Voiceprint feature-based bolt looseness degree quantitative evaluation method

Abstract

The invention relates to the technical field of transmission tower fault monitoring, in particular to a voiceprint feature-based bolt looseness degree quantitative evaluation method, which comprises the following steps of S1, arranging sound signal collectors at a plurality of bolt connection structures of a transmission tower; s2, controllable external excitation is applied to the power transmission tower, and the sound signal collector obtains a sound signal from the bolt connection structure; s3, the sound signals are preprocessed, PLP features and MFCC features are extracted from the preprocessed sound signals, the PLP features and the MFCC features are fused, and PLP-MFCC features are formed; s4, converting the sound signal into a vibration signal, and obtaining a frequency domain variance feature and a frequency domain dispersion feature based on the vibration signal; and S5, constructing and training a bolt looseness quantization depth belief network, and inputting the frequency domain variance feature, the frequency domain dispersion feature and the PLP-MFCC feature into the bolt looseness quantization depth belief network to obtain a bolt looseness evaluation result.

Description

Technical Field

[0001] The present invention relates to the technical field of transmission tower fault monitoring, and in particular to a method for quantitatively evaluating the degree of bolt looseness based on soundprint features. Background Art

[0002] Transmission towers are widely used in transmission lines as supports for overhead conductors. The safety of the towers is directly related to the safe and stable operation of the transmission lines. With the increase in electricity load, a number of new transmission lines are built every year. Transmission lines are very sensitive to wind loads, which are generally recognized as the control loads in transmission line design. Under the action of wind loads, the tower components are prone to fatigue damage or even destruction, resulting in the instability of the tower structure and the truncation of the tower waist or even the overall collapse. For transmission tower structures, bolt connections and cable fixation are the most widely used connection and fixing methods. In the structural vibration caused by long-term wind loads, the bolt connection points will become loose or the bolt holes will crack. If not discovered in time, the damage will be further expanded, and even the collapse of the tower will be caused by strong winds;

[0003] At present, the research on the loosening of bolt connections at home and abroad is mainly concentrated in the general machinery industry. There are few studies on the loosening of tower bolt connections, and they are all in the primary stage. Most of the literature only studies the mechanical properties of bolt connections, the processing and installation technology of bolts, the stress relationship of bolts and other related contents. The existing methods for detecting loose bolts on transmission towers are mostly manual inspections, which are inefficient and cannot guarantee the accuracy of the detection results. Summary of the invention

[0004] In view of the above-mentioned prior art, the present invention provides a method for quantitatively evaluating the degree of bolt looseness based on soundprint features, mainly to solve the technical problems existing in the above-mentioned background technology.

[0005] To achieve the above object, the technical solution of the embodiment of the present invention is implemented as follows:

[0006] The first aspect of the present invention discloses a method for quantitatively evaluating the degree of bolt looseness based on soundprint features, the method comprising the following steps:

[0007] S1. Arrange acoustic signal collectors at multiple bolt connection structures of the transmission tower;

[0008] S2, applying a controllable external excitation to the transmission tower, and obtaining the acoustic signal from the bolt connection structure by the acoustic signal collector;

[0009] S3, preprocessing the acoustic signal, extracting PLP features and MFCC features from the preprocessed acoustic signal, and fusing the PLP features with the MFCC features to form PLP-MFCC features;

[0010] S4, converting the acoustic signal into a vibration signal, and obtaining a frequency domain variance feature and a frequency domain discreteness feature based on the vibration signal;

[0011] S5. Construct and train a deep confidence network for bolt loosening quantification, input frequency domain variance features, frequency domain discrete features, and PLP-MFCC features into the deep confidence network for bolt loosening quantification, and obtain the bolt loosening assessment results.

