RU2749640C1 - System and method for diagnostics of industrial object based on analysis of acoustic signals - Google Patents

Abstract

FIELD: industrial facility diagnostics.SUBSTANCE: invention relates to industrial facility diagnostics based on the analysis of acoustic signals. The substance of the invention lies in the fact that a system for diagnosing an industrial facility based on the analysis of acoustic signals contains at least one microphone configured to receive acoustic signals from an industrial facility, a computer configured to generate a file with acoustic signals received from at least one microphone. signals, validation and transfer of a file to a database, a database made with the possibility of saving the said files and supplementing them with data on at least the connection of acoustic signals and an industrial facility, time, place, conditions for receiving acoustic signals, a module for determining file anomalies made with the ability to make a decision on the normal or abnormal operation of an industrial facility based on a predetermined training sample from data on the normal operation of an industrial facility, a decision-making module configured to make a decision on the presence or absence of anomalies based on additional correction in conditions of increased noise; an in-depth anomaly analysis module configured to identify the cause of anomalies in an industrial facility based on acoustic signals using heuristic dependencies.EFFECT: invention provides ability to diagnose various types of equipment and perform fast horizontal and vertical scaling, while achieving high quality breadth and depth of analysis.8 cl, 2 dwg

Description

The technical field to which the invention relates.

The invention relates to the field of monitoring and diagnosing the state of an industrial facility, in particular, assessing its technical condition.

State of the art.

A known method for diagnosing the technical state of parts, assemblies and drive units of a gas turbine engine and a device for its implementation (patent RU2379645C2, publ. 20.01.2010). For non-contact diagnostics of aircraft gas turbine engines (GTE) and drive units (PA) without disassembling the GTE and PA with the simultaneous possibility of identifying and localizing various defects of parts, assemblies and PA GTE according to the known method for diagnosing the technical condition of parts, assemblies and PA GTE, including measuring and processing of vibration signals from the hull structures of the GTE and PA with obtaining information about the technical condition of the diagnosed parts, assemblies and PA of the GTE, the measurement of vibration signals from the hull structures of the GTE and drive units is carried out in the measurement zones close to the diagnosed parts, units and drive units of the GTE, remotely and contactlessly by means of a laser vibration transducer with measurement and digital processing of vibration signals caused by aerogasdynamic and mechanical processes in the gas-air duct and kinematic pairs in the diagnosed parts, units and drive units of the gas turbine engine and transmitted to the body structures of the gas turbine engine and drives some units, and information on the technical condition of the diagnosed parts, assemblies and drive units of the GTE is obtained by digital processing of vibration signals on a recorder-analyzer with the possibility of digital processing of vibration signals with the calculation of modulation depths on discrete components of the vibration envelope spectrum in the high-frequency range of oscillations of the GTE and PA body structures. The technical result is to diagnose the technical condition of aircraft engines, components and drive units of a gas turbine engine without disassembling the gas turbine engine and drive units, while simultaneously identifying and localizing various defects of parts, assemblies and drive units of the gas turbine engine.

However, in this solution, vibration signals are used for analysis, the signal processing scheme is different, the processing principles are also different, which does not provide sufficient accuracy and speed in determining the state of various industrial facilities.

There is a known method for checking the quality of an active malfunction and intelligent diagnostics of an engine malfunction (CN101839805A, publ. 09/22/2010). The invention discloses a method for checking the quality of an active fault and intelligently diagnosing an engine fault that corresponds to a fault symptom at an appropriate frequency for checking the quality of an active fault when the engine is not disassembled. The method, in particular, includes the following steps: sequential loading of signals with different frequencies from low to high by means of a vibration device into the engine system according to the principle of the relationship between frequencies and a fault; resonate with signals with the same frequency; determination of the symptom of a malfunction generated by signals in the frequency or frequency band; collection of relevant information on faults; extracting the eigenvector of the fault for the collected information about the fault using the wavelet analysis method as an input eigenvector for inheriting and optimizing the input nodes of the wavelet neural network; and obtaining a training sample of the network for diagnosing online measurement data. The test sample of the invention has the advantages of ease of production, high relevance, high accuracy, low cost, and the like. The optimized network has the characteristics of good non-linear display and high convergence rate, and can diagnose motor malfunction effectively.

