Chapter 52

Identification of Key Factors

for a Development of Smart Organization

Lyudmila V. Glukhova, Svetlana D. Syrotyuk, Anna A. Sherstobitova

and Svetlana A. Gudkova

Abstract In the process of intellectualization, informatization, and digitalization

of all types of activity, the problem situations started to form, the ones requiring
the rigorous mathematical justification of the obtained conclusions’ effectiveness.
Along with the search for answers to the question about how to evaluate the obtained
result, how far the effectiveness indicators are from the planned values, what the
analyzed activity risks and the reasons for their occurrence are, it remains vital to
search for answer’s: “What was the obstacle to achieving the preplanned value of a
controlled parameter?” or “How to estimate the value of the influence of the analyzed
structure’s internal environment parameters on the formed external indicator of its
activity?”. Answers to these and many other questions can be obtained in the process
of formalization. However, various deviations from the initial model assumptions
inevitably appear during the process of handling the applied problems; therefore,
the use of the standard mathematical statistics methods can lead to the obtained
conclusions’ distortion. In this regard, it becomes necessary to use new methods
to process information, the ones that are robust to the possible deviations of the
real data characteristics from the expected ones. The article shows a new approach
to the development of the statistical data analysis technologies and the obtained
conclusions quality evaluation methods in relation to the evaluation of a self-learning
organization’s activity.

Keywords Smart organization · Robust statistics methods

L. V. Glukhova
Volzhsky University Named After V.N. Tatischev, Toglyatti, Russia
S. D. Syrotyuk (B) · A. A. Sherstobitova · S. A. Gudkova
Toglyatti State University, Toglyatti, Russia
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2019 595

V. L. Uskov et al. (eds.), Smart Education and e-Learning 2019,
Smart Innovation, Systems and Technologies 144,
https://doi.org/10.1007/978-981-13-8260-4_52
596 L. V. Glukhova et al.

52.1 Introduction

Today, many authors pay attention to the smart organizations’ formation and devel-
opment. For example, works [1–5] contemplate the approaches to formation and
development of such concepts like a smart university, smart pedagogics, and smart
environment.
Despite a fairly large interest in the activities of smart organizations in a broad
aspect of application, an attempt to identify the cumulative influence of factors on
the resulting indicator of a smart-type organizational structure activity remains a
problem. Without doubt, among the analytical methods of studying this subject area
one of the promising methods is the method of factor analysis of the risks arising in
the process of an organization’s activities, when there is an attempt to reduce a risk
to the factors comprising it and to analyze the value of influence of every individual
factor on the resulting indicator [5].
Let us consider using the Taguchi methods for the evaluation of the noise factors’
influence on the formation of a specific quality characteristic typical for a smart
organization [1].
We see a smart organization as a self-learning organization where a team of
employees is ready for the rapid implementation of innovations into the society. A
smart organization concept and the quality indicators allowing to evaluate the smart
organization’s development level are shown in more details in article [2].
Suppose that the basic characteristic of quality in a self-learning organization
(a smart subdivision) is the complex indicator “the smart organization’s intellectual
potential development level.” Many indicators influence the formation of an eval-
uation of the achieved (high, low, or medium) development level. With the help of
the methods of structural analysis and synthesis it is possible to describe the exist-
ing cause-and-effect relationships within the organizational structure of the analyzed
indicators and to identify a greater or lesser degree of influence of each of them on
the formed complex indicator.
The choice of studying the possibilities of evaluating a smart organization devel-
opment level on the basis of robust design is based on the fact that this method
allows to adjust the self-learning technology and to regulate the flow of mastered
new information in order to increase the efficiency of self-learning [3].
A new approach in economics is that Taguchi roughly divided the factors influ-
encing product quality into two categories: uncontrolled factors called noises and
controlled factors called signals. It is possible to improve the activity quality via
reducing noises and amplifying signals. The mathematical theory of robust estima-
tors is interesting, because it is based on the already known approaches, but it has
additional properties allowing to complement and to improve the estimated results
significantly [1, 4, 5]. Taguchi proposes to divide the factors into groups in such a
manner that the factors responsible for the main response (“nominal”) would be in
one group and the factors responsible for variance would be in another group. To
identify these groups, Taguchi introduces a new generalized response that he calls
the “signal-to-noise ratio” (SNR). This ratio is the inverse value of the variation
52 Identification of Key Factors for a Development … 597

