Neural Comput & Applic (2013) 23:913925

DOI 10.1007/s00521-012-1016-0

ORIGINAL ARTICLE

Designing of an effective structure of system for the maintenance

of a technical object with the using information from an artificial
neural network
Stanisaw Duer Konrad Zajkowski
Radosaw Duer Jacek Pas

Received: 6 March 2012 / Accepted: 21 June 2012 / Published online: 7 July 2012
 Springer-Verlag London Limited 2012

Abstract The present article covers a description of the

structures of maintenance systems with a diagnostic artificial neural network as well as a description of those
systems that are without a neural network (organized in a
classical manner). Such a maintenance system that is
organized on the basis of information from an artificial
neural network is a rational system. In this system, maintenance costs are significantly reduced (a reduced time
required for regeneration and reduced expenditures on
preventative activities). Also, theoretical basis is presented
of the modelling of the structure of the maintenance system
of objects in the form of the following models: mathematical (analytical), graphical and descriptive. For the
purpose of the research being conducted, the methodology
was developed of a synthesis of the structure of the
maintenance system of technical objects (continuous
operation), that is such technical objects that require short
shutdown times (aircrafts, radiolocation systems, power
engineering equipment, etc.).

S. Duer (&)  K. Zajkowski

Department of Mechanics, Technical University of Koszalin,
15-17 Raclawicka St., 75-620 Koszalin, Poland
e-mail: stanislaw.duer@tu.koszalin.pl
K. Zajkowski
e-mail: konrad.zajkowski@tu.koszalin.pl
R. Duer
OTICON Polska Production Sp. z.o.o, ul. Lubieszynska 59.
Mierzyn, 72-006 Szczecin, Poland
e-mail: radoslaw.duer@wp.pl
J. Pas
Department of Electronic, Military University of Technology,
Warsaw, 2 Kaliskiego St., 00-950 Warszawa, Poland
e-mail: jpas@wat.edu.pl

Keywords Servicing process  Intelligent system 

System modelling  Expert system  Artificial neural
networks  Knowledge base  Diagnostics information

1 Introduction
Technical objects used in the process of operation may be
found in specific states (operation or servicing) (Fig. 1)
[24, 714, 19, 20]. During operation, a technical object is
exposed to various reactions, there are ageing changes, and
the objects functional elements wear. The changes of the
objects physical properties in relation to their nominal
values result in the lowering value of the quality of functioning and thus the lowering value of the objects operational function during its operation. Then, technical object
passes to a shutdown state in the operation process, which
is determined by the states of non-operation or incomplete
operation states.
Objects do not realize their operational functions in the
shutdown state. In these states, there occurs a location of
defects (non-operation) or maintenance. In order to counteract random changes of the operational quality of an
object and in order to maximize it, technical maintenance
of an object is organized through maintenance (regeneration). Technical objects may possess automatic diagnosing
systems that make use of artificial neural networks; this
issue was described in [2, 3, 5, 12, 13, 25, 26]. For this
class of objects, rational maintenance systems are organized on the basis of information obtained from a neural
network, which are presented in [2, 4, 6, 8]. At the same
time, for those technical objects that are not diagnosed (no
recognition of the present state), the preventative activities
organized for the object are executed on two stages. On the
first stage, the defect (non-operation condition) is located in

123

914

Neural Comput & Applic (2013) 23:913925

interferance
servicing

{1;0}
{2}

The object X(ei,j)

Nonunder
operational
X(w)(ei,j)
repairing
state

Diagnostic
system with Fc
ANN (DIAG
programme)

min

Repairing
of
the object

operation
Fc

max

Fig. 1 Diagram of operation process for technical object utilizing

artificial neural network, where X(ei,j) diagnostic signal in jth element
of ith set; X(w)(ei,j) model signal for X(ei,j) signal; FC function of the

use of the object, W(e(ei,j) = {2, 1, 0}) diagnostic information-value

of state assessment logics for element j within i module of the
object (the states of elements of the object)

the object. This type of preventative activities for an object

is known as unplanned maintenance. It is only on the further stage, that is the removal of the non-operation condition that the object is further regenerated in compliance
with the maintenance programme. For this class of objects
that are not diagnosed in a continuous manner (online), the
times of the objects operation are determined, after which
the maintenance of the object is to be carried out. The
objects maintenance systems organized in this way are
known as planned maintenance.
Operation of technical objects is connected with operational costs. The user of a given technical object undertakes efforts to optimize the costs borne in relation to the
operation; these issues are complex [2, 4, 811, 14]. In
order to minimize the maintenance costs, one should do the
following: know those indices that characterize the operation process of a technical object, know the operational
properties of a given object (how to use this object effectively, on what conditions, when maintenance is to be
organized and how to optimize this process) and to conduct
investigations into the operation process in the aspect of its
effectiveness.

