Dry-type reactor overheat fault detection method based on sound signal analysis

Abstract

The invention discloses a dry-type reactor overheating fault detection method based on sound signal analysis, and the method comprises the steps: S1, forming a dry-type reactor overheating fault detection system through a voiceprint wave sensor, a main control box, and a 4G antenna, and obtaining a first sound signal source generated during the operation of a dry-type reactor; s2, processing the first sound signal source by adopting a time-frequency distribution signal reconstruction technology to obtain a second sound signal source; s3, performing local feature scale decomposition on the second sound signal source to obtain an overheat fault feature parameter of the second sound signal source; and S4, taking the characteristic parameters in the normal state and the overheating fault state as training data to train a diagnosis model, establishing a sound signal overheating fault diagnosis model, and carrying out overheating fault detection on the dry-type reactor. According to the invention, efficient monitoring and early warning of the overheating fault of the dry-type reactor are realized, and the safety and stability of a power system are enhanced.

Description

Technical Field

[0001] The invention relates to the technical field of dry-type reactor detection, and in particular to a dry-type reactor overheat fault detection method based on acoustic signal analysis. Background Art

[0002] With the development of power systems and the growth of electric energy demand, dry-type reactors, as key components in power transmission systems, are crucial to the safety and stability of the entire power grid. However, in actual operation, dry-type reactors may experience overheating failures due to overload, increased ambient temperature, or aging, which will not only affect the normal operation of the equipment, but may even cause serious safety accidents. Traditional detection methods are insufficient in early warning. Therefore, there is an urgent need for a technical means to efficiently and accurately detect overheating failures of dry-type reactors to ensure the safe operation of the power system.

[0003] Traditional dry-type reactor overheating fault detection mainly relies on temperature sensors to directly measure temperature changes, or indirectly judge the equipment status by monitoring changes in electrical parameters such as current and voltage. The advantage of this method is that the principle is simple and easy to implement, but the disadvantages are also obvious: on the one hand, the temperature sensor can only reflect the local temperature situation and cannot fully understand the overall temperature distribution inside the reactor; on the other hand, the change of electrical parameters often lags behind the temperature rise, resulting in the failure to detect the fault in time. In addition, these traditional methods lack effective detection capabilities for some hidden faults, such as the aging of insulating materials.

[0004] In the prior art, there are also non-contact fault diagnosis methods based on vibration analysis or infrared thermal imaging, which have improved the efficiency and accuracy of fault detection to a certain extent. However, they still have limitations. For example, the vibration analysis method is suitable for mechanical fault detection, but it is not very sensitive to overheating faults caused by electrical factors; and although infrared thermal imaging can intuitively display the temperature field distribution, it is easily interfered in complex electromagnetic environments, and it is difficult to penetrate the device casing to obtain internal temperature information. Therefore, the existing fault detection technology still has room for improvement when facing a specific type of power equipment such as dry-type reactors.

[0005] The present invention proposes a dry-type reactor overheating fault detection system and method based on acoustic signal analysis. The method utilizes an array composed of sound print sensors to collect and process the acoustic signals generated during the operation of the dry-type reactor. Through advanced technical means such as time-frequency distribution reconstruction, local feature scale decomposition, and fuzzy entropy calculation, accurate identification of dry-type reactor overheating faults is achieved. Summary of the invention

[0006] The present invention provides a dry-type reactor overheating fault detection method based on acoustic signal analysis, comprising:

[0007] S1. A dry-type reactor overheating fault detection system consists of a voiceprint wave sensor, a main control box, and a 4G antenna to obtain the first sound signal source generated during the operation of the dry-type reactor.

[0008] S2. Process the first sound signal source using the signal reconstruction technology of time-frequency distribution to obtain the second sound signal source.

[0009] S3. Perform local characteristic scale decomposition on the second sound signal source, decompose the second sound signal source into intrinsic scale components, calculate the fuzzy entropy of the intrinsic scale components, analyze the change characteristics of the fuzzy entropy, and obtain the overheating fault characteristic parameters of the second sound signal source.

[0010] S4. Use the characteristic parameters of the sound signal samples of the dry-type reactor in the normal state and the overheating fault state as training data to train the diagnostic model, establish a sound signal overheating fault diagnostic model, and perform overheating fault detection on the dry-type reactor.

[0011] Preferably, in S1, the dry-type reactor overheating fault detection system consists of a voiceprint wave sensor, a main control box, and a 4G antenna.

