Está en la página 1de 12

Neural Comput & Applic (2010) 19:755–766

DOI 10.1007/s00521-009-0333-4

ORIGINAL ARTICLE

Diagnostic system with an artificial neural network


which determines a diagnostic information for the servicing
of a reparable technical object
Stanisław Duer • Radosław Duer

Received: 24 March 2009 / Accepted: 19 December 2009 / Published online: 10 January 2010
 Springer-Verlag London Limited 2010

Abstract In this paper, the features of a diagnostic 1 Introduction


measurement setup have been determined. The purpose of
the setup was registration of diagnostic signals taken from The technical state of the object in a given time of the use
an electronically controlled gasoline-powered engine. determines the possibilities of the realization of its required
Special attention has been paid to the presentation of the functions. It is determined by a subset of its physical
features of an analog–digital converter card used, as well as properties [1–7], which describe a given object. For the
to the possibility of its practical use. Another important purpose of a description of the quality of the use of an
aspect is the design and programming of computer soft- object, the quality which reflects best the functional prop-
ware dedicated for action along with the setup. The results erties is the functional quality function (FC(t)) (Fig. 1) and
of the work have been presented using the example of the the coefficient of the quality of the use of the object (FC),
engine’s microprocessor control module. A diagnostic which is calculated for the boundary value of the function
analysis was conducted, as a result of which sets of the (FC(t)). The index of the attribute of the functional quality
functional elements of the object and its diagnostic signals function can therefore constitute a current assessment of
were determined. Also, the methodology of the diagnostic the state of the object and owing to it can be recognized as
examination of the technical system was presented. The its measure. For practical reasons, to the states of the object
result was a functional and diagnostic model, which con- in the diagnosing process, numerical values are assigned,
stituted the basis for initial diagnostic information, which is which depend from the logics of the classification of the
provided by the sets of information concerning the ele- states applied. For divalent logics, these are states from set
ments of the basic modules and their output signals. {1, 0}, where ‘‘1’’ is the operational state and ‘‘0’’ is the
non-operational state. For the trivalent assessment of the
Keywords Diagnostic systems  Technical diagnostics  classification of states [2, 8, 9], to the states of the object,
Neural networks  Knowledge bases states marked with the values from set {2, 1, 0} were
assigned, where ‘‘2’’—the state of full operation; ‘‘1’’—the
state of incomplete usability; and ‘‘0’’—the state of non-
operation (defect).
Repairable technical objects for which a short time of
their shutdown is required (radar systems, airplanes etc.)
S. Duer are frequently equipped with specialist adjustment systems
Departament of Mechanics, Technical University of Koszalin,
which reconstruct their functional functions to the nominal
15-17 Raclawicka St., 75-620 Koszalin, Poland
e-mail: stanislaw.duer@tu.koszalin.pl level. An adjustment system of the object’s functionality
functions is a sophisticated system of the regeneration of
R. Duer (&) the object, which includes the following subsystems:
Departament of Electronics and Informatics,
diagnosing and maintenance. The purpose of the diagnostic
Technical University of Koszalin, 2 Sniadeckich St.,
75-543 Koszalin, Poland system is current and constant recognition (monitoring) of
e-mail: radoslaw.duer@02.pl the state of the object.

