof,
with all the tubing connected to the interior of the
house. The next stage of full-scale experiments was
conducted in the 80s and 90s at Texas Tech
University on low-rise structures [9-10]. Also, during
the 1980s several structures were instrumented at the
Silsoe Research Institute. A review of these
experiments was published in Ref. [11]. All those
systems used differential pressure transducers
connected by tubing to a box of common reference
[12] pressure, whose location influenced the measured
pressure coefficients.
Another low-rise building was instrumented by the
John Hopkins University. The system used a
complicated system of tubes with drains to avoid
water intrusion and clogging, requiring an entire room
for the equipment [13]. Permanent installations like
these caused Insects, bugs and dirt problems. This lead
to the creation of systems that would be temporarily
installed only during a storm, while at other times
properly kept in a clean place.
Clemson University developed the first of such
systems under the concept of removable, independent
units that can be attached to the roof surface at
pre-defined locations. This avoided drilling holes,
passing tubes, using valuable space inside the building
and keeping maintenance efforts down to a minimum,
which made it an ideal system for instrumenting
private homes [11]. This was the first system that used
absolute pressure transducers. It was still a wired
system, for which large amounts of cables needed to
be permanently installed outside the house, to which

171
the sensors would be connected at the time of
deployment. Unfortunately the permanent installation
of wires experienced weathering problems such as rust
and corrosion causing bad electrical connections
during deployments.
In order to avoid the problems caused by the wires,
the concept of a completely wireless system was
developed at Florida Institute of Technology, using a
processor and a pair of wireless transmitter/receivers
powered by a rechargeable battery. At the beginning,
the data had to be sampled at very low rates due to
difficulties in synchronizing data acquisition and
communication hardware timings, but as technology
progressed the first complete generation of sensors
was released, using one transceiver supporting
bidirectional communication along with a
Programmable Interface Controller (PIC) processor
[14]. A second generation system was created as a
major upgrade over the first one, using a better
processor, better transceiver and improved battery life
capabilities [15] and a better software approach [16].
The scope of this paper is to describe the latest
version of the wireless sensors network system [17].
Section 2 provides a very detailed description of the
wireless sensors system, including explanations on the
hardware, software (communication and data
processing), and calibration. Section 3 describes the
systems performance, including precision, accuracy,
signal range and speed. Here the results of the sensor
network performance tests conducted on three houses
during non-hurricane periods are also discussed. The
conclusions are provided in section 4.
2. System Description
An overview diagram of the system is presented in
Fig. 1. The overall network consists of three major
sub-components:
House InstallationThis sub-component
includes the remote sensor units, base unit and
laptop that are installed on the house premises at
deployment time. This sub-component is responsible

Fig. 1 System diagram.

for data acquisition, with up to 30 pressure sensors
(one of them is connected to a pressure port at some
distance from the house to measure reference
pressure), plus up to two anemometers.
The base unit and the field laptop computer are
placed in a water proof box in the vicinity of the house.
The collection laptop is connected to the 110 VAC
(volts of alternating current) power supply from the
house as well as to a portable external battery for
backup in case of a power outage.
The laptop, connected through Universal Serial Bus
(USB) to the base unit, talks to sensors either directly
or by hopping through routers, devices that allow
expanding the range of the system by retransmitting
signal and acting as a robust intermediary in the
wireless data connection.
Communication NetworkThis sub-component
is the link connecting individual house installations to
a central server, either through Wi-Fi or cellular data
networks (3G3rd Generation). Through this link
passes all the information needed for monitoring,
control and data uploading. In case this fails, all the
data is stored locally in the laptop computer by the
house, which can be retrieved after the storm passes or
when the connection is reestablished.
Central ServerFinally, the third sub-component
is the central server itself, which receives and stores
measured data and enables a network operator to
control the operation of the base and sensor units,
through remote login to the remote laptop. As this is
done through the world-wide web, the remote logging
Remote sensor
units
wireless
card
cellular tower
Public Internet
Cellular data
network
Central server
Public Internet
Laptop +base unit
wireless
card
cellular tower
House installation
Laptop +base unit
Remote sensor
units
Communication network

172
in operation can be done from a computer anywhere in
the world (not restricted to the localized server).
2.1 Hardware
The new third generation hurricane sensors have
several improvements over the previous systems. The
sensors use an over clocked 48 MHz processor
(previous version was running at 20 MHz) and a serial
bus speed of 115,200 bps (bits per second), which
allows direct USB firmware updates, i.e. the board is
directly connected to a computer USB port to change
the software running on the processor.
A new transceiver (Xbee-PRO) was also
incorporated, specially designed for mesh networks
[18] using the IEEE 802.15.4 protocol [19]. The board
also features an embedded battery charger, which
eliminates the need for removing the battery out of the
case for external charging. The sensors can be fully
charged for deployment in 6 hours.
The other features of the sensor include:
Lightweight plastic casing, sealed with standard
O-ring;
Better aerodynamic casing shape (improved
geometry and absence of exterior antenna);
Lithium-Ion (Li-Ion) battery, allowing a
continuous operation of approximately 120 hours
between charges;
12-bit analog-to-digital converter (ADC);
Expansion port, with the possibility to mount
almost any sensor (like accelerometer or strain gage)
with voltage output;
PIC processor flashing and programming via
USB port;
Improved on-the-air characteristics, with
individual control for each sensor. This improves the
network reliability, i.e., if one sensor is turned off the
rest of the system continues to work.
2.1.1 Basic Board Layout
A PIC processor with custom software running on it
is powered by a 3.7 Volt Li-Ion battery [20]. It is
connected through a voltage regulator, which provides
a constant voltage to all the components on the board.
An oscillator working at 48 MHz provides a square
wave to the processor and the transducers at that
frequency, for timing.
The temperature and pressure transducers, which
generate voltages in the range of 0 to 3 Volts, provide
the analog signals to digital converters (ADC) in the
processor, which are translated into digital binary
values. This information is collected and assembled
into packets and sent via a serial link [21] to a wireless
transceiver, devices specially designed to send
information wirelessly.
(1) Transmitter/receiver: To determine the power
required for the transceivers, a basic log-distance path
loss model [22] is used:
I = 1u n log
10
(J) + C
where I is the path loss expressed in dB, n is the
path loss exponent (equal to 2 for free space, 4 for
lossy environments), J is the distance in meters
between transmitter and receiver and C is a constant
that accounts for system losses. For an extreme case
scenario, where the system (negligible system losses)
should be able to sustain a communication between
two points 50 meters away from each other, in an
environment prone to losses:
I = 4u log
10
(Su)
I = 67.9S uBm
This loss can be accommodated by
Using low-power XBee modules: 2 mW transmit
power (i.e., +3 dBm),-96 dBm receiver sensitivity.
That gives a total allowed loss of 99 dBm;
Using XBee-PRO modules: 63 mW transmit
power (i.e., +18 dBm),-102 dBm receiver sensitivity.
That gives a total allowed loss of 120 dBm.
According to the numbers estimated above, the
low-power XBee modules should perfectly be able to
establish a good communication. However, after
extensive testing on these modules, with different
types of antennas, locations and data rates it was
found that in order to push the amount of data
required by a fast sample rate (i.e., 30+ samples per

173
second) the retransmissions must be reduced to almost
zero. As the low-power XBee units do not allow
control of the handshaking/data retransmission
protocols, the only way around this problem is to use
more power. The low-power XBee modules proved
inefficient in a full system, even with the use of
routers. Thus, the XBee-PRO modules were used
subsequently.
The XBee-PRO transceivers create a network with
an address for each unit, allowing the system to
identify each sensor and have precise control over it.
In case of an error, a hardware reset of the PIC can be
performed wirelessly, as long as the XBee-PRO unit is
still active.
(2) Processor: The PIC18LF2553 [23] processor
has a built-in timer to accurately sample the sensors at
an interval specified by the control software. The
timer is used to create high priority interrupts in the
firmware routines. This means that the processor will
halt any other process and take a data sample. This
enables the sensor to have extremely precise timing.
The processor stores the data in a buffer using a delta
compression method, i.e., only the difference with the
previous value is stored. Any value n in the series
can be expressed as
p
n
= p
0
+ i
n
=1

where p
n
is the value of nth sample data, p
0
is the
initial value of the series data stored in every packet,
and i is the each incremental stored value with
respect to the previous value.
Once 60 samples are collected, the buffer content is
sent to the Xbee-PRO module as a packet (along with
other information). The transceiver then puts it in a
queue to be sent to the base. This queue is
automatically arranged by the mesh network that the
base creates, without the need to manually organize it.
Once the base receives the packet, it is sent to the PC
(personal computer) using the Xbee API (Application
Programming Interface) frames. Each frame contains,
in addition to the pressure and temperature data,
information about the sensor that transmitted it, such as
Sensor type: router or endpoint (pressure sensor
or anemometer);
Unique hardware ID (identifier): A 16-digit
number (analog to the Media Access Control (MAC)
address);
Parent network address: In order to identify the
network structure;
Battery level: In percentage;
Sequence ID: A correlative 8-bit unsigned
number (i.e., ranging from 0 to 255) assigned to every
packet. This helps to keep track of packets on the
receiving side and to organize them during
post-processing.
This information is used to update the GUIs
(Graphical User Interface) tree. The software logs the
entire packet as it is received in a comma-separated
values file (.csv extension) for post analysis.
2.1.2 Battery/Voltage Regulation
The battery was originally chosen to last for about 5
days in continuous data collection. The board draws
constantly about 50 mA, therefore with a 6000 mAh
battery the total theoretical life span should be 120
hours. These values were confirmed experimentally.
When the XBee-PRO module was used in the board,
a problem appeared regarding the voltage regulation
and power circuitry; the current required by the
transceiver during transmission was 205 mA during a
very short time. This caused the processor to behave
erratically and even reset itself due to low voltage,
since the voltage regulator circuit was not able to
provide such power. This problem was fixed adding a
47 F capacitor in parallel to the one in the 3 V
circuit.
2.1.3 Pressure Transducer
The pressure transducer used is the Freestream
Electronics MP3H6115A with case 1317A-04
(sensitivity: 27 mV/kPa). The sensor board was
designed to be used for many different applications.
In its generic form, the PIC18LF2553 [23]
processor has ports to accept different sensors. In this
specific application, it measures pressure and

174
temperature using the built-in analog sensors, which
are sampled using the processors onboard 12-bit A/D
converter. The resolution obtained with this
combination is
Rcs
con
=
Ioltogc Spon
2
12
=
SI
2
12
= u.7S242
mI
bit
,
Rcs
pcss
=
Rcs
con
Scnsor Scnsiti:ity
=
u.7S242
mI
bit
,
2.7
mI
mbor
,
= u.27
mbor
bit
,
The temperature resolution was about 1
C/bit.
2.1.4 Pressure Port
The R.M. Young Model 61002 Pressure Port is used
to minimize dynamic pressure errors due to wind.
With this the dynamic pressure error is typically less
than 0.5 hPa (hecta-Pascals) at 20 m/s. This Port is
mounted on the same mast as the anemometer but just
below it. The measured pressure served the purpose of
reference pressure.
2.1.5 Anemometer
The anemometer board is similar to the pressure
sensor board, except for the fact that it does not have
the pressure and temperature transducers and uses the
expansion port to connect the output of the R.M.
Youngs anemometer. This expansion port allows
adding any other type of sensors with voltage output
up to 3 V (for example, strain gauges, accelerometers,
position gyros, humidity sensors, light sensors, etc.).
The firmware running in the processor is different as
compared to the pressure and temperature sensors,
because we are now using the expansion slot to collect
data, and a different logic for assembling the data
packets.
The anemometer board is assembled in a poly vinyl
chloride (PVC) weatherproof box with an on/off plug
and a signal connector.
Due to the higher power and voltage needs of the
R.M. Youngs wind anemometer, it is powered by a
11.1 Volt Lithium-Polymer 5400 mAh battery, which
needs to be charged with a balanced charger.
2.1.6 Routers
These are used to extend the range of the sensors.
They act as a bridge between the base and 1 to 10
endpoints (sensors). The endpoints can also connect
directly to the base with no configuration required.
The only differences between sensors and routers
hardware is the absence of transducers on the router
board and the firmware running on the PIC processor,
which only reports battery life to the base unit.
2.1.7 Base Unit
It consists of three separate XBee-PRO units on
USB to serial converters that are connected to a
commercial USB hub to communicate with the
computer using only one physical port.
Each unit belongs to a predefined system,
determined by the channels to which it is set by
default. This configuration can be changed using the
Digi-Key X-CTU software [24].
2.1.8 Sensor Casing
They were designed to be weatherproof and
aerodynamic. The Xbee-PRO has a built-in antenna
which allows for a more aerodynamic dome shape
without external protrusions that disturb air flow. The
first prototype mold was custom-made using a rapid
prototype printer. Then, urethane plastic [25] was used
to create up to 15 cases from each mold.
The technique was used to make manually 93
sensors for the three systems, although for mass
production plastic injection method is advocated. The
assembly exploded view of the pressure sensor is
shown in Fig. 2. It consists of six basic parts: (1) Base
plate (plastic), (2) Top dome (plastic), (3) Pressure
plug (plastic), (4) Stiffening ring (aluminum), (5)
O-Ring 164.8 mm 3.6 mm-260 BN70 (commercial),
(6) Clear PVC hose 3/32 Inner Diameter, 5/32 Outer
Diameter (commercial).
2.2 Software
The PIC processor built in to the circuit boards runs
on the embedded firmware written for the system. Out
of the three types of hardware boards (bases, routers,
and endpoints) with Xbee-PRO Transceiver modules
only the endpoints and routers have a dedicated
processor with custom firmware running on it. The
endpoints firmware is responsible for sampling data

