PNRPU Bulletin. Electrotechnics, Informational Technologies, Control Systems

Creating projects using deep learning neural networks is often associated with solving a number of technical problems: choosing sensors to obtain high-quality input data at a moderate cost, optimizing the use of hardware platform computing resources (especially when it comes to Internet of Things devices - IoT), preparing high-quality training datasets and realizing their capabilities. The purpose of the study is to formulate principles for solving these problems and to develop a method for improving the quality of recognition for cases when basic solutions have reached the maximum of their effectiveness. In this article, this will be demonstrated on the example of the Sortomat project - a machine for receiving processed packaging from the population from various goods. The research methodology is based on the use of recreated results during a series of experiments and mathematical modeling. The results are a modification of the machine, which, in terms of practical significance, allows, while maintaining the total cost of the technical project, to improve a number of Sortomat characteristics. Cameras are selected as sensors in it, allowing to receive a video signal. This is much cheaper than using more complex devices for analyzing an object, but it allows achieving, in most cases, a similar quality of recognition. The Adam optimizer and loop learning are used to optimize the use of computational resources. This made it possible to reduce the weight of the file describing the weight by half without losing the quality of recognition (while maintaining the level of the number of errors of the first and second order). When creating datasets, an algorithm for searching for a region of interest in an image and a majority object recognition system are used. The region of interest can significantly reduce the amount of "useless" information at the input of systems. This improves the overall quality of the sample (only the desired objects are present). The majority system allows making decisions in situations when an object is similar to two or more different classes of objects. Using the method of weighting factors, it determines the reliability of the results for each of the datasets and makes a general verdict on the belonging of the object to the class.

Relevance and goals: Speech/pause segmentation is one of the most important tasks in speech applications, being detection of the boundaries of the beginning and the end of voiced and unvoiced speech, and pauses. Segmentation accuracy is especially important when analyzing noisy speech, since the performance of almost all speech applications depends on the level of ambient noise. The aim of this work is to improve the efficiency of speech/pause segmentation of noisy speech signals based on the method of empirical mode decomposition. Materials and methods: A unique technology for adaptive decomposition of non-stationary signals, namely, the improved complete ensemble empirical mode decomposition with adaptive noise, has been used in the work. The software implementation of the method was performed in © MATLAB (MathWorks) mathematical modeling environment. Results: A noise-robust method for speech/pause segmentation based on the empirical mode decomposition at the preprocessing stage, and subsequent analysis of zero-crossing rate and short-term speech energy, has been developed. A characteristic feature of the developed method is the formation of a set of new investigated signals containing the most reliable information about the boundaries of the beginning and the end of noisy speech informative sections. Conclusions: Research to assess the influence of the decomposition method, and the duration of the studied signal fragments on the efficiency of noisy speech segmentation has been done. In accordance with the research results, there is a decrease in the 1st and 2nd kind errors by an average of 2.3% and 2.6%, respectively, at different signal-to-noise ratios (from 20 to -5 dB with a step of 5 dB) of noisy speech. Noise-robust method for speech/pause segmentation can be tested under real conditions of “aggressive” noise environment in intelligent assessment systems of a human psycho-emotional state.

In the process of managing technical systems, decision-making problems often arise. To increase the efficiency and reliability, decision-making procedures should be automated and based on objective knowledge about the scope of the system's functioning. Decision-making procedures in such systems are usually multi-criteria, which means that many factors must be taken into account for making a decision. The introduction of such a parameter as the importance of criteria allows the usage of universal decision-making methods in various systems, taking into account the criteria on which this decision depends in different ways. Numerical determination of the importance of criteria for automatic decision-making in complex technical systems is currently an urgent problem. This is due to the development of monitoring and diagnostic systems as a means of increasing the operational reliability of control systems and their elements. If decisions are made incorrectly in diagnostic systems, the operability of the system being diagnosed is jeopardized. The goal of the study is to develop a method for numerically determining the importance of criteria for decision-making in multicriteria problems. Research methods and results: a comparative analysis of the existing methods for the numerical determination of the importance of criteria is carried out, their shortcomings are revealed, the urgency of the problem being solved is indicated. A technique has been developed for the numerical determination of the importance of criteria in relation to the considered subject area - technical diagnostics. On the example of a real problem of distributing the diagnostic load between autonomous elements of the built-in test diagnostics system, a set of criteria and characteristics was determined for which a numerical calculation was made. In the conclusion, the results of the application of the developed technique are analyzed, directions for further research are described.

Processes for the preparation of homogeneous mixtures of powder materials are used in many industries. Purpose of the study: The article discusses the use of electromagnetic rotators for mixing powder materials. In such electromagnetic devices, ferromagnetic active elements act as working bodies. Under the action of a rotating magnetic field, ferromagnetic elements are distributed throughout the entire volume of the working chamber and rotate around the axis, entraining the particles of the dispersed materials being mixed. The advantages of electromagnetic rotators are small size, high reliability and easy control over the mixing process. A review of literary sources has shown that, in addition to mixing powder and fine-grained materials, electromagnetic rotators can provide a number of other technological operations: grinding and dispersing bulk materials, cleaning the surface of products, abrasive processing of the surface of workpieces and products, surface hardening, mixing slurries and suspensions, fuel homogenization and others. Expansion of the spheres of application of such devices makes their modeling and research relevant for the development of their theory. Research methods and results: The paper proposes a mathematical model of an electromagnetic rotator. On the basis of solving the equations of the electromagnetic field in cylindrical coordinates, expressions are obtained for the electromagnetic forces acting in a magnetic field on elementary ferromagnetic particles. The analysis of such expressions makes it possible to explain the behavior of the working bodies in the apparatus under consideration and to obtain recommendations on the choice of the apparatus parameters. In particular, to improve the quality of powder mixing, it is recommended to use a two-pole inductor. In such an inductor, the distribution of working bodies throughout the entire volume of the working chamber is ensured. In the case of multi-pole inductors, the influence of magnetic attraction forces increases, and there are no ferromagnetic elements in the center of the working chamber. In order to check theoretical conclusions and a detailed study of the processes of mixing powders in electromagnetic rotators, studies of experimental installations were carried out. The experiments were carried out both in the absence of powders in the working area and in the presence of powders. The results of such studies confirm the theoretical conclusions. At the same time, the complexity of physical processes in the devices under consideration indicates the need to continue their research.

Submersible idnuction motors are one of the main drive types used in electric driven centrifugal pump installations. The motor power factor data in the catalogs are given at the rated voltage at the motor terminals. However, under industrial conditions, the voltage at the motor terminals, as a rule, differs from the nominal voltage. This leads to a change in the power factor and must be taken into account when calculating the parameters of the electrotechnical complex mode. The investigation purpose is the calculation methodology development that allows you to evaluate the change in the motor power factor depending on the voltage at its terminals. Results: it is the method for the parameters identifying of the electric motor equivalent circuit based on the catalog load characteristics of the motor is proposed. Based on the simulation results, the response surfaces of the motor power factor depending on the voltage at the terminals and the load of the motor for various motor models are constructed. The calculation results show that the voltage reducing at the motor terminals has a positive effect in terms of reducing losses in the cable line only at a motor load of 30% or less.Increasing the voltage at the motor terminals allows reducing cable losses when the motor is loaded from 45%. Polynomials describing the resulting response surfaces are obtained. Practical relevance: the result of the study allows increasing the modeling accuracy of submersible induction electric motors operating as part of the electrotechnical complex of an oil field. With the help of the developed models, it is possible to achieve the rated slip of the motors on the nature of the change in their power factor.