The article presents the results of the development of a data mining information system for complex technical systems failure risk evaluation and modelling, which allows taking into account the probabilities of system elements failure with the analysis of the numerical values of damage to interelement connections. The proposed system implements a risk assessment method, as well as intelligent models for assessing and analyzing the risk of elements failures and components of complex technical systems of various topologies and hierarchies based on simulation modeling, soft computing theory and data mining models. For the holistic storage of the obtained and revealed knowledge from the data in the system, a distributed data warehouse module configuration has been developed, based on the use of trained decision tree models, which makes it possible to automate the processes of generating and analyzing brief metadata on the key parameters of the system elements in dynamics when assessing the risks of their failure. The use of the proposed method and models within the framework of the developed system makes it possible analysis process automating, assessing the performance and reliability of complex technical systems elements and interelement connections in heterogeneous emergency scenarios for various target risk management programs. The use of the developed data mining information system for complex technical systems failure risk evaluation has made it possible to increase the efficiency and effectiveness of managerial decision-making by 10% compared to the existing cognitivesimulation approach without intelligent models.
The efficiency of functioning of complex technical systems (CTS) largely depends on the
reliability of their interconnected, interacting elements, inter-element connections (IEC). Examples of
CTS are telecommunications, energy, transport systems, oil and gas production, mining and
processing industries. Ensuring reliability is possible by using methods, diagnostic tools and
predicting the technical state of systems at the stages of CTS design and operation [
To ensure high indicators of efficiency, reliability, environmental performance, it is required to use
methods of diagnostics and forecasting the technical state of CTS [
Taking into account the described specifics of the CTS, one should take into account the
heterogeneity and large amount of diagnostic data obtained for various subsystems and elements [
Analysis of publications showed the predominant use of deterministic, probabilistic, expert,
combined methods for assessing the failure risk of various types of CTS [
The use of information, analytical and intellectual capabilities in the systems of diagnostics and
forecasting with the determination of reliability level allows for high-quality design and maintenance
CTS equipment [
The analysis of publications [27,28] showed also that the development of information technologies
allows solving the problem of assessing the failure risk of CTS with a large elements number, with the
display of systems in the directed graphs form with structural and functional relationships and
interactions of elements [
Within the framework of the cognitive approach, the basis of the model is a cognitive map in the
form of a directed graph, which includes the elements of the CTS [
The developed theoretical base and the availability of a large range of simulation software with AnyLogic, Extend, Arena and other famous systems [20] contributes to the use of CS. Considering the existing uncertainty, incompleteness and vagueness of the information received by the systems for diagnostics and forecasting CTS technical state, it is advisable in the ISDF, in addition to cognitive model (CM), to use fuzzy-probabilistic models (Fig. 1) [20-23], which will allow the joint use of heterogeneous data for obtaining a reliable assessment failure risk.
The initial data processing module (Fig. 1) in DMIS will allow processing data for further clear cognitive and fuzzy modeling. The diagnostics module will provide the construction of clear cognitive and fuzzy models for diagnostics and predicting in the design and operation CTS failures risk.
There are various approaches to the formation of fuzzy probabilistic models for assessing the risk
of failures of functionally interconnected and interacting CTS. To construct such models, it is
recommended to use the probabilities of loss of performance as input parameters, as well as damage
from the consequences of a risk event in the event of failures of elements and IEC of the CTS [24].
The paper [25] describes the use of fuzzy-probabilistic models for assessing the risk of CTS failures,
which allow early identification of a possible transition of a particular process to an emergency mode
to prevent and risk unacceptable scales. In [26], a model for identifying pre-emergency situations of a
technological process based on a fuzzy logic apparatus with additions by elements is used as a similar
model, which allows taking into account the probability of CTS equipment failure. The use of fuzzy
logic for the construction of fuzzy-probabilistic models for assessing the risk of failure of
interconnected and interacting elements and CTS IEC in various operating conditions allows to
reduce the time spent on researching systems [
When assessing the risks of failure of CTS elements as a component of the analysis of the overall
level of its reliability [29], it is important to take into account that еhe effectiveness of DMIS largely
depends on the results of using the technologies of expert systems using knowledge bases [
DT visualization can be done by displaying a connected directed acyclic graph. The branches of the graph of the constructed decision tree DT can store the values of attributes, which are functional parameters of the elements of the studied CTS, on which its reliability depends, and the values of the failure risk are displayed on the leaves of DT [23,24]. Consisting in the fact that the structure of the DT partition can be expressed as a set of logical rules in production form, the peculiarities of constructing DT models makes it possible to use such models as the basis for KB. There are known works in which the following are proposed: an approach to constructing a fuzzy KB, which is the basis for constructing fuzzy inference systems for operational control; architectural, multi-level KB model for storing information in a subject-oriented intelligent system [26].