penetration into this structure [211]. The accepted division is considered to be sufficient if we distinguish the
basic module-element in the structure of the object. One of
the purposes of the functional-diagnostic analysis is the
determination of the objects state. The objects state is
determined on the basis of an examination of the set of output
(diagnostic) signals {X(ei,j)}(Table 1) [26]. The set of its
functional elements {ei,j} determined during a diagnostic
study of the object constitutes the basis for the list included in
the table of a set of diagnostic signals (Table 1).
The modern of an intelligent diagnostic subsystem [2, 3,
5, 9] (Fig. 2) utilizes not only a measurement A/D converter card with appropriate signal interfaces but also some
computer tool used for proper signal registration as well as
for acquiring and processing data registered. The purpose
of such a process is to build a diagnostic knowledge base
based upon an analysis of both the object and the results of
the measurement stored.
The ANN network developed is presented in Fig. 3. It
consists of three layers: F1 the input layer, F2 the output
layer and the intermediate layer. The input cells of layer F1
process the initial diagnostic information according to the
algorithm of the DIAG programme. The whole of the issue
of information processing by ANN neurons (Fig. 3) [3, 5,
9, 12, 13, 16, 17, 22, 23] takes place in D-dimension
diagnostic space (x) determined by the elementary signal
vectors [Xi,j]. The input signal in the form of Xi;j

T
is being passed to all neurons of
X1;1 ; X1;2 ; . . .; Xi;j ; . . .; XI;J
the ANNs input layer.
The input cells memorize the vectors of signal standards
[Xi]. Based upon that, the neurons from the input layer
determine the measures of similarity between the input
signal vector and its standard, and the length of the input
signal [Xi] to all vectors of weights wi;j w1 ; w2 ; . . .; wn ;T
where i = 1,,N. In the ANN network presented in Fig. 3,
the neuron (i) placed in layer F1 is connected to neuron (j)
placed in layer F2, where: j = 1, 2, , N [2, 3, 13, 16, 17,
18, 2426].
The value of its output function in ANN is derived from
the relation (1)

2 Organization of hybrid intelligent system which

support of maintenance of a technical object which
uses information from an artificial neural network
Repairable technical objects for which a short time of their
shutdown is required (radar systems, airplanes, etc.) are
frequently equipped with specialist adjustment systems that
reconstruct their functional functions to the nominal level.
An adjustment system of the objects functionality functions is a sophisticated system of the regeneration of the
object, which includes the following intelligent subsystems: diagnosing and maintenance. The purpose of the
diagnostic system is current and constant recognition
(monitoring) of the state of the object.
The division of the objects internal structure {ei,j}
accepted in the paper defines explicitly the depth of

123

Neural Comput & Applic (2013) 23:913925

Table 1 Table of objects input
diagnostic signals

Object

Where X(ei,j) diagnostic signal

of jth element in ith assembly,
Eith assembly of the object

The technical
object

{X(ei,j)}
{X(w)(ei,j)}

915

Level of object Ei

Vector of initial diagnostic signals {X(ei,j)}

X(e1,1)

X(ei,j)

X(ei,J)
X(e1,J)

E1

X(e1,1)

X(e1,j)

Ei

X(ei,1)

X(ei,j)

X(ei,J)

EI

X(eI,1)

X(eI,j)

X(eI,J)

Measurement
module

ANN module
with DIAG
programme

Fig. 2 The scheme of structure of an intelligent system DIAG for the

diagnosis of a technical object, where X(ei,j) diagnostic signal in jth
element of ith set; X(w)(ei,j) model signal for X(ei,j), W(e(ei,j)) value of

The classification
of ANN module

{W((ei,j))}

Diagnostic knowledge
base of the object

{W((ei,j))}

state assessment logics for jth element within ith module (from the set
of the accepted three-value logic of states assessment) {2, 1, 0})

Fig. 3 Diagram of neural networks

yl f

K
X

xi;j  Xi

i1

where xi,j weight coefficient.

The results of the objects diagnosis obtained from the
relations are presented in Table 2.
The classification process of the objects states is realized in the output layer of the network. The value of the
output function of the network determined for a given input
signal vector constitutes the basis for the classification of
the states. The output layer neuron possesses a memory
cell, where the diagnostic inference rules of the network

are recorded. Those classes of the objects states that are

recognizable, including the ranges assigned to them of the
changes of the networks output function values, are presented in Fig. 4.
On the final stage of the work of a neural network, a
classification process of the objects states is realized
according to the algorithm. For this purpose, to the values
of the output function as determined by the network, proper
classes of the objects states [2, 3] were assigned according
to the classification diagram (Fig. 4).
It is evident from the analysis (Fig. 5) that the process of
the renovation of an object in an analytical approach

123

916

Neural Comput & Applic (2013) 23:913925

Table 2 Table of objects states

State of object

W(e(O))

State of
module

Number of the
assembly

Vector of states of elementary components {ei,j}

e(e1,1)

..

e(ei,j)

e(ei,J)
W(e(e1,J))

W(e(E1))