coefficient. Consequently, the initial problem is divided into two independent opti-
mization procedures: one for the nominal and the second one for the variance. And
since they are carried out in different spaces, their results are simply “glued together”,
thus giving the desired stable or robust mode [5].
Let us note that the analyzed quality characteristic (QC) should be continuous,
easily measurable and quantitative; positive and having absolute zero (absence of
negative values); additive, or at least monotonous (independence of components);
consisting of individual, smaller, and particular characteristics (providing all the
information); and fundamental (absence of displacement of various physical pro-
cesses and of reaction to external disturbances after optimization). Also, the QC
should have the maximum values of importance and priority. In our case, the studied
QC→max. Let us use the robust design methods in the context of evaluating a smart
organization’s intellectual potential development level.

52.2 Validation of the Noise Factors Selection

The problem is the insufficient instrumental base that allows to judge about the quality
level of the intellectual potential formation. It is necessary to clarify the reason for
the insufficient formation of this indicator. Figure 52.1 shows the Ishikawa diagram
that allows to make conclusions about the influence of four factors on the evaluated
indicator formation: the external and internal environments, the personal factor, and
the financial component.
This diagram was built before the start of the experiment. It reflects the cause-and-
effect relationships of the individual indicators’ influence on the achievement of the
result. It is necessary to determine the quality characteristic (QC) through which the
external consumer will evaluate the smart organization’s activity. In our case, it will
be the evaluation of the employees’ development level in a smart organization. Let
us select the noise factors (NFs) that hinder the achievement of the required level of
the intellectual potential development. There is an important condition: the estimated
noise factors do not depend on the smart organization employees’ activity. All these
factors are external, having a negative effect on the formation of the required quality
characteristic. Selection of the NFs is one of the main stages of planning. The system
robustness is achieved only for the selected NFs.
The noise experiment has the following objectives: (1) to determine the most
important noise factors having the maximum influence on variability; (2) to carry
out the comparative analysis (benchmarking) of the initial production characteristics;
(3) to improve the experimental procedure to minimize the NFs influence through
parametric optimization; and (4) to check the uncontrolled noises values, i.e., to
measure an error.
The noise experiment does not imply modeling the dependence of the QC on
NFs, but allows to find the significant noise factors and directions of their influence.
The factors are united into combinations along the directions of the NFs influence
and are called combined noise factors (CNFs). There are CNFs↑ (connected with
598 L. V. Glukhova et al.

Internal
environment Finances

Stimulating bonus to the salary

Payment for for publication activities
Innovative ideas the teaching
availability Intellectual
infrastructure staff’s work Communication means and infrastructure
availability upgrade, knowledge intellectualization
Constant in-
Employees’ house training
relationships Increase of publication activities
Workplace design in foreign journals

Motivation
Intellectual potential Availability of a group system
of like-minded peop le Evaluation of the smart
availability organization’s intellectual
potential development
level
Smart organization ’s competitive Sociability
advantages Teaching staff’s
competitive
advantages
Stress resistance
Loyalty
Competence
Learning
quality
Publication activities
Mobility Proactivity

External environment – Personal traits

Fig. 52.1 “Reasons result” diagram for our experiment

the increase of QC values) and CNFs↓. The noise experiment is carried out with the
ruling factors (RFs) set at the nominal values; also, the orthogonal matrices (OM)
are applied. Usually, the OM does not exceed the L12 size, so that there are no more
than 11 NFs, since a large number of NFs lead to a large number of errors [1, 4].
In our experiment, we give an example of working with a matrix of dimension
L8.
Usually, the NFs are considered at two levels, with NFs interdependencies having
equal distribution. This distribution makes it possible to evaluate the NFs main effect,
without worrying about its influence on the adjacent matrix columns. The noise
experiment analysis is reduced to finding the average response of each column,
the direction of the QC change, and the QC variance value. Let us take a closer
look at the algorithm described above. We must indicate that we will consider the
structural division of a smart organization on the example of the structural subdivision
“Department”. The experiment was conducted to assess the quality of the activities
of the staff of a smart university unit. The activities of the staff of the department
“Organization Management” of Togliatti State University were studied. The results
obtained allow us to obtain a management tool to reduce the risk situations that arise
with each noise factor.
52 Identification of Key Factors for a Development … 599