175

Fig. 2 Sensor assembly.

from the sensors (one version for anemometers and
another for pressure plus temperature sensors), while
the routers firmware is responsible for only reporting
battery level back to the base.
2.2.1 Firmware Design and Operation.
The endpoint PICs, whether running anemometer or
pressure plus temperature sensor specific firmware,
have 3 operating states that can be commanded by the
control client software (described in the following
section). They are
Sleep Mode: While in this mode, the processor is
set into a low power state. A hardware interrupt is set
to fire once every 5 minutes at which time a routine is
executed that will send a heartbeat message to the
base, indicating remaining battery life;
Idle Mode: While in Idle Mode, a heartbeat
message indicating remaining battery life will be sent
to the base once every 5 seconds;
Data Collection Mode: In the data mode the
firmware establishes a synchronized data pipeline
between the external sensors and the Xbee-PRO
transceiver module. The sensor data is sampled via a
hardware interrupt mechanism that is configured to
fire at a rate specified by the control client application.
A synchronized, shared buffer is filled as new data
samples are taken. Once a certain number of samples
is collected a data packet is assembled (i.e., appends
the contents of the buffer with sensor data along with
other status information in the form of a pre-defined
frame), another constantly running routine (not
interrupt based) sends the packet to the Xbee-PRO
module for transmission to the base. The architecture
is essentially a producer-consumer relationship where
the external sensor interrupt routine is the producer
and the always-running program loop which passes
data to the Xbee-PRO module is the consumer.
For the temperature and pressure sensor firmware
configuration, the message sent to the base while in
Data Mode contains 60 pressure samples, 1
temperature sample, battery percentage, and a
message sequence ID number. For the anemometer
configuration, the message contains 29 wind speed
samples, 14 wind direction samples, battery
percentage, and a sequence ID number.
The sequence ID number is a correlative number
(integer cycling from 0 to 255) written by the
processor after a packet is assembled. This way, the
number reflects an accurate time interval and is used
to keep track of packet generation time for post
processing purposes.
2.2.2 Timing
The interrupt routine is handled by a timer with a 1
ms time resolution. Therefore the time t (Fig. 3) is an
integer in milliseconds. This has a slight implication
on the sample rate chosen: The sample time command
for t sent to the endpoint is truncated into an eight-bit
integer. As an example, lets assume that we set 30
samples per second in the Control Client Application.
This means that every sample should have a duration
of
1uuu
ms
s
Su
somplcs
s
= SS.SSSS
ms
somplc

This number is truncated to 33 in order to make it
an integer word, which would be 21h (in hexadecimal),
and that is the value sent to the sensors, meaning that
the real sample rate will be

176
Fig. 3 Data

S
This diffe
processing o
2.2.3 Con
The Cont
status, cont
display sens
sensor data.
User Interfa
Windows F
real-time gra
for Xbee ne
Messaging P
back-end po
main compo
The X
provides an
Programmin
Xbee-PRO
makes exten
The Xbe
definitions f
communicat
network nod
input/output
the applicati
meaning tha
new serial d
All API fr
Design,
path of sensor
1uuu
ms
s
SS
ms
somplc
=
erence is taken
of the data.
ntrol Client Ap
trol Client ap
trol individu
sor data in re
. The design
face that is
orms, the Ze
aphing, and c
etwork status
Protocol and
ortion of the s
onents:
Xbee-PRO M
easy to use
ng Interface (
modules. T
nsive use of th
ee-PRO Mes
for all of the
te with the
des. This cla
t (I/O) betwe
ion itself. All
at callback ro
ata is availab
frames are rep
, Developmen
r endpoint whil
= Su.SuSu
som
n into accoun
pplication
pplication is
ual or group
eal-time, and
n is a user-fr
written in
edGraph [26
custom-built
s and endpoin
d Back-end c
software is co
Message Bro
, object orie
API) for com
The main a
his class.
ssage Broker
API Frames
base, route
ass also hand
een the base
l serial input
outines are ex
ble on the por
presented as f
nt and Testing
le in data mod
mplcs
s

nt during the
used to mon
p of endpo
generate log
iendly Graph
C# and util
] framework
GUI compon
nt status disp
components:
omprised of t
okerThis c
nted Applica
mmunicating w
application l
r class cont
s that are use
er, and endp
dles all the s
e transceiver
is asynchron
xecuted when
rt.
first-class obj
g of a Wireles

de.
post
nitor
oints,
gs of
hical
lizes
k for
nents
play.
The
three
class
ation
with
layer
tains
ed to
point
erial
and
nous,
never
jects
that
app
auto
buf
for
cata
vari
be
netw
T
inte
the
the
Sof
of c
12 t
the
12.
327
no l
vali
netw
mai
chil
wir
W
rece
rece
to t
that
Mo
data
The
the
Fun
thou
the
Net
logi
ss Multi-Sens
t can be easi
plication-layer
omatically ha
ffer and dese
all API frame
The Xbee-P
alog of all d
ious statistic
looked up b
work address
The topology
ernally by an
tree is the C
host machin
ftware. The C
children route
total child no
same combin
The theoret
768 nodes, bu
larger than 1
id parent net
work address
intain a pare
ld network a
eless network
When the Xb
eipt of a ne
eived from an
the Xbee-PR
t transmitted
nitor based o
a received is
e RF buffer i
firmware run
Application
nctionalityT
ught of as a
Xbee-PRO
twork Monito
ic is impleme
sors Network
ily created an
r code. The M
andle serializ
rialization fr
es.
PRO Network
discovered X
s about thos
based on the
s, or their uniq
of the wirele
N-ary tree d
Coordinator m
ne which is ru
Coordinator c
ers or endpo
odes. Each ro
nation of chil
tical limit fo
ut in practice
2 nodes. Eac
twork addres
s. By having
ent network
address, we
k topology in
bee-PRO Me
w radio freq
n endpoint, t
RO Network
the packet is
on its unique
s added to th
s the assemb
nning on the e
n Logic
The Applica
bridge betwe
Message
or layers (Fi
ented, decidi
k System
nd manipula
Message Brok
zation from o
rom byte buf
k MonitorI
Xbee-PRO m
e modules. M
eir dynamica
que hardware
ess network is
data structure
module that i
unning the C
can have any
ints up to a m
outer child ca
ldren with a
or the size o
the size of ou
ch child node
ss and option
a router or en
address and
can represen
nternally.
essage Broke
quency (RF)
that informati
Monitor clas
looked up in
hardware ID
he internal st
bled packet tr
endpoint node
and User
ation Logic l
een the user i
Broker and
ig. 4). This
ing which ac
ated from the
ker class will
object to byte
ffer to object
It maintains a
modules, and
Modules can
ally assigned
e addresses.
s represented
. The root of
s attached to
Control Client
combination
maximum of
an itself have
maximum of
of the tree is
ur network is
must have a
nally a child
ndpoint node
optionally a
nt the actual
r reports the
data packet
ion is passed
ss. The node
n the Network
D, and the RF
torage nodes.
ransmitted by
es.
r Interface
layer can be
interface and
Xbee-PRO
is where the
tions to take
e
l
e
t
a
d
n
d
d
f
o
t
n
f
e
f
s
s
a
d
e
a
l
e
t
d
e
k
F
.
y
e
e
d
O
e

177

Fig. 4 Abstract software architecture of control and
display client.

upon receipt of various messages such as heartbeats,
data payloads, node identifiers and more from router
and endpoint nodes.
The callback routines for user interface (UI) buttons
and commands are also present in this layer. For
instance, when the user clicks the OK button on the
Start Session screen, a callback function is executed
that uses the Xbee-PRO Message Broker to issue
commands that will cause all router and endpoint
nodes to send a node-identifier message back to the
base.
2.2.4 Data Logging
All sensor data is logged in CSV (Comma
Separated Value) tabular format in a plain-text file. All
rows have the same number of columns, and each
column represents the same field. A
microsecond-resolution timestamp is prefixed to each
log entry by the Control Client Application indicating
the time of the receipt of the data being logged.
All logging is automatically broken up into separate
files each consisting of 5 minutes worth of data. The
user is allowed to name the log file before starting a
new data collection session.
2.2.5 Graphical User Interface
The Graphical User Interface is meant to be an easy
to use method of monitoring and controlling the
wireless sensor network. There are 2 main dialogs that
the user interacts with. One is the Start Session Screen
where various pre-session configuration items are
specified such as
The Com Port to use for communicating with the
base;
The name and path of the log file that will be
generated;
The sample rate of the sensors;
The path to the configuration file which
associates a node unique hardware address with a
given alias.
The other dialog is the main program display and
control screen Here we have a visual representation of
the network topology in the form of a TreeView
control on the left hand side, a real-time graph of
sensor data that shows minimum, maximum, and
average values over time of pressure data on the right
hand side, and a status panel that shows session
averages for temperature and pressure data, as well as
information about battery life and node operating
states.
Individual or groups of nodes can be commanded
by right-clicking on their icon in the TreeView control
on the left of the screen. Right-clicking on a router
presents various commands to issue to all nodes under
the router, right-clicking on a specific endpoint node
presents commands specific to that sensor, and
right-clicking on the base node presents commands
that can be issued to the entire network.
The state of each endpoint on the wireless network
is represented by a colored ball next to each endpoint
on the TreeView control. A yellow icon indicates an
Idle Mode state, a green icon means Data Mode, while
a red icon means the node is either sleeping (due to
low battery hardware shutdown) or lost.
In summary, the user can quickly get an idea of how
the wireless sensor network is performing based on the
various visual displays provided in the user interface.
2.3 System Configuration
Each complete system consists of up to 35 units:
30 sensors;
Network
Topology
Tree and
Statistics
Serial Port
Xbee Message Broker Xbee Network Monitor
Application Logic
Graphical User Interface

178
2 anem
3 router
For the s
they are se
ZigBee stan
Fig. 5), usin
Each sub
least 2 cha
interference
XBee-PRO
default, the
energy scan
channels an
data transmi
many nodes
assigned to
other. Mor
equivalent t
have to be
coordinators
2.3.1 Cha
The cha
configuration
base and the
channel usin
The SC p
with each b
energy scan
starting from
on how to fo
For a full
set to at leas
2.3.2 Data
A data pro
in order to
collected by
A series o
extract data
The codes th
into Matlab
command, m
Design,
mometers;
rs.
ake of separ
et to differe
ndard uses 16
g 5 MHz of b
system (i.e.,
annels apart
. This is done
units at the
XBee-PRO
n during star
nd will choos
ission. For th
s it is crucia
specific ch
reover, the
to Service S
set also to
s.
annel and PAN
annel assig
n of the XBee
e endpoints/r
ng the Scan Ch
parameter is a
it representin
n will take i
m right to lef
orce a unit to
l system, the
st two channe
a Processing
ocessing code
o sort and
y the systems
of software t
from the data
hat execute t
b C execu
meaning that t
, Developmen
rating the ope
ent operating
6 channels fro
bandwidth ea
set of 10 s
from each
e through con
moment of
is configure
rtup through
se the most
his particular
al that all the
hannels, far a
PAN IDs
et Identifier
constrain n
N ID Assignm
gnment is
e units, both t
routers, to the
hannels param
a 16 bit hexa
ng a possible
into consider
ft. Fig. 6 disp
use channel n
three subsys
els apart from
e was develop
convert all
into data file
tools were wr
a files for the
these operatio
utables throu
they can run o
nt and Testing
erating chann
g channels.
om 11 to 26
ach.
sensors) is se
other to av
nfiguration of
f installation.
ed to perform
the 16 poss
suitable one
application w
e subsystems
apart from e
(network
(SSID in W
nodes to spe
ment
done thro
the correspon
e same opera
meter (SC).
adecimal num
channel that
ration at star
plays an exam
number 13.
stems have to
m each other.
ped in MATL
the informa
s.
ritten in orde
e post process
ons are comp
ugh the m
on any compu
g of a Wireles
nels,
The
(See
et at
void
f the
. By
m an
sible
e for
with
s are
each
ID,
WiFi)
cific
ough
nding
ating
mber
t the
rtup,
mple
o be

LAB
ation
er to
sing.
piled
mcc
uter.
Fig.