[0012] Optionally, a noise elimination method based on wavelet transform and adaptive threshold selection is used to eliminate noise from the acoustic signal to implement preprocessing of the acoustic signal, which specifically includes:

[0013] S201, using wavelet transform of multi-resolution analysis to decompose the acoustic signal S(n) into sub-band signals of different frequencies, and setting the number of wavelet decomposition layers;

[0014] S202, setting an initial soft threshold or hard threshold by the following formula:

[0015]

[0016] Where σ is the standard deviation of the noise, and N is the length of the acoustic signal S(n);

[0017] S203, using the initial threshold to perform soft threshold or hard threshold processing on the wavelet coefficients of each layer, and then perform inverse wavelet transform to form a reconstructed signal;

[0018] S204, calculating the mean square error between the reconstructed signal and the acoustic signal S(n), and if the mean square error cannot reach the minimum value, continuously iteratively adjusting the threshold to find the optimal soft threshold or the optimal hard threshold that minimizes the mean square error;

[0019] S205, using the optimized optimal soft threshold and the optimal hard threshold to perform threshold processing on the wavelet coefficients of each layer, remove the noise part, obtain the processed wavelet coefficients, use the processed wavelet coefficients to perform inverse wavelet transform, and reconstruct the clean sound signal S(N).

[0020] Optionally, threshold processing is performed on the wavelet coefficients of each layer using the optimized optimal soft threshold and the optimal hard threshold, specifically including:

[0021] The wavelet coefficients whose absolute values ​​are less than the optimal soft threshold are set to zero, and the wavelet coefficients whose absolute values ​​are greater than or equal to the optimal soft threshold are subtracted from the threshold to implement soft threshold processing on the wavelet coefficients of each layer;

[0022] The wavelet coefficients whose absolute values ​​are less than the optimal hard threshold are set to zero, and the wavelet coefficients greater than or equal to the optimal hard threshold remain unchanged, so that the wavelet coefficients of each layer after soft threshold processing are subjected to hard threshold processing.

[0023] Optionally, the PLP features are fused with the MFCC features to form PLP-MFCC features, specifically including: fusing the PLP features and the MFCC features in series to obtain a series fusion matrix K, performing principal component analysis on the series fusion results to obtain a transformation matrix P, and multiplying the series fusion matrix K by the transformation matrix P to obtain PLP-MFCC features.

[0024] Optionally, converting the sound signal into a vibration signal specifically includes:

[0025] S401. Measure the stress areas of bolts of different types and construct axial vibration equations of bolts of different types;

[0026] S402, forming an expression for the force caused by the sound wave on the bolt surface based on the expression of the sound signal;

[0027] S403, obtaining a homogeneous solution, a special solution, a general solution and a constant of the axial vibration equation of the bolt based on the force expression;

[0028] S404. Substitute the homogeneous solution, the special solution, the general solution and the constant into the axial vibration equations of the bolts of different types to form a vibration signal expression.

[0029] Optionally, based on the plurality of vibration signals, obtaining frequency domain variance features and frequency domain dispersion features specifically includes:

[0030] S411, numerically differentiating the vibration signal expressions at the multiple bolt connection structures twice respectively to obtain multiple vibration accelerations, and calculating the average values ​​of the multiple vibration accelerations;

[0031] S412, solving the frequency domain variance of the vibration acceleration based on multiple vibration accelerations;

[0032] S413, solving the frequency domain variance and average value of the vibration acceleration based on the multiple vibration accelerations, and calculating the frequency domain dispersion, wherein the frequency domain dispersion is calculated by the following formula: C=s / B, wherein B represents the average value, and s represents the frequency domain variance;

[0033] S414: forming a frequency domain variance feature and a frequency domain dispersion feature based on the frequency domain variance and the average value.

[0034] Optionally, frequency domain variance features and frequency domain dispersion features are formed based on the frequency domain variance and the average value, specifically including:

[0035] The frequency domain variance feature is formed by the following formula:

[0036]

[0037] Where s1 represents the variance in the current bolt state, and s1 represents the variance when the bolt is in a tightened state;

[0038] The frequency domain discreteness feature is formed by the following formula:

[0039]

[0040] Where C1 represents the frequency domain discreteness of the current bolt state, and C2 represents the frequency domain discreteness of the bolt in the tightened state.

