Timely diagnostics of the technical state and technical support of complex multifaceted systems that provide comprehensive automation of control processes requires the development of new, high-precision and reliable troubleshooting methods. Using one of these methods may include a neural network apparatus.
The level of preparedness of the Armed Forces of Ukraine (AFU) to perform assignment directly depends on the availability of state-of-the-art weapons and military equipment, but neither weapons nor any equipment will be able to ensure effective performance of combat missions without quality command and weapons. Modern control and communication systems must not only have high combat readiness, bandwidth, stability, mobility, accessibility, controllability, but also ensure trouble-free operation of the system and guarantee the quality of information exchange.
In recent years, the AFU have been expanding and improving the system of
communication and the automatic control of forces. With the introduction of the widespread
use of modern digital radio stations of the Harris, Aselsan and Elbit companies, the
performance and functionality has increased, while the element base of electronic
equipment has become more complicated [
The requirement of reliable operation of electronic equipment in the system of
complex automation of control processes with the use of complex multifaceted systems,
communication and automation tools (CAT) is especially acute. Therefore, the
development of new, high precision and reliable faultfinding methods is becoming
increasingly urgent. Considering the complexity of modern CAT, there is a need to automate
the process of diagnostics for troubleshooting of equipment. One such method may be
the development and use of a neural network apparatus [
In connection with the use in the Armed Forces of modern CATs, the improvement of their diagnostics should carried out based on the use of scientifically grounded criteria, modern diagnostic tools and the necessary specialized software. During the Joint Forces Operation (JFO), the search for expedient ways of creating and improving a scientifically grounded, economically feasible system of diagnostics of the CAT is continued, aimed at ensuring that the accumulated knowledge, scientific achievements work primarily to ensure national security and defense of Ukraine.
The purpose of the article is to increase the efficiency of the diagnosis of CAT by using a basic method of the neural network tool to assess the reliability and ensure the guaranteed quality of operation of the means of this class. 3
A variety of research methods can used in the diagnostic system, which include theories
of graphs, decision support systems, fuzzy sets, neural networks, multi-criteria
optimization methods, expert methods [
According to the graph theory, it is possible to solve many problems in the field of
economics, technology and many other areas of human activity. An abstract oriented
graph is the basis for planning the actions of a human operator, changing the order of
the sequence of his actions, performing various operations. The setting of network
planning tasks begins with the identification of the nodal moments (events) of the operation,
which determine the beginning and end of its main stages, after which the measures
(work) are defined that lead to the implementation of each event and are accompanied
by the cost of time and resources. Then the events and the corresponding stages of work
are presented in the form of some logical sequence, namely a project graph that reflects
all the ways to achieve the goal of the operation. In the general case, the network graph
(Fig. 1) is a graph which nodes а x is events and oriented arcs Axy is the stages of work,
x y . Thus in the graph any arc reflects only one stage of work, and in a case of a
situation when two stages of work have the same output and input nodes (events) an
additional fake stage of work is added, not related to the time and other resources.
One possible sequence of events and workflows from the original event to the final
event is the path of the project graph, and the path that has the maximum execution time
of each of the work stages and the total execution time of the entire operation is a critical
path. Decision Support Systems (DSS) are special interactive information systems that
use hardware, software, data, model databases and manager work to support all stages
of decision making in the analytical modeling process. In other words, DSS is a set of
software tools that includes a set of different decision support algorithms, a model base,
a database, ancillaries, and a management program (Fig. 2). The leading program
provides the decision-making process based on the specificity of the problem [
The method of expert evaluation gives an objective description of the qualitative and quantitative aspects of the object of forecasting based on processing and analysis of a set of individual opinions of experts.
The quality of the experts’ review, its reliability and validity are crucially dependent on the developed methodology and processing of individual expert values, which includes the following steps: Selection of experts and assessment of their competence; Preparation of questionnaires for expert interviews; Obtaining expert opinions; Assessing the consistency of expert opinions; Evaluation of the reliability of the results.
The advantage of expert methods is their relative simplicity for predicting almost any situation, including in the context of incomplete information. An important feature of these methods is the ability to predict the qualitative characteristics of the process for the diagnosis of CAT. The disadvantages of expert methods include the subjectivity of the opinions of experts and the limitations of their judgment.
Multi-criteria optimization or programming is the process of simultaneously optimizing two or more conflicting target functions in each given areas. A multi-criteria problem is often understood not as a verbal description of the problem itself, but rather by its model, namely: a multi-criteria task is a mathematical model of making the optimal decision by several criteria. These criteria may reflect evaluations of the various qualities of the object or process of diagnosis of a protected area that is being decided.
Formal instrument for processing expert information is a mathematical apparatus of fuzzy sets, which allows you to formulate decision rules for the diagnosis of CAT. It is possible to formalize the relevant information using the linguistic variables "A", "B", "C", "D". The linguistic variables "A", "B", "C" are the input variables, the linguistic variable "D" is the output variable. Affiliation functions are built for each linguistic variable, a fuzzy knowledge base is formed based on production rules, which is necessitated by the need to account for the real time scale and convenience of presenting information on procedures and conditions of their execution. Therefore, the solution of Input signals
Output layer
the problem of determining the order of diagnostics of system nodes is based on the use
of the instrument of fuzzy sets theory [
Neural networks and fuzzy logic are universal approximations of complex (nonlinear) functional dependencies in many intellectual cybernetics problems.