Fig.

S
accu
con
F
ano
2
T
tran
fact
the
60
and
T
ban
una
boa
resu
A
whe
P
o
x
coll
b
T
raw
swi
1
1
2
4
0
0
ss Multi-Sens
. 5 Channel c
. 6 SC param
Since each tr
urately, the d
nverted into p
For resamplin
other program
2.3.3 Pressure
The Freestrea
nsducers hav
tory expresse
temperature
degrees Cels
d V
s
the supply
This transfer
nd of 1.5
acceptable fo
ard transfer fu
ults that meet
A linear funct
ere:
PAbsolute
o
1
Calibrati
xRaw bina
lection softwa
b
1
Calibratio
The procedure
The specifi
The data c
w values are
itch;
2
4
0
5
2
4
1
0
2
4
1
5
2
4
2
0
2
4
2
5
2
4
3
0
1
1
1
2
1
3
1
4
1
5
1
6
2
4
1
2
2
4
1
7
2
4
2
2
2
4
2
7
1234
sors Network
center frequenc
meter usage.
ransducer tra
data collecte
hysical quant
ng, the phys
m.
e Calibration
am Electroni
ve a standard
ed by I
out
=
range used in
sius). Here V
y voltage.
function has
kPa for
or the applic
unction gives
t the needs.
tion is assume
P = o
1
x
pressure in m
on Coefficien
ary value o
are;
on Coefficien
e for calibrati
c sensor to be
collection sof
e displayed
2
4
3
0
2
4
3
5
2
4
4
0
2
4
4
5
2
4
5
0
2
4
5
5
1
6
1
7
1
8
1
9
2
0
2
1
2
4
3
2
2
4
3
7
2
4
4
2
2
4
4
7
2
4
5
2
56789
k System
cies distributio
ansfer functio
d by each se
tities.
sical data is
cs MP3H611
d transfer fu
I
s
(u.uu9 P
n the measure
V
out
is the ou
s an absolute
any given
ation. Quant
s stable and m
ed in the form
x + b
1

mbar;
nt (slope);
obtained fro
nt (offset).
ion is as follo
e calibrated i
ftware is sta
using the T
2
4
5
5
2
4
6
0
2
4
6
5
2
4
7
0
2
4
7
5
2
4
8
0
2
1
2
2
2
3
2
4
2
5
2
6
2
4
5
7
2
4
6
2
2
4
6
7
2
4
7
2
1
0
1
1
1
2
1
3
on.
on is defined
ensor can be
run through
15A pressure
unction from
-u.u9S) in
ements (up to
utput voltage
e uncertainty
transducer,
tifying every
more accurate
m:
m the data
ows:
s turned on;
arted and the
Toggle Units
Mhz
2
4
8
0
ZigBeeChannel#
2
6
Mhz
WiFiChannel#
d
e
h
e
m
n
o
e
y
,
y
e
a

e
s

179
Using a relative pressure calibrator with higher
order accuracy differential pressures are applied in a
relatively wide range (e.g. from-50 to +50 mbar
relative pressure) and the pressure and the
corresponding raw output values are recorded;
The data is plotted and a linear regression fit (Fig.
7) is performed;
Once completed, the results are plotted to
confirm they lie on a straight line and no outliers are
present.
2.3.4 Temperature Calibration
For the temperature transducer calibration the same
approach as in pressure calibration was used,
subjecting the sensor to different known temperatures
and finding the coefficients o
2
and b
2
:
I = o
2
x + b
2

2.3.5 Anemometer Calibration
The R.M. Youngs Mechanical Anemometer model
05103V with analog 0-5 V DC (volts direct current)
output is supplied from factory with a guaranteed
calibration sheet and there is no need for recalibration.
3. Sensor System Performance Tests
In order to quantify the accuracy and precision of
the sensors several tests were conducted. The two
most relevant tests are presented here.
3.1 Melbourne Airport (MLB) Comparison
The first test performed was a comparison between
the Melbournes International Airport weather station
and the system. This comparison gave satisfactory
results (Fig. 8), considering that the two locations
were about 3 miles away. These measurements were
only used to qualitatively identify the general trend
but not to quantify their attributes, since other effects
appear to be present (probably light local wind) and
pressure differences due to location mismatch.
3.2 Link Building
The latest test conducted to quantify the accuracy
and precision of the sensors was a comparison between
Fig. 7 Calibration line for sensor number 9.

Fig. 8 MLB airport weather station and system
comparison.

the system and a Paroscientific MET3A
meteorological station [27] located on the Link
Building at the Florida Institute of Technologys
Melbourne campus. This station is a part of Suominet
[28], a network used to measure water vapor in the
atmosphere from Global Positioning System (GPS)
measurements.
The pressure transducer in the MET3A station is
certified for 0.08 mbar with a maximum drift better
than 0.1 mbar per year. These figures are considered
to be very accurate for our measurements since the
range of pressure differences (between outside and
inside a building) that produce damage are generally
two orders of magnitude larger.
Three sensors (sensor numbers 1, 2, 3) were
deployed near the meteorological station on the roof,
at a distance of about four meters from the MET3A
120
100
80
60
40
20
0
20
40
60
80
2900 3000 3100 3200 3300 3400 3500 3600 3700
R
e
l
a
t
i
v
e
P
r
e
s
s
u
r
e
(
m
b
a
r
)
rawvalue
0 5 10 15 20 25 30 35 40 45 50
1010
1010.5
1011
1011.5
1012
1012.5
1013
1013.5
1014
1014.5
Time (h) since Aug 13
th
,2010 at 18:00 ET (GMT -5:00)
P
r
e
s
s
u
r
e

(
m
b
a
r
)

Sensor Pressure Data
Airport Pressure Data

180
station and attached with screws to a wooden rack.
The test started at 9:47 AM on March 31st, 2011
and recorded data for 120 hours (five days), at a
sampling rate of 10 samples per second for each
sensor. Fig. 9 shows the complete time series record
(low pass filtered with cutoff frequency of 0.1 Hz),
including the data from the Paroscientific station,
which was collecting one sample (average) every 30
minutes.
Two hours after the system was deployed a strong
thunderstorm occurred, lasting over a day. It resulted
in different pressure records for the first 30 hours of
data since the sensors were at different distances from
a wall while the station was located above the wall.
Once the storm dissipated and the wind calmed down
the desired test condition was established.
In order to quantify the system performance correctly,
the record used starts at hour 45 and ends in hour 105.
In order to evaluate the accuracy, the spread
(precision) has to be eliminated and only the mean
value of the measurements has to be taken into account.
This way, a 10-second average is taken of the original
data and compared to the MET3A station for each
sensor. Then, the precision was evaluated as the
standard deviation of the raw data with subtracted mean,
graphically expressed for sensor number 2 in Fig. 10.
Accuracy for the three sensors (on average) was
obtained to be 0.0404 mbar. This includes possible
breezes and light winds that might have created the
observed variations. It is important to note that
accuracy is usually correctable through the calibration
curves, due to the fact that it is most likely associated
to systematic error [29].
Precision, on the other hand, was found to have a
standard deviation of 0.3085 mbar, meaning that any
sample taken is found within 0.61 mbar of the true
value with a confidence level of 95.45%. This
confidence level was selected since outliers can be
easily recognized and filtered from the data, since
major changes in pressure between two consecutive
samples is not expected at high sampling rates.

Fig. 9 Link building test.

Fig. 10 Spread and average comparisonsensor 2.

Several other field tests were conducted during the
development phase of the system. Most of them were
performed on a house instrumented in Satellite Beach,
FL. Other range tests were conducted in the FITs
campus and a house in Palm Bay, FL.
3.3 Network System Deployment
A test with three systems working simultaneously
on three different houses was performed before the
2011 hurricane season, during the last days of May.
All three houses were in Brevard County, in the
vicinity of Melbourne.
The three houses are members of the FCMP program
(Munger, 28 6'48.35N, 8034'32.57W Miller,
2814'22.20N, 8041'23.40W, and Tracy
2810'32.64N, 8035'36.16W). The three houses are
in a suburban residential environment, two of them
(Tracy and Munger) close to the beachside of the barrier
island and the third inland in Melbourne (near Viera).
0 20 40 60 80 100 120
1002
1004
1006
1008
1010
1012
1014
1016
1018
1020
Time (h) since 03/31/2011 9:47 AM
P
r
e
s
s
u
r
e

(
m
b
a
r
)

Sensor 1
Sensor 2
Sensor 2
Paroscientific MET3A

181
Data collection for the three systems began at
System 1 (Miller): May 30th, 2011 at 3:29 PM
Eastern T;
System 2 (Tracy): May 30th, 2011 at 6:05 PM
Eastern Time;
System 3 (Munger): May 29th, 2011 at 6:00 PM
Eastern Time.
The data was post processed and averaged over 300
seconds (5 minutes), since the general pressure trend
for all the systems provides more meaning than the
raw data.
In Fig. 11, the pressure time history is shown for all
sensors in Millers house. Fig. 12 displays the
anemometer data through the same record. Notice that
the velocities are low since the anemometer, was close
to the roof. Moreover, the mean wind direction was
southerly around 180 degrees, as evidence of
continuous change in direction.
Fig. 13 is the pressure history for Tracys house. Fig.
14 displays the wind velocity history. The general
same trend of the pressure (mostly static) time history
at all three houses confirms the proper functioning of
the network system.
Notice the higher spread in the data in certain areas.
Since the wind speeds are low and almost constant in
magnitude through the entire range, this effect does
not come from pressure. As these days were sunny
and it was right before Floridas summer, the
temperature on the sensors reached over 60 degrees
Celsius on direct sunlight, drifting from the constant
response area of the pressure transducer for which the
accurate calibration is performed. However, during a
hurricane the sun is usually covered by clouds and the
temperature would not exceed 60 degrees Celsius
inside the sensor.
If the sensors have to be used for any test on direct
sunlight, the pressure needs to be corrected according
to the temperature response.
Out of 60 sensors deployed, plus the nine routers,
only three sensors crashed during the entire test and
were not able to be recovered. The anemometers and
the routers never failed.

Fig. 11 Pressure time historyMillers house.

Fig. 12 Velocity/direction time historyMillers house.

Fig. 13 Pressure time historyTracys house.