[0012] The voiceprint sensor obtains valid signal sources and invalid signal sources; the valid signal sources and invalid signal sources are distinguished by calculating the sound signal data scores; the sensors corresponding to the sound signals with data scores exceeding the threshold in the sound signal data are regarded as abnormal sensors, and the invalid signal sources are obtained from the abnormal sensors; the sensors corresponding to the sound signals with data scores lower than or equal to the threshold in the sound signal data are regarded as normal sensors, and the valid signal sources are obtained.

[0013] The 4G antenna transmits the real-time detection result of the overheating fault of the dry-type reactor to the remote control center. Preferably, the calculation of the sound signal data score specifically includes: forming a matrix with the obtained first sound signal source and projecting the high-dimensional sound signal data into a low-dimensional space through principal component analysis; in the low-dimensional space, normal data is in a clustering or distribution pattern, while abnormal data deviates from the clustering or regular distribution. By calculating the distance from the data point to the hyperplane formed by the principal components, setting a threshold T, and determining the data points with a distance exceeding the threshold T as abnormal data points.

[0014] The sound signal data calculates the covariance matrix, and the formula is: where X is the sound signal data matrix collected by m sensors at n time points, and calculate the eigenvalues λ of the sound signal data of the covariance matrix C i and eigenvectors ξ i , select the eigenvector ξ i1 , ξ i2 ,…, ξ ik as the principal components to form the principal component matrix P = [ξi1 , ξ i2 , …, ξ ik , the acoustic signal data matrix X is projected onto the principal component space to obtain the projection matrix Y. The formula is Y = XP. For the data point y j (j = 1, 2, …, n j ) calculate the Mahalanobis distance d j , and the formula is: where is the mean vector of the projected data, Σ is the covariance matrix of the projected data, and calculate the anomaly score s j , and the formula is: where max(d) is the maximum value of the Mahalanobis distance. If the anomaly score s j exceeds the threshold T, it is an abnormal data point.

[0015] Preferably, after obtaining the effective signal source, a preprocessing operation is performed; the preprocessing includes denoising the effective signal source, removing background noise and non - relevant interference signals, and generating the first acoustic signal source.

[0016] Preferably, in S2, the first acoustic signal source is processed using a signal reconstruction technique of time - frequency distribution to obtain the second acoustic signal source. Specifically, it includes:

[0017] The first acoustic signal source s1(t) is transformed into the time - frequency domain through short - time Fourier transform to obtain the time - frequency distribution S1(t, f). The formula is: where h(t) is the transformation function, l is the imaginary unit, τ is the time variable, t is the time, and f is the frequency; the obtained time - frequency distribution S1(t, f) is enhanced through the weighting function w(t, f) to obtain the enhanced time - frequency distribution S2(t, f). The formula is: S2(t, f) = w(t, f)S1(t, f); the enhanced time - frequency distribution S2(t, f) is subjected to inverse time - frequency transformation to return to the time domain, and the second acoustic signal source s2(t) is obtained. The formula is: where S2(t, f) is the enhanced time - frequency distribution.

[0018] Preferably, when the obtained time - frequency distribution S1(t, f) is enhanced through the weighting function w(t, f) to obtain the enhanced time - frequency distribution S2(t, f), the weighting function w(t, f) sets high weights for the known fault - characteristic frequency bands and zero weights for the irrelevant frequency bands.

[0019] Preferably, in S3, the second acoustic signal source is subjected to local characteristic scale decomposition, the second acoustic signal source is decomposed into intrinsic scale components, and the local characteristic contour of the acoustic signal data is formed by identifying all local extreme value points in the second acoustic signal source, including maximum value points and minimum value points; the acoustic signal data is divided into a plurality of continuous sub-intervals with adjacent extreme value points as the boundary, and the upper and lower envelopes are respectively constructed in the sub-intervals by the cubic spline interpolation method, covering the acoustic signal data in the sub-intervals, and the mean curve of the upper and lower envelopes is calculated, and the mean curve is subtracted from the second acoustic signal source to obtain a new acoustic signal data sequence as a candidate intrinsic scale component. Check whether the candidate intrinsic scale component satisfies the condition that the number of zero crossing points and extreme value points of the intrinsic scale component is equal or differs by at most one and the local average value of each point is zero. Once an intrinsic scale component is obtained, it is separated from the original signal as an independent component. Repeat the above process for the remaining signal, and stop calculating the fuzzy entropy when the amplitude of the remaining signal is lower than the preset threshold and cannot be decomposed any further or the number of iterations reaches the set value. The residual signal can be regarded as the trend term of the signal, and the entire signal can be expressed as the sum of the intrinsic scale components plus the trend term. The intrinsic scale components and the trend term together constitute the overheating fault characteristic parameters of the second acoustic signal source.