However, in this solution, vibration signals are used for analysis, the signal processing scheme is different, the processing principles are also different, which does not provide sufficient accuracy and speed in determining the state of various industrial facilities.

Known selected as a prototype solution, describing a method using MFCC and CELP for detecting malfunctions of gas turbine engines (US8655571B2, publ. 02/18/2014). A known method for detecting and diagnosing faults for gas turbine engines includes collecting a sensor signal from an acoustic or vibration sensor in a gas turbine engine, preprocessing the sensor signal to remove the predicted background, and extracting a set of characteristics from the sensor signal using Mel-Algorithms of Frequency Cepstral Coefficients (MFCC) and / or Code Excited Linear Prediction (CELP) algorithms. Fault and normal states are reported based on a comparison between a set of functions and a library of fault and non-faulty symptom profiles corresponding to the faulty and normal states of a gas turbine engine.

However, in this solution, vibration signals are used for analysis, the signal processing scheme is different, the processing principles are also different, which does not provide sufficient accuracy and speed in determining the state of various industrial facilities.

Disclosure of the invention.

In one aspect of the invention, a system for assessing the condition of an industrial facility is disclosed, comprising:

- at least one microphone configured to receive acoustic signals from an industrial facility;

- a computer capable of generating a file with acoustic signals received from a set of microphones, validating and sending the file to the database;

- a database capable of storing said files and supplementing them with data on at least the relationship between acoustic signals and an industrial facility, time, place, conditions for receiving acoustic signals;

- a module for determining file abnormality, configured to make a decision on the normal or abnormal operation of an industrial facility based on a predetermined training sample from data on the normal operation of an industrial facility;

- a decision-making module configured to make a decision on the presence or absence of anomalies based on additional correction in conditions of increased noise;

- a module for in-depth analysis of anomalies, made with the ability to identify the cause of anomalies in an industrial facility based on acoustic signals using heuristic dependencies.

In additional aspects, it is disclosed that the decision-making module is configured with the additional possibility of taking into account expert opinions when making a decision; the file with audio files contains data with a sampling rate of at least 16 kHz; the computer is made with the additional ability to filter the frequency range around the frequency of the industrial facility, and the size of the frequency range is selected from the condition that the amplitudes of frequencies N, 2N, 3N, 4N, 5N fall within the range from -5 to +5 dB from the amplitude of the fundamental frequency, where N is frequency of rotation of an element of an industrial facility; the module for determining the anomalousness of the file is configured to use convolutional neural networks, the input of which is a MEL spectrogram; the module for in-depth analysis of anomalies is configured to build heuristic rules based on the analysis of the ratio of amplitude harmonic coefficients determined in the time-frequency space; the module for in-depth analysis of anomalies is configured to use for analysis those frequencies from N, 2N, 3N, 4N, 5N, 6N, 7N, 8N, 9N, 10N, the amplitude of which is in the range from -3 to +3 dB from the amplitude of the fundamental frequency, where N is the frequency of rotation of an element of an industrial facility.

In another aspect, a method for assessing the state of an industrial facility is disclosed, comprising the steps of:

- receive acoustic signals from an industrial facility using at least one microphone;

- preliminary processing of acoustic signals in order to determine the presence of anomalies in the operation of an industrial facility using a computer based on a training sample from data on the normal operation of an industrial facility;

- using the web interface, the results of determining anomalies are corrected using expert opinions;

- determine the cause of the anomaly in an industrial facility on the basis of heuristic dependencies using a computer based on the analysis of acoustic signals in which the anomaly was detected.

In another aspect, an apparatus for assessing the condition of an industrial facility is disclosed, comprising:

- a microphone capable of receiving acoustic signals from an industrial facility;

- a primary processing module capable of generating a file with acoustic signals received from a microphone, validating and transferring the file to a memory unit;

- a memory module made with the possibility of storing said files and supplementing them with data on at least the connection of acoustic signals and an industrial facility, time, place, conditions for receiving acoustic signals;

- a module for determining file abnormality, configured to make a decision on the normal or abnormal operation of an industrial facility based on a predetermined training sample from data on the normal operation of an industrial facility;

- a decision-making module configured to make a decision on the presence or absence of anomalies based on additional correction in conditions of increased noise;

- a module for in-depth analysis of anomalies, made with the ability to identify the cause of anomalies in an industrial facility based on acoustic signals using heuristic dependencies.