Fig. 14 Velocity/direction time historyTracys house.
25 30 35 40 45 50 55 60 65 70
1008
1009
1010
1011
1012
1013
1014
1015
Time (h) since May 29
th
, 2011 at 6:00 PM ET
P
r
e
s
s
u
r
e

(
m
b
a
r
)

S71
S72
S73
S74
S75
S82
S84
S85
S86
S87
S88
S89
S90
S93
S94
S98
S99
20 30 40 50 60 70 80
0
1.5
3
4.5
6
7.5
9
Time (h)
S
p
e
e
d

(
m
/
s
)
20 30 40 50 60 70 80
N (0)
NE (45)
E (90)
SE (135)
S (180)
SW (225)
W (270)
NW (325)
N (360)
D
i
r
e
c
t
i
o
n

(
d
e
g
)
25 30 35 40 45 50 55 60 65 70
1007
1008
1009
1010
1011
1012
1013
1014
Time (h) since May 29
th
, 2011 at 6:00 PM ET
P
r
e
s
s
u
r
e

(
m
b
a
r
)

S36
S37
S38
S39
S40
S41
S42
S43
S51
S53
S54
S58
S59
S60
S61
S62
S63
S64
S65
20 30 40 50 60 70 80
0
1.5
3
4.5
6
7.5
9
Time (h)
S
p
e
e
d

(
m
/
s
)
20 30 40 50 60 70 80
N (0)
NE (45)
E (90)
SE (135)
S (180)
SW (225)
W (270)
NW (325)
N (360)
D
i
r
e
c
t
i
o
n

(
d
e
g
)

182
Five sensors (numbers 5, 6, 7, 8 and 10) did suffer
signal loss after hour 53, with unknown causes. For
the rest of the sensors, the packet loss was minimal
(<< 1%).
4. Conclusions
The main goal of creating a portable, reliable,
robust, easy to use wireless pressure sensors network
system was accomplished. The performance, stability
and characteristics of the wireless pressure sensors
system were assessed, leaving at the end of this
project three fully functional ready-to-deploy systems,
very superior to the previous systems. Overall, this
generation of the sensor system resulted in major
improvements, including reliability, robustness, ease
of use, and expansion possibilities. In particular, speed
has been doubled: the most reliable sample rate
recorded is 36 samples per second for a full set of 30
sensors, while the previous generation was not
exceeding 20 for the same conditions.
It tested favorably in remote communication and
ability to regain control of the remote sensors without
physical presence at the deployment site. It was
concluded that XBee modules, used in the previous
system, are more sensitive to location and signal
losses due to the higher operating frequency (2.4
GHz); this issue was corrected by increasing
transmission power using XBee-PRO modules.
Tests also included comparisons with known
higher-order precision instruments, laboratory tests
and simulated field tests. The accuracy and precision
were quantified, giving figures of precision an order
of magnitude lower than the range of the measurand.
Additionally, data extracted from a three-house
network deployment was processed. This test allowed
evaluating and confirming the robustness of three
systems collecting data at the same time while
uploading it to a central server. It also tested the
remote communications and the ability to gain control
over them without physical presence at the
deployment site.
In summary, the current sets of sensors would allow
simultaneous deployment on three residences in case
of a hurricane landfall in Florida. The ultimate goal is
to collect data to better understand and quantify wind
structure interaction. The wireless sensor network
system can be adapted to other environments like lab
setting of the wall of wind or IBHS or even for normal
wind tunnels. Also the capability of the system allows
to expand the technology to any other type of
measurements.
References
[1] FCMP, Florida Coastal Monitoring Program, available
online at: http://fcmp.ce.ufl.edu, accessed: September 9,
2008.
[2] L.S. Cochran, J.E. Cermak, Full and model-scale
cladding pressures on the Texas Tech University
experimental building, Journal of Wind Engineering and
Industrial Aerodynamics 43 (1992) 1589-1600.
[3] E.A. Dearhart, Comparison of field model wind pressures
on residential buildings in tropical storm winds, M.S.
Thesis, Civil Engineering Department, University
Clemson, Clemson, SC, USA, 2003.
[4] H.W. Tieleman, M.R. Hajj, T.A. Reinhold, Wind tunnel
simulation requirements to assess wind loads on low-rise
buildings, Journal of Wind Engineering and Industrial
Aerodynamics 74-76 (1998) 675-685.
[5] S. Cao, et al., Reproduction of wind velocity history in a
multiple fan wind tunnel, Journal of Wind Engineering
and Industrial Aerodynamics 90 (2002) 1719-1729.
[6] Z. Liu, et al., Assessment of wind storm facility at the
insurance center for building safety research, in: 7th
Asia-Pacific Conference on Wind Engineering, Taipei,
Taiwan, November 8-12, 2009.
[7] B. Yu, A.G. Chowdhury, F.J. Masters, Hurricane wind
power spectra, cospectra, and integral length scales,
Boundary Layer Meteorology 129 (3) (2008) 411-430.
[8] K.J. Eaton, J.R. Mayne, The measurement of wind
pressures on two-story houses at Aylesbury, Journal of
Ind. Aerodynamics 1 (1975) 67-109.
[9] M.L. Levitan, K.C. Mehta, Texas Tech field experiments
for wind loads: Part I. Building and pressure measuring
system, Journal of Wind Engineering and Industrial
Aerodynamics 43 (1992) 1565-1576.
[10] M.L. Levitan, K.C. Mehta, Texas Tech field experiments
for wind loads: Part II. Meteorological instrumentation
and terrain parameters, Journal of Wind Engineering and
Industrial Aerodynamics 43 (1992) 1577-1588.
[11] B.J. Michot, Full-scale wind pressure measurement

183
utilizing unobtrusive absolute pressure transducer
technology, M.S. Thesis, Civil Engineering Department,
Clemson University, SC, USA, 1999.
[12] M.L. Levitan, Analysis of reference pressure systems
used in field measurements of wind loads, Ph.D.
Dissertation, Texas Tech University, Lubbock, TX, USA,
1993.
[13] M.L. Porterfield, N.P. Jones, The development of a field
measurement instrumentation system for low-rise
construction, Wind and Structures, An International
Journal 4 (3) (2001) 247-260.
[14] C. Subramanian, et al., A wireless multipoint pressure
sensing system: design and operation, IEEE Sensors
Journal 5 (2005) 1066-1074.
[15] F. Kreit, et al., Performance testing of the wireless
network system for hurricane monitoring, in: 1st
International Conference on Sensor Device Technologies
and Applications, Venice, Italy, 2010.
[16] A. Velasquez, WINDS-HM Wireless networking sensors
for hurricane monitoring, M.S. Thesis, Civil Engineering
Department, Florida Institute of Technology, FL,
Melbourne, 2009.
[17] L. Gabriel, Hurricane wind effects: instrumentation,
measurement and analysis, M.S. Thesis, Aerospace
Engineering Florida Institute of Technology, Melbourne,
Florida, August 2011.
[18] C. Li, Y. Wang, The application research of wireless
sensor network based on ZigBee, in: 2nd International
Conference on MultiMedia and Information Technology,
Hong Kong, 2010, pp. 89-92.
[19] Part 15.4: Wireless Medium Access Control (MAC) and
Physical Layer (PHY) Specifications for Low-Rate
Wireless Personal Area Networks (WPANs), IEEE Std.
802.15.4-2006.
[20] G. Nazri, G. Pistoia, Lithium Batteries: Science and
Technology, Springer, 2004.
[21] A. Osborne, An Introduction to Microcomputers: Volume
1. Basic Concepts, McGraw-Hill, Berkeley, 1980.
[22] J.S. Seybold, Introduction to RF propagation, Wiley,
Hoboken, 2005.
[23] Microchip Technology Inc. PIC18F2458/2553/4458/4553
Data Sheet, 2009.
[24] Digi International, X-CTU Configuration & Test Utility
Software: Users Guide, available online at:
http://ftp1.digi.com/support/documentation/90001003_A.
pdf, accessed: August 20, 2008.
[25] Smooth-On, available online at:
http://www.smooth-on.com/.
[26] ZedGraph, Free Software Foundation Inc., available
online at: http://www.zedgraph.org.
[27] Paroscientific Manual, available online at:
http://www.paroscientific.com/pdf/MET.pdf, accessed:
November, 2001.
[28] UCAR Community Programs, Real Time Integrated
Atmospheric Water Vapor and TEC from GPS, available
online at: http://www.suominet.ucar.edu/, accessed: July
2009.
[29] Joint Committee for Guides in Metrology, International
Vocabulary of Metrology-Basic and General Concepts
and Associated Terms (VIM), JGCM, 2008.


Distributed Robot Control System Based on the
Real-Time Linux Platform
Goran Ferenc, Zoran Dimi, Maja Lutovac, Vladimir Kvrgi and Vojkan Cvijanovi
Robot Controllers, Lola Institute, Belgrade 11030, Serbia

Received: November 04, 2011 / Accepted: November 18, 2011 / Published: March 25, 2012.

Abstract: This paper presents a development of a new distributed control system for robots control. The system is based on the
real-time Linux platform on which open architecture software OROCOS (Open RObot COntrol Software) is set designed specifically
for creating applications of this type. The system consists of components of various structures and with different roles. They can be
distributed to different hardware platforms with the same operating system and to communicate through the CORBA (Common
Object Request Broker Architecture) protocol. The system is designed as a modular solution that allows easy connection and
disconnection of the components with various roles depending on the needs of applications including those needed to communicate
with different types of hardware. That way the structure of the system remains the same.

Key words: OROCOS (open robot control software), real-time robot control, distributed systems, CORBA (common object request
broker architecture).

1. Introduction

Industrial robots are included in a wide range of
applications today, so there are many configurations,
physical shapes and drive robot systems. As the
demand for increased performances is constantly
growing, requirements for better and more complex
controllers are on the rise.
A solution to deal with this problem has two ways.
First, it is possible to retrofit robots, which includes
replacement of some working parts and the
development of all software features. These are major
changes and it is expensive and demands a long time.
An open architecture appears as a second way to solve

Zoran Dimi, M.Sc., research fields: robot manipulators,
machine tools, robotics, embedded software systems.
Maja Lutovac, M.Sc., research fields: application
programming, robot languages programming, robotics.
Vladimir Kvrgi, Ph.D., research fields: robotics, robot
manipulators, machine tools, kinematics and dynamics.
Vojkan Cvijanovi, M.Sc., research fields: application
programming, information technologies.
Corresponding author: Goran Ferenc, M.Sc., research
fields: real-time control systems, computer controlled systems,
robotics, electronics. E-mail: goran.ferenc@li.rs.
this problem. In this way, almost all features in the
design can be changed by everyone as well as by the
original manufacturer. The price of the development
becomes less expensive and development demands
less time. These benefits are achieved by using open
architecture and open source software platforms.
Development of a new control system for Lola
Institutes robots was started on a previous version of
Lola Institutes robot controller developed on the OS-9
real-time platform. Unlike the previous version which
requires a specific set of hardware the new version of
the system is developed on the real-time Linux platform,
where the Linux kernel is patched with Xenomai.
OROCOS (Open RObot COntrol Software) libraries
are installed on this platform and designed specifically
for the development of applications for robot control. In
this way, the real-time part of the control system is
completely changed from the real-time part of the
previous version. New control system has open
architecture and it is modernized according to the
development of similar control systems in the world,
achieving greater efficiency and speed [1].
D
DAVID PUBLISHING
Distributed Robot Control System Based on the Real-Time Linux Platform