[0020] Preferably, the calculating of the fuzzy entropy of the intrinsic scale component to obtain the overheating fault characteristic parameter of the second acoustic signal source specifically includes:

[0021] The similarity of the acoustic signal data of the intrinsic scale component at different time points is quantified by a fuzzy function, and the formula is: Where u(C i (n c ),m,r) are fuzzy functions, C i (n c ) is the intrinsic scale component, n c is a discrete time point, m is a similarity tolerance parameter, and r is a width parameter of the fuzzy function; through the fuzzy function u(C i (n c ),m,r) to calculate the fuzzy entropy F(C i ), the formula is: Where N is the finite length data.

[0022] Preferably, in S4, the fuzzy entropy change characteristics are analyzed to obtain the overheating fault characteristic parameters of the second sound signal source as input data of the deep learning neural network.

[0023] Preferably, in S4, the characteristic parameters of a normal dry-type reactor and the characteristic parameters of a dry-type reactor in an overheating fault state are obtained, status labels are marked, and they are divided into a training set, a validation set, and a test set, which are input into a convolutional neural network model; the characteristic parameters of the training set are input into the model, and after operations in the hidden layer and processing by an activation function, a prediction result is output. The connection weights and biases are adjusted according to the difference from the actual labels. After each round of training, the validation set is used to evaluate and adjust the parameters. After multiple iterations, the optimal result is achieved in the validation set, and an acoustic signal overheating fault diagnosis model is formed.

[0024] Compared with the prior art, the technical solution of the present application has the following technical effects:

[0025] In the present invention, an array composed of acoustic sensors is arranged distributively around the dry-type reactor to widely capture the sound characteristics of the dry-type reactor at different positions and angles, providing a rich and accurate data basis for subsequent fault analysis, thus ensuring early warning and accurate identification of internal overheating faults of the equipment, and greatly improving the comprehensiveness and accuracy of fault diagnosis.

[0026] In the present invention, by converting the first acoustic signal source into the time-frequency domain and enhancing the known fault characteristic frequency bands through a weighting function, the problem of insensitivity to specific fault characteristics in traditional time-frequency analysis is solved. The specific frequency components related to overheating faults are enhanced, and the noise in irrelevant frequency bands is effectively suppressed, making the second acoustic signal source more focused on fault characteristics, greatly facilitating the subsequent process of extracting fault characteristic parameters, significantly improving the sensitivity and specificity of fault detection, and providing strong support for quickly and accurately judging the health status of the dry-type reactor.

[0027] In the present invention, by performing local characteristic scale decomposition on the second acoustic signal source and calculating the fuzzy entropy of the intrinsic scale components to obtain overheating fault characteristic parameters, the technical problem that it is difficult to quantify and characterize the hidden fault information in complex acoustic signals in the prior art is solved. The change characteristics of fuzzy entropy are used to identify and quantify fault characteristics, and even under weak signal conditions, the signs of overheating faults can be stably captured, thus realizing efficient monitoring and early warning of overheating faults of dry-type reactors, and greatly enhancing the safety and stability of the power system.

[0028] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, so as to be implemented in accordance with the content of the specification, and in order to make the above and other purposes, features, and advantages of the present application more obvious and understandable, the following describes the preferred embodiments of the present application in detail in conjunction with the drawings.

[0029] From the following detailed description of the specific embodiments of the present application in conjunction with the drawings, those skilled in the art will become more clear about the above and other purposes, advantages, and features of the present application. Brief Description of the Drawings

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0031] Figure 1 It is a flowchart of the dry-type reactor overheating fault detection method based on acoustic signal analysis of the present invention;

[0032] Figure 2 It is a composition diagram of the dry-type reactor overheating fault detection system based on acoustic signal analysis of the present invention;

[0033] Figure 3 It is an installation schematic diagram of the dry-type reactor overheating fault detection system based on acoustic signal analysis of the present invention;

[0034] Figure 4 It is a comparison diagram of acoustic wave patterns of the dry-type reactor in normal and overheating fault conditions based on acoustic signal analysis of the present invention;

[0035] Figure 5 It is the intrinsic scale component of the dry-type reactor under normal conditions based on acoustic signal analysis of the present invention;

[0036] Figure 6 It is the intrinsic scale component of the dry-type reactor under overheating fault conditions based on acoustic signal analysis of the present invention.

[0037] Figure 7 It is a verification curve graph of the dry-type reactor overheating fault detection method based on acoustic signal analysis of the present invention; Detailed Description of the Embodiments

[0038] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. In the following description, specific details such as specific configurations and components are provided only to help a comprehensive understanding of the embodiments of the present application. Therefore, those skilled in the art should clearly understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. Additionally, descriptions of known functions and structures are omitted for clarity and conciseness.

