An intelligent needle-listening system for abnormal sound of power equipment and its diagnosis method

Through the automatic frequency division and machine learning method of the abnormal sound intelligent listening needle system of the power equipment, the problem of relying on manual judgment of the fault detection of power equipment is solved, the intelligent positioning of abnormal sound sources and the efficiency of fault diagnosis is improved, and the training of professionals is reduced Cost and time.

CN113984191BActive Publication Date: 2024-11-08SUZHOU VEIZU EQUIPMENT DIAGNOSIS TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111256753.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-10-27
Publication Date
2024-11-08
Estimated Expiration
2041-10-27

AI Technical Summary

Technical Problem

The fault detection of existing power equipment mainly relies on manual judgment, which leads to the impact of the diagnosis results being experienced and knowledge level of professionals. It takes several years and is costly to cultivate experienced professionals, and it is difficult to accurately identify the source of different sounds under complex strange sound conditions. This results in untimely troubleshooting.

Method used

The intelligent listening needle system for different sounds of power equipment is adopted, including equipment information management and maintenance module, data acquisition module, online frequency modulation module, online listening module, different sound training module and different sound source intelligent recognition module. Through automatic frequency division and machine learning methods, Realize intelligent positioning and fault diagnosis of abnormal noise sources.

Benefits of technology

It improves the efficiency of power equipment fault diagnosis, reduces the training cycle and cost of professional and technical personnel, reduces the dependence of manual experience, and realizes accurate identification and rapid positioning of different sound sources.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113984191B_ABST
    Figure CN113984191B_ABST
Patent Text Reader

Abstract

An intelligent listening system for abnormal sounds of power equipment and a diagnostic method thereof, comprising an equipment information management and maintenance module, a basic information configuration module, a data acquisition module, an online frequency modulation module, an online listening module, an abnormal sound training module, and an abnormal sound source intelligent identification module; the collected data signal is decomposed into a plurality of frequency bands by means of octave analysis; the frequency bands and sensitive characteristic indicators containing abnormal sound information are automatically screened out from the plurality of frequency bands by means of a machine learning method, and then the intelligent positioning of the abnormal sound source of the power equipment is realized by means of the screened frequency bands and sensitive characteristic indicators, thereby improving the efficiency of power equipment fault diagnosis, reducing the training cycle of professional and technical personnel, and saving training costs.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment fault detection, and in particular to an intelligent needle-finding system for abnormal sound of power equipment and a diagnosis method thereof. Background Art

[0002] Abnormal noises in power equipment are often accompanied by abnormal operation or failure of the power equipment. The existing detection of abnormal noises in power equipment is mainly based on manual judgment, and the diagnosis results are directly affected by the experience and knowledge level of professionals. Under the current model, it takes several years or even longer to train an experienced professional, and the training cycle and economic cost are high. In addition, when the complexity of the abnormal noise is higher than the processing level of professionals, technicians will not be able to accurately identify the source of the abnormal noise, which will lead to untimely fault handling, causing long-term equipment shutdown at the least, and even large-scale power outages or disasters in severe cases. Summary of the invention

[0003] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide an intelligent listening system for abnormal sounds of power equipment and a diagnostic method thereof, so as to realize the intelligent positioning of the abnormal sound source of power equipment, improve the efficiency of fault diagnosis of power equipment, reduce the training cycle of professional and technical personnel, and save training costs.

[0004] In order to achieve this object, the present invention adopts the following technical solutions:

[0005] An intelligent listening system for abnormal sound of electric power equipment comprises an equipment information management and maintenance module, a basic information configuration module, a data acquisition module, an online frequency modulation module, an online listening module, an abnormal sound training module, and an abnormal sound source intelligent identification module; the equipment information management and maintenance module creates and maintains basic information and measuring point information of electric power equipment involved in locating abnormal sound sources, so that a mutual identification relationship is established between data and measuring points, sets sensor parameters, acquisition card parameters, sampling parameters, and algorithm parameters for each measuring point through the basic information configuration module, and collects sound data of electric power equipment according to the above configuration information through the data acquisition module;