E1

W(e(e1,1))

W(e(e1,j))

W(e(Ei))

Ei

W(e(ei,1))

W(e(ei,j))

W(e(EI))

EI

W(e(eI,1))

W(e(eI,j))

W(e(eI,J))

Where W(e(ei,j)) value of state assessment logics for jth element within ith module (from the set of the accepted three-value logic of states
assessment) {2, 1, 0}), [ symbol complementing the size of table

Fig. 4 The scheme of hybrid

intelligent system which support
of a maintenance of a technical
object, where FC max. value of
the function of the use of the
object, (W(e(ei,j)) value of state
assessment logics for jth
element within ith module (from
the set of the accepted threevalue logic of states
assessment) {2, 1, 0})

consists in the transfer of the objects usability features

from the level of the plane of the current use (x) to the
level of the plane of nominal usability features (ME) [211,
14]. The function that renovates the object in the servicing
system is presented in the form of the following
dependence:
 
         
 
ME ei;j f M ei;j ; W Z ei;j ; Al ei;j ; Rr ei;j ; ME ei;j
2
where {M(ei,j)} structure of the technical object, W(z(ei,j))
the maintenance structure of the object, {Rr(ei,j)} the set of
rules for maintenance (repairing), {Al(ei,j)} the set of preventive activities, {ME(ei,j)} the maintenance system produces a set of maintenance information.
A particularly important element of the maintenance
system is the knowledge base (Fig. 5). It can be defined as
specialized set of the objects maintenance information that
is determined by the following: the maintenance structure
of the object {Wz(ei,j)}, the set of rules for maintenance
(repairing) {Rr} and the set of preventive activities {A(ei,j)}
[2, 4, 68].

123

As a result, the intelligent maintenance system produces

a set of maintenance information {ME(ei,j)}, which will be
used for organization of the objects rational (optimised)
maintenance system.
A set of the objects servicing information that constitutes the basis in the process of designing of the structure of
the servicing system is presented in the form of the following dependence:
       

ME ei;j M ei;j ; W z ei;j Al ; Rr :
3
As a result, the maintenance system produces a set of
maintenance information {ME(ei,j)} that is presented in
Table 3.
A methodology of the construction of a maintenance
system was developed in connection with an increasingly
more popular phenomenon of the use of the computer in
the process of organization (designing) of a maintenance
system of technical objects. The purpose of this technology is
also to guarantee a practical application of the material prepared during the construction of the structure of the maintenance system of an object [4, 6, 811, 14, 15, 19, 2123].

Neural Comput & Applic (2013) 23:913925

917

Fig. 5 Diagram of a servicing knowledge base of the object on the

basis information from an intelligent system with the use of an
artificial neural network. Where X(ei,j) diagnostic signal in jth element
of ith set; X(w)(ei,j) model signal for X(ei,j) signal, Rr(ei,j)} the set of

rules for maintenance (repairing), {Ap(ei,j)} the set of preventive

activities, {ME(ei,j)} the maintenance system produces a set of
maintenance information, {M(ei,j)} the set of elements of internal
structure of the object, FC function of the use of the object

Table 3 Structure of the servicing information of the object

Object

Servicing levels
of object

Number of the
assembly

Vector of the servicing information of the object [ME(ei,j)]

ME(e1,1)

ME(ei,j)

ME(ei,J)
ME(e1,J)

E1

ME(e1,1)

ME(e1,j)

Ei

ME(ei,1)

ME(ei,j)

ME(ei,J)

EI

ME(eI,1)

ME(eI,j)

ME(eI,J)

Where ME(ei,j) servicing information of jth element in ith assembly, Eith assembly of the object

3 Organization and description of the structures

of maintenance systems in an intelligence system
on the basis information from an artificial neural
network
A maintenance system of technical objects which is not
diagnosed in a continuous manner by a diagnostic system is
not able to react systematically to the changes of the states
occurring in it. If there is a defect in the object, it will be in a
non-operation state. For this state of the object, an unplanned
repair is to be organized [2, 4, 6, 811, 14, 19, 2124].
A technical object in the state of unplanned repair is
serviced on two stages. On the first stage, the object is
diagnosed: in the case when a defect has occurred. It is
only on the second stage that it is serviced through preventative activities.
The structure of the objects maintenance system in the
state of unplanned repair consists of the service and
diagnostic structures. The object serviced, when staying in
the maintenance system, is regenerated with probability
(Pq) (Fig. 1). An example form of the maintenance system
structure is presented in Fig. 6.
The characteristics that describe the structure of the
maintenance system in the state of unplanned repair can be

developed on the basis of a diagram of the structure of a

maintenance system. For this purpose, a diagram of the
structure of the maintenance system was developed with
the use of information included in Table 3.
The time of the removal of the objects unfitness (Ta) is
the time when an unfitness is removed in an object located
in a given maintenance system. In the maintenance system
presented in Fig. 6, two ways are possible of the location of
an unfitness, which were divided into linear and sequential
considering the possible ways of transition between decision (service) levels. Each of the ways of the removal of an
unfitness of the object (Ta) can be realized in one of the
accepted logics of the recognition of states, here, in a
bivalent and trivalent evaluation of states.
The average time of the objects maintenance (regeneration) on a given level of the maintenance structure is the
total average time during which the activities of preventative servicing are carried out, within the confines of
which the object is regenerated in time (TOD) on a given
level of the maintenance structure or the whole system of
the objects maintenance.
In such a maintenance system of an object which is
organized in a classical manner, that is one that does not
include an expensive automatic system to diagnose this