The main tasks solved by the claimed invention are to determine the technical state of an industrial facility on the basis of acoustic data received from this facility, to automate the process of analyzing the state of an industrial facility.

The essence of the invention lies in the fact that with the help of at least one microphone, the sound emanating from a working industrial facility (sound that emanates from an engine and other mechanical objects) is recorded, the recorded data is processed in two stages. At the first stage, the presence or absence of anomalies is determined, at the second stage, a specific malfunction is determined through in-depth analysis.

The technical result achieved by the solution is the ability to quickly scale horizontally and vertically, while achieving high quality breadth and depth of analysis, that is, the ability to adapt the invention to various types of equipment and conduct in-depth analysis of different types of equipment.

Brief description of the drawings.

1 shows a diagram of a system for diagnosing the state of an industrial facility.

2 shows a method for diagnosing a condition of an industrial facility.

Implementation of the invention.

Terms and Definitions:

Anomaly is a situation in which the values of the parameters of an industrial facility differ from the normal values for a given operating mode of an industrial facility.

Diagnostics of an industrial facility - determining the state of an industrial facility and the cause of abnormal operation, if any.

Retraining is a tuning process for extending the anomalous properties of the recorded acoustic data file.

Violation of operating modes is a type of incident characterized by improper operation of a working industrial facility due to incorrect actions of a human operator.

Failure is a type of incident characterized by the transition of an industrial facility from an operational state to an inoperative state.

Retraining is a tuning process to change the anomalous properties of the recorded acoustic data file.

Hardware and software complex - a device or system consisting of software and hardware, a particular example may be a microprocessor system, a server, a specialized integrated circuit, a general-purpose computer, etc.

Industrial facility - an object related to the field of transport, mechanical engineering, production, particular examples may be a train, an airplane, a factory, a production line, a conveyor, a power plant, an industrial pump, etc.

An element of an industrial facility is a component, block, part, unit, unit that is part of an object that is more complex in its design and functions, examples can be a train or plane engine, a separate line or conveyor machine, a unit assembly, etc.

The solution for technical analysis of equipment acoustic data consists of hardware (AF) for collecting, processing and transmitting acoustic data and software (FC) for processing and presenting data.

The frequency response consists of at least one acoustic sensor - a measuring microphone, in some versions, installed directed to the most important elements of the analyzed equipment, means of digitization, processing and transmission of signals.

The digitizing tool is an ADC (analog to digital converter) that can be built into a microphone or processing tool.

The processing means are at least one of a computer, laptop, controller, processor, specialized circuit, specialized device for collecting and processing acoustic data. The processing means can be of any form, but must have the functionality to process the signals as needed.

The transmission medium is any known data transmitter, wired or wireless.

The inverter can be implemented in any programming language, which does not matter for the declared solution, the main requirement is effective data processing and solving problems to determine the state of the industrial facility under study.

The VFD also implements the functionality for additional training and / or retraining of the module for determining the state of an industrial facility, which makes it possible to increase the accuracy of detecting anomalies.

The differences between the proposed technical solution and those known from the prior art are:

1. Multilevel architecture that allows you to adapt systems for different types of equipment and implement fast horizontal and vertical scaling, while achieving high quality breadth and depth of analysis.

2. The presence of the module for calculating the anomalousness of the file and the module for making decisions allows you to accurately diagnose the abnormality of the equipment operation by adjusting it to a specific industrial facility or its unit.

The proposed system 100 for assessing the state of an industrial facility is shown in Fig. 1.

System 100 includes at least one microphone 101 configured to receive acoustic data in the desired frequency range, with the required accuracy. The microphone can be directed to receive data from a specific site in an industrial facility. In some embodiments, a microphone array may be used.

The single board computer 102 is configured to validate the audio data received from the microphone and send the validated data to the database 103. Validation includes checking the integrity and the presence of correct metadata: at least the time of the recording, the duration of the recording. A single board computer can be implemented using a special purpose integrated circuit (ASIC), processor, controller, or other suitable means.

The database 103 stores data received from the computer 102. The database can be located on a server or be local.

