A method for identifying the health and fault characteristics of rotating machinery based on audio technology

Abstract

The present invention discloses a method for identifying the health and fault characteristics of rotating machinery based on audio technology, which relates to the field of equipment operation detection and includes the following steps: S1: Collect the acoustic fingerprint information of rotating machinery on-site to establish a database; S2: Construct a training feature neural network model, establish a data processing platform, and comprehensively coordinate telecommunication signals, telemetry, remote sensing data, and acoustic time-series data; S3: Deploy several audio collectors on-site to collect data; S4: Import the data into the feature neural network model; S5: Feature recognition and abnormal warning; S6: Archive abnormal data and correct the database. Import the audio information into the feature neural network model for health and fault feature recognition, detect the operation status of the rotating machinery, and at the same time collect the abnormal warnings to correct the database. Calculate and judge a large amount of data in real time, reduce the detection response time after a fault occurs, and automatically monitor the collection, automatically analyze the operation status, and automatically give warnings. The entire process does not require staff to enter the dangerous space.

Description

Technical Field

[0001] The present invention relates to the field of equipment operation detection, specifically to the field of audio detection of equipment operation, and discloses a method for identifying the health and fault characteristics of rotating machinery based on audio technology. Background Art

[0002] With the rapid development of science and technology, the degree of industrial mechanization is getting higher and higher, and mechanical failures have correspondingly increased, and bearing damage accounts for a large proportion of mechanical failures. Due to the limitations of the working environment and conditions, it is very inconvenient to inspect, judge, maintain and repair the health status of bearings. Also, they are all in rotating parts, and the operation is extremely unsafe, and personal injury may be caused accidentally. It not only affects the production efficiency of enterprises, but also endangers the personal safety of workers, which has always been an urgent problem for enterprises to solve.

[0003] Traditional inspection methods rely on workers to listen with their ears, feel with their hands, and judge the quality of bearings based on experience. It is neither accurate nor convenient, and is particularly dangerous. Moreover, rotating machinery is often in a confined space, and its working conditions are often complex and harsh, often including high temperature and pressure, chemical pollution, noise pollution, toxic and harmful substances, mechanical damage, etc., which affect the health of workers. There are still a large number of non-standard factory scenarios in the industrial field that rely on workers' hearing, etc., which are collectively referred to as experience in the industry. Among them, sound is one of the most important comprehensive representations of the health status of motor equipment. In the quality inspection link of motor products, an old master needs to listen to the sound of the equipment on site, and the quality of the product depends on the experience level polished by the old master for many years.

[0004] There are the following problems. On the one hand, it is difficult to form a unified standard through manual quality inspection, and the subjective randomness is large, and there are risks of missed inspections and false inspections; on the other hand, it is difficult to trace and feedback the discovery of defects, and the current product process flow cannot be effectively improved. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for identifying the health and fault characteristics of rotating machinery based on audio technology in order to solve the above technical problems.

[0006] The present invention specifically adopts the following technical solutions to achieve the above purpose:

[0007] 1. A method for identifying the health and fault characteristics of rotating machinery based on audio technology, comprising the following steps;

[0008] S1: Collect the sound pattern information of the rotating machinery on site through an audio collector, establish a sound pattern database by collecting the sound spectra within N groups of fixed time, the database contains the sound pattern characteristics of normal operation and the sound pattern characteristics of classic faults, and preprocess the data after collection;

[0009] S2: Construct a neural network model, establish a training model, repeatedly perform arithmetic training on the training model through the spectrums of multiple collected classic sounds to be recognized, obtain a feature neural network model, optimize the recognition parameters, and the feature neural network model recognizes the normal operating voiceprint features and classic fault voiceprint features. At the same time, establish a data processing platform to comprehensively coordinate telemetry, telemetering, remote sensing data, and acoustic time-series data;

[0010] S3: Deploy several audio collectors on-site, use audio collection equipment resistant to corrosion, high temperature, and high pressure, attach a wireless communication module to the audio collectors, synchronously collect on-site audio information according to the time gradient, and preprocess the data after collection;