123
756 Neural Comput & Applic (2010) 19:755–766

Ei 2 The structure of diagnostic system for the diagnosis


ME of a technical object

Fc Fc(xi,j) The modern diagnostic system [2, 8–15] (Fig. 3) utilizes


1
not only a measurement A/D converter card with appro-
X(w)i,j
Δxi,j ω priate signal interfaces but also some computer tool used
0,75
for proper signal registration as well as for acquiring and
ω (ei,j) processing data registered. The purpose of such a process is
5
0,5
Ei to build a diagnostic knowledge base based upon an anal-
4 Xi,j
3
ysis of both the object and the results of the measurement
0,25 2
stored.
1 The measurement system (Fig. 3) was implemented and
ej
used within the diagnostic module that recognizes the
0 1 2 3 4 5 6 7 8 9 10 states of the object. The system is composed of the fol-
lowing elements:
Fig. 1 Distribution of changes of object’s states during operating
time (example), where x—the plane of actual usability features of 1. A measurement structure of the investigated object
the object; ME—the plane of the nominal usability features of the
received from the functional analysis. As an effect of
object; Fc(Xi,j)—the value of usage function; Xi,j—the vector of
actual diagnostic signal; X(w)i,j—the standard vector of diagnostic such an analysis, it is possible to establish the set of the
signal; and DXi,j—the vector of differential metric of diagnostic object’s elements (modules) {ei,j} with an adequate
signal output signal along with the set of diagnostic signals
necessary to measure {X(ei,j)}.
2. A signal track assuring that the levels of the signals
The maintenance subsystem regenerates an object in the measured match the range of an A/D converter inputs.
states of shutdown through a reconstruction of its func- 3. An A/D converter to measure and acquire the values of
tionality properties to the nominal level [1–3, 6, 7]. An diagnostic signals {X(ei,j)},
adjustment system presented in this manner can perform its 4. A PC application (software) to control the operation of
function if such a diagnostic system has been developed an A/D converter and to register measurement data.
which will recognize the object’s states in the values of The task of the PC application is also to prepare the
trivalent logics {2, 1, 0}. A diagram of the abovementioned expert knowlegde base {W(e(ei,j))} in the form of a
control process of the operation process by the system of matrix (Table 2).
adjustment of the object’s functionality function is pre-
sented in Fig. 2.
A functional and diagnostic analysis constitutes the 3 Functional and diagnostic analysis of a technical
basis for the designing of every maintenance system for object
any technical object. The result is information obtained
about the object, including usability, diagnostic, functional, The technical object used for tests in the present study is a
maintenance-specialist and any other data. reparable complex technical object of an analog class.

Fig. 2 Diagram of operation Fc min


process for technical object influance/interaction
utilizing diagnostic module, maintenance of the object
where X(ei,j)—diagnostic signal
in jth element of ith set; (ei,j) Creation of the {ME}
The object Repairing of
X(w)(ei,j)—model signal for The object expert knowledge
the object
Fc max
to repairing the states base of the object
X(ei,j) signal; and FC—function {1,0}
of the use of the object Fc max
{X(ei,j)} {X(w)(ei,j)}

Diagnostic
module of {W( ε(ei,j))}
the object
operation/
non-operation
of the object
the states {2}

123
Neural Comput & Applic (2010) 19:755–766 757

functional element in the object. It was assumed in the


Measurement
of knowledge paper that such an element is understood as a basic element
base {x(ei,j)}
in the object where there is an output (diagnostic) signal on
its output.
Diagnostic
Measurement knowledge If object has been divided into i structural levels, and in
Object {ei,j} Computer
{X(ei,j)} module base W( ε(ei,j)) each of them, there are j basic elements, then each of the
of the object
object’s structural levels constitutes a set of operating
Diagnostic elements {ei,j}, which was presented in the form of the
{X(w)(ei,j)}
programme following dependence:
"DIAG"
    