123

918

Neural Comput & Applic (2013) 23:913925

Fig. 6 Diagram of the q-level model of the structures maintenance system with an artificial neural network

object, the regeneration time is additionally increased by

the location time of unfitness on a given level of the
maintenance structure or in the whole system of the object.
The present article covers the organization of a maintenance system with an artificial neural network. A technical object is regenerated on the basis of diagnostic
information from an artificial neural network, which is
converted to the form of a set of maintenance information
{ME(ei,j)} [see (Table 3); (Fig. 3)]. The methodology of
the determination of the objects maintenance information
was presented in the introductory section of this article.
A graphical presentation of the structure of a maintenance system was presented in Fig. 6. When analysing the
structure of the maintenance system presented in Fig. 6,
one can notice that the object on the first level of its
maintenance structure is regenerated with probability (P1)
in the time of the objects regeneration (T1OD). If there was

123

a defect during the regeneration of the object on the first

maintenance level, then the unfitness of the object is
located with probability (1P1) in time (T1LU). Once the
defect has been removed, the object is regenerated in time
(T1LU ? T1OD) and goes onto regeneration on the second
regeneration level with probability (P2) in time (T2OD). If
there was a defect during the regeneration of the object on
the second maintenance level, then the unfitness of the
object is located with probability (1P2) in time (T2LU).
Once the defect has been removed, the object is regenerated in time (T2LU ? T2OD) and goes onto regeneration on
another (ith) regeneration level with probability (Pi) in time
(Ti OD).
If there occurred a defect during the regeneration of the
object on the (ith) level of its maintenance structure, then
the objects unfitness is located with probability (1Pi) in
time (TiLU). Once the defect has been removed, the object

Neural Comput & Applic (2013) 23:913925

919

is further regenerated in time (TiLU ? TiOD) and goes onto

regeneration on another regeneration level with known
probability (PI) in time (TI OD).
During a regeneration of an object in a maintenance
system, there might occur an event that consists in the fact
that the object being regenerated is damaged. Then,
objects state changes and it goes onto the unfitness state.
The preventative activities performed at that time in a
given structure of the objects maintenance system consist
in the location of the unfitness. For this reason, those
preventative activities that consist in the location of unfitness do not belong to the determined set {al} of preventative activities described with dependence (4). Once the
unfitness has been located and removed, the process of
the objects regeneration continues on this (ith) level of the
maintenance system where the defect occurred.
Once the time required for the removal of the objects
unfitness (Tla) had been determined, the total time was
calculated which was required for the regeneration of the
object in a given regeneration system on the basis of the
following dependence:
Ta Tal TOD :

Once the time required for the removal of the objects

unfitness (Tla) had been determined, a possible increment of
the time required for the removal of the objects unfitness
(DTlasq) was calculated depending of the logics applied of the
evaluation of states on the basis of the following dependence:
l
l
DTasq
Ta1
 Ta

where DTlasq an increment of the time required for the

removal of the objects unfitness resulting from an application a diagnosis in the maintenance system expressed in
the accepted valence of the evaluation of states.
The time required for the removal of the objects
unfitness (Tla) on the (qth) decision level was determined on
the basis of the following dependence:
Tal Dq
J
X

TLU;1;j TOD;1;j TLU;2;j TOD;2; ;    ; TLU;q;j

j1

TOD;q;j ;    ; TLU;Q;j TOD;Q;j

!
:

6
The determined structure of the maintenance system,
which was presented in a graphical form (Fig. 6) and in a
tabular form (Table 3), can be described with the aid of
similar indices that characterise the previously constructed
and described models for the state of an unplanned repair.
In the model under consideration with a q-level maintenance structure of the object (Fig. 6), the time of the
duration of the objects regeneration was expressed in the
form of the following dependence:

TODq

J 
X

Td:q;j Tw:q;j Tsp:q;j Tr:q;j Tdig;k

j1

where Td.q,j the average time of the delivery of an unfit

basic element in the (qth) level and in the (ith) layer; Tw.q,j
an average time connected with the replacement of an
unfit basic element in the (qth) level and in the (ith) layer;
Tsp.q,j an average time connected with the verification of
the basic element replaced in the (qth) level and in the
(ith) layer; Tr.q,j an average time connected with the
adjustment of the basic element replaced in the (qth) level
and in the (ith) layer; Tdig,k an average time connected
with the execution of the objects control diagnostics after
its regeneration.
If there occurred a defect in an object being regenerated
in a maintenance system, the algorithm presented in
(Fig. 6) is to be followed. The objects unfitness is to be
removed on the first stage, and only later, the objects
preventative activities can begin in compliance with
the objects maintenance programme developed. The
duration of the objects unfitness in this model was determined similarly as in (7) on the basis of the following
dependence:
l
Taq