185
The paper consists of five sections: In the next
section, OROCOS overview is presented, including
Real-Time Toolkit (RTT) and component interface.
Section 3 presents the architecture of the control
system, while section 4 describes distribution of the
components. At the end, conclusions are given.
2. OROCOS
OROCOS is a European project, started on
September 1st, 2001. Three laboratories are
participating: the Katholieke Universiteit Leuven
(KULeuven, Belgium, project contractor), the
Laboratory for Analysis and Architecture of Systems
(CNRS/LAAS, France) and Kungl Tekniska Hgskolan
(KTH, Sweden). Many other European laboratories
participate in the discussions and design [2].
The project aims at producing an open source
software framework by providing a functional basis
for general robots control. The software is intended to
be platform independent, which is also thought
as application independent.
The OROCOS codebase is divided into modules or
libraries. There are generally three types of modules:
Support moduleThis module is software
without functional robotics content, but it is needed to
build a working environment for robot control system
such as real-time operating system, software
component configuration tools, 3D visualization and
simulation, numerical libraries, inter-process
communication, documentation writing tools, etc.;
Robotic moduleThis module is software that
implements specific robotics algorithms, kinematics
and dynamics of kinematic chains, servo controllers,
motion planners for mobile robots, serial and parallel
manipulators, Bayesian and Neural Network
estimators, etc. The robotic module makes use of one
or more supporting modules;
ComponentsThese are CORBA (Common
Object Request Broker Architecture) objects with their
IDL (Interface Description Language) descriptions.
These components are constructed out of the previous
two types of modules and they are the building blocks
with which users assemble their robot control software
environment.
Open standards are very important for efficient
sharing of software and for its cooperative development.
Only an open source project can guarantee the optimal
use of open standards. The most important standards
that the OROCOS project supports are:
The CORBA IDL for object embedding. Through
the IIOP (Internet Inter-Orb Protocol), all different
implementations of CORBA are interoperable.
CORBA also has a real-time extension, which is very
important for robot control;
The structured document language XML
(eXtensible Markup Language) for configuration and
data files;
Modelica for modeling of dynamical systems;
DocBook and LATEX for documentation. Both
are platform independent, open source and can be
translated into many other formats and as such they
are the preferred text processing vehicles.
OROCOS does not want to impose a standard
programming language. The philosophy behind this
strategy is that the best tool for the job should be used.
The CORBA middleware is able to work in a
heterogeneous programming language environment.
The OROCOS software system uses support from
other open source software projects. Some of the
projects that OROCOS can be built on are:
Octave is a very rich numerical library that uses
the same foundations as, and in its scripting form is
very compatible with MATLAB;
Real-Time Linux and/or RTAI (Real-Time
Application Interface), which are real-time kernel
extensions to Linux; or the Linux-independent eCos
(embedded Configurable operating system);
Xenomai is a real-time development framework
cooperating with the Linux kernel, in order to provide
a pervasive hard real-time support to user-space
applications [3];
Comedi is a library for real-time device drivers,

186
such as AD (Analog to Digital) or DA (Digital to
Analog) cards;
Various packages and open formats for 3D
visualization: VRML (Virtual Reality Modeling
Language), Java3D, OpenGL with OpenInventor or
Coin3D, etc.
The OROCOS project supports four C++ libraries:
the Real-Time Toolkit, the Kinematics and Dynamics
Library, the Bayesian Filtering Library and the
OROCOS Component Library.
The OROCOS Real-Time Toolkit (RTT) is not
an application in itself, but it provides the
infrastructure and the real-time functionalities to build
robotics applications in C++.
The OROCOS Component Library (OCL)
provides some ready to use control components.
The OROCOS Kinematics and Dynamics Library
(KDL) is a C++ library which allows calculating
kinematic chains in real-time.
The OROCOS Bayesian Filtering Library (BFL)
provides an application independent framework for
inference in Dynamic Bayesian Networks, recursive
information processing and estimation algorithms
based on Bayes rule, such as Kalman Filters, Particle
Filters (Sequential Monte methods), etc.
The OROCOS applications are composed of
software components, which form an application
specific network. When using OROCOS, engineers
can choose to use predefined components, or build
their own components, using the OROCOS Real-Time
Toolkit. Components work together, cooperatively.
For example, the application template for motion
control contains components for path planning,
position control, hardware access and data reporting.
The components are chosen as such that their
interfaces are compatible.
A single component may be well capable of
controlling a whole machine, or is just a small part in
a whole network of components, for example an
interpolator or kinematic component. The components
are built with the RTT and optionally make use of any
other library (like a vision or kinematics toolkit). Most
users will interface components through their (XML)
properties or command/method interface in order to
configure their applications. Besides defining the
above component communication mechanisms,
OROCOS allows the Component or Application
Builder to write hierarchical state machines which use
these primitives. This is OROCOS way of defining
application specific logic.
2.1 OROCOS Real-Time Toolkit
The Real-Time Toolkit is referred to as middleware
because it lies between the application and the
operating system as shown in Fig. 1. It allows setup,
distribution and the building of real-time components
and also takes care of real-time communication and
execution of software components [4].
The Real-Time Toolkit is structured in layers on the
top of operating system and the I/O (Input/Output)
devices as shown in Fig. 2.
Considering an OROCOS component is built upon
RTT, it allows designers to build their components
which are configurable using XML files, accessible
over a network and listen to a scripting interface,
which allows components to be controlled using text
commands. OROCOS Device Interface (DI) defines
how to interact with analog and digital I/O and
encoders. A component which accesses I/O devices
can use the OROCOS DI. Components can make use
of external libraries as well.
OROCOS components which only use the
Real-Time Toolkit are portable over different
processor architectures and Operating Systems (OS).
OROCOS has an internal OS abstraction which allows
the components to run on any supported architecture.
When some component uses an external library
portability depends on these libraries.
2.2 OROCOS Component Interface
The system is designed as a modular solution that
consists of components. Components are basic units

187

Fig. 1 OROCOS real-time toolkit as middleware layer.

Fig. 2 OROCOS real-time toolkit layers.

which executes one or more action with a specific
activity. They are created using RTT OROCOS
libraries. In this way, OROCOS provides an interface
for components design. Components can also use the
native libraries of the operating system.
There are four policies for the control of component
activities [2]:
NonPeriodicActivity;
PeriodicActivity;
SequentialActivity;
SlaveActivity.
Each component inherits a public interface from its
base class (TaskContext), which defines primitives for
component interactions (as shown in Fig. 3):
Events;
Methods;
Commands;
Properties;
Data Port.
In the system which is developing at Lola Institute,
basic components of OROCOS OCL library are used:

Fig. 3 OROCOS component interface.

TaskBrowser;
DeploymentComponent;
ReportingComponent.
The TaskBrowser is a component used to interact
with other components. Each OROCOS component
inherits a standard interface through the TaskContext
base class, which is used by the TaskBrowser to
interact. It works like a text console, receiving
command lines and executing them.
The DeploymentComponent is a component used to
load and configure other components from
descriptions in XML files. The system configuration
is performed by connecting the components with their
peers and their Data Ports. With all connections done,
the Properties of each component, as well as the
component activity are updated from that file.
In the ReportingComponent is used to monitor and
capture data exchanged by other components. It works
by connecting as a peer of the components to be
monitored and attaching to the specified Data Ports.
3. Control System Architecture
The components designed by the developer make
the basic architecture of the system. The system is
based on the components model. Each component
model can be instantiated in a component by using the
specific parameters of a given robot. The basic block
model of the system is given in Fig. 4 [5-10].

188

Fig. 4 Basic block model of the control system.

The Sampler is the component that generates the
sampling rate of the control loop and synchronizes the
other components. It generates an Event which is
received by the other components, automatically
triggering a control cycle. The Sampler is configured
in the XML file to generate periodic Event every 1
ms.
The Interpolator component is a trajectory generator.
It calculates the reference position values that have to
be sent to the Controller component. New values are
written in Data Port DesiredPosition every 5 ms. It is
one of the most important component in the system.
In the development of our new system OROCOS,
KDL library is not used. The Interpolator component
implements the algorithms developed at the Lola
Institute [7]. The interface of this component to the
rest of the system required a change in the original
form of these algorithms, so this component can be
included into system that is designed on OROCOS
platform.
The Controller component model abstracts the
system controller. At every 1 ms, this component
sends reference velocity values that have to be reached.
It writes velocities into its Data Port OutputVelocity.
When the Sensor component writes position values
from the sensor to SensorPosition Data Port (every 1
ms) it triggers a call to the Controllers virtual
function which should compute the control signal and
write it to the OutputVelocity Data Port. It means that
the Controller has five changes to drive the Actuator
component to the position sent by the Interpolator.
In this project, the nAxesControllerPos component
from the OROCOS OCL library is used. It uses a
position feedback law to calculate the velocity output
from measured and desired positions.
The Sensor component model abstracts the sensors
of the system. It has a Data Port where it writes the
values which are read from the sensors. This Data Port
is represented by a vector which size depends on the
number of joints of the robot.
The Actuator component model abstracts the
system actuator. Similar to the Sensor component
model it has a virtual member function which is called
in response to an Event from the Sampler. It has a
read-only port where the component can read values
sent by the Controller component. Those values
should be applied to the robot.
The above models are the basic components of the
control system. In addition to these systems include
more models. One of them is the Joint component
model. This model represents a component that
communicates with the hardware of robot joint and the

189
number of components that are derived from this
model depends on the number of them. Each of them
is associated with components Sensor and Actuator
which Data Ports give signals of all robots axes in the
vector form, and because of that it is necessary to
create components for multiplexing the signals from
the N joints and for demultiplexing the vector output,
so the components can be connected to each other.
4. Distribution of the Components
As the presented system is designed as a modular
solution and built from components, it can be
distributed to different hardware platforms with the
same operating system and to communicate with each
other over the network. Components are linked using
CORBA deployer application. Connecting data flow
ports of components is done by defining connections.
When components are distributed using the CORBA
deployment component, it is necessary to declare a
proxy component in one of the XML files and connect
to a port of that proxy. It is necessary to setup a
specific connection in one XML file and the other
files do not need to repeat the same information.
5. Conclusions
The goal of a new Lola Institutes control system
development is to provide all necessary functionality
for robots, machine tools and similar devices control
in a more efficient way. Also, the idea is to create
modular solution so that the different functionality
could be included or excluded in accordance with the
needs of applications without changing the structure
of the system. This allows faster and easier possible
system upgrades. The system can be distributed to
different hardware platforms. During its development
modern methods and open source packages are used
which make development easier and less expensive.
Acknowledgments
This work was created within the research project
Development of the devices for pilots training and
dynamic flight simulation of modern combat aircraft:
3 DoF centrifuge and 4 DoF spatial disorientation
trainer that is supported by the Ministry of Science
and Technological Development, Republic of Serbia.
References
[1] M. Milievi, J. Vidakovi, Z. Dimi, S. Trgovevi,
Modern open architecture control systems for machine
tools and robots control, in: Proceedings of the 36th
JUPITER Conference, 2010, pp. 4.41-4.46.
[2] Available online at: http://www.orocos.org.
[3] Available online at: http://www.xenomai.org.
[4] Available online at:
http://people.mech.kuleuven.be/~orocos/.
[5] M. Milievi, V. Kaplarevi, Z. Dimi, V. Kvrgi, V.
Cvijanovi, Development of new control system for
robots and multi-axis machining systems, in: Proceedings
of the 4th International Conference on Manufacturing
Engineering (ICMEN), 2011, pp. 451-457.
[6] M. Milievi, V. Kaplarevi, Z. Dimi, V. Cvijanovi, M.
Buan, Development of distributed control system for
robots control based on real-time Linux platform, in:
Proceedings of the 10th Anniversary International
Conference on Accomplishments in Electrical and
Mechanical Engineering and Information Technology
(DEMI), 2011, pp. 813-818.
[7] V. Kvrgi, Development of intelligent systems for
industrial robots control and programming, Ph.D. Thesis,
Faculty of Mechanical Engineering, University of
Belgrade, 1998.
[8] G. Pritschow, Y. Altintas, F. Jovane, Y. Koren, M.
Mitsuishi, S. Takata, et al., Open controller
architecturepast, present and future, CIRP
AnnalsManufacturing Technology 50 (2001) 463-470.
[9] C. Brecher, A. Verl, A. Lechler, M. Servos, Open control
systems: state of the art, Prod. Eng. Res. Devel. 4 (2010)
247-254.
[10] M. Pavlovi, V. Kvrgi, D. Velaevi, L-IRL: high level
programming language for robots, in: Proceedings of the
European Robotics and Intelligent Systems Conference,
Malaga, Spain, 1994.

J ournal of Mechanics Engineering and Automation 2 (2012) 190-196

Hybrid Algorithms for Multiobjective Optimization
of Mechanical and Hydromechanical Systems
Valeriy D. Sulimov and Pavel M. Shkapov
Department of Theoretical Mechanics, Bauman Moscow State Technical University, Moscow 105005, Russia

Received: J anuary 21, 2012 / Accepted: February 06, 2012 / Published: March 25, 2012.