Module 104 for calculating anomalies of the acoustic data file performs processing to detect anomalies in the received data. For this, an intelligent software component is used that uses pre-recorded data on the normal operation of an object or its unit. This component compares the data on the normal operation of the object with the measured data and makes a conclusion about the presence or absence of an anomaly.

The decision-making module 105 is configured to make a decision on the normal or abnormal operation of the equipment, and this module also implements the possibility of retraining and retraining the model using expert opinions that amend the conclusions obtained as a result of the operation of module 105. The expert can indicate data containing anomalies, or vice versa, not containing them, if module 104 incorrectly determined the presence or absence of an anomaly. Thus, additional training or retraining of the model occurs.

Module 106 adjusting the model is configured to reduce false positive conclusions about the presence of anomalies, in conditions of increased external noise impacts by generalizing data for training by augmentation of noise accumulated in the system.

Module 107 for in-depth analysis of anomalous files is configured to perform in-depth analysis of the cause of the anomaly using heuristic dependencies obtained on the basis of GOST / ISO recommendations and vibroacoustic simulation of equipment, the result of the operation of module 107 is the probabilistic characteristic of the detected cause of the anomaly.

A method for assessing the state of an industrial facility is shown in Fig. 2.

In part of the method, the problem is solved, and the technical result is achieved by the fact that the diagnostics and monitoring of industrial equipment is carried out as follows. At step 201, an audio file is recorded using the measurement microphone. At step 202, the audio file is validated, and its further transfer via a wired or wireless communication channel to the storage server, then the file is stored on the long-term storage server and additional file verification takes place, additional entities are formed that allow the system to store additional information about the connection of the recorded audio file and industrial facility, time, place and conditions of recording. Next, the anomalousness of the file is calculated using the intelligent component; initially, for the operation of the intelligent component, a training sample is formed from the recorded audio files of the normal operation of the equipment. Next, a decision is made on the normal or abnormal operation of the equipment, and in this module the possibility of retraining and additional training of the model is realized with the help of expert opinions, which amend the conclusions obtained as a result of the system's operation. Further, the model is automatically adjusted to avoid false positives in conditions of increased external noise influences. If after step 202 the file is considered abnormal, then it goes to step 203, where an in-depth analysis of the cause of the anomaly using heuristic dependencies is performed, the result of the algorithm operation at step 203 is the probabilistic characteristic of the detected cause of the file anomaly.

Further, at step 204, the data of the analysis of the state of the industrial facility is transmitted to the devices of specialists carrying out the repair of the industrial facility for inspection, diagnostics or repair.

In one of the embodiments, the proposed solution is implemented in the form of a device in which a microphone or microphones, signal storage and processing units, and an analysis result output unit are located in a single housing.

The processing unit can be a set of modules (102-106) or a single software and hardware module that implements the functionality of modules 102-106, for any form of hardware implementation, the functionality for the described processing of acoustic signals is essential.

The device can be a specialized tool or a tool implemented on the basis of a laptop or PC, if their functionality allows you to receive acoustic signals, process them and transmit the processing results.

DETAILED DESCRIPTION OF IMPLEMENTATION OPTIONS .

To analyze the anomalousness of an industrial facility, an audio file is used, preferably in the wav format; the sampling frequency when digitizing acoustic data must be at least 16 kHz. For the analysis, the frequencies N, 2N, 3N, 4N, 5N are used, where N is the fundamental frequency (the frequency of the industrial facility, for example, the rotational speed of an electric motor), while the amplitudes of these frequencies should be in the range from -5 dB to + 5 dB relative to fundamental frequency.

For in-depth analysis, a wav file with a sampling frequency of at least 16 kHz and a frequency range corresponding to two conditions is used:

1. the frequency for analysis is one of N, 2N… 10N, where N is the proposed fundamental frequency of the industrial facility;

2. The amplitude of these frequencies is in the range from -3 dB to + 3 dB relative to the fundamental frequency.

The single board computer 102 is preferably a Linux operating system and may have different parameters for different condition assessment system configurations. The recorded file is checked based on the file size and its sampling rate. If the file has not passed the check for these parameters, then it is not sent further to check for anomalies.

In preferred embodiments, the PostrgesQL relational database 103 is used. The long-term file storage system differs depending on the system configuration.