[0006] The online frequency modulation module selects the frequency division method for the sound data of the power equipment obtained by the data acquisition module. The online listening module captures the characteristic indicators of several frequency ranges obtained after the frequency division processing. The obtained characteristic indicator library is used as the input of the abnormal sound training module and model learning is carried out. The frequency range information and sensitive characteristic indicators output by the abnormal sound training module are used as the input of the intelligent identification module of the abnormal sound source. The trend graph is drawn using the sensitive characteristic indicators of the selected frequency range, and the position of the abnormal sound source is located through trend analysis.

[0007] The intelligent listening system for abnormal sounds of power equipment also includes a historical data management module and a precision analysis module. The historical data management module manages and extracts the stored historical sound data, and the precision analysis module extracts the sound data of interest through the historical data management module for precise diagnosis.

[0008] The basic information in the basic information configuration module includes the sensitivity configuration of the sensor, the power supply mode and coupling mode of the acquisition card, the sampling time, sampling frequency, characteristic value calculation interval, inspection data storage time, the percentage of sound source prompt sensitivity, the filtering mode, frequency division mode and unilateral interception time configuration in the algorithm parameters.

[0009] The signal frequency division modes in the online frequency modulation module include double frequency division, 1 / 3 double frequency division and cochlear implant frequency division modes.

[0010] The online listening module is used to extract characteristic indicators, which are divided into absolute indicators and relative indicators, including peak-to-peak value, kurtosis, root mean square value, decibel value, relative peak-to-peak value, relative kurtosis, relative root mean square value and relative decibel value.

[0011] The foreign sound source intelligent identification module uses the selected frequency band and characteristic index to draw a trend chart of the index change under the frequency band, and finds the position of the foreign sound source through the trend chart.

[0012] The historical data management module is used to save historical databases for different types of power equipment.

[0013] The precision analysis module performs time domain analysis, frequency domain analysis, frequency band analysis and time-frequency analysis on the sound data.

[0014] A diagnostic method using an intelligent needle-listening system for abnormal sound of electric power equipment includes the following steps:

[0015] Step 1: Based on octave analysis and cochlear implant theory, select a frequency division method from the online frequency modulation module to automatically divide the collected sound data of the power equipment. The specific frequency division process is as follows:

[0016] Octave method: For any octave frequency division method, the upper and lower cutoff frequencies and center frequencies of each frequency range are expressed by the following formula:

[0017]

[0018] Among them, f upper,i 、f lower,i and f center,i They represent the upper cutoff frequency, lower cutoff frequency and center frequency of the frequency range respectively. The center frequency of the frequency range is specified in the standard of GB / T 3240-1982 as follows:

[0019]

[0020] The above formula takes 1000Hz as the reference frequency, f center,i is the center frequency of the corresponding frequency range; the octave division and 1 / 3 octave division used are the scenarios where N is 1 and 3;

[0021] Cochlear implant method: The sound data is divided into critical frequency bands for analysis. In the entire audio range, the law of how the critical frequency band bandwidth changes with the center frequency is expressed as follows:

[0022] △ f G =25+75×(1+1.4f 2 ) 0.69 ,

[0023] Where Δf G represents the critical frequency band in Hz, and f represents the center frequency in kHz;

[0024] Step 2: Collect L groups of sound data without abnormal sound as OK samples through the data acquisition module and record the labels, then perform frequency division processing on the L groups of OK samples to obtain the OK sample database M. Where X LS represents the Sth frequency band of the Lth group of samples;

[0025] Step 3: Use the online listening module to extract the absolute and relative characteristic indicators of each frequency range after the frequency division and filtering. The calculation formula of the absolute characteristic indicator is as follows:

[0026] Peak-to-peak value: x pp =max(x)-min(x), where x represents a set of sampled numbers;

[0027] Kurtosis index value: where x n represents the sampling point, and Num represents the number of sampling points;