Q 

X
Pl1  TaD1 Pl2  TaD2 Pl3  TaD3 ;;Plq  TaDq :
q1

8
Once the time required to remove the objects unfitness
(Tlaq) had been removed, the total time was determined required for the regeneration of the object in a
given maintenance system on the basis of the following
dependence:
l
Ta Taq
TOD :

After the quantity of the time required to remove the

objects unfitness (Tla) had been calculated, the assumed
increment (gain) of the time was determined according to
dependence (9) which was required for the removal of the
objects unfitness (DTlas) on the basis of the following
dependence:
l
DTas
q

Q 
X

Pl1  TaD1 Pl2  TaD2 Pl3  TaD3  Plq  TaDq

q1

 Ta :

10

Owing to the use of dependence (10), one can demonstrate

the influence of a given valence of the evaluation of states
(bivalent or trivalent) on the increment (gain) of the costs
of the regeneration of a technical object during an
examination of a maintenance system with an artificial
neural network.

123

920

4 The examination of the quality of the regeneration

of an object in a maintenance system organized
on the basis information from an intelligent system
The method presented in the present paper concerning the
control of the exploitation of a technical object on the basis
of its state was verified on the example of a reparable
technical object, which is a radar system. A functional and
diagnostic analysis of the object was carried out for this
purpose. A functional model was prepared and described of
the object: a missile homing station of an antiaircraft
missile set, which was presented in Fig. 7.
As a result of the described manner of the division of the
objects internal structure, the object was subject to a threelevel division of its structure. As a consequence of this
division of the internal structure (Fig. 7), seven functional
assemblies were distinguished (E1, E2,, E7) (Table 4),
and up to five basic elementsmodules [2, 4, 6, 811] were
distinguished in each one of them. As a result of the
analysis carried out, a functional and diagnostic diagram
was developed, on the basis of which a set of operational
elements and a set of output (diagnostic) signals were
established.
The presented method of a servicing of technical objects
requires the use of a uniform compliance of the designation
of the elements of the objects structure. For this reason,
the basic elements, modules of the object included in its
functional and diagnostic model, must be addressed in
the following manner (ei,j), where j is the number of the
element in a given assembly, and (i) is the ith number of
this assembly of the object.
The method of the evaluation of the influence of
selected expert inference methods on the quality of the
preventive procedures conducted was presented in an
example for a radar system. The basis for the design of
structures operating systems of the object is diagnostic
knowledge, which is determined by the diagnostic subsystem

Fig. 7 The functional scheme of air-defence radar device. where E1

steering (synchronization) unit, E2 transmitter unit (channel I or
channel II), E3 receiving unit, E4 permanent echo suppression unit, E5
display indicator unit, E6 precise display indicator unit, E7 electric
power supply unit of the station

123

Neural Comput & Applic (2013) 23:913925

DIAG. DIAG Diagnostic Module solution is a modern

application that uses neural network algorithm to work. The
DIAG program uses well-known and quite widely used
diagnostic approach for comparing the diagnostic signal from
their masters (similaritybased methods SBM). The end
result of the network diagnostic (DIAG program) is presented
diagnostic information in the form of an array An array of
object states.
DIAG computer diagnostic system constitutes the basis
for this qualitative assessment method, in which an artificial neural network (Fig. 3) was used. The object examined
possesses a complex internal structure; it also possesses a
mixed functional and information (analoguedigital) structure. For this reason, it is a good subject of the research
considering the various approaches of the expert when conducting a maintenance study: creating a specialist knowledge
base.
As a result, the maintenance system produces a set of
maintenance information {ME(ei,j)}, which will be used for
organization of the objects, which is presented in Table 5
and Fig. 8.
Conducting an experimental evaluation process of the
preventive procedures of a technical object required a
preparation of a plan for a testing experiment due to the
manners of the work (decision making) of the expert [1, 4,
611, 2123, 25, 26]. The basis of this examination was
setting up on the grounds of the same algorithm [7], a set of
maintenance information for the three examined methods
of the approach (inference) on the part of the specialist. In
this manner, three maintenance information sets were
obtained: {ME(e1,1)I}, {ME(e1,1)II}, {ME(e1,1)III}, on the
basis of which the vectors were prepared of the input
maintenance information. This constituted the basis for the
execution of the preventive procedures of the object. The
issues concerning the determination of maintenance information were presented in the following papers [7].
The final form of the maintenance knowledge base was
presented in the form of the information included in
Table 4. This table includes a set of maintenance rules for
the object. Each rule included in this table determines
explicitly which element of a given assembly of the object
must be subject to regeneration, and what range of prevention activities (from the set of the activities) is to be
performed on it. The execution of the determined set of
maintenance rules will result in the regeneration of the
whole element, assembly and the whole object. The technical object, once the maintenance has been performed, is
subject to a control check-up (testing) of the state, and in
the case of a negative result of the control, the object is
once again referred to prevention.
The examination of the quality of the regeneration of an
object in a maintenance system organized on the basis of
maintenance information was conducted for the two