[0039] It should be understood that the "one embodiment" or "this embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, the "one embodiment" or "this embodiment" that appears throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner.

[0040] In addition, this application may repeat reference numerals and / or letters in different instances. Such repetition is for the purpose of simplicity and clarity, and does not in itself indicate the relationship between the various embodiments and / or arrangements discussed.

[0041] The term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another association object relationship, indicating that there can be two relationships. For example, A / and B can represent: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0042] The term "at least one" in this article is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, at least one of A and B can represent: A exists alone, A and B exist simultaneously, and B exists alone.

[0043] It should also be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion.

[0044] Embodiment 1

[0045] This embodiment details a dry-type reactor overheating fault detection method based on acoustic signal analysis, as Figure 1 shown, including the following steps:

[0046] S1. A dry-type reactor overheating fault detection system is composed of a voiceprint wave sensor, a main control box and a 4G antenna to obtain a first acoustic signal source generated during the operation of the dry-type reactor;

[0047] S2. The first acoustic signal source is processed by using a signal reconstruction technology of time-frequency distribution to obtain a second acoustic signal source;

[0048] S3, performing local characteristic scale decomposition on the second acoustic signal source, decomposing the second acoustic signal source into intrinsic scale components, calculating the fuzzy entropy of the intrinsic scale components, analyzing the fuzzy entropy change characteristics, and obtaining the overheating fault characteristic parameters of the second acoustic signal source;

[0049] S4. Using characteristic parameters of the acoustic signal samples of the dry-type reactor in the normal state and the overheating fault state as training data to train the diagnostic model, establish an acoustic signal overheating fault diagnostic model, and perform overheating fault detection of the dry-type reactor.

[0050] like Figure 2 As shown, the dry-type reactor overheating fault detection system includes 10 acoustic ripple wave sensors, a main control box and a 4G antenna; the 10 acoustic ripple wave sensors are distributed in a ring, tightly surrounding the middle height position of the dry-type reactor body, and the spacing is set at a specific ratio based on the diameter of the reactor to ensure accurate capture of the high-frequency characteristic acoustic signals generated by the winding area inside the reactor.

[0051] like Figure 3 As shown, the dry-type reactor overheating fault detection system is installed at the middle position of the inner wall of the dry-type reactor housing to shield the influence of environmental noise on the dry-type reactor overheating fault detection system.

[0052] Furthermore, in S1, valid signal sources and invalid signal sources are obtained through the voiceprint sensor; valid signal sources and invalid signal sources are distinguished by calculating the sound signal data score; the sensor corresponding to the sound signal whose data score in the sound signal data exceeds the threshold is regarded as an abnormal sensor, and the abnormal sensor obtains an invalid signal source; the sensor corresponding to the sound signal whose data score in the sound signal data is lower than or equal to the threshold is regarded as a normal sensor, and the obtained signal is a valid signal source.

[0053] Furthermore, the calculation of the acoustic signal data score specifically includes: projecting the high-dimensional acoustic signal data into a low-dimensional space by principal component analysis using the first acoustic signal source matrix obtained; in the low-dimensional space, normal data is clustered or distributed in a regular manner, while abnormal data deviates from the clustered or regular distribution, and by calculating the distance from the data point to the hyperplane formed by the principal component, a threshold value T is set, and the data point whose distance exceeds the threshold value T is determined as an abnormal data point;

[0054] The acoustic signal data is calculated by the covariance matrix, the formula is: Where X is the acoustic signal data matrix collected by m sensors at n time points, and the eigenvalue λ of the acoustic signal data of the covariance matrix C is calculated i and the eigenvector ξ i , select the eigenvector ξ i1 ,ξ i2 ,…,ξ ik As the principal component, the principal component matrix P = [ξi1 , ξ i2 , …, ξ ik , project the acoustic signal data matrix X onto the principal component space to obtain the projection matrix Y. The formula is Y = XP. For the data point y j (j = 1, 2, …, n j ) calculate the Mahalanobis distance d j , and the formula is: where is the mean vector of the projected data, Σ is the covariance matrix of the projected data, and calculate the anomaly score s of the data point j , and the formula is: where max(d) is the maximum value of the Mahalanobis distance. If the anomaly score s j exceeds the threshold T, it is an abnormal data point.

[0055] Furthermore, after obtaining the effective signal source, perform preprocessing operations; the preprocessing includes noise reduction on the effective signal source, removing background noise and non - relevant interference signals, and generating the first acoustic signal source.