[0028] RMS value: where x n represents the sampling point, and Num represents the number of sampling points;

[0029] Decibel value: where p i represents the measured sound pressure, p 0 is the reference sound pressure, p 0 =0.00002;

[0030] The relative index is the ratio of the absolute characteristic index of the frequency range to the absolute characteristic index of the total frequency signal. The specific calculation formula is as follows:

[0031] Relative peak-to-peak value: where x pp Represents the peak-to-peak value of the divided frequency signal, X PP Represents the peak-to-peak value of the total frequency signal;

[0032] Relative kurtosis index value: Where kurt represents the kurtosis index value of the frequency division signal, and Kurt represents the kurtosis index value of the total frequency signal;

[0033] Relative RMS: Where m represents the RMS value of the divided frequency signal, and M represents the RMS value of the total frequency signal;

[0034] Relative decibel value: where n dB Represents the decibel value of the divided frequency signal, N dB Represents the decibel value of the total frequency signal;

[0035] Step 4: Using the absolute and relative characteristic indicators in step 3, construct the characteristic vector F of the frequency range, F = [x pp ,kurt,m,n dB ,R_X pp ,R_Kurt,R_M,R_N dB ];

[0036] Step 5: According to the feature vector F of the frequency range in step 4, calculate the feature vector of each frequency range in the OK sample database, so as to construct the OK sample feature matrix G. where F LS The characteristic vector representing the Sth frequency range of the Lth group of sample signals;

[0037] Step 6: Use the out-of-phone training module to process the eigenvalues ​​of each column in the OK sample feature matrix G through the 3σ criterion to construct the upper and lower limits of the feature index envelope. The calculation formula is as follows:

[0038]

[0039]

[0040] In the above formula, Feat j_upper and Feat j_lower Represent the upper and lower envelopes of the j-th eigenvalue, and σ j Represent the mean and standard deviation of the j-th eigenvalue, where j = 1, 2, 3, …, 8;

[0041] Step 7: Use the data acquisition module to collect abnormal sound samples of power equipment, namely NG samples, and perform frequency division filtering according to the frequency division method in the first step. Then construct a feature vector for each frequency range according to the third and fourth steps, and arrange them in rows in sequence according to the arrangement method in the fifth step to obtain the NG sample feature vector. Finally, compare the NG sample feature vector with the upper and lower limits of the feature index envelope obtained in the sixth step, and use the feature index and its corresponding frequency range that exceed the upper and lower limits of the envelope the most as the sensitive feature index and its corresponding frequency range for the final screening;

[0042] Step 8: Use the sensitive feature indicators selected in the seventh step as detection indicators, and draw a trend chart of the detection indicators in real time. When the data acquisition module gradually approaches the source of the abnormal sound, the detection indicator gradually rises, and when the data acquisition module gradually moves away from the source of the abnormal sound, the detection indicator drops, thereby realizing the function of locating the source of the abnormal sound.

[0043] The existing power equipment fault troubleshooting is mainly done by professionals manually judging based on the listening results, and the results are directly affected by the experience and knowledge level of the professionals. It takes several years or even longer to train an experienced professional, and the training cycle and economic cost are high. The advantages of the present invention are:

[0044] ① The present invention uses the psychoacoustic principle to automatically divide the frequency of the collected sound signal, eliminating the process of manual spectrum analysis;

[0045] ② The present invention uses machine learning methods to screen sensitive characteristic indicators and their corresponding frequency ranges, eliminating the process of manual characteristic indicator and frequency range selection;

[0046] ③ The intelligent auscultation system for abnormal sound and its diagnosis method of the present invention have no professional quality requirements for operators and do not require manual experience judgment. They can reduce the training costs and time costs of professional listeners, improve the efficiency of abnormal sound source diagnosis, and achieve cost reduction and efficiency improvement for industrial enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The figure is a flow chart of the system and the diagnostic method thereof of the present invention.

[0048] Figure 2 Schematic diagram of frequency range screening according to an embodiment of the present invention.