52.3 Planning the Noise Experiment

The following factors are the most significant ones for the entire smart organization:
• factor 1 (NF1)—absence of the skills of writing scientific works in a foreign
language;
• factor 2 (NF2)—insufficiency of the performance indicators during the knowledge
transfer process; and
• factor 3 (NF3)—insufficient level of identification in the external environment.
We need to take into account that each of the factors has its own condition that
may be evaluated through an expert method, so we will identify this condition and
summarize the identified indicators in one table (Table 52.1).

52.3.1 Identification of Noise Factors

The noise factor itself, its condition, and a conventional name for each of the condi-
tions are described in Table 52.1.
To calculate the number of realizations and to carry out the experiment, we need
to know the following parameters:
• a—the number of noise factors;
• b—the number of ruling factors;
• m—the number of levels; and
• n—the number of degrees of freedom (m−1).
The number of realizations, in this case, is calculated according to formula (52.1)

K = m ∗ a + n ∗ b + 1. (52.1)

The data in Table 52.1 shows that there will be three noise factors, each of them
having two levels of the identified condition (m = 2).
Let b = 1 (the ruling factor). In this case, the number of degrees of freedom is
equal to 1. Therefore, we deduce that K = 2*3 + 1*1 + 1 = 8.

52.3.2 Combination of Noise Factors

Combination of noise factors is determined by an orthogonal matrix.

 600

Table 52.1 Noise factors

General QC NF1 NF1 condition NF2 NF2 condition NF3 NF3 condition
A smart Absence of the A—insufficient Insufficiency of the C—absence of a Insufficient level of E—Effectiveness
organization’s skills of writing knowledge of a performance personnel’s identification in the of interaction with
intellectual scientific works in foreign language indicators during readiness for external competitors and
potential a foreign language B—absence of the knowledge knowledge transfer environment professional
development level knowledge about transfer process D—absence of interest-based
how to build a motivation for communities
scientific knowledge K—a base of
publication’s exchange indicators of the
structure external
environment
interaction
effectiveness is not
formed
L. V. Glukhova et al.
52 Identification of Key Factors for a Development … 601

52.3.3 Conduction of the Experiment

The experiment is conducted twice for each realization (replica 1 and replica 2).

52.3.4 Calculation of Average Values

The average value of the QC for each realization is calculated according to formula
(52.2), and a sum of squared deviations (Se2 ) is calculated according to formula (52.3)


n
yi
i=1
ȳi = (52.2)
n

n
Se2 = (yi − ȳ)2 . (52.3)
i=1

52.3.5 Calculation of Average Values for All Experiments

The average value of the QC is calculated according to formula (52.4), and the
average deviation for all experiments is calculated according to formula (52.5)


n
ȳi
i=1
ȳ = ; (52.4)
n

n
Se2
i=1
S̄e2 = . (52.5)
n
The outcomes of calculations are presented in Table 52.2.
Table 52.2 presents (1) the results of conducting the experiment (replica 1, replica
2), (2) the average value for all experiments, and (3) the sum of squares of the obtained
deviations from the average value. The average values for each level of a noise factor
are presented in Table 52.3.
Graphic representation of the experiment results. Figure 52.2 presents the
indicators of the segment’s beginning and end, denoting them as a1 and a2 for NF1;
as b1 and b2 for NF2; and as c1 and c2 for NF3.
602 L. V. Glukhova et al.

Table 52.2 The outcomes of noise experiment

Realizations NF1 NF2 NF3 Replica 1 Replica 2 Average Sum of
value y i squared
deviations
Se2
1 A C E 0.7 0.6 0.65 0.001359766
2 A C K 0.5 0.8 0.65 0.001359766
3 A D E 0.65 0.55 0.6 0.000172266
4 A D K 0.6 0.75 0.675 0.003828516
5 B C E 0.6 0.82 0.71 0.009384766
6 B C K 0.5 0.74 0.62 4.72656E-
05
7 B D E 0.6 0.35 0.475 0.019078516
8 B D K 0.5 0.55 0.525 0.007766016
 y= 2
Se =
0.613125 0.0053746