[0011] S4: The computer imports the on-site audio data into the feature neural network model. The data processing platform comprehensively coordinates telemetry, telemetering, remote sensing data, and acoustic time-series data to perform health and fault feature recognition. At the same time, the computer displays the collected spectrum and frequency-domain feature data as real-time charts on the display screen of the data processing platform;

[0012] S5: According to the feature recognition, judge whether the equipment is operating normally. If an abnormality occurs, give an alarm and store the abnormal record at the same time;

[0013] S6: After each alarm, according to whether it is a false alarm and its severity level, the system automatically records the monitoring data as a sample in the database and corrects the database again; for more serious accidents that may occur, an alarm can be given under a lower confidence level, record and display the faults on the data processing platform, and at the same time perform comprehensive processing on the operating information of each device on the data processing platform, comprehensively evaluate the health degree of the device, and push information in real time through the Internet, associated apps, text messages, and phone channels, and realize online diagnosis of difficult problems through the Internet to facilitate further decision-making by managers.

[0014] Furthermore, in step S1, the voiceprint features take sound intensity and sound pressure as the main features, the collection time is 3 s, and after collection, perform logarithmic transformation based on time-series signal processing.

[0015] Furthermore, in step S1, collect the voiceprint information of rotating machinery. During the collection of normal operating voiceprint samples, noise is inevitably mixed in. Remove part of the noise through the Deep Residual Network (ResNet) method, that is, use different thresholds to separate different noises, notice unimportant features through the attention mechanism, and set them to zero through the soft threshold function; notice important recognition features through the attention mechanism and retain them, so as to enhance the ability of the deep neural network to extract useful features from noisy signals, that is, repeatedly analyze and compare the spectrum, remove noise, and improve the accuracy.

[0016] Furthermore, in step S2, optimizing the recognition parameters includes merging multiple groups of normal operation voiceprint features and classic fault voiceprint features into a test group, comparing and verifying the test group with the normal operation voiceprint features and classic fault voiceprint sample features in the neural network model, capturing the feature spectrum, and using the automatic parameter search method. If the verification passes, the feature neural network model is obtained; if not, the neural network model construction in step S2 is continued to establish a training model until the accuracy requirements are met.

[0017] Furthermore, in step S3, after collecting the field audio information, the data is preprocessed in the same way as in step S1, where a part of the noise is removed by using the deep residual network ResNet method.

[0018] Furthermore, in the process of noise removal, human voiceprints and commonly used maintenance voiceprints are additionally introduced to remove interference from human voices and maintenance work sounds.

[0019] Furthermore, in step S3, multiple audio collectors synchronously collect audio, and collect audio synchronously according to a collection time of 3 seconds each time, and perform logarithmic transformation based on time series signal processing after collection.

[0020] Furthermore, in step S5, the warning method is a reasonable combination of text message, phone call, light, and audio.

[0021] Furthermore, in step S1, when the initial database information is insufficient, the data is expanded by linearly combining the collected data of multiple audio collectors and adding noise to increase the data volume.

[0022] The beneficial effects of the present invention are as follows:

[0023] 1. The present invention collects voiceprint information of corresponding rotating machinery and equipment on site, processes it and establishes a voiceprint database, and at the same time, according to the content of the voiceprint database, trains a feature neural network model for identifying the voiceprint features of normal operation of rotating machinery and the voiceprint features of classic faults, and at the same time deploys multiple audio collectors on site, imports the processed audio information into the feature neural network model through the unlimited communication module, and performs health and fault feature recognition to detect the operation of the rotating machinery. At the same time, the abnormal alarm is collected and the database is corrected, so as to achieve unmanned automatic monitoring and collection, automatic operation status analysis, automatic early warning, abnormal detection and collection, and a cycle conversion and data fusion are performed in three seconds. Large quantities of data are calculated and judged in real time, which improves the efficiency of data processing and judgment, reduces the detection response time after the fault occurs, and greatly improves the guarantee for the safe operation of the equipment. The whole process does not require staff to enter the dangerous space. At the same time, the detection standard is more accurate, and the detection results are archived for analysis. The whole process realizes unmanned monitoring, automatic early warning, and a high degree of automation;