The module 104 for calculating file anomalies includes an ensemble of several (depending on the task) convolutional neural networks, to the input of which a pre-calculated MEL spectrogram is supplied. Next, neural networks trained on MEL spectrograms of the normal operation of the facility are used to determine abnormal performance based on the MEL spectrograms of measured acoustic data from the industrial facility.

Decision module 105 takes into account the initial threshold introduced after the initial training of the system (determined on the basis of technical recommendations for the condition and data on the last maintenance of equipment), amendments made by experts, and concludes whether there is an abnormality in the operation of the industrial facility.

The model adjusting module 106 includes algorithms for reducing false positives by changing the threshold value obtained after the decision module. False positives determine the threshold for sensitivity to small changes in acoustic data, the change in the threshold value is carried out based on the recommendations of the maintenance expert, after a technical inspection, after the fact after an incident occurred, or in any other suitable way.

The in-depth analysis module 107 is implemented based on the Winger-Will (Weil) transform, which provides increased resolution in both the frequency and time domains. The acoustic data that have been subjected to this transformation is analyzed based on the analysis of the ratio of the amplitude harmonic coefficients determined in the time-frequency space.

The heuristic rules are based on the analysis of the ratio of the amplitude harmonic coefficients determined in the time-frequency space. The values of the coefficients are determined with the accuracy required for a given type of element of an industrial facility on the basis of a statistical analysis of the available information about the operation of a known good model, in particular, described in the recommendations of GOST / ISO on the basis of repeated experiments. The heuristic dependences for the frequency domain selected from the recommendations are further transformed for use in the time-frequency domain.

A single audio source can have both a leading and a minor role. The leading role means that the microphone is aimed at the industrial object under study or its unit and records the acoustic data generated by a specific industrial object. The auxiliary role implies non-directional use for recording and subsequent filtering of background noise. The unification of the leading and auxiliary audio files is carried out at the stage of calculating the file anomaly.

To reduce the diagnostic error, the model is trained by an expert, which occurs in two ways:

1. Manual indication of the anomaly by an expert by designating the audio files on which the anomaly was observed, as well as the anomaly's belonging to a group of elements of an industrial facility. For files marked by an expert as an anomaly, the absolute, relative values of all parameters characterizing the operation of the specified group of elements of an industrial facility (as well as their rate of change, variance values and correlation coefficients) are estimated with similar values for areas not marked by an expert as an anomaly.

For example, if, when analyzing the operation of an asynchronous electric motor, an expert discovers an imbalance of the output shaft, he can mark the audio files containing this anomaly, and with a small amount of these files, he can give a command to retrain the model, with a large number of anomalous files, a command to retrain the model can be given.

2. Verification by an expert of the anomaly found by the model can be carried out both at the stage of calculating the file anomaly and at the stage of in-depth analysis in order to clarify the cause of the file anomaly.

At the stage of in-depth analysis, if the anomaly is confirmed by an expert, the deviation of the actual value of the output parameter from the predicted value for this parameter is marked as unacceptable, and if it is exceeded in the future, such areas will be marked as an incident.

When the user refutes the anomaly, the deviation value is marked as permissible, and in the future areas with similar deviations are not marked as anomalies, due to the adjustment of the anomaly threshold at the decision-making stage

The estimation of the residual resource of an element of an industrial facility is carried out by solving the problem of finding the most plausible value of the parameter that best describes the historical data for the observation period using machine learning methods. If it is necessary to use a calculated (not measured at the industrial facility itself) parameter to assess the residual resource, the model is supplemented with a separate block that solves the problem of determining the empirical pattern of changes in the calculated parameter at given points. The analyzed parameter is the largest ratio of the amplitude harmonic coefficients of the fundamental frequency of rotation and multiple harmonics generated by the object under study, determined in the time-frequency space.

Depending on the amount of data and the nature of their change, the task of predicting the residual life can be carried out by the following method: with a sufficient amount of data and their sufficient discreteness, the assessment is made on the basis of the empirical distribution function of the failure rate depending on the frequency of detected anomalies.

The use of the proposed two-stage approach to assessing the state of an industrial facility and finding the causes of malfunctions makes it easy to adapt the claimed solution to different types of facilities, and also simplifies in-depth analysis and obtaining a final assessment of the state of the facility in comparison with a hypothetical one-stage method implemented on similar principles. In addition, the result of the analysis of the technical condition of the object is obtained faster due to the possibility of obtaining the result after the first stage of processing.