Abstract: In two recent decades, approaches with the use of vectorial criteria as mathematical models of choosing find a wide utility
in numerous applications. It will suffice to mention some problems of identification and diagnostics of complex systems, optimal
design, control, saving of safety. The goal of multiple criteria numerical strategies is to compute a set of non-dominated solutions
which constitute Pareto-optimal front approximations. Generally individual criteria are multiextremal not everywhere differentiable
functions. Under these conditions searching for a global solution for the criterion is an independent challenging problem. Account
must be taken of that the effectiveness of deterministic global optimization algorithms is quickly reduces with increasing the
dimension of the search space. On the other hand, stochastic global optimization algorithms are computation-intensive. These
considerations motivate a topicality of the development of hybrid methods for solving multiobjective optimization problems with
multiextremal nonsmooth criteria. Two novel hybrid vector optimization algorithms combining a Metropolis-based stochastic
algorithm and deterministic gradient techniques or a space-filling curve method are introduced. Implementations of the hybrid
algorithms are discussed. Results of successful computational experiments are presented to illustrate the efficiency of the approach.

Key words: Vectorial criterion, non-dominated solution, multiextremal function, global optimization, hybrid algorithm.

1. Introduction

In many practical cases optimization problems must
be formulated as problems with multiple criteria. One
of the approaches suggests converting the initial
vector optimization problem into a scalarized
single-objective problem. More general methods are
founded on the concept of Pareto-dominance, where
they need to find trade-off solutions for criteria being
in conflict. When individual criteria are multiextremal
not everywhere differentiable functions, the vector
optimization problem becomes highly complicated
and computationally expensive one [1-2]. Most
problems of this type cannot be solved exactly owing
to highly complex and large dimensional search space.
The goal is to find the set of non-dominated solutions
[3]. The vector linearization method is one of the

Corresponding author: Pavel M. Shkapov, professor,
Dr.Sci.Tech., research fields: two-phase flow hydromechanics,
optimization of hydromechanical systems. E-mail:
spm@bmstu.ru.
effective methods for numerical solutions of the
multiobjective problems [4]. An introduction of
biparametric smoothing approximations makes it
possible to extend the method on the class of
nondifferentiable problems [5]. In the context of the
vector problem under consideration, it is necessary to
solve global optimization subproblems for individual
criteria. Properties of deterministic global
optimization algorithms are well-studied. It is
common knowledge that the performance of
deterministic algorithms essentially depends on the
problem dimension. On the other hand, stochastic
global optimization algorithms are generally
computation-intensive. A hybrid global optimization
algorithm NMPCA that combines the recent stochastic
algorithm PCA and Nelder-Mead simplex method is
presented in Ref. [6]. The NMPCA (Nelder-Mead
simplex method combined with Particle Collision
Algorithm) was applied to a nuclear reactor core
design optimization problem. During processing two
D
DAVID PUBLISHING
Hybrid Algorithms for Multiobjective Optimization of Mechanical and Hydromechanical Systems

191
procedures are performing: A wide search in the
solution space using the stochastic PCA and a local
search in the promising areas with the deterministic
simplex algorithm. The local search procedure is
performed iteratively until a certain number of fitness
function evaluations being reached. It is pertinent to
note that the convergence theory for Nelder-Mead
simplex method is far from completion; so the method
can fail to converge or converge to non-stationary
points [7]. By this is meant that the question of
reliability of the algorithm as a whole is still an open
question. As an alternative to the NMPCA a novel
hybrid algorithm may be introduced. In this version of
the global optimization algorithm the local search
mechanism is a standard deterministic linearization
method. Inverse problems are considered to be
substantially difficult because of the kinks connected
with presence of the repeated or very close
frequencies in registered spectra for the computational
model under updating. The difficulty motivated the
development of algorithms for the solution of the
minimization problem via some smooth
approximation, which could be minimized by using
any of the efficient classical approaches for smooth
optimization. Several approximations to smooth out
the kinks may be introduced. One of them results in a
continuously differentiable approximate fitness
function, whereas another one leads to a twice
continuously differentiable approximate function.
These approximations replace the original function in
some neighborhoods of directional differentiability
points. Moreover, this approach preserves such
important property of the original function as its
convexity. It is clear that the approach makes it
possible to implement efficient gradient techniques in
the solution process [8]. Computational experiments
show the principal applicability of the proposed
hybrid algorithm PCAHS (PCA Hybridized with
gradient method and Smoothing approximations) for
solving the global optimization problems. Some
powerful algorithms for multiextremal non-convex
optimization problem are based on reducing the initial
multi-dimensional problem to the equivalent problem
of one dimension. This reduction can be executed by
applying Peano-type space-filling curves mapping a
unit interval on the real axis onto a multi-dimensional
hypercube [9]. The approach needs not any derivatives
of the function to be minimized with updating
parameters.
The plan of the remainder of this paper is as follows:
The section followed contains statement of the
multiobjective optimization problem. Section 3
provides brief description of hybrid vector optimization
algorithms for solving problems with multiextremal
nonsmooth criteria. In section 4 successful
computational experiments for two model
multiobjective optimization problems are presented to
illustrate peculiarities of the proposed approach.
Section 5 gives conclusions and discussion on further
work.
2. Formulation of the Problem
Assume that the functions ( ), x f
i

, R
n
x

M i , ... , 2 , 1 =
constitute a vectorial criterion
( ) ( ) ( ) ( ) x f x f x f
M i
, ... , = for a multiobjective
optimization problem subject to
{ } , 0 ) ( : R J j x g x X x
j
n
= , where x stands
for the vector of control variables; ( ) x g
j
stands for
j th function of constraints; the corresponding index
set is defined as { } ..., , 1 : k j j J = = . It is supposed
that in the problem under consideration both
individual criteria and constraints are represented by
multiextremal nonsmooth functions. In addition to
main definitions [3] let introduce two ones.
Definition 1 [1]: For a given vector optimization
problem the Pareto-optimal set is defined as
{ } , :
* *
x x X x X x S
P
= (here the sign
denotes dominance).
Definition 2 [1]: For a given vector optimization
problem and the Pareto-optimal set
P
S the Pareto
front is defined as
{ } : ) (
P P
S x x f F =
.

192
It should be noted that a Pareto-optimal set is
always a non-dominated set. But there may exist
non-dominated sets that contain some Pareto-optimal
solutions and some non-Pareto-optimal solutions [2].
So it is necessary to realize that the non-dominated
solutions found by an optimization algorithm are not
certain to be able to present the true Pareto-optimal set.
The next section presents algorithms for solving the
multiobjective optimization problem that realize the
vectorial optimization method [4].
3. Hybrid Multiobjective Optimization
Algorithms
3.1 The Hybrid Algorithm V-PCALMS
Let introduce biparametric local smoothing
approximations for every individual criterion and
every constraint function [8]. Let associate every point
n
x R with the following auxiliary problem
( ) ( ) , , , , ,
~
:
2
1
min
2
,
I i w x q p f w
i
w

( ) ( ) ( )

+ J j x q p g w x q p g
j j
, 0 , ,
~
, , ,
~
(1)
where w is a vector standing for the search direction;
is a parameter; ( ) x q p f
i
, ,
~
stands for the gradient
of the smoothed approximated criterion
i
f
~
which is
computed in an admissible point; { } ..., , 1 : M i i I = = ;
the set J has been defined earlier. Let introduce the
following assumptions. Let exists 0 > N such that
for
n
x R
0
and certain I i the set
( ) ( ) ( ) {
( )}
0
0
, ,
~

, ,
~
, ,
~
, ,
~
:
x q p G N
x q p f x q p G N x q p f x
i i N
+
+ =

is restricted;
( ) ( ) ( ) { } x q p g x q p g x q p G
k
, ,
~
, ... , , ,
~
, 0 max , ,
~
1
=

gradients of the functions
( ) ( ) , , , ,
~
, , , ,
~
J j x q p g I i x q p f
j i
in
N

satisfy the Lipschitz condition with a constant L ;
there are exist Lagrange multipliers of the problem (1)
, , J j v
j
such that

J j
j
N v , and the problem is
solvable with respect to
n
w R for each
N
x . If
for certain 1 0 , < < , it is chosen , 1 0 such
that L N / ) 1 /( ) 1 ( 0 + , then the following
inequality is true [4]:
( ) ( ) ( ) ( ) w x q p G N x q p f w x q p f
i i
I i
+ + +
, ,
~
, ,
~
, ,
~
max
( )
2
, ,
~
w x q p G N (2)
The algorithm for solving the vector optimization
problem consists of the following main steps [4-5].
Let
0
x be an initial approximation and certain
number 1 0 , < < , and parameters of the
approximations 0 , 0 > < q p are chosen. Let the
current point
k
x , is obtained, then:
Step 1: Solve the auxiliary problem (1) when
k
x x = .
Step 2: Find the first value , ... , 1 , 0 = s such that the
inequality (2) holds for ( )
s
2 / 1 = ; if the appropriate
0
s s = found, then set
0
2
s
k
= , .
1 k
k
k k
w x x + =
+

Optimization of the individual criteria is performed
by use of the hybrid global optimization algorithm
combining the stochastic algorithm PCA [6] and the
linearization method during the local search. For the
functions that are not everywhere differentiable the
smoothing approximations are introduced. The
implementation of the recent stochastic algorithm
PCA is based on the analogy with physical processes
of absorption and scattering of particles during nuclear
reactions. The algorithm PCA uses only one particle
for scanning the search space. The hybrid algorithm
for solving a global optimization subproblem for an
individual criterion is proposed in Ref. [8]. The
convergence of the algorithm is proved for the case of
simple constraints. The hybrid algorithm V-PCALMS
(Vectorial PCA combined with Linearization Method
and Smoothing approximations) implements the above
main steps and makes it possible to find global
minima of individual criteria when solving the
multiobjective problem.
3.2 The Hybrid Algorithm V-PCASFC
The second hybrid algorithm V-PCASFC (Vectorial
PCA combined with Space-Filling Curve method) for
multiobjective optimization implements the
deterministic space-filling curve method for solving
the global optimization subproblem for the individual

193
criterion. The mentioned algorithm is based on the
algorithm PCA in combination with the deterministic
method of the space-filling curve [9]. Some powerful
algorithms for multi-extremal non-convex
optimization problem are based on reducing the initial
multi-dimensional problem to the equivalent problem
of one dimension. This reduction can be executed by
applying Peano-type space-filling curves mapping a
unit interval on the real axis onto a multi-dimensional
hypercube. The Peano curve development maps the
segment [ ] 1 , 0 of the real axis
1
R into the hypercube
1
R X . Actually, this is the case of continuous
single-valued mapping that offers finding point
( ) ( ) ( ) ( ) X z x z x z x
T
n
= , ... ,
1
for each point z [ ] 1 , 0 :
( ) ( ) z x x f
z
n i
X x
1 0
1
min ,..., min

=

So, the initial multi-dimensional minimization
problem is equivalent to the above one-dimensional
problem of finding the global minima of the
discontinuous multiextremal function ( ) z . The
Hilbert technique is used here for building the
development of the Peano space-filling curve
depending on parameter m that stands for the
number of subdivision levels (the development
density). The approach needs not any derivatives of
the function to be minimized with updating
parameters. Some disadvantage of this approach is in
the fact that one-dimensional problem obtained by the
above reduction leaks some information on the
closeness of iteration points in the initial
multi-dimensional space.
Algorithm SFC: reduction of the problem
dimension using the space-filling curve method.
(1) Set formally the point
*
x of local minimum.
Let two small numbers , 0
1
> , 0
2
> and the
vector of constraints
n
R be given. Let
1 , 0 = = m k .
(2) The local minimization phase: find new point
*
x (after l sub-iterations); set l k k + = . If
( ) 0 =
k
x f then go to step 3. If ( )
1
<
k
x f and
i
k
i
x > for some ,
x
I i { }, , ... , 2 , 1 n I
x
= then build
the effective sub-vector
k
eff
x , reduce the
multi-dimensional problem to the one-dimensional
form and go to step 3. Else, go to step of stochastic
scanning.
(3) Set 1 + = m m . If
max
m m then go to step 4.
Else, go to step 5.
(4) Define the current value
*
z using the
development of the Peano curve and calculate the
current approximation of ( )
*
z . If ( )
2 *
z then go
to step 3.
(5) Reconstruct
*
x and ( )
*
x f using
*
z and ( )
*
z .
(6) Define the point of local minimum
*
x x
LM
= . Stop.
The pseudo code brief description of the hybrid
PCASFC algorithm that combines the PCA and the
deterministic space-filling curve method is as follows:
(1) Generate an initial solution Old_Config.
(2) For 0 = n to #iterations
Generate a stochastic perturbation of the solution
If Fitness (New_Config) >Fitness (Old_Config)
Old_Config :=New_Config
Local search ( )
Else
Scattering ( )
End If
End For
(3) Local search ( )
Apply procedure of local search using the
Space-Filling Curve Method
Return
(4) Scattering ( )
=1
scatt
p ( Fitness (New_Config))/(Best Fitness)
If
scatt
p >random(0, 1)
Old_Config :=random solution
Else
Exploration ( )
End If
Return
4. Computational Results
In this section, two numerical examples of hybrid
algorithms V-PCALMS and V-PCASFC applications
to model multiobjective optimizations problems are