Structure of diagnostic system fOg ) fEi g ) ej ¼ ei;j ð1Þ

Fig. 3 Scheme of diagnostic system with measurement module where {O}—object’s internal structure, )—relation of
result (division), Ei—ith functional assembly of the object,
ej—jth subassembly in ith assembly of the object, {ei,j}—
While preparing a diagnostic model of this class of an set of basic elements in the object (structure of the object).
object, its internal structure was divided into four levels of
the maintenance structure (Fig. 4): level one: object, level
two: assemblies (in object), level three: subassemblies (in 4 The structure of measurement system
each assembly {Ei}), level four: modules-basic elements for the diagnosis of a reparable technical object
(in each subassembly, of each assembly of the object). The
first level of the maintenance structure of the object is Registration and visualization of the signals measured were
constituted by the object itself. It is a set of functional realized using a NI M-series USB-6221 A/D converter card
assemblies {Ei}. The assemblies of the object constitute the (Fig. 5). It allows performing up to 16 measurements at
second level of the object’s maintenance structure, while one time. The USB-6221 device (NI M-series family)
each of them is a set of operation subassemblies. Subas- contains 16 differential analog inputs (16-bit, 250 kS/s),
semblies in assemblies constitute the third level of the two analog outputs (16-bit, 833 kS/s), 24 digital I/O with a
object’s maintenance structure. The lowest level, i.e., the 1-MHz timer plus two 32-bit fast counters. M-series
fourth level of the structure, is constituted by the basic devices are equipped with an NI-STC 2 module, an NI-
elements: modules. PGIA 2 amplifier, and NI-MCal calibration technology that
As the basic element, modules, the smallest distin- assures a higher number of highly efficient I/O channels
guished (as a result of the division) functional element in plus a better resolution and a higher speed [8].
the object on the output where there occurs at least one The USB connectivity feature makes these devices
diagnostic signal, are defined. If there are a larger number appropriate for a broad spectrum of applications, especially
of signals on the output of a given element, then one automatic testing, process management or sensing and
generalized signal is determined. There is a rule in the measurements. NI signal streaming technology allows
present study that to each element of an object, only one generation and/or registration of different signals on both
diagnostic signal is assigned. the analog and digital I/O channels simultaneously. A
Each functional subassembly of the object consists of configuration of a signal track was universal. It consisted of
basic elements, which are the smallest and indivisible an A/D converter, analog multiplexers operating in

Fig. 4 Functional and


diagnostic model of the object, e1 x1,1
x1,2
where Ei—i jth functional
assembly in the object, ei—j jth Y1
e1,1 x2,1 e1,2
E1
subassembly or functional
element in a given assembly, e1
Functional and e1
Y1,2,3—input signals in the x2,1 x2,2 X4,1 Diagnostic
diagnostic
object, Xi,j—output signals in analysis of the Y2
e2,1 e2,2 e4,1 knowledge base
the object E2 {X(ei,j)}
object E4

e1 x2,1
x3,1 x3,2

Y3
e3,1 e3,2
E3
Structure of the object

123
758 Neural Comput & Applic (2010) 19:755–766

Fig. 5 Scheme of measurement


module of the object
e1 x1,1
x1,2
Y1
e1,1 x2,1 e1,2
E1

e1
e1
x2,1 x2,2
Computer
Y2
e2,1 e2,2 e4,1 X4,1 programme with
E2
E4 measuring
e1
x3,1
x2,1
x3,2
card
Y3
e3,1 e3,2
E3
Structure of the object

{X(ei,j)}

Structure of measurement system of the object

Measurement of knowledge base of the object


Vector of initial diagnostic signals {X(e i,j)}
X(e1,1) ... X(ei,j) ... X(ei,J)
X(e1,1) ... X(e1,j) ... X(e1,J)
... ...

...
...

...
X(ei,1) ... X(ei,j) ... ∅
... ...

...
...

...
X(eI,1) ... X(eI,j) ... X(eI,J)

different modes according to the characteristic of the cur- Once the whole package of data is acquired (and pro-
rently registered signal. A high performance amplifier was cessed if necessary), it is immediately passed to the des-
placed between A/D converter and multiplexer. The ranges tination VI through the output terminals (control and
of each analog input voltage were set independently for indicator terminals) and program nodes.
each channel. It was also necessary to use special wires: The measurement program consists of two applications.
they were shielded and dedicated for operating with high The first one (Fig. 7) performs data acquisition using NI
frequency signals. DAQ Assistant and writing to a measurement file as the
Data registration control with NI 6221 M was realized in second reads data from a measurement file and writes data
the LabVIEW environment. In contrast with code pro- into a spreadsheet file in MS Excel format (Fig. 7). It also
gramming languages, the G-language used within the derives an amplitude spectrum of registered signal and
LabVIEW environment is a graphical language. Program- writes it into a separate spreadsheet file.
ming is being realized via icons that represent functions The configuration and calibration of the A/D converter
and/or procedures. All of them are translated into a is performed with its driver DAQmx using the DAQ
machine code during the program execution. Assistant (Fig. 7). The range of the input analog voltage is
Each program is treated as a virtual device (VI—virtual settled for each input channel separately by the software.
instrument) due to the fact that it simulates the operation of Thus, the DAQ Assistant is an interface for configuration
a physical device such as oscilloscopes and multimeters. arguments and acquired data between the A/D converter
The virtual instrument consists of three elements: the front (hardware) and the LabVIEW application.
panel, the block diagram, and the icon with the connectors Figure 7 presents the DAQ Assistant window. It is used
pane. The front panel serves as the user interface. The for setting the A/D converter configuration: any parameter
block diagram contains the graphical source code that of a signal track such as the number of samples, the range
defines the functionality of the VI. The icon and the con- of the measured Signac amplitude (for NI M6221 up to
nector pane identify the interface to the VI so that you can ±10 V). The number of samples was fixed to 50 k. Such a
use the VI in another VI. A VI within another VI is called a number seems to be a reasonable maximum number con-
subVI. A subVI corresponds to a subroutine in text-based cerning the sampling frequency and the computer resources
programming languages [8]. limitation.
The measurement program (Fig. 6) realizes the control The writing element (Fig. 8) is used for writing mea-
of the data flow. Every time the A/D converter driver (NI sured samples into LabVIEW measurement file (.lvm). For
DAQmx) icon is invoked with the complete set of input the configuration of such an element, it is necessary to set
arguments, it triggers the data acquisition task. the filename and the path. Figure 8 depicts also