[0024] 2. By continuously collecting warning data and grading it according to whether it is a false alarm and its severity level, the system automatically records the monitoring data as a template in the database and corrects the database again. For more serious accidents that may occur, an alarm can be issued with a lower confidence level. Fault records and displays are carried out simultaneously on the data processing platform. At the same time, the operation information of various devices is comprehensively processed on the data processing platform to comprehensively evaluate the health of the devices, and information is pushed in real time through the Internet, associated apps, text messages, and phone channels. Online diagnosis of difficult problems is achieved through the Internet, which is convenient for managers to make further decisions, continuously improving the accuracy of the system and the operation safety level of rotating machinery.

[0025] 3. The computer will display the collected spectrum and frequency-domain characteristic data as real-time charts on the display screen of the data processing platform, which is more intuitive and convenient. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic flow chart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated herein can be arranged and designed in various different configurations.

[0028] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0029] Embodiment 1

[0030] As Figure 1 shown, this embodiment provides a method for identifying the health and fault characteristics of rotating machinery based on audio technology, which is characterized by including the following steps;

[0031] S1: Collect the sound pattern information of the rotating machinery on-site through an audio collector, establish a sound pattern database by collecting the sound spectra within N groups of fixed time, where the database contains the sound pattern characteristics of normal operation and the sound pattern characteristics of classic faults, and preprocess the data after collection;

[0032] S2: Construct a neural network model, establish a training model, repeatedly perform arithmetic training on the training model through the spectrums of multiple collected classic sounds to be recognized, obtain a feature neural network model, optimize the recognition parameters, the feature neural network model recognizes the normal operating voiceprint features and classic fault voiceprint features, and at the same time establish a data processing platform to comprehensively coordinate telemetry, telemetering, remote sensing data, and acoustic time-series data;

[0033] S3: Deploy several audio collectors on-site, using audio collection equipment resistant to corrosion, high temperature and high pressure. The audio collectors are attached with a wireless communication module, and the wireless communication module can be one or several of Bluetooth, wifi, and mobile network. The audio collectors can also use wired transmission to synchronously collect on-site audio information according to the time gradient, and preprocess the data after collection;

[0034] S4: The computer imports the on-site audio data into the feature neural network model. The data processing platform comprehensively coordinates telemetry, telemetering, remote sensing data, and acoustic time-series data for health and fault feature recognition. At the same time, the computer displays the collected spectrum and frequency-domain feature data as real-time charts on the display screen of the data processing platform;

[0035] S5: According to the feature recognition, judge whether the equipment is operating normally. If an abnormality occurs, issue an alarm and store the abnormal record at the same time;

[0036] S6: After each alarm, according to whether it is a false alarm and its severity level, the system automatically records the monitoring data as a sample in the database and corrects the database again; for more serious accidents that may occur, an alarm can be issued under a lower confidence level, and fault records and displays are performed on the data processing platform at the same time. At the same time, the data processing platform comprehensively processes the operating information of each device, comprehensively evaluates the health of the device, and real-time information is pushed through the Internet, associated apps, text messages, and phone channels, and online diagnosis of difficult problems is realized through the Internet to facilitate further decision-making by managers.

[0037] In step S1, the voiceprint features take sound intensity and sound pressure as the main features, the collection time is 3s, and after collection, logarithmic transformation based on time-series signal processing is performed.

[0038] In step S1, collect the voiceprint information of rotating machinery. During the collection of normal operating voiceprint samples, noise will inevitably be mixed in. Use the Deep Residual Network ResNet method to remove part of the noise, that is, use different thresholds to separate different noises, notice unimportant features through the attention mechanism, and set them to zero through the soft threshold function; notice important recognition features through the attention mechanism and retain them, so as to enhance the ability of the deep neural network to extract useful features from noisy signals, that is, repeatedly analyze and compare the spectrum, remove noise, and improve the accuracy.