[0056] Furthermore, in S2, process the first acoustic signal source using the signal reconstruction technology of time - frequency distribution to obtain the second acoustic signal source, specifically including:

[0057] Convert the first acoustic signal source s1(t) to the time - frequency domain through short - time Fourier transform to obtain the time - frequency distribution S1(t, f). The formula is: where h(t) is the conversion function, l is the imaginary unit, τ is the time variable, t is the time, and f is the frequency; enhance the obtained time - frequency distribution S1(t, f) through the weighting function w(t, f) to obtain the enhanced time - frequency distribution S2(t, f). The formula is: S2(t, f) = w(t, f)S1(t, f); perform inverse time - frequency transform on the enhanced time - frequency distribution S2(t, f) to return to the time domain and obtain the second acoustic signal source s2(t). The formula is: where S2(t, f) is the enhanced time - frequency distribution.

[0058] Furthermore, enhance the obtained time - frequency distribution S1(t, f) through the weighting function w(t, f) to obtain the enhanced time - frequency distribution S2(t, f). The weighting function w(t, f) sets high weights for the known fault - characteristic frequency bands and zero weights for the irrelevant frequency bands.

[0059] Further, in S3, the second acoustic signal source is subjected to local characteristic scale decomposition, and the second acoustic signal source is decomposed into intrinsic scale components. By identifying all local extreme points in the second acoustic signal source, including maximum points and minimum points, a local characteristic profile of the acoustic signal data is formed; the acoustic signal data is divided into a plurality of continuous sub-intervals with adjacent extreme points as the boundary, and upper and lower envelopes are respectively constructed in the sub-intervals by cubic spline interpolation method, covering the acoustic signal data in the sub-intervals, and the mean curve of the upper and lower envelopes is calculated. The mean curve is subtracted from the second acoustic signal source to obtain a new acoustic signal data sequence as a candidate intrinsic scale component. Check whether the candidate intrinsic scale component satisfies the conditions that the number of zero crossing points and extreme points of the intrinsic scale component is equal or differs by at most one and the local average value of each point is zero. Once an intrinsic scale component is obtained, it is separated from the original signal as an independent component. Repeat the above process for the remaining signal, and stop calculating the fuzzy entropy when the amplitude of the remaining signal is lower than the preset threshold and cannot be decomposed any further or the number of iterations reaches the set value. The residual signal can be regarded as the trend term of the signal, and the entire signal can be expressed as the sum of the intrinsic scale components plus the trend term. The intrinsic scale components and the trend term together constitute the overheating fault characteristic parameters of the second acoustic signal source.

[0060] Furthermore, the fuzzy entropy of the intrinsic scale component is calculated to obtain the overheating fault characteristic parameters of the second acoustic signal source, including:

[0061] The similarity of the acoustic signal data of the intrinsic scale component at different time points is quantified by the fuzzy function, and the formula is: Where u(C i (n c ),m,r) are fuzzy functions, C i (n c ) is the intrinsic scale component, n c is a discrete time point, m is a similarity tolerance parameter, and r is a width parameter of the fuzzy function; through the fuzzy function u(C i (n c ),m,r) to calculate the fuzzy entropy F(C i ), the formula is: Where N is the finite length data.

[0062] Furthermore, in S4, the fuzzy entropy change characteristics are analyzed to obtain the overheating fault characteristic parameters of the second sound signal source as input data of the deep learning neural network.

[0063] Further, in S4, by obtaining the characteristic parameters of a normal dry-type reactor and the characteristic parameters of a dry-type reactor in an overheating fault state, status labels are marked and divided into a training set, a validation set, and a test set, which are then input into the convolutional neural network model; the characteristic parameters of the training set are input into the model, and after operations in the hidden layer and processing by the activation function, a prediction result is output. The connection weights and biases are adjusted according to the difference from the actual label. After each round of training, the validation set is used to evaluate and adjust the parameters. After multiple iterations, the optimal result is achieved in the validation set, forming an overheating fault diagnosis model for acoustic signals.

[0064] This embodiment utilizes an array composed of acoustic sensors to collect and process the acoustic signals generated during the operation of the dry-type reactor. Through technical means such as time-frequency distribution reconstruction, local characteristic scale decomposition, and fuzzy entropy calculation, accurate identification of overheating faults in the dry-type reactor is achieved.

[0065] Embodiment 2

[0066] This embodiment details the local characteristic scale decomposition of the second acoustic signal source, decomposing the second acoustic signal source into intrinsic scale components, specifically including;

[0067] For local characteristic scale decomposition, the second acoustic signal source is s2(t). All local extreme points (including maximum points and minimum points) in s2(t) are identified. These extreme points constitute the local characteristic profile of the acoustic signal data, which reflects the change trend and characteristics of the signal within a local range.