Thus, the further development of methods of cognitive-imitation and fuzzy-probabilistic modeling, the use of DT models as the basis for KB, is relevant when constructing the DMIS of the CTS risk failure of interrelated and interacting elements and EMI.
The method of diagnostics and forecasting involves the determination of an assessment of the risk of CTS failures based on heterogeneous information of different volume and content. The method is implemented in the DMIS scheme (Fig. 3), which includes cognitive and fuzzy models for assessing the failure risk of CTS, KB and DB elements. In the proposed DMIS, the knowledge base uses the results of fuzzy and cognitive modeling, as well as scenarios for the development of the risk of CTS failures arising during system designing and portioning. CM is based on Bayesian analysis and Bayesian networks. A Bayesian network is a directed graph. The nodes of the network are many random variables. The vertices are connected in pairs by oriented edges. The complex indicator of the reliability of CTS elements is the risk of failure ( R ) - a combination of the probabilities of failures ( P ) and losses from failures ( D ), Conseq, means consequence.
R Conseq . P Event D Conseq .
Time Time Time R { Pi(j) ,Di(j) ,q i(j) ,Pij ,Dij ,q ij }, i,j 1,2 ,...,N , (1) (2) where i – the number of the vertex from which the edge CM emerges; j - the number of the vertex that contains the edge CM; Pi(ij) - element failure risk probability (IEC) CTS; Di(ij) - damage from the consequences of the risk of failure of an element (IEC) CTS; qi(ij) - weight i( j) for each N element failure risk (IEC) CTS within 0 ... 1 under the conditions qi(j) 1(i, j 1, N ) and i1,j1 М qij 1(ij 1, М ) ; N - amount of elements CTS, M – amount of IEC. ij1
Cumulative estimates of the risk of failures, taking into account the relationships and interactions of elements and IEC CTS
N М N ,М R i1 Ri qi j1 R j q j i, j1 Ri, j qi, j . (3)
The model is based on the structural representation of the CTS in the form of a directed graph,
which is negatively influenced by external impulses. Each directed graph is reflected using an
incidence matrix, the columns of which are the edges of the directed graph CM CTS, and the rows are
the vertices. The construction of a model for assessing the structural and functional risk of failures by
elements and IEC for CTS is carried out in 3 stages. The first stage of CS is the development of a CM,
taking into account the type of the EMI resource in the form of a cognitive map and an oriented graph
with N vertices and M edges [
To assess the structural performance of the CTS and the associated damage, CM uses striking
simulation impulses (SSI). The proposed SSI propagation procedure is close to the approach
considered in [
Uncertainty, incompleteness and fuzziness of information in assessing the structural and functional risk of failure of interrelated and interacting components designed and operated by CTS requires the ISDF to use the apparatus of a fuzzy probabilistic model. The developed structure of the fuzzyprobabilistic model includes input linguistic variables: D - damage, P - probability of failure and the output value R - risk of failure. The Takagi-Sugeno model is used to assess the risk of CTS failures based on a fuzzy probabilistic model. This allows us to obtain the final fuzzy set for the output variable of the failure risk assessment with a membership function of the form μxi (f Ri,qi ) max [μxPi (f R ),μxDi (f R ),μxqi (f R )] ,
YR (4) where xPi ( fR) - fuzzy set of element failure probabilities CTS; xDi ( fR) - fuzzy set of damage from failures of CTS elements; xqi ( fR) - fuzzy set of weights of the failure risks of CTS elements.
The production rules were entered on the basis of the Harrington generalized desirability function
in accordance with the method of assessing the structural and functional risk of CTS failures. As a
membership function, the Gauss distribution function was used, implemented in Matlab in the form of
gaussmf to set smooth symmetric membership functions. At the fuzzification stage, the input
variables of the fuzzy-probabilistic model for assessing the failure risk of designed and operated CTS
are set in the form of the probabilities of failures and damages from failures of elements and EMI. To
describe the key ontological information about heterogeneous datasets of large volumes, as well as to
provide a more compact form of their presentation with subsequent processing and analysis, the
structure of partitioning decision trees can be expressed as a set of logical rules in production form.
This makes it possible to use such models as the basis for KB in ISDF. It is proposed to formalize the
structure of the DT model using the CART algorithm. The use of the algorithm is due to its focus on
building binary DTs, each individual node of which, in the process of logical partitioning, forms only
two descendants. CART is implemented by splitting a lot of data by gender at each step. On one
branch of the tree there are examples where the logical rule is executed successfully, and on the other
it is not successful [
The process of conducting experiments was carried out on the basis of several stages:
collection and extraction of data on the studied CTS structure, which is a transport engine
cooling system [
To study the DMIS of the risk of CTS failures in emergency scenarios, as an example, we studied a system consisting of 12 elements and 26 IEC (Fig. 4), 17 of which are energy carriers (indicated by solid lines), 9 are carriers of matter (indicated by dashed lines). To determine the damage to the edges and vertices of the directed graph CM DMIS of the failure risk CTS, we set the weight values Svi for vertices (Table 1) and Saj - for edges (Table 2). Values of the probabilities of failure of vertices and edges of the directed graph CM CTS pvi(t) and paj(t) are presented in tables 1,2.