[0039] In step S2, optimizing the recognition parameters includes merging multiple groups of normal operation voiceprint features and classic fault voiceprint features into a test group, comparing and verifying the test group with the normal operation voiceprint features and classic fault voiceprint sample features in the neural network model, capturing the feature spectrum, and using the automatic parameter search method. If the verification passes, the feature neural network model is obtained. If not, the neural network model construction in step S2 is continued to establish a training model until the accuracy requirements are met.

[0040] In step S3, after collecting the field audio information, the data is preprocessed in the same way as in step S1, where a portion of the noise is removed by using the deep residual network ResNet method. In the process of removing the noise, additional human voiceprints and common maintenance voiceprints are introduced to remove the interference of human voices and maintenance work sounds.

[0041] In step S3, multiple audio collectors synchronously collect audio, and collect audio synchronously according to the collection time of 3 seconds each time, and perform logarithmic transformation based on time series signal processing after collection.

[0042] In step S5, the warning method is a reasonable combination of text message, phone call, light, and audio.

[0043] In step S1, when the initial database information is insufficient, the data is expanded by linearly combining the collected data of multiple audio collectors and adding noise to increase the data volume.

[0044] By continuously collecting alarm data, the system automatically records the monitoring data as a sample into the database according to whether it is a false alarm and the severity level, and then corrects the database again; for the more serious accidents that may occur, the alarm can be issued at a lower confidence level, and the faults are recorded and displayed on the data processing platform at the same time. At the same time, the operating information of various equipment is comprehensively processed on the data processing platform, and the health of the equipment is comprehensively judged. Information is pushed in real time through the Internet, related apps, SMS, and telephone channels, and difficult online diagnosis is achieved through the Internet, which is convenient for managers to make further decisions, continuously improve the accuracy of the system, and improve the operating safety level of rotating machinery.

[0045] Example 2

[0046] The difference between Example 2 and Example 1 lies in the noise processing method. A comparison method is used. Generally, the frequency spectrum of rotating machinery is relatively regular and stable whether it is in a fault or normal operation state, while other noises are relatively messy. Other noises can be filtered out through a filter, and most characteristic voiceprint signals can be retained to eliminate noise interference.

[0047] Implementation principle: The present invention collects the voiceprint information of corresponding rotating mechanical equipment on-site, processes it and establishes a voiceprint database. At the same time, according to the content of the voiceprint database, a feature neural network model for identifying the voiceprint characteristics of normal operation and classic fault voiceprints of rotating machinery is trained. At the same time, multiple audio collectors are deployed on-site, and the processed audio information is imported into the feature neural network model through a wireless communication module to perform health and fault feature identification to detect the operation status of the rotating machinery. At the same time, abnormal warnings are collected to correct the database, achieving unmanned automatic monitoring and collection, automatic operation status analysis, automatic early warning, abnormal detection and collection, with a cycle conversion and data fusion every three seconds, and a large amount of data is calculated and judged in real time, improving the efficiency of data processing and judgment, reducing the detection response time after a fault occurs, greatly improving the guarantee for the safe operation of the equipment. The entire process does not require staff to enter the dangerous space, and at the same time, the detection standard is more accurate, the detection results are filed and analyzed, and the entire process realizes unmanned monitoring, automatic early warning, and high automation.