Table 52.3 Average values Noise factor Factor level A smart organization’s
in noise experiment intellectual potential
development level
NF1 A =(0.7 + 0.5 + 0.65 + 0.6 + 0.6
+ 0.8 + 0.55 + 0.75)/8 = 0.64
B =(0.6 + 0.5 + 0.6 + 0.5 + 0.82
+ 0.74 + 0.35 + 0.55)/8 = 0.58
NF2 C =(0.7 + 0.5 + 0.6 + 0.8 + 0.6
+ 0.5 + 0.82 + 0.74)/8 = 0.66
D =(0.65 + 0.6 + 0.55 + 0.75 +
0.6 + 0.5 + 0.35 + 0.55)/8 =
0.59
NF3 E =(0.7 + 0.65 + 0.6 + 0.6 + 0.6
+ 0.55 + 0.82 + 0.35)/8 = 0.60
K =(0.5 + 0.6 + 0.5 + 0.5 + 0.8
+ 0.75 + 0.74 + 0.55)/8 = 0.62

52.3.6 Generalization of Experiments’ Outcomes

Let us generalize the factors and verify the results of the noise experiment.
52 Identification of Key Factors for a Development … 603

0,7
0,69
0,68
0,67
0,66
0,65
0,64
0,63
0,62
0,61
0,60
0,59
0,58
0,57
0,56
0,55
0,54
0,53
0,52
0,51
0,50

а1 а2 b1 b2 c1 c2

Fig. 52.2 Results of the experiment

52.3.6.1 NF Level’s Reference

Each NF level has a reference to one or two categories—either for increasing the QC
value (CNFs↑) or for lowering the QC value (CNFs↓), specifically:
• CNFs↑ (A2, B2, C1),
• CFs↓(A1, B1, C2).

52.3.6.2 Prediction Values of QC

The predicted values of QC for CNFs↑ and CNFs↓ can be calculated as follows:
     
Y ↑= Ȳ + Yi↑ − Ȳ + Yi↑ − Ȳ + Yi↑ − Ȳ
     
Y ↓= Ȳ + Yi↓ − Ȳ + Yi↓ − Ȳ + Yi↓ − Ȳ
Y ↑= 0.613 + (0.64 − 0.613) + (0.66 − 0.613) + (0.62 − 0.613) = 0.694
Y ↓= 0.613 + (0.58 − 0.613) + (0.59 − 0.613) + (0.60 − 0.613) = 0.544

52.3.6.3 Verification of QC Values

Verification is the comparison of the predicted (planned) value of QC versus the

actual value of QC obtained from experiments. In this case, the selected factors are
604 L. V. Glukhova et al.

Table 52.4 Verification of CNFs Prediction Point

the noise experiment
CNFs1 0.694 0.697
CNFs2 0.544 0.547

the noise factors if the actual experimental values differ from the predicted ones with
no more than the value of S2e (Table 52.4).
The verification of the obtained experimental data and the noise experiment show
that the analyzed factors are the noise ones and their influence should be minimized.

52.4 Selection of the Ruling Factors and Planning

the Experiment

Let us describe the process of selection of the ruling factors (RFs). The ruling factors
are the parameters ensuring the system’s additivity and reliability. They are selected
depending on which factors depending on a human could influence the QC. We
identify the following ruling factors as follows:
• RF1: constant in-house training;
• RF2: increase of publication activities in foreign journals; and
• RF3: stimulating bonus to the salary for publication activities.
The conditions for those factors are presented in Table 52.1 as well.
The main experiment is carried out according to the same form; however, in this
case, the signal-to-noise ratio is calculated additionally according to formula (52.6)
 
 1 1
S N j = −10 · lg . (52.6)
n i yi2

The average value of QC is calculated according to formula (52.4). The average

value of the S/N indicator for all the experiments is calculated as presented in formula
(52.7):


n 
S Nj
 i=1
S N= . (52.7)
n
The maximum value of S/N—S/Noptimal —is determined as presented in
Table 52.5.
At the next step of the algorithm, we will analyze the main experiment data and
then we will verify it. Let us calculate (1) average value of QC, and (2) the S/N
ratio for each level of a ruling factor. The range of values is calculated according to
formula (52.8).
52 Identification of Key Factors for a Development … 605