[0041] Optionally, the specific steps of constructing and training a deep confidence network for bolt loosening quantification include: using a restricted Boltzmann machine to construct a deep confidence network for bolt loosening quantification, introducing a stochastic gradient descent method to optimize the deep confidence network for bolt loosening quantification, using training set data to train the deep confidence network for bolt loosening quantification, initializing each layer of the network through unsupervised learning, and using the training results as input to a higher layer of the network, wherein the final output result of the deep confidence network for bolt loosening quantification is expressed as healthy / unhealthy.

[0042] The second aspect of the present invention discloses a quantitative evaluation system for bolt looseness based on soundprint features, the system is used to implement the evaluation method, the evaluation system includes a vibration exciter, a sound signal acquisition device, and a signal processing device, the vibration exciter is used to generate an adjustable external excitation acting on a transmission tower, the sound signal acquisition device is connected to the signal processing device, the signal processing device includes a preprocessing module, a first feature extraction module, a second feature extraction module, and an evaluation module.

[0043] The preprocessing module is configured to remove noise from the acoustic signal using a noise removal method based on wavelet transform and adaptive threshold selection;

[0044] The first feature extraction module is configured to extract PLP features and MFCC features from the preprocessed acoustic signal, and fuse the PLP features with the MFCC features to form PLP-MFCC features;

[0045] The second feature extraction module is configured to convert the preprocessed acoustic signal into a vibration signal expression, and obtain a frequency domain variance feature and a frequency domain discreteness feature based on the vibration signal expression;

[0046] The evaluation module includes a bolt loosening quantification deep confidence network, which obtains the bolt loosening evaluation result based on frequency domain variance features, frequency domain discreteness features, and PLP-MFCC features.

[0047] The beneficial effects of the present invention are: obtaining an acoustic signal at a bolt by applying an external force, and extracting PLP features and MFCC features from the acoustic signal respectively, and fusing the PLP features with the MFCC features to form PLP-MFCC features, which can more comprehensively reflect the vibration characteristics of the bolt and enhance the feature discrimination. Furthermore, by converting the acoustic signal into a vibration signal and extracting the frequency domain variance features and the frequency domain discreteness features, the feature set is enriched, and more information support is provided for the deep learning model. In obtaining the evaluation results, the frequency domain variance features, the frequency domain discreteness features and the PLP-MFCC features of the bolt connection structure area are input into the bolt loosening quantification deep confidence network, which can automatically learn and identify the complex pattern of bolt loosening, and realize the accurate evaluation of the degree of bolt loosening. The bolt loosening evaluation result is used to judge whether the bolt connection in the bolt connection structure area is in a healthy state or an unhealthy state. The present invention not only improves the efficiency and accuracy of bolt loosening detection, but also provides strong technical support for the safe operation and maintenance of transmission towers. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Schematic diagram of the process of quantitatively evaluating the degree of bolt looseness based on voiceprint features in an embodiment of the present application. DETAILED DESCRIPTION

[0049] The technical solution of the present invention is further elaborated in detail below in conjunction with the drawings and specific embodiments of the specification. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention. In the following description, the expression "some embodiments" is related to a subset of all possible embodiments, but it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0050] In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is apparent to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present invention, some technical features well known in the art are not described.

[0051] It should be understood that the present invention can be implemented in different forms and should not be interpreted as being limited to the embodiments proposed herein. On the contrary, providing these embodiments will make the disclosure thorough and complete, and the scope of the present invention will be fully conveyed to those skilled in the art. And the purpose of the terms used herein is only to describe specific embodiments and is not intended to be a limitation of the present invention. When used herein, the singular forms of "one", "one" and "said / the" are also intended to include plural forms, unless the context clearly indicates another way. It should also be understood that the terms "compose" and / or "include" when used in this specification determine the presence of the features, integers, steps, operations, elements and / or parts, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, parts and / or groups. When used herein, the term "and / or" includes any and all combinations of the relevant listed items.

[0052] It should also be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "inside", "outside", "left", "right" and similar expressions used herein are for illustrative purposes only and are not intended to be the only implementation method.

[0053] In order to fully understand the present invention, a detailed structure will be proposed in the following description to illustrate the technical solution proposed by the present invention. The optional embodiments of the present invention are described in detail as follows, but in addition to these detailed descriptions, the present invention may also have other implementations.