[0068] Taking adjacent extreme points as boundaries, the acoustic signal data is divided into multiple continuous sub-intervals. Within each sub-interval, upper and lower envelope lines are constructed through cubic spline interpolation. Let E u (t) be the upper envelope line function and E l (t) be the lower envelope line function, covering all the acoustic signal data s2(t) within the sub-interval. Calculate the mean curve of the upper and lower envelope lines, and the formula is: where m z (t) is the mean of the upper and lower envelope lines.

[0069] Through s n (t) = s2(t) - m z (t), a new acoustic signal data sequence s n (t) is obtained, removing the low-frequency trend component in the signal and highlighting the local fluctuation characteristics of the signal.

[0070] Then, the above operations of extreme point identification, sub-interval division, envelope line construction, and difference calculation are repeated for the new sequence s n (t) to obtain s n+1 (t) = s n (t) - m z+1 (t) (where m z+1 (t) is sn (the mean curve of the upper and lower envelope of (t)), and iterate continuously. Assume that after k z iterations, s k (t) is obtained. When |s k (t)| ≤ ε (ψ is a preset threshold) or (k z = K) (K is the set number of iterations), stop the iteration. At this time, s k (t) is an intrinsic scale component C1(t). The entire formula for the second sound signal source is: (where r n (t) is the residual component).

[0071] This embodiment details that the local characteristic scale decomposition can accurately extract the intrinsic scale component from the second sound signal source. By calculating its fuzzy entropy and analyzing the change characteristics, the overheating fault characteristic parameters can be effectively obtained, thus significantly improving the accuracy and sensitivity of the overheating fault detection of dry-type reactors and providing a key and reliable basis for fault diagnosis.

[0072] Embodiment 3

[0073] Based on Embodiment 1, this embodiment details the specific implementation effects, including:

[0074] Obtain the first sound signal source through the acoustic array arrangement;

[0075] Select a voiceprint sensor with a sensitivity of 0.1 mV / Pa and a frequency response range of 20 Hz to 20 kHz. Install the overheating fault detection system for dry-type reactors at the middle position of the inner wall of a 10 kV dry-type reactor shell. 10 voiceprint wave sensors receive the acoustic signals inside the dry-type reactor from different directions. During operation, the first sound signal source is obtained through these sensors. The signal waveform collected shows a stable waveform with a certain frequency and amplitude during normal operation, while the waveform is distorted during overheating faults, as Figure 3 shown.

[0076] The first sound signal source s1(t) collected is transformed into the time-frequency domain using the short-time Fourier transform to obtain S1(t, f). During the transformation, the Hanning window is selected as the transformation function h(t), and the window length is 0.05 seconds; for the frequency range known to be related to overheating faults (such as 500 Hz to 1000 Hz), a weighting function w(t, f) is designed, with a weight of 2 set in this frequency band and 0 set in irrelevant frequency bands. The time-frequency distribution is enhanced through S2(t, f) = w(t, f)S1(t, f), and then the second sound signal source s2(t) is obtained through the inverse time-frequency transform. By comparison, it is found that after signal reconstruction, the signal energy in the fault-related frequency band is significantly enhanced, making it easier for subsequent analysis.

[0077] Perform local characteristic scale decomposition on the second sound signal source s2(t), identify its local extreme points, divide sub-intervals with adjacent extreme points, construct upper and lower envelope lines using cubic spline interpolation method, calculate the mean curve and find the difference to obtain a new sequence, iterate until the remaining signal amplitude is lower than 0.01 or iterate 3 times, and decompose to obtain 10 intrinsic scale components C1(t), C2(t)…, C 10 (t), as Figure 5 and Figure 6 shown.

[0078] When calculating fuzzy entropy, take the similarity tolerance parameter m = 2 and the fuzzy function width parameter r = 0.1, and calculate the fuzzy entropy of each intrinsic scale component according to the formula. Calculate the fuzzy entropy of the 10 decomposed intrinsic scale components, and take the first 3 fuzzy entropies as characteristic parameters as the input data of the neural network. For example, when a certain acoustic wave sensor of a voiceprint is operating normally for a dry-type reactor, the fuzzy entropy of F(C1) is 0.0074, F(C2) is 0.0147, and F(C3) is 0.6; when the dry-type reactor has an overheating fault, the fuzzy entropy of F(C1) becomes 1.8, F(C2) becomes 1.5, and F(C3) becomes 1.2.

[0079] Collect a large number of acoustic signal samples of dry-type reactors in normal and overheating fault states, such as 100 normal samples and 80 fault samples. After extracting the characteristic parameters, divide them into a training set, a validation set, and a test set according to a ratio of 7:2:1, and input them into the convolutional neural network model. During the training process, select ReLU as the activation function, with an initial learning rate of 0.001. After 100 rounds of iteration, the accuracy of the model on the validation set reaches 95%, and finally the accuracy on the test set is 82%, as Figure 7 shown, and a reliable overheating fault diagnosis model is established.