123
Neural Comput & Applic (2010) 19:755–766 759

Fig. 6 Application for data


acquisition using NI USB-6221
A/D converter card

Fig. 7 DAQ assistant window

‘‘Waveform Graph’’ elements: they are responsible for the determination of the object’s state. The object’s state is
data presentation in the reconfigurable graph. determined on the basis of an examination of the set of
The division of the object’s internal structure {ei,j} output (diagnostic) signals {X(ei,j)} (Table 1) [2, 4, 5, 8,
accepted in the paper defines explicitly the depth of 9, 14, 16–19].
penetration into this structure. The accepted division is The set of its functional elements {ei,j} determined
considered to be sufficient if we distinguish the basic during a diagnostic study of the object constitutes the basis
module-element in the structure of the object. One of the for the list included in the table of a set of diagnostic
purposes of the functional-diagnostic analysis is the signals (Table 1).

123
760 Neural Comput & Applic (2010) 19:755–766

Fig. 8 Application for data


processing and writing into MS
Excel file

Table 1 Table of object’s input diagnostic signals signal {Xi} to all vectors of weights wi;j ¼
½w1 ; w2 ; . . .; wn T ; where i = 1,…, N. In the ANN network
Object Level of Vector of initial diagnostic signals {X(ei,j)}
presented in Fig. 8, the neuron (i) placed in layer F1 is
object Ei
X(e1,1) … X(ei,j) … X(ei,J) connected to neuron (j) placed in layer F2, where
j = 1,2,…, N.
O E1 X(e1,1) … X(e1,j) … X(e1,J)
Neuron (i) sends the signal of value (xi) with the con-
: : … : … :
necting strength (wi,j) of the activation function. Following
Ei X(ei,1) … X(ei,j) … X(ei,J)
the literature of the subject [8, 14, 18–21], the Minkowski’s
: : … : … :
measure is used for the analysis of the measures of signal
EI X(eI,1) … X(eI,j) … X(eI,J)
vectors.
where X(ei,j)—diagnostic signal of jth element in ith assembly The Minkowski’s measure can be expressed by the
following relation (2):
!1=a
5 The structure of diagnostic program DIAG using XN
a
an artificial neural network (ANN) for the diagnosis DM ðXi ; XðwÞi ; aÞ ¼ jXi  XðwÞi j ð2Þ
i¼1
of a technical object
where DM—the standard deviation of the signal measure
The ANN network developed is presented in Fig. 9. It vector.
consists of three layers: F1—the input layer, F2—the output In the comparative analysis of diagnostic signals, a
layer, and the intermediate layer. The input cells of layer F1 special case of Minkowski’s measure was applied, with
process the initial diagnostic information according to the parameter a = 2. Then, the relation (2) becomes the
algorithm of the DIAG program presented in Fig. 10. The Euclidean measure and can be used with the ANN net-
whole of the issue of information processing by ANN work [14, 17, 20]. For this reason, in the process of
neurons (Fig. 9) [11] takes place in D-dimension diag- input data processing, a transformation of input data is
nostic space (x) determined by the elementary signal performed to reduce too high initial disproportions
vectors (Xi). The input signal in the form of Xi ¼ between the initial values in particular dimensions. One
½x1 ; x2 ; . . .; xn T is being passed to all neurons of the ANN’s of certain input data transformation methods, which is at
input layer. the same time quite effective, is the normalization of
The input cells memorize the vectors of signal standards input data which transforms the values into range of [0,
{Xi}. Based upon that, the neurons from the input layer 1].
determine the measures of similarity between the input For the ANN presented in Fig. 9, neuron (i) is connected
signal vector and its standard, and the length of the input with neuron (j), so it transmits a signal of value (Xi) with