[0054] Please refer to the attached Figure 1 In a first aspect, the present invention discloses a method for quantitatively evaluating the degree of bolt looseness based on soundprint features, the method comprising the following steps:

[0055] S1. Arrange acoustic signal collectors at multiple bolt connection structures of the transmission tower;

[0056] For example, the sound signal collector mainly includes a bone conduction sound sensor, which senses the sound vibration signal of the detected object, collects the sound vibration signal and converts it into an electrical signal. Unlike traditional microphones that collect sound through air conduction, bone conduction sound sensors use the characteristics of sound propagation through solid media and are not easily disturbed by ambient sound. They can collect sound with high clarity in a very noisy environment, thereby reducing the attenuation of sound signals.

[0057] S2, applying a controllable external excitation to the transmission tower, and obtaining the acoustic signal from the bolt connection structure by the acoustic signal collector;

[0058] Exemplarily, the controllable external excitation is generated by external force hitting the transmission tower, wherein the external force is generated by a force hammer, an exciter and other excitation devices. When the transmission tower is hit by the external force, a low-frequency and high-energy elastic wave is generated. The elastic wave propagates in the metal solid and is finally detected and received by an acoustic signal collector arranged at the bolted connection structure.

[0059] S3, preprocessing the acoustic signal, extracting PLP features and MFCC features from the preprocessed acoustic signal, and fusing the PLP features with the MFCC features to form PLP-MFCC features;

[0060] S4, converting the acoustic signal into a vibration signal, and obtaining a frequency domain variance feature and a frequency domain discreteness feature based on the vibration signal;

[0061] S5. Construct and train a deep confidence network for bolt loosening quantification, input frequency domain variance features, frequency domain discrete features, and PLP-MFCC features into the deep confidence network for bolt loosening quantification, and obtain the bolt loosening assessment results.

[0062] In an optional implementation, a noise elimination method based on wavelet transform and adaptive threshold selection is used to eliminate noise from the acoustic signal to implement preprocessing of the acoustic signal, which specifically includes:

[0063] S201, using wavelet transform of multi-resolution analysis to decompose the acoustic signal S(n) into sub-band signals of different frequencies, and setting the number of wavelet decomposition layers, among which commonly used mother wavelets include Daubechies, Symlet, Coiflet, etc., and 3-5 layers of decomposition are usually selected to cover the main frequency components of the signal;

[0064] S202, setting an initial soft threshold or hard threshold by the following formula:

[0065]

[0066] Where σ is the standard deviation of the noise, and N is the length of the acoustic signal S(n);

[0067] S203, using the initial threshold to perform soft threshold or hard threshold processing on the wavelet coefficients of each layer, and then perform inverse wavelet transform to form a reconstructed signal;

[0068] Mean square error: If the mean square error cannot reach the minimum value, the threshold is adjusted iteratively to find the optimal soft threshold or optimal hard threshold that minimizes the mean square error.

[0069] S204, calculating the mean square error between the reconstructed signal and the acoustic signal S(n), wherein the method for calculating the mean square error is common knowledge among those skilled in the art and will not be specifically explained in this embodiment;

[0070] S205, using the optimized optimal soft threshold and the optimal hard threshold to perform threshold processing on the wavelet coefficients of each layer, remove the noise part, obtain the processed wavelet coefficients, use the processed wavelet coefficients to perform inverse wavelet transform, and reconstruct the clean sound signal S(N).

[0071] Specifically, firstly, the initial hard threshold and soft threshold are set and calculated, wherein the sizes of the hard threshold and the soft threshold are different, and the hard threshold and the soft threshold are used to perform soft threshold or hard threshold processing on the wavelet coefficients of each layer respectively, and then an inverse wavelet transform (IWT) is performed to obtain the reconstructed signal, and the mean square error between the reconstructed signal and the original signal is calculated to determine whether the mean square error is minimized. If it is not minimized, the hard threshold and the soft threshold are iteratively adjusted to find the optimal hard threshold and soft threshold that minimize the MSE. The noise elimination method based on wavelet transform and adaptive threshold selection can balance the noise reduction effect and signal loss, effectively extract the bolt loosening sound signal, and eliminate the surrounding environmental interference sounds such as wind, human voice, animal calls, and thunder and lightning.