[0080] In this embodiment, through the carefully designed acoustic array layout, acoustic signals are comprehensively obtained, which can better reflect the overall state of the reactor compared with traditional detection methods. The time-frequency distribution signal reconstruction technology effectively enhances the signals in the fault characteristic frequency band and improves the sensitivity of fault detection. The combination of local characteristic scale decomposition and fuzzy entropy analysis accurately extracts the fault characteristic parameters, and the convolutional neural network model trained based on a large number of samples can diagnose overheating faults with high accuracy, providing a strong guarantee for the safe operation of dry-type reactors.

[0081] The above are only the preferred embodiments of the present invention, and it does not limit the protection scope of the present invention. For those skilled in the art, the present invention can have various changes and modifications; within the spirit and principle of the present invention, any changes, modifications, substitutions, integrations, and parameter changes made to these embodiments through conventional substitutions or capable of achieving the same functions without departing from the principle and spirit of the present invention fall within the protection scope of the present invention.

Claims

1. A dry-type reactor overheating fault detection method based on acoustic signal analysis, characterized in that: include: S1, a dry-type reactor overheat fault detection system is composed of an acoustic ripple wave sensor, a main control box and a 4G antenna to obtain the first acoustic signal source generated when the dry-type reactor is running; S2, processing the first sound signal source using a signal reconstruction technology of time-frequency distribution to obtain a second sound signal source; S3, performing local characteristic scale decomposition on the second acoustic signal source, decomposing the second acoustic signal source into intrinsic scale components, calculating the fuzzy entropy of the intrinsic scale components, analyzing the fuzzy entropy change characteristics, and obtaining the overheating fault characteristic parameters of the second acoustic signal source; S4. Using characteristic parameters of the acoustic signal samples of the dry-type reactor in the normal state and the overheating fault state as training data to train the diagnostic model, establish an acoustic signal overheating fault diagnostic model, and perform overheating fault detection of the dry-type reactor.

2. The dry-type reactor overheat fault detection method based on acoustic signal analysis according to claim 1 is characterized in that: The dry-type reactor overheat fault detection system in S1 is composed of an acoustic ripple wave sensor, a main control box and a 4G antenna; The voiceprint sensor obtains valid signal sources and invalid signal sources; the valid signal sources and invalid signal sources are distinguished by calculating the sound signal data scores through the main control box; The sensor corresponding to the acoustic signal whose data score in the acoustic signal data exceeds the threshold is regarded as an abnormal sensor, and the abnormal sensor obtains an invalid signal source; the sensor corresponding to the acoustic signal whose data score in the acoustic signal data is lower than or equal to the threshold is regarded as a normal sensor, and the obtained signal is a valid signal source; The 4G antenna transmits the real-time detection result of the overheating fault of the dry-type reactor to the remote control center.

3. The dry-type reactor overheat fault detection method based on acoustic signal analysis according to claim 2 is characterized in that: The calculation of the acoustic signal data score specifically includes: projecting the high-dimensional acoustic signal data into a low-dimensional space by principal component analysis using the acquired first acoustic signal source composition matrix; in the low-dimensional space, normal data is clustered or distributed regularly, while abnormal data deviates from the clustered or regular distribution, and by calculating the distance from the data point to the hyperplane formed by the principal component, a threshold T is set, and the data point whose distance exceeds the threshold T is determined as an abnormal data point; The acoustic signal data is calculated by covariance matrix, the formula is: Where X is the acoustic signal data matrix collected by w sensors at n time points, and the eigenvalue λ of the acoustic signal data of the covariance matrix C is calculated i and the eigenvector ξ i , select the eigenvector ξ i1 ,ξ i2 ,…,ξ ik As the principal component, the principal component matrix P = [ξ i1 ,ξ i2 ,…,ξ ik ], project the acoustic signal data matrix X into the principal component space to obtain the projection matrix Y, the formula is Y = XP, for data point y j (j=1,2,…,n j ) Calculate the Mahalanobis distance d of the data points j , the formula is: in is the mean vector of the projected data, Σ is the covariance matrix of the projected data, and the anomaly score s of the data point is calculated. j , the formula is: Where max(d) is the maximum value of the Mahalanobis distance. If the anomaly score s j If the data point exceeds the threshold T, it is considered an abnormal data point.

4. The dry-type reactor overheat fault detection method based on acoustic signal analysis according to claim 1, 2 or 3, characterized in that: After obtaining the effective signal source, performing a preprocessing operation; The preprocessing includes performing noise reduction on the effective signal source, removing background noise and non-correlated interference signals, and generating a first sound signal source.