123
Neural Comput & Applic (2010) 19:755–766 761

Fig. 9 Diagram of neural w11 n


networks
x1
w12
D
y1 = ∑ υ1i xi
i =1
w1n

w21 n
w22 y 2 = ∑ υ 2 i xi
x2 D i =1
w2n
The diagnostic
DIAG knowledge base of
Diagnostic module programme an object
{W( ε(ei,j))}
of the object
wj1

wj2 n
D y j = ∑ υ ji xi
wjn i =1

wN1
wN2 n

xn D
y n = ∑ υ Ni xi
wNn i =1

Fig. 10 The algorithm of Measurment of knowledge base of the object


Diagnostic knowledge base of the object
diagnostic DIAG program Vector of initial diagnostic signals {X(e i,j)} Vector of states of elementary components {ei,j}
X(e1,1) ... X(ei,j) ... X(ei,J) DIAG programme ε(e1,1) ... ε(ei,j) ... ε(ei,J)
X(e1,1) ... X(e1,j) ... X(e1,J) with ANN module W(ε(e1,1)) ... W(ε(e1,j)) ... W(ε(e1,J))
... ...
...
...
...

K
yl = f ∑ υ i , j ⋅ X i ... ...

...
...
...
X(ei,1) ... X(ei,j) ... ∅
i =1 W(ε(ei,1)) ... W(ε(ei,j)) ... ∅
... ...
...

...
...

... ...

...
...
...
X(eI,1) ... X(eI,j) ... X(eI,J)
W(ε(eI,1)) ... W(ε(eI,j)) ... W(ε(eI,J))

weight coefficient (wi,j) and the activation function, rep- On the final stage of the work of a neural network, a
resented by the relation: classification process of the object’s states is realized
X
K according to the algorithm (Fig. 10). For this purpose, to
fl ðx; wÞ ¼ wi;j  Xi ð3Þ the values of the output function as determined by the
i¼1 network, proper classes of the object’s states [2] were
The value of its output function is s derived from the assigned according to the classification diagram (Fig. 11).
relation (4) The results of the object’s diagnosis are presented in
Table 2.
X
K
yl ¼ f ti;j  Xi ð4Þ
i¼1
7 Research results of the diagnostic system
where ti,j—weight coefficient. for the diagnosis of a car engine

The method for the diagnostic knowledge base determi-


6 The diagnostic rules for the classification nation presented will be verified on the example of a rep-
of the object’s states arable technical object, which is an analog controller unit
for a combustion automotive engine with its peripheries
The determination of the value of the network’s output (Fig. 12).
function {yl} made it possible to explicitly determine the Research setup was developed on the basis of a spark
set of elementary vectors of signals which describe the ignition engine with multi-point injection. The object was
diagnostic space of the object. In view of the fact that the subject to a diagnostic development, as a result of which a
purpose of a diagnosis of the object [10–15, 18–21] is the functional-diagnostic diagram was developed. In the
recognition of its state in the values of the accepted logic of example, an object was used whose internal structure
states’ assessment, the results obtained in the form of (Fig. 13) is composed of seven modules (E1, E2,…, E7)
relation (4) were subject to the classification process (Table 3), and in each one of them, up to five elements
according to the diagram presented in Fig. 11. were distinguished [1, 2, 8].