Neural Comput & Applic (2013) 23:913925

921

Table 4 Internal structure of the object

Assembly of object

Structure of the object

E1

e1,1

e1,2

E2

e2,1

e2,2

e2,3

e2,4

e2,5

e2,6

E3

e3,1

E4

e4,1

e4,2

e4,3

e4,4

E5

e5,1

E6

e6,1

e6,2

E7

e7,1

e7,2

e7,3

e7,4

e7,5

Table 5 Structure of the servicing information of the object

Object

Number of the
assembly

Servicing levels
of object

Vector of the servicing information of the object [ME(ei,j)]

ME(e1,1)

ME(ei,j)

ME(ei,J)

E1

ME(e1,1)

ME(e1,j)

ME(e1,J)

:
ME(ei,J)

Ei

ME(ei,1)

ME(ei,j)

EI

ME(eI,1)

ME(eI,j)

ME(eI,J)

Where ME(ei,j) servicing information of jth element in ith assembly

Fig. 8 The structure of a set of

the maintenance information
{ME(ei,j)} of the radar system.
Where {1 to 10} number of
preventive activities of the
object: {1} control check-up,
{2} servicing, {3} adjustment,
{4} tuning, {5} regeneration,
{6} conservation, {7}
lubrication, {8} cleaning, {9}
refilling of working fluids, {10}
replacing the element with a
new one

123

922

Neural Comput & Applic (2013) 23:913925

method creation of the maintenance information of the

object:
1.

2.

The maintenance information of the object which

describes the structure of a maintenance system
without a diagnostic artificial neural network-a classical manner (model I).
The maintenance information of the object which
describes the structure of a maintenance system with
an artificial neural network (model II).

The core of the qualitative process of the preventive

procedures of a technical object involves a two-stage
examination. On the first stage, a regeneration (model I)
procedure of the object is conducted on the basis of the
maintenance information which is determined with the use
of a given expert inference method. It is only on the second
stage that the quality of these preventive procedures is
assessed on the basis of control diagnosing of the object
after the preventive procedures have been conducted. The
diagnosing of a control object is performed in a diagnostic
system with the use of the DIAG neural network. It was
assumed in the experimental examination of the (model I)
procedures of the object that each of the three inference
methods has an influence on the preventive procedures in
such a way that it guarantees a full regeneration of the
object.
The first method of the objects procedures (model I) is
performed with the use of a deductive manner of expert
inference. From there, the testing vector was developed of
the maintenance information being determined for the
deductive method of expert inference (model I), which
accepts the following form (9):
     
 
 
ME ei;j I ME e1;1 I ; . . .; ME ei;j I ; . . .; ME eI;J I
11


where ME ei;j I servicing information of jth element in ith

assembly for model I.
On the basis of this information: dependence (11), the
maintenance system was organized (Fig. 6), in which the
technical object was regenerated. Once the preventive
procedures have been performed, the technical object
undergoes control diagnosing with the use of the diagnostic
artificial neural network in DIAG program. The results
obtained of the diagnosing of the object (for model I) were
presented in the form of Table 6 Fig. 9.
Further, similarly as described above, the maintenance
was performed of the technical object for the inductive
method of expert inference. The preventive procedures of
the object were performed with the use of the testing vector
of maintenance information that was determined for the
inductive method of expert (model II), and which was
presented in the following analytical form (12):

123

     
 
 
ME ei;j II ME e1;1 II ; . . .; ME ei;j II ; . . .; ME eI;J II
12


where ME ei;j II servicing information of jth element in ith

assembly for (model II).
After the (model I) procedures have been performed, the
technical object, as in the previous case, underwent control
diagnosing with the use of DIAG program. The results
obtained of the diagnosing of the object for (model II) were
presented in the form of Table 7 and Fig. 10.
A function of the qualitative evaluation of the maintenance process of the object was developed. The qualitative
function of the control qualitative evaluation of the preventive procedures process of a technical object (Fr) in a
maintenance system is the quantity that determines the set
of fully regenerated functional elements of a technical
object in relation to the set of all those elements of the
object which have undergone a maintenance process. This
quantity was presented in the form of dependence (13):
Fr

nk
 100%
n

13

where n the set of all those elements of the object which

underwent the maintenance processa set of those elements that determine the objects maintenance structure
{ei,j}, k a set of the fully regenerated (with the fitness states
being recognized) primary elements of the object.