[0072] Furthermore, in step S205, the optimized optimal soft threshold and the optimal hard threshold are used to perform threshold processing on the wavelet coefficients of each layer, specifically including:

[0073] The wavelet coefficients whose absolute values ​​are less than the optimal soft threshold are set to zero, and the wavelet coefficients whose absolute values ​​are greater than or equal to the optimal soft threshold are subtracted from the threshold to implement soft threshold processing on the wavelet coefficients of each layer;

[0074] The wavelet coefficients whose absolute values ​​are less than the optimal hard threshold are set to zero, and the wavelet coefficients greater than or equal to the optimal hard threshold remain unchanged, so that the wavelet coefficients of each layer after soft threshold processing are subjected to hard threshold processing.

[0075] In an optional implementation, the PLP feature is fused with the MFCC feature to form a PLP-MFCC feature, wherein the process of extracting the PLP feature from the acoustic signal is as follows:

[0076] Perform short-time Fourier transform on the acoustic signal S(N) to obtain the spectrum x(ω), and calculate the power spectrum p(ω) of the spectrum x(ω):

[0077] p(ω)=R x [x(ω)] 2 +I m [x(ω)] 2

[0078] The power spectrum p(ω) is subjected to critical band analysis, and the power spectrum p(ω) is normalized to the Bark domain with its frequency axis as follows:

[0079] Ω(ω)=6ln{ω / 1200π+[(ω / 1200π) 2 +1] 1 / 2}

[0080] Where ω is the angular frequency and Ω is the Bark domain frequency;

[0081] After the power spectrum is normalized to the Bark domain, the power spectrum is convolved with the simulated auditory frequency critical curve Ψ(Ω). The formula of the critical band curve is:

[0082]

[0083] The Ψ(Ω) curve approximately simulates the auditory filtering characteristics of the human ear and is basically stable in the Bark domain. The Bark domain sampling step is set to 1, and 24 critical bands are divided. Through the discrete convolution of Ψ(Ω) and P(Ω), the critical band power spectrum of the sampled signal is obtained:

[0084]

[0085] In the formula, i is 1, 2, ..., 24, θ[Ω i ]’s sampling interval is 1;

[0086] The equal loudness curve obtained based on the simulation of human ear characteristics is used to perform equal loudness pre-emphasis processing on the signal frequency components in the Fourier frequency domain:

[0087]

[0088] Γ[Ω(ω)]=E(ω)θ[Ω(ω)]

[0089] In the formula, ω is the signal frequency component, and E(ω) approximately simulates the human ear's ability to perceive sound signals of different frequencies;

[0090] In order to simulate the nonlinear relationship between objective signal strength and subjective loudness, it is necessary to control the amplitude transformation range of the critical band spectrum and compress the critical band power spectrum amplitude by taking the cube root. The calculation formula is as follows:

[0091]

[0092] To obtain Perform inverse Fourier transform to obtain the signal short-time autocorrelation function R(τ), solve the 12th-order linear prediction coefficient through Durbin formula, perform spectral analysis on the 12th-order linear prediction coefficient, and obtain 12-dimensional PLP feature parameters and their first-order differential feature parameters, a total of 24-dimensional feature parameters, which are the PLP features.

[0093] Furthermore, the step of obtaining the MFCC feature parameters is: performing a frame operation on the acoustic signal S(N), and performing high frequency pre-emphasis and windowing processing on each frame of the acoustic signal S(N);

[0094] Perform discrete Fourier transform on each frame of acoustic signal S(N) to obtain discrete power spectrum X(k):

[0095] Pass X(k) through a filter bank H of M bandpass filters m (n) filtering, and obtaining M power values ​​p m , where m is 0, 1, …, M-1;

[0096] The power value p m Perform logarithmic operation to obtain the logarithmic energy L m , where m is 0, 1, …, M-1;

[0097] The logarithmic energy L m After discrete cosine transform, the Mel cepstrum feature parameter D is obtained m :

[0098]

[0099] In the formula, m is 0, 1,…,M-1.