123
762 Neural Comput & Applic (2010) 19:755–766

Fig. 11 Range of variability of


diagnostic signal features, States of the object
where (y1l , y2l )—the range of yl
non-significant changes of the "0"
20
outputs; (y1’ 1 2
l , yl ) and (yl , yl ))—
the range of significant changes
0
yl2'
of the outputs, ((-?, y1l ) and
20
(yl , ??))—the range of
"1" Diagnostic knowledge base of the object
inadmissible changes of the
outputs Vector of states of elementary components {ei,j}
yl2 ε(e1,1) ... ε(ei,j) ... ε(ei,J)
W(ε(e1,1)) ... W(ε(e1,j)) ... W(ε(e1,J))
ANN module
... ...

...
...

...
"2" W(ε(ei,1)) ... W(ε(ei,j)) ... ∅
yl nom ... ...

...
...

...
W(ε(eI,1)) ... W(ε(eI,j)) ... W(ε(eI,J))
yl1

"1"

yl1'

"0"

Table 2 Table of object’s states


State of object State of module Vector of states of elementary components {ei,j}
e(e1,1) … e(ei,j) … e(ei,J)

W(e(O)) W(e(E1)) W(e(e1,1)) … W(e(e1,j)) … W(e(e1,J))


: : … : … :
W(e(Ei)) W(e(ei,1)) … W(e(ei,j)) … [
: : … : … :
W(e(EI)) W(e(eI,1)) … W(e(eI,j)) … W(e(eI,J))
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

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 object was subject to a diagnostic development, as a
result of which the following was developed: a functional-
diagnostic diagram, on the basis of which a set of main-
tenance elements was determined (Table 3).
This makes data processing application universal and
very convenient. Similarly, the processing measurements
results in MS Excel are highly effective. Table 4 presents
examples of registered data representing the values of the
Fig. 12 Photograph of an electronic controller for a car engine registered signal amplitude. Each package of data contains
50,000 samples.
The use of the DIAG software requires a preparation
The presented method of diagnosing technical objects of input diagnostic information on the basis of a func-
requires the use of a uniform compliance of the designation tional and diagnostic analysis of a given object. A
of the elements of the object’s structure. For this reason, functional and diagnostic model of an object needs to be
the basic elements, modules of the object included in its made.

123
Neural Comput & Applic (2010) 19:755–766 763

X(e1,1) X(e1,2) X(e1,3) X(e1,4) X(e4,2)


e1,1 e1,2 e1,3 e1,4 e1,5 e4,2
ECU

E1 from engine X(e4,1)


e4,1

X(e1,5)
ECU
E4
ECU X(e2,1) X(e2,2)
e2,1 e2,2
E2
X(e5,1)
Engine e5,1 e5,2 e5,3

ECU X(e3,1) X(e3,2) ECU ECU E5


e3,1 e3,2

X(e3,3) X(e6,1) ECU


e3,3 ECU e6,1 e6,2
E3 e7,1 E6
E7

Fig. 13 Diagram of an electronic controller for an automotive starting circuit: e4,1—combustion knocking sensor e4,2—coolant
engine, where E1—ignition module: e1,1—automotive alternator, temperature sensor; E5—power supply circuit: e5,1—oxygen sensor
e1,2—voltage regulator, e1,3—battery, e1,4—coil ignition, e1,5— (1), e5,2—catalyser, e5,3—oxygen sensor (2); E6—engine block:
sparking plug; E2—fueling module: e2,1—fuel tank ventilation valve, e6,1—crank shaft position sensor, e6,2—EGR valve: E7—e7,1 elec-
e2,2—fuel injector; E3—air-feeding module: e3,1—air flow meter, tronic control unit
e3,2—throttle position sensor, e3,3—idle run position controller; E4—

Table 3 Elements of diagnostic structure of the object Table 5 Matrix of measures of diagnostic signals from the object
Level of object Ei Structure of the object {ei,j} Level of object Ei Vector of initial diagnostic signals {X(ei,j)} [V]
Ei e1 e2 e3 e4 e5 Ei e1 e2 e3 e4 e5

E1 e1,1 e1,2 e1,3 e1,4 e1,5 E1 12.12 12.11 11.93 47.95 47.98
E2 e2,1 e2,2 [ [ [ E2 14.01 48.09 [ [ [
E3 e3,1 e3,2 e3,3 [ [ E3 2.99 3.03 12.02 [ [
E4 e4,1 e4,2 [ [ [ E4 12.04 1.04 [ [ [
E5 e5,1 e5,2 e5,3 [ [ E5 3.10 2.01 3.10 [ [
E6 e6,1 e6,2 [ [ [ E6 4.04 12.09 [ [ [
E7 e7,1 [ [ [ [ E7 5.09 [ [ [ [