Claims

1. A method for identifying the health and fault characteristics of rotating machinery based on audio technology, characterized in that Including the following steps; S1: On-site, use an audio collector to collect the acoustic fingerprint information of rotating machinery. By collecting the sound spectra within N groups of fixed time intervals, establish an acoustic fingerprint database, which contains the acoustic fingerprint characteristics of normal operation and classic fault acoustic fingerprints. After collection, preprocess the data; S2: Construct a neural network model, establish a training model, repeatedly perform arithmetic training on the training model through multiple collected classic sound spectra to be recognized, obtain a feature neural network model, optimize the recognition parameters. The feature neural network model recognizes the acoustic fingerprint characteristics of normal operation and classic fault acoustic fingerprints. At the same time, establish a data processing platform to comprehensively coordinate telemetry, remote sensing, remote control data, and acoustic timing data; S3: Deploy several audio collectors on-site, using audio collection equipment that is corrosion-resistant, high-temperature-resistant, and high-pressure-resistant. The audio collectors are attached with wireless communication modules, and synchronously collect on-site audio information according to the time gradient. After collection, preprocess the data; S4: The computer imports the on-site audio data into the feature neural network model. The data processing platform comprehensively coordinates telemetry, remote sensing, remote control data, and acoustic timing data to perform health and fault feature recognition. At the same time, the computer displays the collected spectrum and frequency domain feature data as real-time charts on the display screen of the data processing platform; S5: According to the feature recognition, determine whether the equipment is operating normally. If an abnormality occurs, issue an alarm, and at the same time store the abnormal record; S6: After each alarm, according to whether it is a false alarm and its severity level, the system automatically records the monitoring data as a sample in the database and revises the database again; for more serious accidents that may occur, an alarm can be issued at a lower confidence level. At the same time, perform fault recording and display on the data processing platform, and comprehensively process the operation information of each device on the data processing platform, comprehensively evaluate the health status of the device, and push information in real time through the Internet, associated apps, text messages, and phone channels, and achieve online diagnosis of difficult problems through the Internet to facilitate further decision-making by managers; In step S1, when collecting the acoustic fingerprint information of rotating machinery, noise is inevitably mixed in the process of collecting normal operation acoustic fingerprint samples. Use the Deep Residual Network ResNet method to remove part of the noise, that is, use different thresholds to separate different noises, notice unimportant features through the attention mechanism, and set them to zero through the soft threshold function; notice important recognition features through the attention mechanism and retain them, so as to enhance the ability of the deep neural network to extract useful features from noisy signals, that is, repeatedly analyze and compare the spectra, remove noise, and improve the accuracy.

2. The method for identifying the health and fault characteristics of a rotating machine based on audio technology according to claim 1, wherein In step S1, the main acoustic fingerprint features are sound intensity and sound pressure, and the collection time is 3s. After collection, perform logarithmic transformation based on time-series signal processing.

3. A method for identifying the health and fault characteristics of rotating machinery based on audio technology according to claim 1, characterized in that In step S2, optimizing the recognition parameters includes merging multiple groups of normal operation voiceprint features and classic fault voiceprint features into a test group, comparing and verifying the test group with the normal operation voiceprint features and classic fault voiceprint sample features in the neural network model, capturing the feature spectrum, and using the automatic parameter search method. If the verification passes, the feature neural network model is obtained. If not, the neural network model construction in step S2 is continued to establish a training model until the accuracy requirements are met.

4. A method for identifying the health and fault characteristics of rotating machinery based on audio technology according to claim 1, characterized in that, In step S3, after collecting the field audio information, the data is preprocessed in the same way as in step S1, where a part of the noise is removed by using the deep residual network ResNet method.

5. The method for identifying the health and fault characteristics of a rotating machine based on audio technology according to claim 4, characterized in that, In the process of noise removal, human voiceprints and commonly used maintenance voiceprints are additionally imported to eliminate interference from human voices and maintenance work sounds.

6. The method for identifying the health and fault characteristics of a rotating machine based on audio technology according to claim 1, wherein In step S3, multiple audio collectors synchronously collect audio, and collect audio synchronously according to the collection time of 3 seconds each time, and perform logarithmic transformation based on time series signal processing after collection.

7. A method for identifying the health and fault characteristics of rotating machinery based on audio technology according to claim 1, characterized in that, In step S5, the warning method is a reasonable combination of text message, phone call, light, and audio.

8. A method for identifying the health and fault characteristics of rotating machinery based on audio technology according to claim 1, characterized in that, In step S1, when the initial database information is insufficient, the data is expanded by linearly combining the collected data of multiple audio collectors and adding noise to increase the data volume.

Images