[0100] Remove D m The DC component D0 in the equation is D1, D2, ... D k As the static parameters of the MFCC feature parameters, the differential coefficients are obtained through the static parameters, and the MFCC features d(n) can be obtained by combining the static parameters with the differential coefficients.

[0101]

[0102] In the formula, k is a constant, indicating the order of the differential coefficients taken, c(n) represents the Mel cepstrum feature parameters, and finally 12-dimensional MFCC feature parameters and their first-order differential feature parameters can be obtained, a total of 24-dimensional MFCC feature parameters.

[0103] After obtaining the PLP feature and the MFCC feature, the PLP feature and the MFCC feature are fused in series, wherein the expression of the series fusion is: K = (α*β), wherein α is the PLP feature, β is the MFCC feature, and the principal component analysis is performed on the series fusion result to obtain the conversion matrix P, and the series fusion matrix K is multiplied by the conversion matrix P to obtain the PLP-MFCC feature.

[0104] In an optional implementation, converting the acoustic signal into a vibration signal specifically includes:

[0105] S401. Measure the stress areas of bolts of different types, and construct axial vibration equations of bolts of different types, where the axial vibration equation can be expressed as:

[0106]

[0107] Where ρ is the material density of the transmission tower, A is the bolt force area, u(x, t) is the displacement, f(x, t) is the external force, and E is the Young's modulus of the transmission tower.

[0108] S402: Based on the expression of the acoustic signal, an expression of the force caused by the acoustic wave on the bolt surface is formed, wherein the acoustic signal, especially the impact elastic wave signal, can be expressed as:

[0109] p(t)=P0ξ(t-t0)

[0110] Where P0 is the impact intensity, ξ(t-t0) is the Dirac function representing the instantaneous impact at t = t0;

[0111] The force F(t) caused by the impact elastic wave on the bolt surface can be expressed as: F(t) = A*P0ξ(t-t0), where A is the force-bearing area of ​​bolts of different models;

[0112] S403, based on the force expression, obtain the homogeneous solution, special solution, general solution and constant of the axial vibration equation of the bolt, wherein the process of obtaining the homogeneous solution, special solution, general solution and constant is a conventional method in the art, and will not be specifically explained in this embodiment;

[0113] S404. Substitute the homogeneous solution, special solution, general solution and constant into the axial vibration equations of bolts of different types to form a vibration signal expression, which describes the displacement change of the bolt under the impact elastic wave with time and space.

[0114] Specifically, when converting the acoustic signal into a vibration signal, the type of bolts used on the transmission tower is first retrieved from the database to obtain the corresponding force-bearing area. The axial vibration equations of different bolts are obtained based on the corresponding force type. Based on the same acoustic signal expression, the homogeneous solution, special solution, general solution and constant of the axial vibration equations of different bolts are respectively obtained. The homogeneous solution, special solution, general solution and constant are respectively substituted into the axial vibration equations of different types of bolts to form vibration signal expressions for different bolts.

[0115] In an optional implementation, based on the plurality of vibration signals, obtaining frequency domain variance features and frequency domain dispersion features specifically includes:

[0116] S411, numerically differentiating the vibration signal expressions at the multiple bolt connection structures twice respectively to obtain multiple vibration accelerations, and calculating the average values ​​of the multiple vibration accelerations;

[0117] Specifically, the transmission tower is struck multiple times by external force, and the vibration signal expressions at multiple bolt connection structures can be obtained after each strike. The vibration signal expressions are numerically differentiated twice to obtain multiple vibration accelerations, and the average values ​​of the multiple vibration accelerations are calculated. For example, for the bolt connection structure A, after the transmission tower is struck multiple times by external force, multiple vibration accelerations in the bolt connection structure A area are obtained within a certain time interval, and the average values ​​of the multiple vibration accelerations are calculated.