5 Results and discussion

The results obtained from the examination in the experiment were the subject of an analysis and a study with the
use of dependence (13). If the time is known of the diagnosis (recognition of the states) of the individual elements
of the object, then the function can be established of the
qualitative regeneration of the object in the maintenance
system (Fr = f(td(ei,j))), where td is the diagnosing time of
the states of a technical object for jth element in ith unit of
the object (ei,j). The quantities obtained are graphically
presented in Fig. 11.
It is evident from the analysis of the regeneration quality
in the maintenance system organized on the basis of the
testing maintenance information for (models I and II)
(Fig. 11) that the most reliable expert maintenance information was obtained for (model I) = 92 %. The quality of
the maintenance information is slightly worse for (model
I) = 92 %. The quality of the maintenance information for
(model II) = 85 % substantially departs from these two
models. Therefore, it can be stated that for the three
manners of expert inference examined: deductive, probabilistic and inductive, the first one of them can be of the
greatest practical significance.

Neural Comput & Applic (2013) 23:913925

923

Table 6 Table of control states of the object (model I)

State of the object
0

Number of the assembly

State of assembly

Vector of states of elements in the object {e(ei,j)}

E1

E2

E3

E4

E5

E6

E7

Fig. 9 The final results

obtained of DIAG diagnostic
programmeTABLE OF
STATES OF THE OBJECT
(model I)

Table 7 Table of control states of the object (model II)

State of the object
1

Number of the assembly

State of assembly

Vector of states of elements in the object {e(ei,j)}

E1

E2

E3

E4

E5

E6

E7

Many factors have an impact on the quality of the preventive procedures performed. These include the following: the reliability of the maintenance (inference) rules
developed, the quality of the experts work, the reliability
of the accepted (proposed) preventive activities and their
assigning to the particular maintenance elements of the
object, the quality of the work of the measuring automated
device of diagnostic signals, the permissible and boundary
values for the diagnostic signals: diagnostic inference rules,
the preparation of the maintenance personnel, the means of
maintenance being at ones disposal, etc.
For the purpose of the operation of such technical
objects that are characterised by a short time of their repair,

various solutions are sought in the optimization process of

their qualitative operation function. The authors practice
related to the operation of radiolocation systems proves it
that automatic adjustment and control systems for the
operation process of this class of technical objects may be
particularly useful. Such an adjustment system includes a
reliable diagnostic subsystem that makes use of an artificial
neural network (a comparing system) [27] and a subsystem of operation (a system to adjust the operation functions
of an object). The use of trivalent diagnostic information
from an artificial neural network constitutes the basis for
designing (organizing) the structure of the objects operation system.

123

924

Neural Comput & Applic (2013) 23:913925

Fig. 10 The final results

obtained of DIAG diagnostic
programmeTABLE OF
STATES OF THE OBJECT
(model II)

6 Conclusions

Fig. 11 Diagram of the qualitative function of the regeneration of the

elements of a technical object (Fr) by the maintenance system. Where
1 the diagram for the vector of the testing maintenance information
(model II); 2 the diagram for the vector of the testing maintenance
information (model I); td(ei,j) the regeneration time of the jth primary
element in the ith functional units of the object

In the organization process of every servicing system,

the costs constitute the factor that determines this process.
The results obtained in the research indicate that an optimisation of the costs of the prophylaxis of technical objects
can be achieved in two ways. The first one of them indicates that only those elements of an object should be renovated which require that these are elements in the state of
an incomplete operation and in the non-operation state
(therefore, there is a problem of the designing of the
structure of the servicing system). According to the second
manner of decreasing the costs of regeneration, the objects
elements need to be regenerated in a selective manner by
performing not all of the regeneration activities but only
prophylactic activities as selected by an expert (an expert
knowledge base), which are suitable for the state in which a
given element is of an object of a given class of the servicing element.

123

This article presents a description of maintenance systems

with a neural network and without a network. The set of
information determined by an artificial neural network
constitutes the basis of the description proposed. On the
basis of this information, the set of the objects maintenance information is determined (Table 5). The methodology of the determination of this information set was
presented in [2, 4, 611]. The maintenance structure of the
object that has been set on the basis of this information
consists of a set of those elements of the object which need
to be regenerated, a set of preventative activities that
regenerate the elements of the object, and a set of maintenance rules.
The idea of this prevention of technical objects, which
already has practical applications, yet is continuously being
developed, is as follows. The technical object is normally
used: it performs its required function (its tasks). During its
use, it is subject to constant diagnosing: examination of the
state by for example a neural network or in any other
manner. When a diagnostic system or a neural network
recognizes in the object a state of an incomplete usability,
that is state-{1} or a non-operation state-{0}, this means
that the time has been determined when the preventionregeneration of the object is to commence.
The paper also presents a process of the development of
the structure of a maintenance system (in tabular (matrix)
graphical and analytical forms). The process of the determination of a maintenance system without a neural network consists of two stages, where the system of unplanned
and planned maintenance is set. The presentation of the
methodology to set the structure of a maintenance system
constitutes an important element of this paper.
The article covers a description of those quantities that
characterize the maintenance systems presented, including

Neural Comput & Applic (2013) 23:913925

the following: an average time of the regeneration of an

object on a given maintenance level and in the whole
system, an average time of the location and removal of
unfitness in an object, and an average time of the regeneration of an object within the confines of unplanned
servicing.