[0049] Figure 3 This is a trend diagram of sensitive characteristic indicators of an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The technical solution of the present invention is further described below in conjunction with embodiments and drawings.

[0051] Reference Figure 1, an intelligent listening system for abnormal sound of power equipment, including equipment information management and maintenance module, basic information configuration module, data acquisition module, online frequency modulation module, online listening module, abnormal sound training module, abnormal sound source intelligent identification module, historical data management module and precision analysis module; the equipment information management and maintenance module creates and maintains basic information and measuring point information of power equipment involved in abnormal sound source positioning, so that a mutual identification relationship is established between data and measuring points, sets sensor parameters, acquisition card parameters, sampling parameters and algorithm parameters for each measuring point through the basic information configuration module, and collects power equipment sound data according to the above configuration information through the data acquisition module;

[0052] The online frequency modulation module selects the frequency division method for the sound data of the power equipment obtained by the data acquisition module. The online listening module captures the characteristic indicators of several frequency ranges obtained after the frequency division processing. The obtained characteristic indicator library is used as the input of the abnormal sound training module and the model learning is carried out. The frequency range information and sensitive characteristic indicators output by the abnormal sound training module are used as the input of the intelligent identification module of the abnormal sound source. The trend chart is drawn using the sensitive characteristic indicators of the selected frequency range, and the position of the abnormal sound source is located through trend analysis.

[0053] The historical data management module manages and extracts the stored historical sound data. The precision analysis module extracts the sound data of interest through the historical data management module for precise diagnosis, thereby assisting in the intelligent positioning of the abnormal sound source.

[0054] The basic information in the basic information configuration module includes the sensitivity configuration of the sensor, the power supply mode and coupling mode of the acquisition card, the sampling time, sampling frequency, characteristic value calculation interval, inspection data storage time, the percentage of sound source prompt sensitivity, the filtering mode, frequency division mode and unilateral interception time configuration in the algorithm parameters.

[0055] The signal frequency division modes in the online frequency modulation module include double frequency division, 1 / 3 double frequency division and cochlear implant frequency division modes.

[0056] The online listening module is used to extract characteristic indicators, which are divided into absolute indicators and relative indicators, including peak-to-peak value, kurtosis, root mean square value, decibel value, relative peak-to-peak value, relative kurtosis, relative root mean square value and relative decibel value.

[0057] The foreign sound source intelligent identification module uses the selected frequency band and characteristic index to draw a trend chart of the index change under the frequency band, and finds the position of the foreign sound source through the trend chart.

[0058] The historical data management module is used to save historical databases for different types of power equipment to implement subsequent operations for users.

[0059] The precision analysis module performs time domain analysis, frequency domain analysis, frequency band analysis and time-frequency analysis on the sound data.

[0060] A diagnostic method using an intelligent needle-listening system for abnormal sound of electric power equipment includes the following steps:

[0061] The first step is to select the 1 / 3 frequency automatic division mode from the online frequency modulation module and determine the number of divisions to be 23. The frequency division principle is as follows:

[0062] For the 1 / 3 octave frequency division method, the upper and lower cutoff frequencies and center frequencies of each frequency range can be expressed by the following formula:

[0063]

[0064] Among them, f upper,i 、f lower,i and f center,i They represent the upper cutoff frequency, lower cutoff frequency and center frequency of the frequency range respectively; the center frequency of the frequency range can be specified as follows according to the standard in GB / T 3240-1982:

[0065]

[0066] The above formula takes 1000Hz as the reference frequency, f center,i is the center frequency of the corresponding frequency range;

[0067] In the second step, the actual motors in the power plant are used as sampling objects. 30 groups of sound data without abnormal sounds are collected as OK samples through the data acquisition module and the labels are recorded. Then, the 30 groups of OK samples are divided by the frequency division method in the first step to obtain the OK sample database M. Where X 30,23 represents the 23rd frequency band of the 30th group of samples;