The results of the impact of the SSI on the CM CTS were established by research in the developed software package based on the cross-platform Python language. The structure of the CTS in the form of a directed graph is represented using the dot language. The technical state of the CTS elements in CM is reflected in the JSON format. Graphviz is used to visualize graphs. Analysis of the simulation results was carried out in MS Office and Open Office.
Figure 4: Oriented graph of CM elements and EMI CTS
From the results of the research it follows that the place of application of the SSI has a significant impact on the process of distribution of the SSI along the directed graph CM CTS. Taking into account such structural features of a directed graph as connectivity, the presence of contours, the vulnerability of its vertices and the type of EMI resource, the most connected components of a directed graph, which are affected by a number of EMI of various resource types, have been identified.
The values of estimates of the structural risk of failures established by modeling for the vertices of the CM CTS and their ranking are shown in Fig. 5. In the course of research, it has been confirmed that if the place of application of the SSI does not belong to the structural component of strong connectivity, then the passage of the SSI is less associated with the risk of CTS failure. In the studied CTS, the components of strong connectivity are elements 3, 8 and IEC 2, 5, the performance of which determines the risk of failures of adjacent elements and IEC. If IEC 2 is damaged, risks arise during the operation of elements 1, 5, 7-12, and if element 3 is damaged, risks arise for all CTS elements.
To effectively store data for different types of system elements, an add-on was created for distributed data warehouse based on the use of Microsoft Azure Data Lake Storage Service, which will allow scaling the system's capabilities.
The developed CM reflects the direct dependence of the risk of failures of elements and IEC on their positions in the system structure, as well as the dependence of the entire CTS on the selected topology at the design stage. The advantage of the method is its simplicity, clarity and applicability for assessing the risk of failures of a wide class of CTS. The procedures of the method implementation are easily formalized and transformed into a computational algorithm and model for assessing the risk of failure, which is important for CTS with a large number of elements and IEC. Considering that the objects of technical condition diagnostics often consist of various CTS, for some of which the use of CM for assessing the risk of element failures and IEC is difficult due to the uncertainty, incompleteness and fuzziness of information received by the diagnostic systems and predicting their technical condition, it is advisable in DMIS additionally in addition to cognitive modeling, use fuzzy probabilistic models. An example of such a CTS can be a ship power plant (SPP), the diagnostic information from which is characterized by uncertainty, incompleteness and indistinctness [30]. As a fuzzy model, the Takagi-Sugeno model is used, the advantages of which include a more compact size of the rules repository compared to the Mamdani fuzzy inference models, as well as lower computational complexity due to a simpler defuzzification procedure [24,37]. Fuzzy rules have the structure “If damage or probability then risk”, where damage and probability are the linguistic value of the antecedent, determined by the fuzzy membership function, which is a consequent stochastic variable equal to one of the values in the range from 0 to 1.
The result of the fuzzification stage is the establishment of a correspondence between the specific value of a separate input variable of the fuzzy inference system and the value of the membership function of the corresponding term of the input linguistic variable.
The aggregation operation is based on the maximization method implemented in Matlab as max. The weight values for all input variables of the fuzzy-probabilistic model were taken equal to one. During accumulation, the max-disjunction method was used to combine all degrees of truth of the conclusions of the rules. The developed fuzzy-probabilistic model contains 25 logical rules. Defuzzification was carried out using the center of gravity method implemented in Matlab.
The damage to the elements and IEC of the SPP was set using the linguistic variables Not significant (0-0.25), Low (0.12-0.36), Medium (0.3-0.6), High (0.55-0.8), Critical (0.7-1). On the basis of the developed fuzzy-probabilistic model, it is possible to minimize the time of assessing the risk of failures of the designed and operated SPP. This is achieved due to the mechanism of fuzzy rules incorporated into the developed model for assessing the risk of failure of the SPP using the Matlab Rule Viewer module. Such a mechanism makes it possible to identify and quantitatively reflect the degree of influence of specific values of damage and the elements and IEC probabilities failure on the risk of failure of the SPP as a whole. In particular, with the value of damage equal to 0.234 and the value of the probability of failure equal to 0.5, the value of the output variable of the risk of failure was 0.157, i.e. about 16%, which testifies to the low degree of danger of the SPP efficiency loss.