Table 4 Measurement matrix of diagnostic signals


No. Time [s] Amplitude of signal [V]
High voltage coil Connecting rod sensor Idle gear mechanism Fuel injector

1 0 -0.003542 -1.055235 0.024681 -1.143795


2 0.00002 -0.001595 -0.765549 0.02533 -1.142822
3 0.00004 -0.000622 -0.461914 0.025005 -1.143471
4 0.00006 -0.002893 -0.141086 0.024681 -1.143471
5 0.00008 -0.003866 0.200828 0.025005 -1.143471
6 0.0001 -0.005488 0.564152 0.02533 -1.143471
7 0.00012 -0.003542 0.949535 0.02533 -1.143147
8 0.00014 -0.003866 1.439373 0.025005 4.547419
9 0.00016 -0.003866 1.966193 0.025654 3.893109
10 0.00018 -0.00419 2.400884 0.024681 3.397431

123
764 Neural Comput & Applic (2010) 19:755–766

Fig. 14 Matrix of measures of


diagnostic signals from the
object

The results of measurements for choosen elements of the The final results obtained with the DIAG program were
object are presented in Table 5 and Fig. 14. presented in the form of a table of states (Fig. 15).
The elements from this set were grouped into subsets of The article presents methods applied for the assistance
classes using the classification method proposed in the of the prevention of the devices which control the work of a
paper. The state of the object was determined on the basis car engine. This object was chosen considering costs.
of the measurements of the diagnostic signal features However, this is a device with a complex internal structure
processed. They were processed and analyzed by an arti- as it includes assemblies and subassemblies of various
ficial neural network. classes, including mechanical and electric classes, which is

Fig. 15 Matrix of diagnostic


information of the object: the
result form of DIAG program
‘‘Table of object’s states’’