References
1. Buchannan B, Shortliffe E (1985) Rule-based expert systems.
Addison-Wesley Publishing Company, Boston p 387
2. Duer S (2009) Artificial neural network-based technique for
operation process control of a technical object. Def Sci J 59(3):
305313
3. Duer S, Duer R (2010) Diagnostic system with an artificial neural
network which determines a diagnostic information for the servicing of a reparable technical object. Neural Comput Appl
19(5):755766
4. Duer S (2010) Investigation of the operation process of a
repairable technical object in an expert servicing system with an
artificial neural network. Neural Comput Appl 19(5):767774
5. Duer S (2010) Diagnostic system with an artificial neural network
in diagnostics of an analogue technical object. Neural Comput
Appl 19(1):5560
6. Duer S (2010) Qualitative evaluation of the regeneration process
of a technical object in a maintenance system with an artificial
neural network. Neural Comput Appl. doi:10.1007/s00521-0100418-0
7. Duer S (2010) Expert knowledge base to support maintenance of
a radar system. Def Sci J 60(5):531540
8. Duer S (2011) Qualitative evaluation of the regeneration process
of a technical object in a maintenance system with an artificial
neural network. Neural Comput Appl 20(5):741752. http://www.
springer.com/home?SGWID=0-0-1003-0-0&aqId=1320967&
checkval=ad10dea9b0005e4b1523cfd0443fbf7d
9. Duer S (2011) Assessment of the quality of decisions worked out
by an artificial neural network which diagnoses a technical object.
Neural Comput Appl. doi:10.1007/s00521-011-0725-0
10. Duer S (2011) Modelling of the operation process of repairable
technical objects with the use information from an artificial
neural network. Expert Syst Appl 38:58675878. http://dx.doi.
org/10.1016/j.eswa.2010.11.036

925
11. Duer S (2011) Examination of the reliability of a technical object
after its regeneration in a maintenance system with an artificial
neural network. Neural Comput Appl. doi:10.1007/s00521-0110723-2
12. Duer S (2011) Applications of an artificial intelligence for servicing of a technical object. Neural Comput Appl. doi:10.1007/
s00521-011-0788-y
13. Duer S (2012) Artificial neural network in the control process of
objects states basis for organization of a servicing system of a
technical objects. Neural Comput Appl 21(1):153160
14. Dhillon BS (2006) Applied reliability and quality, fundamentals,
methods and procedures. Springer, London, p 186
15. Hayer-Roth F, Waterman D, Lenat D (1983) Building expert
systems. Addison-Wesley Publishing Company, Boston, p 321
16. Hojjat A, Hung S-L (1995) Machine learning, neural networks,
genetic algorithms and fuzzy systems. Wiley, Hoboken, p 398
17. Gupta MM, Jin L, Homma N (2003) Static and dynamic neural
networks, from fundamentals to advanced theory. Wiley, Hoboken, p 718
18. Mathirajan M, Chandru V, Sivakumar AI (2007) Heuristic
algorithms for scheduling heat-treatment furnaces of steel casting
industries. Sadahana 32(5):111119
19. Nakagawa T (2005) Maintenance theory of reliability. Springer,
London, p 264
20. Nakagawa T, Ito K (2000) Optimal inspection policies for a
storage system with degradation at periodic tests. Math Comput
Model 31:191195
21. Palkova Z, Okenka I (2007) Programovanie. Slovak University of
Agriculture in Nitra, p 203
22. Palkova Z (2010) Modeling the optimal capacity of an irrigation
system using queuing theory, 55th edn. Warsaw University of
Life Sciences Press, Warszawa, pp 511
23. Palkova Z, Rodny T (2010) Modeling of object-related database
structures in the development of artificial irrigation system. In
Technika v technologiach agrosektora 2010 [elektronicky zdroj]:
zbornk vedeckych prac, Nitra, Nov 2010. Slovak University of
Agriculture in Nitra http://www.slpk.sk/eldo/2011/zborniky/0411/palkova-rodny.pdf
24. Tang L, Liu J, Rong A, Yang Z (2002) Modeling and genetic algorithm solution for the slab stack shuffling problem when implementing steel rolling schedules. Int J Prod Res 40(7):272276
25. Wiliams JM, Zipser D (1989) A learning algorithm for continually running fully recurrent neural networks. Neural Comput
1:270280
26. Zurada IM (1992) Introduction to artificial neural systems. West
Publishing Company, St. Paul, p 324

123