The proposed system can at least partially be implemented on the basis of web technologies, that is, after receiving and digitizing acoustic data from an industrial facility at the location of this facility, these data are transmitted through communication networks to remote computers or servers, where they are further processed into including with the participation of experts who, through the web interface, can retrain or retrain the software part, an interface for entering a description of the anomalies identified after diagnostics / analysis / repair is also provided at the location of the industrial facility.

IMPLEMENTATION OPTIONS.

Option 1 . Asynchronous and synchronous motors with direct load connection to the output shaft.

Asynchronous and synchronous motors with direct load connection to the output shaft can be susceptible to various types of malfunctions. The monitoring and diagnostics system relies on an acoustic signal generated by an industrial facility or its unit during operation. For industrial facilities not equipped with internal control systems, troubleshooting can take a long time and require partial or complete disassembly of the industrial equipment. Condition monitoring is carried out by the system based on audio data collected during normal operation of the equipment. When the state of the equipment changes, the audio signal changes, including at frequencies inaudible to the human ear, so, for example, the sound emitted when a bearing is worn out is at frequencies greater than 20 kHz. The appearance of anomalous audio artifacts causes an increase in the anomalousness of the audio file in the anomaly calculation module. When the threshold is crossed, the decision algorithm sends the audio file for in-depth analysis.

In the case of monitoring and diagnostics of asynchronous and synchronous electric motors with direct connection of the load to the output shaft, the causes of malfunctions can be the shaft support bearings, stator windings, position and condition of the output shaft. The main disadvantage of existing systems is the need to install sensors directly on the body of an industrial facility. The acoustic system for troubleshooting will allow recording equipment to be installed at a distance of 5-200 cm in the direction of the output shaft. Diagnostics can be carried out both in real time and by recording audio data on removable media for further analysis.

Option 2. Reducer.

To determine anomalies in the operation of the gearbox, it is necessary to install the sound recording equipment at a distance of 20-200 cm from the gearbox housing, the entire housing must be included in the directivity pattern of the microphone. After detecting the anomaly using the smart component, the anomalous audio file is transferred to the deep analysis stage. A wide frequency range is required to accurately identify a defect, since defects in gear pairs are determined by the gear frequency, which in turn depends on the number of teeth. The use of a wide frequency range requires increased processing power of the server side.

Option 3. Compressor.

The system allows detecting anomalies in the operation of compressor units of all types, but the determination of defects can be carried out for compressor units of the type:

1. Screw

1. Cam

1. Paddle

Option 4 . Turbine.

Sound recording equipment is installed outside the air flow created by the turbine, at a distance of 20-200 cm. The system allows detecting anomalies in the operation of turbine installations of all types, defects that can be identified:

1. Defects of turbine blades

1. Fight of the primary and secondary shaft of the turbine

1. Defects in bearings

To accurately identify a defect, a wide frequency range is required, since defects in turbines are determined by the blade frequency, which in turn depends on the number of blades. The use of a wide frequency range requires increased processing power of the server side. The frequency range when diagnosing turbines is the widest, since the speed of the input shaft can reach 100,000 rpm.

Option 5 . ICE.

The system allows detecting anomalies in the operation of the internal combustion engine on the basis of the same approaches as described for the electric motor or turbine, but taking into account the specificity of the internal combustion engine mechanism and the specificity of the acoustic signals emanating from it.

The embodiments are not limited to the embodiments described herein, to a person skilled in the art based on the information set forth in the description and knowledge of the prior art, other embodiments of the invention will become apparent without departing from the spirit and scope of the present invention.

Elements mentioned in the singular do not exclude the plurality of elements, unless otherwise specified.

The functional connection of elements should be understood as a connection that ensures the correct interaction of these elements with each other and the implementation of one or another functionality of the elements. Specific examples of functional communication can be communication with the ability to exchange information, communication with the ability to transmit electric current, communication with the ability to transmit mechanical movement, communication with the ability to transmit light, sound, electromagnetic or mechanical vibrations, etc. The specific type of functional connection is determined by the nature of the interaction of the mentioned elements, and, unless otherwise indicated, is provided by widely known means using principles widely known in the art.