123
Neural Comput & Applic (2010) 19:755–766 765

significant for the verification of the methods for objects of 9 Conclusions


different classes. This object was particularly useful at the
initial stage of the development of the method. On this The signal track is a highly important part of the diag-
device, too, initial tests were carried out during the nostics system. Development of methods and equipment
development of the method, designing of the measuring for automotive measurements, usually computer aided and
track, records and analysis of the measuring results, the controlled, allows faster data acquisition and diagnostic
structure of the diagnostic system, and at the stage of signals analysis with a higher accuracy and precision. Due
the construction of the knowledge base which assists the to that, the diagnostic and servicing of technical objects can
regeneration of this object. One may wonder whether this be done in quasi-real-time.
object (as proposed in the initial version of the article) has This article presents the application of a high-class
been well chosen for this method. measurement module with a high-speed and high-precision
A/D converter card as its heart to the diagnostics of an
analog class object. For this purpose, the technical object
8 Results and discussion was subjected to a diagnostic study. An important stage of
the work is a functional and diagnostic analysis of the
The basis of the presented method is the determination object. Knowing the diagnostic structure of the object, it
diagnostic information concerning the recognized states of was possible to point the most important diagnostic signals,
the object. Diagnostic information is developed in a diag- and to measure their values.
nostic system; in this method, an artificial neural network
was used for the purpose of diagnosing the object. For the
needs of the diagnosing process, the technical object in
question was additionally supplemented with an important References
aspect for the method presented, which is a description of
the functional and diagnostic analysis of the object of the 1. Be˛dkowski L, Da˛browski T (2006) Podstawy eksploatacji cz. 2.
research. In the article, this problem is presented with the Wyd. WAT, Warszawa, p 187 (in polish)
2. Duer S (2004) The concept of assistant system for analogue class
description of the model of the technical object. Also, a
technical object servicing. In: Sixth international conference on
method of the division of the object’s internal structure was unconventional elektromechanical and electrical system
presented. As a result of this division, a set of the basic UEES0 04. Alushta, The Crimea, Ukraine, pp 687–690
elements and a set of diagnostic signals were determined. 3. Dhillon BS (2006) Applied reliability and quality, fundamentals,
methods and procedures. Springer—Verlag London Limited, p 186
The issues presented in the article of the building of a set
4. Gercbach JB, Kordoński CB (1972) Modele niezawodności obi-
of diagnostic information concerns various fields of ektów technicznych. WNT, Warszawa (in polish)
knowledge, including technical diagnostics, the theory of 5. Gniedenko BW, Bielajew JK, Sołowiew AD (1968) Metody
operation, information technology, expert systems, fuzzy matematyczne w teorii niezawodności. WNT, Warszawa (in
polish)
sets, and artificial neural networks. Each of these fields is
6. Nakagawa T (2005) Maintenance theory of reliability. Springer—
well and broadly studied in the literature. It is the author’s Verlag London Limited, p 264
opinion that one can claim with full responsibility that even 7. Nakagawa T, Ito K (2000) Optimal inspection policies for a
the basic problem, i.e., the use of diagnostic information storage system with degradation at periodic tests. Math Comput
Model 31:191–195
obtained in the diagnosing process of a technical object in
8. Duer S, Duer R, Duer P, Płocha I (2009) Measurement system for
the designing and organization of the operation process, is the diagnosis of analogue technical objects with the use of arti-
being constantly developed in various aspects (directions). ficial neural networks. Academic Journals, Poznan University of
At present, the direction of the applications of neural net- Technology, s. Electrical Engineering, Published by Poznan
University of Technology, No. (59), pp 61–72
works, among others in the diagnostics of technical objects,
9. Spath H (1980) Cluster analysis for data reduction and classifi-
is being intensively developed. However, new solutions cation of object. Ellis Horwood Ltd., Chichester, UK
and possibilities are constantly being sought; hence, the 10. Buchannan B, Shortliffe E (1985) Rule—based expert systems.
author’s papers and studies are presented concerning a Wesley Publishing Company, Addison, p 387
11. Hayer-Roth F, Waterman D, Lenat D (1983) Building expert
practical application of a trivalent evaluation (classifica-
systems. Wesley Publishing Company, Addison, p 321
tion) of the object’s states [2, 8, 9]. However, there is no 12. Jackson P (1990) Introduction to expert systems. Wesley Pub-
full description in the literature of methods to develop ways lishing Company, Addison, p 526
and algorithms for the processing of diagnostic information 13. Mathirajan M, Chandru V, Sivakumar AI (2007) Heuristic
algorithms for scheduling heat-treatment furnaces of steel casting
obtained by diagnostic systems: an artificial neural network
industries. Sadahana 32(5):111–119
etc. to the form of an expert knowledge base of a main- 14. Sinha M, Gupta MM, Zadeh LA (1999) Soft-computing and
tenance system, presented in a computer programming intelligent control systems, theory and applications. Academic
language. Press, New York, p 321

123
766 Neural Comput & Applic (2010) 19:755–766

15. Tang L, Liu J, Rong A, Yang Z (2002) Modeling and genetic 18. Wiliams JM, Zipser D (1989) A learning algorithm for continually
algorithm solution foe the sjab stack shuffing problem when running fully recurent neural networks. Neural Comput 1:270–280
implementing steel rolling schedules. Int J Prod Res 40(7):272– 19. Zurada IM (1992) Introduction to artificial neural systems. West
276 Publishing Company, St. Paul, p 263
16. Hojjat A, Shih-Lin Hung (1995) Machine learning, neural net- 20. Białko M (2000) Podstawowe właściwości sieci neuronowych i
works, genetic algorithms and fuzzy systems. John Wiley End hybrydowych systemów ekspertowych. Wyd. Politechniki
Sons, Inc., p 398 Koszalińskiej, Koszalin (in polish)
17. Gupta MM, Jin L, Noriyasu H (2003) Static and dynamic neural 21. Gupta MM (2001) Fuzzy tests, fuzzy logic and fuzzy systems, in
networks, from fundamentals to advanced theory. John Wiley encyclopedia of physical science and technology. Meyer RA (ed),
End Sons Inc., Hoboken, New Jersey, p 718 Academic Press, San Diego, p 452

123

También podría gustarte