[0068] The third step is to use the online listening module to extract the absolute and relative characteristic indicators of each frequency range after the frequency division and filtering processing. The calculation formula of the absolute characteristic indicator is as follows:

[0069] Peak-to-peak value: x pp =max(x)-min(x), where x represents a set of sampled numbers;

[0070] Kurtosis index value: where x n represents the sampling point, and Num represents the number of sampling points;

[0071] RMS value: where x n represents the sampling point, and Num represents the number of sampling points;

[0072] Decibel value: where p i represents the measured sound pressure, p 0 is the reference sound pressure, p 0 =0.00002;

[0073] The relative index is the ratio of the absolute characteristic index of the frequency range to the absolute characteristic index of the total frequency signal. The specific calculation formula is as follows:

[0074] Relative peak-to-peak value: where x pp Represents the peak-to-peak value of the divided frequency signal, X PP Represents the peak-to-peak value of the total frequency signal;

[0075] Relative kurtosis index value: Where kurt represents the kurtosis index value of the frequency division signal, and Kurt represents the kurtosis index value of the total frequency signal;

[0076] Relative RMS: Where m represents the RMS value of the divided frequency signal, and M represents the RMS value of the total frequency signal;

[0077] Relative decibel value: where n dB Represents the decibel value of the divided frequency signal, N dB Represents the decibel value of the total frequency signal;

[0078] The fourth step is to construct the frequency feature vector F using the absolute and relative feature indices in the third step, F = [x pp ,kurt,m,n dB ,R_X pp ,R_Kurt,R_M,R_N dB ];

[0079] The fifth step is to calculate the feature vectors of each frequency range in the OK sample database according to the feature vector F of the frequency range in the fourth step, so as to construct the OK sample feature matrix G. where F 30,23 The eigenvector representing the 23rd frequency range of the 30th group of sample signals;

[0080] In the sixth step, the out-of-phone training module is used to process the eigenvalues ​​of each column in the OK sample feature matrix G through the 3σ criterion to construct the upper and lower envelope limits of the feature index. The calculation formula is as follows:

[0081]

[0082]

[0083] In the above formula, Feat j_upperand Feat j_lower Represent the upper and lower envelopes of the j-th eigenvalue, and σ j Represent the mean and standard deviation of the j-th eigenvalue, where j = 1, 2, 3, …, 8;

[0084] In the seventh step, a group of abnormal sound samples of the power equipment, namely, NG samples, are collected by using the data acquisition module, and frequency division filtering is performed according to the frequency division method of the first step. Then, a feature vector is constructed for each frequency range according to the third and fourth steps, and the NG sample feature vector is obtained by arranging them row by row according to the arrangement method in the fifth step. Finally, the NG sample feature vector is compared with the upper and lower limits of the feature index envelope obtained in the sixth step, and the feature index and its corresponding frequency range that exceed the upper and lower limits of the envelope the most are used as the sensitive feature index and its corresponding frequency range to be finally screened; in this embodiment, the sensitive feature index screened out is an absolute decibel value, and the corresponding frequency range is 20 (5300-6400Hz), such as Figure 2 and 3 As shown;

[0085] Step 8: Use the absolute decibel value selected in step 7 as the detection index and draw a real-time trend chart of the detection index, such as Figure 3 As shown, when the data acquisition module gradually approaches the source of the abnormal sound, the detection index gradually increases, and when the data acquisition module gradually moves away from the source of the abnormal sound, the detection index decreases, thereby realizing the function of locating the source of the abnormal sound.

[0086] The present invention takes the octave analysis idea in psychoacoustics as the theoretical basis of the signal frequency division method, and decomposes the collected sound data into several frequency bands with the help of octave analysis; through the machine learning method, the frequency bands and sensitive characteristic indicators containing abnormal sound information are automatically screened out from the several frequency bands, and then with the help of the screened frequency bands and sensitive characteristic indicators, the intelligent positioning of the abnormal sound source of the power equipment is realized, the efficiency of the power equipment fault diagnosis is improved, the training cycle of professional and technical personnel is reduced, and the training cost is saved.