194
presented. First example is devoted to the
isoperimetric problem of determining the optimal
configuration of the rod system. In the second
example the updating of the computational model of
the coolant two-phase flow dynamics in the primary
circuit is carried out.
Example 4.1: The computations were performed in
connection with the problem of choosing the optimal
parameters for the shaped rod system with a
compensator of axial deformations. The system
consists of uniform straight rod elements with two
fixed lower nodes. Two rods (left and right) are
vertical. The middle rod rigidly bounded with them is
in a horizontal position. The system has a full length
of 2.5 m and is made of steel (Youngs modulus
200 = E GPa and mass density 7800 = kg/m
3
).
The rods have ring-shaped sections with the outer
diameter of 0.018 m and the wall thickness of 0.0014
m. The compensator of axial deformations is modeled
by a spring element with a stiffness of 1.0 N/m. Free
oscillations of the system are investigated. The first
criterion ) (
1
x f is a height of the system, and the first
(the smallest) eigenfrequency is the second criterion
) (
2
x f . It should be noted that the first criterion is the
first variable
1
x (undimensional) in the problem
simultaneously. The second variable
2
x is the
undimensional distant parameter determining a
position of the compensator on the horizontal rod. A
dependence of the two lower eigenfrequencies ) (
1
x p ,
) (
2
x p corresponding to out-of-plane oscillations and
in-plane oscillations of the system from the second
variable is shown in Fig. 1.
The lower envelope of the two curves in Fig. 1
represents the second criterion of the problem. The
goal is to maximize the criteria being in conflict. It is
clear that the second criterion has two local and one
global maxima. So it is necessary to solve
subproblems for global optimization of the second
criterion when solving this multiobjective
optimization problem with two criteria. Individual
normalized criteria are determined as follows:

Fig. 1 The second criterion of the multiobjective problem
as a lower envelope of eigenfrequencies that correspond to
in-plane and out-of-plane oscillations of the system.

( ) ( ) ( ) 2 , 1 , / ) (
min max max
= = i f f x f f x F
i i i i i

The iteration history for the solution of the current
subproblem is shown in Fig. 2 ( m stands for the
development density in the space-filling curve method).
The solution of the multiobjective optimization
problem for normalized criteria in the criterion space
is presented in Fig. 3. Exact (theoretical) global
Pareto-optimal front and approximated Pareto-optimal
front ( ( ) x F
2
and ( ) x a F
2
correspondingly) are rather
closely related. The maximal computational error here
is no more than 3.9%.
Example 4.2: The computations were performed in
connection with the problem of identification of the
coolant phase constitution in the VVER-1000 primary
circuit [10]. Appearance of the second phase is
possible: in a coolant heating zone (pressure tank of
the pressurizer), in an exit volume of the reactor
pressure vessel (RPV), in a core barrel of the RPV, in
exit volumes of main circulating pumps. In order to
formulate the inverse problems some vectors of
relative acoustic velocities in a coolant flowing throw
the specified zones may be introduced. Let now the
anomalous coolant state constitution be characterized
by the only controlling variable: ] 100 , 70 [
1
x (in %).
The error function is determined using ten lower
spectral components. The goal is to maximize the first
spectral component (the first eigenfrequency of
coolant oscillations) and to minimize the seventh
spectral component simultaneously. The criteria are
defined as follows:


195

Fig. 2 The convergence of the solution of the current
global optimization subproblem when the first criterion is
fixed ) 8 . 0 (
1
= x .

Fig. 3 The exact and approximated global Pareto-optimal
fronts.

( ) ( ) ( )
min
1
max
1 1
max
1 1
/ ) ( p p x p p x F =

( ) ( ) ( ) ( )
min
7
max
7
min
7 7 2
/ p p p x p x F =

Both the components are mostly sensitive to the
considered state of the coolant constitution. The
corresponding bicriterial problem is formulated as the
problem of the appropriate correction of the
computational model of the two-phase coolant
according to current abnormal spectrum. The iteration
history for the solution of the current subproblem is
shown in Fig. 4.
The solution of the bicriterial optimization problem
for the criteria in the criterion space is presented in Fig.
5. Exact (theoretical) global Pareto-optimal front and
approximated Pareto-optimal front ( ( ) x F
2
and
( ) x a F
2
correspondingly) are rather closely related.
The maximal computational error here is about 4.0%.
5. Conclusions
Two novel hybrid algorithms for multiobjective
optimization problems combining a Metropolis-based

Fig 4 Number of final iterations Niter of the global
optimization algorithm vs. the problem criteria values.

Fig. 5 The exact and approximated global Pareto-optimal
fronts.

algorithm and deterministic linearization method or
space-filling curve method for local search are
presented. Smoothing approximations are introduced
during the local search that makes it possible to
expand the V-PCALMS algorithm on the class of
non-differentiable problems. The V-PCASFC
algorithm being introduced here does not require any
gradient information. Both the algorithms were used
for solving bicriterial model problems connected to
choosing parameters of shaped rod system with a
compensator of axial deformations and with
computational model updating for the two-phase
coolant flow in the nuclear reactor primary circuit.
Numerical experiments show the principal
applicability of the proposed hybrid algorithms for
solving the above model multiobjective optimization
problems. The future work will be devoted to
increasing the computational efficiency of tools for
solution the model updating problems with regard to
noisy data using multiple criteria representation.

196
Acknowledgment
This research is partially supported by the program
Leading Scientific Schools of Russian Federation
(Grant NSh-5271.2010.8).
References
[1] J . Gao, J. Wang, A hybrid quantum-inspired immune
algorithm for multiobjective optimization, Applied
Mathematics and Computation 217 (2011) 4754-4770.
[2] C. Gil, A. Mrques, R. Baos, M.G. Montoya, J . Gmez,
A hybrid method for solving multi-objective global
optimization problems, J ournal of Global Optimization
38 (2007) 265-281.
[3] V.V. Podinovskij, V.D. Nogin, Pareto-Optimal Solutions
of Multicriterial Problems (in Russian), 2nd ed.,
FIZMATLIT, Moscow, 2007.
[4] B.N. Pschenichnyj, R.B. Sosnovskij, A linearization
method for multiple criteria problem solution (in Russian),
Kybernetika 6 (1987) 107-109.
[5] V.D. Sulimov, P.M. Shkapov, A smoothing
approximation in vector non-differentiable optimization
problems for mechanical and hydro-mechanical systems
(in Russian), Vestnik MGTU, Estestvennye Nauki 2 (21)
(2006) 17-30,.
[6] W.F. Sacco, H.A. Filho, N. Henderson, C.R.E. Oliveira,
A metropolis algorithm combined with Nelder-Mead
simplex applied to nuclear reactor core design, Annals of
Nuclear Energy 35 (2008) 861-867.
[7] K.I.M. McKinnon, Convergence of the Nelder-Mead
simplex method to a non-stationary point, SIAM J .
Optimization 9 (1998) 148-158.
[8] V.D. Sulimov, A local smoothing approximation in a
hybrid algorithm for optimization of hydro-mechanical
systems (in Russian), Vestnik MGTU, Estestvennye
Nauki 3 (38) (2010) 3-14.
[9] R.G. Strongin, Y.D. Sergeev, Global optimization:
Fractal approach and non-redundant parallelism, J ournal
of Global Optimization 27 (2003) 25-50.
[10] V.G. Kinelev, P.M. Shkapov, V.D. Sulimov, Application
of global optimization to VVER-1000 reactor diagnostics,
Progress in Nuclear Energy 43 (2003) 51-56.

J ournal of Mechanics Engineering and Automation 2 (2012) 197-202

Preventive Maintenance of Passengers Cars Driving in
the Territory of the Republic of Kosovo
Xhemajl Mehmeti, Naser Lajqi, Bashkim Baxhaku, Shpetim Lajqi and Hajredin Tytyri
Department of Construction and Design, Faculty of Mechanical Engineering, University of Prishtina, Prishtina 10 000, Republic of
Kosovo

Received: February 09, 2012 / Accepted: February 24, 2012 / Published: March 25, 2012.

Abstract: Preventive maintenance management enables the extension of the life expectancy of the vehicle equipment and
significantly contributes in reducing of maintenance cost. This paper analyzes necessary period for change of filters, respectively
period of changing of the oil filter, air filter and fuel filter that are fitted in the passengers cars by internal combustion engine.
Criteria for determining of the period (passed kilometres of the vehicle) for changing the filters in the territory of the Republic of
Kosovo are climatic condition, road state, average speed of the vehicles, fuel and oil quality, as well as level of air pollution. Also
here is an investigation made for period of change of filters for some types of passengers cars in auto services in Kosovo. The paper
presents real situation of preventive maintenance of the passengers cars in territory of Kosovo and the given conclusions will help
drivers of passengers cars and auto services to determine the optimal period for changing filters which will be best suits of
exploitation conditions for passengers cars to have better life expectancy of the engine.

Key words: Maintenance, filters, service life, interval, environmental pollution.

1. Introduction

A mechanical system to be technically in working
condition which means working normally, is essential
to have quality maintenance. This is important
especially in those systems that failures appear
consistently and mechanical system is not able to
work regularly. In order to prevent failures, a suitable
maintenance is necessary. As a concrete case of
mechanical systems, motor vehicles have applications
in many fields such as transport, agriculture,
construction as well as in different branches of
economy and industry.
Quality maintenance of motor vehicles represents as
an important stage of their life expectancy (service
life). This life expectancy needs to be examined as a
summary of all activities during the service life of
systems, starting from the initial phase of exploitation

Corresponding author: Shpetim Lajqi, assistant, Ph.D.
student, research field: automotive engineering.
E-mail: shpetim.lajqi@uni-pr.edu.
until the end of period of using the passengers cars.
Motor vehicles are one of the systems which are
considered enough complexes, which consist from a
large number of subsystems. The rapid development
of the technology in vehicle has prevented the
development of quality maintenance in vehicle. These
difficulties have influenced vehicle maintenance to
rank in the group of complex technical systems. This
requires a development process and professional
research of the maintenance systems in vehicle and
finding the necessary changes for improving vehicle
maintenance.
As a conclusion of all of this, the purpose to have
quality maintenance management should be focused
on improving the life expectancy of vehicles, vehicle
safety, readiness of the vehicle and environmental
protection [1].
The paper is organized as follows: Section 2
discusses the influence of filters in life expectancy of
the vehicles engine. Section 3 introduces the interval
D
DAVID PUBLISHING
Preventi ve Maintenance of Passengers Cars Dri ving in the Territory of the Republic of Kosovo

198
of change of filters in vehicles that driving in the
territory of Kosovo. Section 4 gives conclusions.
2. Influence of Filters in Life Expectancy of
the Vehicles Engine
Most of the vehicles experts believe the parts that
have the greatest impact on the life expectancy of the
engine are filters. Through regular changing of the
filters which has a low cost but has impact in better
life expectancy of the internal combustion engine.
During driving the vehicle across different
environments, the engine absorbs polluted air with
dust which is very abrasive (wasting) and in most
cases presents the main causes of wear of moving
parts of the engine.
Therefore, taking the importance of changing the
filters in the vehicles into consideration followed are
analyzed for finding optimal period for changing the
air filter, fuel filter, oil filter, motor oil as well as their
influence on the life expectancy of engine.
2.1 The Air Filter
Practice demonstrates that even in clean
environments, considerable amounts of abrasive
particulate matter are found in air. Therefore, to
prevent pollutants found in air, it is needed to install
the air filter in intake manifold of engine. Air filter has
influence in improving the life expectancy of the
engine (Fig. 1).