The probability of SPP elements and IEC failure was set using the linguistic variables: Not significant (0-0.15), Low (0.1-0.31), Average (0.22-0.6), High (0.45-0.75), Critical (0.6-1). The terms Minimum risk (0-0.2), Acceptable risk (0.1-0.35), Significant risk (0, 3-0.65), Critical Risk (0.63-1). In the course of modeling, a three-dimensional visualization model’s surface for assessing the risk of failure of elements and IEC of the SPP was obtained (Fig. 6).
The results fuzzy-probabilistic model studying make it possible to establish that the probability of failure of elements and EMI has a more significant effect on the assessment of the risk of failure of the EMI.
According to the obtained three-dimensional visualization fuzzy-probabilistic assessment model’s surface the greatest increase in the level of risk of EEC is observed with values of the probability of failures in the range from 0.455 to 1. In order to use KB for possible scenarios of actions when making control decisions and forming a visual structure for a data set from CTS, the DT graph is used (Fig. 7).
The composition of the logical rules records for KB based on the interpretation of this model can be expressed as: IF StructRisk> 0.435 AND Average cost <= 1600 AND Maintainability = “High” THEN Damage is Middle. Additional clarification of this rule is possible by using the priorities of features or weight values (degree of significance) for each of them. An example of a selection of options for the values of input and output parameters obtained from the results of cognitive and fuzzyprobabilistic modeling is shown in table. 3.
Table 3 is supplemented with a column of logical inference obtained based on the use of KB for possible target alternatives (action scenarios) when making management decisions out of 4 possible options: prevention, repair, replacement, no action). The received recommendations can be assessed by a decision-maker to make changes to the KB structure both in manual and automatic mode, which provides feedback to the model and its adaptability Table 3 Selections of options for values of input and output parameters
No 1 2 3 4 5 6 7 8 9 10
A data mining system has been developed for diagnostics, predicting failure risk assessments of designed and operated interconnected and interacting elements, inter-element connections of complex technical systems, based on the application of the method and cognitive-imitation, fuzzy-probabilistic models of failure risk assessments.
The knowledge base of the proposed system is based on the methodology of cognitive and fuzzyprobabilistic modeling, the decision tree model, which makes it possible to take into account the scenarios of the development of the risk of failure of elements and inter-element connections of systems.
The use of the developed methods and models in the intellectual system makes it possible to assess the reliability of systems based on the results of the effects of the damaging effect of each element in emergency scenarios on the structure of the system, as well as to predict the consequences of failure of elements and inter-element connections of systems. The practical application of the developed system can be performed by automating the tasks of collecting and processing data from various CTS, especially in the transport sector.
The use of machine learning algorithms in the created system makes it possible to form linked and structured metadata sets based on risk assessments of failure of elements and interelement links in various scenarios.
Application of the developed method and models will make it possible to make effective management decisions to ensure the reliability of systems, both in the design and in the operation of complex technical systems.
[19] A. Zykov, Fundamentals of graph theory. University book, Moscow, 2014. [20] S. Shtovba, O. Pankevich, A. Nagorna, Analyzing the criteria for fuzzy classifier learning,
Automatic Control and Computer Sciences, 49 (2015) 123–132. [21] A. Rotshtein, Selection of Human Working Conditions Based on Fuzzy Perfection, Journal of
Computer and Systems Sciences International, 57 (2018) 927–937. [22] YU. Kudinov, Fuzzy input models in expert systems, Izv. Theory and Control Systems, 5 (2017) 75–83. [23] Zhedunov R.R. The system of identification of pre-emergency situations of the technological process, using the apparatus of fuzzy logic and data of probabilistic deviations, Bulletin of the Astrakhan State Technical University, 38 (2017) 169–173. [24] G. Luger, Artificial intelligence: structures and strategies for complex problem solving, Addison
Wesley Publishing Company, United States, 2019. [25] D.Tupikov, A. Rezchikov, An approach to building a fuzzy knowledge base for determining firehazardous situations, Mathematical methods in engineering and technologies, 62 (2014) 125-127. [26] V. Nechayev, M. Koshkarev, Architecture of a pre-oriented knowledge base of an intelligent system, Educational resources and technologies, 1 (2015) 156-163. [27] Yu. Antokhina, V. Balashov, E. Semenova, A. Varzhapetyan, Computer simulation of processes in technical systems, Journal of Physics: Conference Series. 1691 (2020) doi:10.1088/17426596/1691/1/012069 [28] V. Devyatkov, Methodology and technology of simulation studies of complex systems: current state and prospects of development, Vuzovsky textbook, Moscow, 2014. [29] I. Nazarenko, M. Delembovskyi, O. Dedov, A. Onyshchenko, I. Rogovskii, M. Nazarenko, I.
Zalisko, Study of reliability of technical systems reliability, Dynamic processes in technological technical systems 14 (2021) 110-139. doi: 10.15587/978-617-7319-49-7.ch7