The methods disclosed herein comprise one or more steps or steps to achieve the described method. The steps and / or steps of the method can be substituted for each other without departing from the scope of the claims. In other words, if a specific order of steps or actions is not defined, the order and / or use of specific steps and / or actions may vary without departing from the scope of the claims.

The application does not indicate specific software and hardware for the implementation of blocks in the drawings, but a person skilled in the art should understand that the essence of the invention is not limited to a specific software or hardware implementation, and therefore any software and hardware known in the art can be used to implement the invention. prior art. So hardware can be implemented in one or more specialized integrated circuits, digital signal processors, digital signal processing devices, programmable logic devices, user-programmable gate arrays, processors, controllers, microcontrollers, microprocessors, electronic devices, other electronic modules made with the ability perform the functions described in this document, a computer, or a combination of the above.

Although not specifically mentioned, it is obvious that when it comes to storing data, programs, etc., a computer-readable storage medium is meant, examples of computer-readable storage media include read only memory, random access memory, register, cache memory , semiconductor storage devices, magnetic media such as internal hard disks and removable disks, magneto-optical media and optical media such as CD-ROMs and digital versatile disks (DVDs), and any other storage media known in the art.

While exemplary embodiments have been described in detail and shown in the accompanying drawings, it should be understood that such embodiments are illustrative only and are not intended to limit the broader invention, and that the invention should not be limited to the specific arrangements and structures shown and described. as various other modifications may be apparent to those skilled in the art.

The features mentioned in various dependent claims, as well as the implementations disclosed in various parts of the description, can be combined with the achievement of beneficial effects, even if the possibility of such a combination is not explicitly disclosed.

Claims (18)

1. A system for diagnosing an industrial facility based on the analysis of acoustic signals, containing: - at least one microphone configured to receive acoustic signals from an industrial facility; - a computer capable of generating a file with acoustic signals received from at least one microphone, validating and transferring the file to the database; - a database capable of storing said files and supplementing them with data on at least the connection between acoustic signals and an industrial facility, time, place, conditions for receiving acoustic signals; - a module for determining file abnormality, configured to make a decision on the normal or abnormal operation of an industrial facility based on a predetermined training sample from data on the normal operation of an industrial facility; - a decision-making module configured to make a decision on the presence or absence of anomalies based on additional correction in conditions of increased noise; - a module for in-depth analysis of anomalies, made with the ability to identify the cause of anomalies in an industrial facility based on acoustic signals using heuristic dependencies. 2. The system according to claim 1, in which the decision-making module is made with the additional possibility of taking into account expert opinions when making a decision. 3. The system of claim 1, wherein the audio file contains data with a sampling rate of at least 16 kHz. 4. The system according to claim 1, in which the computer is made with the additional possibility of filtering the frequency range around the frequency of the industrial facility, and the size of the frequency range is selected from the condition that the amplitudes of frequencies N, 2N, 3N, 4N, 5N fall within the range from -5 to + 5 dB from the amplitude of the fundamental frequency, where N is the frequency of rotation of an element of an industrial facility. 5. The system of claim 1, wherein the file abnormality determination module is configured to use convolutional neural networks, the input of which is a MEL spectrogram. 6. The system according to claim 1, in which the module for in-depth analysis of anomalies is configured to build heuristic rules based on the analysis of the ratio of amplitude harmonic coefficients determined in the time-frequency space. 7. The system according to claim 6, in which the module for in-depth analysis of anomalies is configured to use for analysis those frequencies from N, 2N, 3N, 4N, 5N, 6N, 7N, 8N, 9N, 10N, the amplitude of which is in the range from - 3 to +3 dB from the amplitude of the fundamental frequency, where N is the frequency of rotation of an element of an industrial facility. 8. A method for diagnosing an industrial facility based on the analysis of acoustic signals, comprising stages at which - receive acoustic signals from an industrial facility using at least one microphone; - preliminary processing of acoustic signals in order to determine the presence of anomalies in the operation of an industrial facility using a computer based on a training sample from data on the normal operation of an industrial facility; - using the web interface, the results of determining anomalies are corrected using expert opinions; - determine the cause of the anomaly in an industrial facility on the basis of heuristic dependencies using a computer based on the analysis of acoustic signals in which the anomaly was detected.

Images