[0087] The above embodiments are only to illustrate the basic principles and characteristics of the present invention. The present invention is not limited by the above examples. Without departing from the spirit and scope of the present invention, the present invention may undergo various changes and modifications. All deformations that can be directly derived or associated with the contents disclosed by ordinary technicians in this field should be considered to be within the scope of protection of the present invention.

Claims

1. An intelligent needle-hearing system for detecting abnormal noises in power equipment, characterized in that: It includes equipment information management and maintenance module, basic information configuration module, data acquisition module, online frequency modulation module, online listening module, abnormal sound training module, and abnormal sound source intelligent identification module; the equipment information management and maintenance module creates and maintains basic information and measuring point information of power equipment involved in locating abnormal sound sources, establishes a mutual identification relationship between data and measuring points, sets sensor parameters, acquisition card parameters, sampling parameters and algorithm parameters for each measuring point through the basic information configuration module, and collects power equipment sound data according to the above configuration information through the data acquisition module; The online frequency modulation module selects the frequency division method for the sound data of the power equipment obtained by the data acquisition module. The online listening module captures the characteristic indicators of several frequency ranges obtained after the frequency division processing. The obtained characteristic indicator library is used as the input of the abnormal sound training module and the model learning is carried out. The frequency range information and sensitive characteristic indicators output by the abnormal sound training module are used as the input of the intelligent identification module of the abnormal sound source. The trend chart is drawn using the sensitive characteristic indicators of the selected frequency range, and the position of the abnormal sound source is located through trend analysis. The basic information in the basic information configuration module includes the sensitivity configuration of the sensor, the power supply mode and coupling mode of the acquisition card, the sampling time, sampling frequency, characteristic value calculation interval, inspection data storage time, the percentage of sound source prompt sensitivity, the filtering mode, frequency division mode and unilateral interception time configuration in the algorithm parameters.

2. According to claim 1, the intelligent needle-hearing system for detecting abnormal noise of power equipment is characterized by: It also includes a historical data management module and a precision analysis module. The historical data management module manages and extracts the stored historical sound data, and the precision analysis module extracts the sound data of interest through the historical data management module for precision diagnosis.

3. According to claim 1, the intelligent needle-hearing system for detecting abnormal noise of power equipment is characterized by: The signal frequency division modes in the online frequency modulation module include double frequency division, 1 / 3 double frequency division and cochlear implant frequency division modes.

4. According to claim 1, the intelligent needle-hearing system for detecting abnormal noise of power equipment is characterized by: The online listening module is used to extract characteristic indicators, which are divided into absolute indicators and relative indicators, including peak-to-peak value, kurtosis, root mean square value, decibel value, relative peak-to-peak value, relative kurtosis, relative root mean square value and relative decibel value.

5. According to claim 1, the intelligent needle-hearing system for detecting abnormal noise of power equipment is characterized by: The foreign sound source intelligent identification module uses the selected frequency band and characteristic index to draw a trend chart of the index change under the frequency band, and finds the position of the foreign sound source through the trend chart.

6. According to claim 2, the intelligent needle-hearing system for detecting abnormal noise of power equipment is characterized by: The historical data management module is used to save historical databases for different types of power equipment.

7. The intelligent needle-hearing system for detecting abnormal noise of electric power equipment according to claim 2 is characterized by: The precision analysis module performs time domain analysis, frequency domain analysis, frequency band analysis and time-frequency analysis on the sound data.