The main feature of neural networks is their ability to learn, which is implemented using
specially designed algorithms. No prior information about the structure of the requested
functional dependency is required for training the neural network. A training sample in
the form of experimental input-output pairs is. The advantage of fuzzy logic is the
ability to use expert knowledge of the structure of an object in the form of linguistic
expressions: if "inputs", then "output". However, the apparatus of fuzzy logic does not
contain learning mechanisms, so the results of fuzzy inference depend heavily on the
kind of membership functions that formalize fuzzy terms. The result of combining
fuzzy logic with neural networks is a neural-fuzzy network has two major intellectual
properties: linguistics, that is, the use of natural language knowledge, and the ability to
learn in real time [
It is possible to speed up and simplify the diagnosis process by using the neural network apparatus, which have several important advantages: High performance of complex logic operations; The possibility of solving difficult formalized problems that share data of logically incompatible nature (incomplete, incorrect); Sustainability of work that is compatible with the expansion, transformation and improvement of knowledge; The reliability that is provided by the many paths to logical inference and the ability to recover lost data; Ability to build self-learning and self-configuring systems; It is well combined with the traditional "computing" information processing algorithms, which allows building complex control systems with maximum reliability, adaptability and minimum resources used.
To solve the problems logically, it is sufficient to use one of the following functions,
which determine the value of V excitation of a neuron, depending on the magnitude of
Vі and excitation of related neurons, weight ωi of these connections and threshold h [
V : ( іVіі h ) , (x) = x, if x 0 ; 0 , otherwise
V : ( іVіі h ) , V : if V A then A else V ;
V : іVіі , V : if V h then V else
0 ;
V : іVіі , V : if V h V 1 then V h else if V 1then 1 h else 0 ;
V : 1n іVіі ,
V : if V h then V else 0 ; where n is the amount of neuron’s active inputs.
Taking into account the advantages of the NN instruments, the task of diagnostics in TM of different types of CAT can reduced to the process of pattern recognition, where the image is the state of CAT as a whole or its individual components. In this case, the decision on the status of the CAT is made using the NN apparatus and it is possible to carry out both a quantitative assessment of the parameters and to predict its state for a certain period.
Depending on the formulation of the problem, the input signals can be the basic parameters (transmitter power, receiver sensitivity, etc.) and auxiliary (frequencies, currents, and voltages at different control points of individual blocks) depending on the problem statement. The output layer of the neural network characterizes the different states of the system. In the course of training NN, the values of individual parameters and the state of a particular CAT are generally memorized (image creation). The use of this image in subsequent TM is the criterion for evaluation of "serviceable - faulty". 5
As an example, a neural network (Fig. 4) is built to diagnose communication equipment (an example of constructing a network for a task). The method of support vector networks or tracing is used during the construction. As inputs (standards) working effects are used: those that get on the device during its intended use, and those that are its main output parameters (you can also use specially generated test effects). At the output layer, the expected response to the defined input standard is formed.
Take, for example, the neural network shown in Fig. 4. Suppose that all the same
neurons perform a single transfer function, and that the scales and thresholds realize
equal and general capabilities. We introduce the transfer function of an arbitrary j-th
neuron with the number of inputs m [
m 0 if х 0 V : (V jij hi ), (x) ,
j1 х if х 0 where Vj is the amount of excitation (of another neuron) that arrives at the j-th input; ωij is weight of connections; hi – thresholds.
For example, let us define that ωij = 0,8; hi = 0,2 and submit the situation to the input (standart) {А1, В1, С2} is "distortion of the signal of the 1st path within the normal range". According to the inputs of neurons 2, 3, 8, they receive single impulses. The energy data of impulses of the output neurons is calculated then considering the weighted connections between the neurons. Because of the calculation, we obtain that the maximum excitation value (1, 4) reached the output neuron 11 (Exit 11).
Based on the calculations for the construction of the network, we compose a Table 1, which reflects the correct operation of the network when making different decisions. Analyzing the terms of the table, which correspond to reliable situations, it can seen that the maximum excitation is determined correctly enough. This allows us to assert that the neural network, designed directly for the task, is built correctly.
In [
For realization of this complex, it is necessary to use the device of neural networks in carrying out diagnostics of other CATs with the use of a single package of applications. Thus, one workplace will provide diagnostics during TM and troubleshooting.
The analysis showed that the Armed Forces of Ukraine are undergoing a gradual reequipment of units for CATs, which allows providing a variety of information and telecommunication services. Timely diagnostics of the technical state and technical support of complex multifaceted systems that provide comprehensive automation of control processes requires the development of new, high-precision and reliable troubleshooting methods. Using one of these methods may include a neural network apparatus.
Technically, the process of diagnostics of various CATs can implemented by a single measuring and diagnostic complex based on a personal computer (PC). With a switching device that connects the outputs of the equipment to the PC input, the status recognition process takes seconds. The results of the construction of a neural network for solving the problem of diagnostics of modern CAT are given, which will allow reducing the time for conducting TM and the number of measuring instruments involved.
In the future, it is necessary to create a single package of applications for diagnostics of different types of communication devices based on neural networks, as well as a switching device by which the outputs of the equipment should connected with the input of the PC of the operator's workplace.