Fig. 1 Air filter [2].
The air absorbed from engine has a lot impurity and
is depended on climatic condition, road infrastructure,
the level of air pollution, etc. These impurities should
be stopped in air filter and do not be allowed to enter
cylinder of the engine. The situation that the impurity
enters cylinder of engine will cause serious damage in
engine and have impact on reducing life expectancy of
the engine.
2.2 The Oil Filter
Lubrication system plays important role in vehicle
engine. It is designed to deliver clean oil at the correct
temperature and press to every moving part of engine.
During operation process of the engine, oil could be
polluted by carbon (during the burning process of
fuel), dust, particular metal piece (from wear of
moving engine parts) as well as water. All of these
will cause increasing wear of moving part of the
internal combustion engine. To prevent the wear of
moving parts, engine should be equipped with an
efficient oil filtration system. So, the oil filter should
be installed to prevent wear of engine (Fig. 2).
A good oil filter should be able to stop the
penetration of very small materials and at the same
time not prevent the flow of oil through the filter and
flows in oil channels of the engine.
2.3 The Fuel Filter
For normal operation of the engine, the fuel supply
system plays crucial role. Significant part of the fuel
supply system is also the fuel filter (Fig. 3).

Fig. 2 Oil filter [2].
P
a
p
e
r

f
i
l
t
e
r

m
e
m
b
e
r

Filtered oil
Unfiltered oil
Unfiltered oil
Center pipe & filtered oil

199

Fig. 3 Fuel filter [2].

The primary task of the fuel filter is to stop the
various impurities in fuel and prevent them not to
move in sensitive parts of feeding system of engine.
Feeding system is completely closed system which
presents that any fault observed through the decrease
or increase of pressure (when the filter is blocked by
impurities) in fuel supply system.
Impurities in fuel are the main cause of filter blocking.
Such problem is very difficult to isolate, so the vehicle
engine should be supplied with quality fuel and the
frequency of fuel filter should be as often as possible.
3. The Interval of Change of Filters in Vehicles
that Driving in the Territory of Kosovo
This section gives recommendations on necessary
period for changing the filters and motor oil for
internal combustion engine, which is based on the
operation conditions of passengers cars that are
driving in territory of the Republic of Kosovo.
3.1 Determining the Required Interval for Changing
Filters and Motor Oil
Determining the optimal period for changing filters
and motor oil is one of the challenges auto services in
Kosovo faced every day. These difficulties are due to
different operation conditions because passengers
cars are being driven in different areas (e.g., area with
variable climatic condition, poor road infrastructure,
etc.) that makes it difficult determining optimal time
to changes the filters.
In Table 1, the actual and recommended period for
changing filters and motor oil for passengers cars
driven in Kosovo are given. Setting the interval to
change the filters and motor oil are obtained from
analysis of data from approximately 25000 vehicles
manufactured after 2000 and driven in Kosovo during
the period J anuary 2011 until October 2011.
From Table 1, it is easy to conclude that the
recommended period for changing of the filters and
motor oil for passengers cars driven in Kosovo is
smaller than current situation. Vehicle manufacturers
usually provide two recommendations period for
preventive maintenance, such as for the normal and
heavy driving conditions. One of the vehicle
manufacturers: Land Rover gives these
recommendations [3].
Manufacturers recommend that maintenance should
be performed after every 5000 km (driving in hard
conditions), if the vehicle will use in these exploitative
conditions [3]:
Frequent starting and stopping;
Driving in dusty conditions;
Climate variation condition from 25
C up to +
25
C;
Frequent short trip;
Extensive idling;
Engine most of the time work in small number of
rotation, etc.
These types of driving conditions, in adequate way
respond to the conditions of Kosovo, therefore, the
interval of change of filters to be realized after
passengers cars pass 5000 km is recommended.

Table 1 Actual and recommended period for changing
filters and motor oil for passengers cars that are being
driven in Kosovo.
Change of
Average of actual
changing of filter and
motor oil (km)
Recommendation for
changing of filter and
motor oil (km)
Oil filter 14.500 5000
Air filter 15.540 5000
Fuel filter 18.500 5000
Motor oil 11.100 5000

P
a
p
e
r

f
i
l
t
e
r

m
e
m
b
e
r

Inlet
Outlet
Water
drain screw

200
A large number of factors have contributed to
reduce the interval of changing filters and motor oil in
Kosovo, such as poor quality of motor oil, fuel, road
infrastructure as well as environmental pollution (dust)
which is 3 up to 10 times higher than that allowed by
International Standards [1]. Starting from a study done
by the Institute of Public Health of Kosovo conducted
in 2009 for air pollution, particular matter (particles
suspended) in some cases is 18 times greater than the
maximum value permitted by laws.
A very important fact should be noted that the
average speed of driving of the passengers cars in
town Prishtina is approximately 15 km/h which affects
the engine of the vehicle. The vehicle will work more
hours and will pass fewer kilometres [1]. In these
circumstances of driving of the vehicles, it is
necessary to utilize the most frequent change of filters
and motor oil.
Recommended period for changing of the fuel filter
is based on the fact that in most cases in market, you
can find poor quality of fuel and another argument is
that most of the passengers cars are driven with
minimum amount of fuel in the tank. A research
conducted in some points for supply with fuel carried
out during 2011 [1] shows that over 80% of drivers
vehicle supply with amount of fuel around 10.00 ,
which is equivalent from 7.5 to 8.5 litters. This
research has also found that over 80% of vehicles are
driven with fuel tank filled with less than 10% of total
capacity; this means that the vehicles are driven with
minimum amount of fuel.
When the amount of fuel in the tank is minimal, the
space above the fuel is filled with air. As is known, air
contains a small amount of moisture and this moisture
of air condenses inside the fuel tank, especially at low
temperatures. The amount of condensed water can
cause corrosion of the metal tank and will mix with
fuel tank, causing reduction in the efficiency and life
expectancy of the fuel feeding system of engine.
Another reason to keep full tank is that corrosion
formed from the air humidity tends to move to the
bottom of the tank by forming a layer of sediment.
Thus, other impurities can be not absorbed from the
pump if they have enough fuel. And if the fuel level in
tank is very low, the supply of engine with fuel will be
from bottom of the tank and fuel mixed with dirt will
cause the reduction of performance of the fuel feeding
system of engine, filter blocking, pipes blocking,
damage of the injectors, etc. Some tanks have installed
the filter that prevents these impurities to enter in
feeding system of engine, but if the cars are driven
continuously with an empty tank, filter may be
blocked.
3.2 Factors that Impact in Wear of the Vehicles Engine
Studies have shown that in most cases, defects of
the engine are as a result of wear of motion parts. To
prevent the wear of the engine, it is necessary to
continually make improvements in the filtration
system. Practice has shown that the life expectancy of
the engines is in function of the purity of air entering
the combustion chamber [1].
The engine wear is very complex to analyze because
it is depended from many factors such as [1, 4]:
Forms of pollutants;
Size of the pollutants;
Type of pollutants;
Concentration of pollutants;
Vehicle operation conditions;
Road infrastructure;
Engine load conditions;
Performance of filters (air, oil as well as fuel
filters), etc.
For a better life expectancy of the engine, it is
required to be adequate protection from abrasive
pollutants that may enter in the combustion chamber
thought the suction manifold.
3.3 Investigation of Changing Filters of the
Passengers Cars in Territory of Kosovo
The following diagrams present current situation of
the changing the filters for about 25000 investigated

201
vehicles in Kosovo for various types, such as VW,
Mercedes, Audi, Land Rover, and other vehicles.
These types of the vehicles are the mostly present in
Kosovo. Investigations are carried out for period
J anuary-October 2011 through auto services given in
Ref. [5].
Fig. 4 shows the total number and percentage of the
changing filters in auto services in Kosovo [5] for all
types of investigated vehicles.
Figs. 5-9 present the number and percentage of the
changing filters for passengers cars such as VW,
Mercedes, Audi, Land Rover and some other types of
passengers cars in auto services in Kosovo mentioned
in Ref. [5]. As seen from presented diagrams, changes
of the filters are not carried out according to the
manufacturers recommendations.
Vehicle drivers primary attention was paid only in
changing of oil filters which is in percentage 99.06%
of the investigated vehicle. Second care is dedicated
on the changing of the air filters which is 74.08%,
while, changing of fuel filter is only 44.67%.

Fig. 4 Diagram of situation for changing filters for all
types of investigated cars.

Fig. 5 Diagram of situation for changing filters for VW
passengers cars.

Fig. 6 Diagram of situation for changing filters for
Mercedes passengers cars.

Fig. 7 Diagram of situation for changing filters for Audi
passengers cars.

Fig. 8 Diagram of situation for changing filters for Land
Rover passengers cars.

Fig. 9 Diagram of situation for changing filters for other
types of the passengers cars.

202
Therefore, it can be concluded that changing the
filters is a worrying fact, which can be seen that air
filters have a substitute smaller than oil filters, while
based on the recommendations of manufacturers, air
filter should be replaced at the same time with oil
filter or more often. This is based on the fact that the
air filter is exposed to most dust and dirt in the air.
Another important fact received from the above
diagram is that the fuel filter has less percentage of
change than air and oil filters. This difference would
be justified if the quality of fuel is high (in accordance
with International Standards). As known, the quality of
fuel in Kosovo is not considered as a high quality and
vehicle drivers drive their vehicles with minimal
amount of fuel in the tank. Based on these arguments,
we can conclude that this change is not regular and
does not comply with the manufacturers
recommendations and presents a permanent risk to
damage the fuel feeding system of the engine. This is
justified that devices of feeding system of engine work
with a high precision. So it is recommended to change
the fuel filter in same time with oil filter and air filter.
4. Conclusions
Starting from investigation made in this paper for
analyzing the period of changing of filters and motor
oil for passengers cars driven in territory of Republic
of Kosovo, we can conclude that
The recommended period for changing the filters
and motor oil for territory of Kosovo is much smaller
than the current period;
Analyses of investigated data show that the
changing of filters is not in accordance with
manufacturers recommendation and will present
permanent risk for reduction of life expectancy of the
vehicle engine and
Vehicles driver pays more attention to changing
the oil filter with a percentage of 99.06%, compared to
air filter with 74.08% and fuel filter with 44.67%.
Research carried out in the paper is a good
suggestion for auto services in Kosovo and so on. We
recommend that the current situation should be
changed. So, the preventive maintenance to be
changed are motor oil, oil filter and air filter while
fuel filter after second preventive maintenance of
course. This will be in consideration during summer
season. While during winter season, we recommend
all filters to be changed at the same time.
References
[1] XH. Mehmeti, Diagnostic model in function of engine
maintenance systems, Master Thesis, University of
Prishtina, 2011.
[2] R. Fisher, R. Gscheidle, U. Heider, B. Hohmann, W. Keil,
J . Mann, B. Schlogl, A. Wimmer, G. Wormer, Fachkunde
Kraftfahrzeugtechnik, Europa Lehrmittel, Germany,
2009.
[3] Available online at: www.landrover.com, accessed May
20, 2011.
[4] D. Mann, A Comprehensive Technical Handbook on
Motor Oil, Lubrication and Filtration, 2008.
[5] Companies-Auto Services: LTG Land Rover, Prosche,
Vali, Reforma, 2011.

Open Architecture Robot Control System

Enviado por

Dados do documento

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Open Architecture Robot Control System

Enviado por

Direitos autorais:

Formatos disponíveis

Volume 2, Number 3, March 2012 (Serial Number 9)

(q) + S(q, q )_ q (t) + 0(q)

(q) + S(q, q )_ q (t) + 0(t)

(t) + 0(q) (19)

(q) + S(q, q ) s(t) + E

(t) which is defined as follows:

(q) + S(q, q ) s(t) =

(t)in Eq. (35), we can represent the derivative of the

(t) u . This implies the

(q) + S(q, q ) s(t) = E

(q) +S(q, q )_ s(t) +

(t) in Eq. (13) can be

(t) in Eq. (66) shows the

Você também pode gostar