8. A diagnostic method using the intelligent needle-hearing system for detecting abnormal noise of electric power equipment according to claim 1, characterized in that: The following steps are involved: Step 1: Based on octave analysis and cochlear implant theory, select a frequency division method from the online frequency modulation module to automatically divide the collected sound data of the power equipment. The specific frequency division process is as follows: Octave method: For any octave frequency division method, the upper and lower cutoff frequencies and center frequencies of each frequency range are expressed by the following formula: Among them, f upper,i 、f lower,i and f center,i They represent the upper cutoff frequency, lower cutoff frequency and center frequency of the frequency range respectively. The center frequency of the frequency range is specified in the standard of GB / T 3240-1982 as follows: The above formula takes 1000Hz as the reference frequency, f center,i is the center frequency of the corresponding frequency range; the octave division and 1 / 3 octave division used are the scenarios where N is 1 and 3; Cochlear implant method: The sound data is divided into critical frequency bands for analysis. In the entire audio range, the law of how the critical frequency band bandwidth changes with the center frequency is expressed as follows: △f G =25+75×(1+1.4f 2 ) 0.69 , Where Δf G represents the critical frequency band in Hz, and f represents the center frequency in kHz; Step 2: Collect L groups of sound data without abnormal sound as OK samples through the data acquisition module and record the labels, then perform frequency division processing on the L groups of OK samples to obtain the OK sample database M. Where X LS represents the Sth frequency band of the Lth group of samples; Step 3: Use the online listening module to extract the absolute and relative characteristic indicators of each frequency range after the frequency division and filtering. The calculation formula of the absolute characteristic indicator is as follows: Peak-to-peak value: x pp =max(x)-min(x), where x represents a set of sampled numbers; Kurtosis index value: where x n represents the sampling point, and Num represents the number of sampling points; RMS value: where x n represents the sampling point, and Num represents the number of sampling points; Decibel value: where p i represents the measured sound pressure, p0 is the reference sound pressure, p0=0.00002; The relative index is the ratio of the absolute characteristic index of the frequency range to the absolute characteristic index of the total frequency signal. The specific calculation formula is as follows: Relative peak-to-peak value: where x pp Represents the peak-to-peak value of the divided frequency signal, X PP Represents the peak-to-peak value of the total frequency signal; Relative kurtosis index value: Where kurt represents the kurtosis index value of the frequency division signal, and Kurt represents the kurtosis index value of the total frequency signal; Relative RMS: Where m represents the RMS value of the divided frequency signal, and M represents the RMS value of the total frequency signal; Relative decibel value: where n dB Represents the decibel value of the divided frequency signal, N dB Represents the decibel value of the total frequency signal; Step 4: Using the absolute and relative characteristic indicators in step 3, construct the characteristic vector F of the frequency range, F = [x pp ,kurt,m,n dB ,R_X pp ,R_Kurt,R_M,R_N dB ]; Step 5: According to the feature vector F of the frequency range in step 4, calculate the feature vector of each frequency range in the OK sample database, so as to construct the OK sample feature matrix G. where F LS The characteristic vector representing the Sth frequency range of the Lth group of sample signals; Step 6: Use the out-of-phone training module to process the eigenvalues ​​of each column in the OK sample feature matrix G through the 3σ criterion to construct the upper and lower limits of the feature index envelope. The calculation formula is as follows: In the above formula, Feat j_upper and Feat j_lower Represent the upper and lower envelopes of the j-th eigenvalue, and σ j Represent the mean and standard deviation of the j-th eigenvalue, where j = 1, 2, 3, …, 8; Step 7: Use the data acquisition module to collect abnormal sound samples of power equipment, namely NG samples, and perform frequency division filtering according to the frequency division method in the first step. Then construct a feature vector for each frequency range according to the third and fourth steps, and arrange them in rows in sequence according to the arrangement method in the fifth step to obtain the NG sample feature vector. Finally, compare the NG sample feature vector with the upper and lower limits of the feature index envelope obtained in the sixth step, and use the feature index and its corresponding frequency range that exceed the upper and lower limits of the envelope the most as the sensitive feature index and its corresponding frequency range for the final screening; Step 8: Use the sensitive feature indicators selected in the seventh step as detection indicators, and draw a trend chart of the detection indicators in real time. When the data acquisition module gradually approaches the source of the abnormal sound, the detection indicator gradually rises, and when the data acquisition module gradually moves away from the source of the abnormal sound, the detection indicator drops, thereby realizing the function of locating the source of the abnormal sound.