Table 52.5 The main experiment

Realizations RF1 RF2 RF3 Replica 1 Replica 2 Average S/N
value of Y
1 Y R S 0.82 0.6 0.71 −2.97
2 Y R T 0.64 0.79 0.715 −2.91
3 Y P S 0.55 0.75 0.65 −3.74
4 Y P T 0.6 0.85 0.725 −2.79
5 N R S 0.63 0.72 0.675 −3.41
6 N R T 0.56 0.79 0.675 −3.41
7 N P S 0.68 0.55 0.615 −4.22
8 N P T 0.56 0.69 0.625 −4.08
 4.74 −27.5
y= S/N =
0.67375 −3.4

Table 52.6 The outcomes of experiment: average values of S/N

Factor Characteristics S/N S/N y y
RF1 Y −3.10 0.3386 0.7000 0.0263
N −3.7832 0.6475
RF2 R −3.1791 0.2654 0.6938 0.0200
P −3.7100 0.6538
RF3 S −3.5882 0.1437 0.6625 0.1437
T −3.3009 0.6850

S/N = S/Ni max − S/Ni min . (52.8)

The outcomes of the experiment are presented in Table 52.6.

The optimal value of the QC is calculated according to formula (52.9).

S/Noptimal = S/N + (S/NY ↑ − S/N ) + (S/N R↑ − S/N ) + (S/N T ↑ − S/N ),

(52.9)
S/Noptimal = (−3.4446) + ((−3.1059) − (−3.4446)) + ((−3.179)
− (−3.4446)) + ((−3.009) − (−3.4446))
= −2.7047.
S/N predicted = −2.7932.

The experiment is considered to be true if the value of S/Noptimal is close to

S/Npredicted.
606 L. V. Glukhova et al.

Table 52.7 The main Realizations CNFs1 CNFs2 S/N

experiment verification
1 0.82 0.6 −2.97
2 0.64 0.79 −2.91
3 0.6 0.85 −2.79

A series of experiments (five realizations) is carried out with optimal ruling factors
with CNFs↑ and CNFs↓. In this case, the average value of the S/N indicator in all the
experiments should differ from the optimal S/N with no more than by 5% (Table 52.7).
In terms of an objective function, we should strive to max. Therefore, the results
of the first or the third realizations were selected.

52.5 Conclusions and Further Application Perspectives

1. The research outcomes clearly show the “signal-to-noise” ratio should be used
as an objective function or output statistics in the design of smart organization.
2. The set of parameters at which this ratio takes the greatest value is considered
to be the optimal one. In this case, smart organization’s existing activity and its
working processes become insensitive to noise.
3. The quality of a self-learning organization’s activity increases without the
increase of the costs for its functioning.
4. The robust design of the personnel’s knowledge transformation process in a
smart organization is an approach allowing to ensure the transferred knowledge
stability against external influencing factors.
5. The knowledge stability is seen as a possibility of preserving the knowledge
structure, i.e., the structural stability.

References

1. Taguchi, G., Yokohava, Y., Wu., Y.: Quality Engineering Series, vol. 2. Taguchi Methods, On-
Line Production. ASI Press (1994)
2. Uskov, V.L., Bakken, J.P., Howlett, R.J., Jain, L.C. (eds.): Smart Universities: Concepts, Systems,
and Technologies, p. 421. Springer, Cham (2018). ISBN 978-3-319-59453-8. https://doi.org/10.
1007/978-3-319-59454-5
3. Serdyukova, N.A., Serdyukov, V.I., Uskov, A.V., Slepov, V.A, Heinemann, C.: Algebraic For-
malization of Sustainability Ranking Systems for Evaluating University Activities: Theory and
Practice, SEEL2017, Smart Education and Smart e – Learning, Smart Innovation, Systems and
Technologies, vol. 75, pp. 459– 474. Springer, Cham (2017)
52 Identification of Key Factors for a Development … 607

4. Charls, R.: Hicks Fundamental Concepts in the Design of Experiments, 5th edn. Oxford Uni-
versity Press, Oxford (1999)
5. Glukhova, L.V., Syrotyuk, S.D.: Application of the full factorial experiment for validating the
processes efficiency indicators. Sci. Vector Tolyatti State Univ. Ser. Pedagog. Psychol. 4(19),
160–164 (2014)