5. The dry-type reactor overheat fault detection method based on acoustic signal analysis according to claim 1, characterized in that: The S2 uses a signal reconstruction technology of time-frequency distribution to process the first sound signal source to obtain a second sound signal source, specifically including: The first sound signal source s1(t) is converted into the time-frequency domain by short-time Fourier transform to obtain the time-frequency distribution S1(t,f), and the formula is: Wherein h(t) is the conversion function, l is the imaginary unit, τ is the time variable, t is time, and f is frequency; the obtained time-frequency distribution S1(t,f) is enhanced by the weighting function w(t,f) to obtain the enhanced time-frequency distribution S2(t,f), the formula is: S2(t,f)=w(t,f)S1(t,f); the enhanced time-frequency distribution S2(t,f) is subjected to inverse time-frequency transformation and returned to the time domain to obtain the second sound signal source s2(t), the formula is: Where S2(t,f) is the enhanced time-frequency distribution.

6. The dry-type reactor overheat fault detection method based on acoustic signal analysis according to claim 5, characterized in that: The obtained time-frequency distribution S1(t,f) is enhanced by a weighting function w(t,f) to obtain an enhanced time-frequency distribution S2(t,f), wherein the weighting function w(t,f) sets a high weight for a known fault characteristic frequency band and sets a zero weight for an irrelevant frequency band.

7. The dry-type reactor overheat fault detection method based on acoustic signal analysis according to claim 5, characterized in that: In S3, the second acoustic signal source is subjected to local characteristic scale decomposition, the second acoustic signal source is decomposed into intrinsic scale components, and all local extreme value points in the second acoustic signal source, including maximum value points and minimum value points, are identified to form a local characteristic profile of the acoustic signal data; the acoustic signal data is divided into a plurality of continuous sub-intervals with adjacent extreme value points as boundaries, upper and lower envelopes are respectively constructed in the sub-intervals by using a cubic spline interpolation method to cover the acoustic signal data in the sub-intervals, the mean curve of the upper and lower envelopes is calculated, and the mean curve is subtracted from the second acoustic signal source to obtain a new acoustic signal data sequence as a candidate intrinsic scale component; Check whether the intrinsic scale component candidates satisfy the conditions that the number of zero crossing points and extreme value points of the intrinsic scale component is equal or differs by at most one and the local average value of each point is zero; once an intrinsic scale component is obtained, separate it from the original signal as an independent component; repeat the above process for the remaining signal, and stop calculating the fuzzy entropy when the amplitude of the remaining signal is lower than the preset threshold and cannot be decomposed any further or the number of iterations reaches the set value; the remaining signal can be regarded as the trend term of the signal, and the entire signal can be expressed as the sum of the intrinsic scale components plus the trend term, and the intrinsic scale components and the trend term together constitute the overheating fault characteristic parameters of the second acoustic signal source.

8. The dry-type reactor overheat fault detection method based on acoustic signal analysis according to claim 7, characterized in that: The calculating of the fuzzy entropy of the intrinsic scale component to obtain the overheating fault characteristic parameter of the second acoustic signal source specifically includes: The similarity of the acoustic signal data of the intrinsic scale component at different time points is quantified by a fuzzy function, and the formula is: Where u(C i (n c ),m,r) are fuzzy functions, C i (n c ) is the intrinsic scale component, n c is a discrete time point, m is a similarity tolerance parameter, and r is a width parameter of the fuzzy function; through the fuzzy function u(C i (n c ),m,r) to calculate the fuzzy entropy F(C i ), the formula is: Where N is the finite length data.

9. The dry-type reactor overheat fault detection method based on acoustic signal analysis according to claim 1 or 8, characterized in that: In S4, the fuzzy entropy change characteristics are analyzed to obtain the overheating fault characteristic parameters of the second sound signal source as input data of the deep learning neural network.

10. The dry-type reactor overheat fault detection method based on acoustic signal analysis according to claim 1, characterized in that: In the S4, characteristic parameters of a normal dry-type reactor and characteristic parameters of a dry-type reactor under an overheating fault state are obtained, state labels are marked, and the labels are divided into a training set, a validation set, and a test set, which are input into a convolutional neural network model; the characteristic parameters of the training set are input into the model, and a prediction result is output after hidden layer operation and activation function processing, and the connection weight and bias are adjusted according to the difference with the actual label. After each round of training, the validation set is used to evaluate and adjust the parameters, and the optimal value is reached in the validation set after multiple iterations, thereby forming an acoustic signal overheating fault diagnosis model.

Images