[0118] S412, solving the frequency domain variance of the vibration acceleration based on the multiple vibration accelerations, wherein the calculation formula of the variance belongs to the common knowledge of those skilled in the art and is not specifically explained in this embodiment;

[0119] S413, solving the frequency domain variance and average value of the vibration acceleration based on the multiple vibration accelerations, and calculating the frequency domain dispersion, wherein the frequency domain dispersion is calculated by the following formula: C=s / B, wherein B represents the average value, and s represents the frequency domain variance;

[0120] S414: forming a frequency domain variance feature and a frequency domain dispersion feature based on the frequency domain variance and the average value.

[0121] Furthermore, frequency domain variance features and frequency domain dispersion features are formed based on the frequency domain variance and the average value, specifically including:

[0122] The frequency domain variance feature is formed by the following formula:

[0123]

[0124] Where s1 represents the variance in the current bolt state, and s1 represents the variance when the bolt is in a tightened state;

[0125] The frequency domain discreteness feature is formed by the following formula:

[0126]

[0127] Where C1 represents the frequency domain discreteness of the current bolt state, and C2 represents the frequency domain discreteness of the bolt in the tightened state.

[0128] Specifically, it obtains the frequency domain variance characteristics and frequency domain discreteness characteristics of each bolt connection structure area.

[0129] In an optional embodiment, the specific steps of constructing and training a deep confidence network for bolt loosening quantification include: using a restricted Boltzmann machine (RBM) to construct a deep confidence network for bolt loosening quantification, introducing a stochastic gradient descent method to optimize the deep confidence network for bolt loosening quantification, using training set data to train the deep confidence network for bolt loosening quantification, initializing each layer of the network through unsupervised learning, and using the training results as input to a higher layer of the network, wherein the final output result of the deep confidence network for bolt loosening quantification is expressed as healthy / unhealthy.

[0130] Furthermore, the training process is divided into two steps: pre-training and fine-tuning.

[0131] (1) Pre-training. The whole pre-training process is unsupervised. Each RBM trained is a new representation of the input data. Assume that the input data is a vector v, and the output vector obtained after an RBM hidden layer unit is activated is h. The output vector h is used as the input training data of the next RBM to train a new RBM. The weights of each RBM obtained can independently extract features from the output of the previous layer.

[0132] Fine-tuning process: Use the RBM pre-trained weight parameters as the initial parameters of the network, and fine-tune all weights in a supervised manner through the stochastic gradient descent method until the training is completed.

[0133] In summary, when the present invention is implemented, the transmission tower is hit multiple times by external force, and the acoustic signal at each bolt connection structure area is collected by the acoustic signal collector after each hit, and the PLP feature and MFCC feature are extracted from the acoustic signal respectively, and the PLP feature and the MFCC feature are fused to form a PLP-MFCC feature, and the acoustic signal is converted into a vibration signal at the same time, and based on the vibration signal, the frequency domain variance feature and the frequency domain discreteness feature of the bolt connection structure area are obtained, and finally the frequency domain variance feature, the frequency domain discreteness feature and the PLP-MFCC feature of the bolt connection structure area are input into the bolt loosening quantification deep confidence network to obtain the bolt loosening assessment result, and the bolt loosening assessment result is used to judge whether the bolt connection in the bolt connection structure area is in a healthy state or an unhealthy state.

[0134] The second aspect of the present invention discloses a quantitative evaluation system for bolt looseness based on soundprint features, the system is used to implement the evaluation method, the evaluation system includes a vibration exciter, a sound signal acquisition device, and a signal processing device, the vibration exciter is used to generate an adjustable external excitation acting on a transmission tower, the sound signal acquisition device is connected to the signal processing device, the signal processing device includes a preprocessing module, a first feature extraction module, a second feature extraction module, and an evaluation module.

[0135] The preprocessing module is configured to remove noise from the acoustic signal using a noise removal method based on wavelet transform and adaptive threshold selection;

[0136] The first feature extraction module is configured to extract PLP features and MFCC features from the preprocessed acoustic signal, and fuse the PLP features with the MFCC features to form PLP-MFCC features;

[0137] The second feature extraction module is configured to convert the preprocessed acoustic signal into a vibration signal expression, and obtain a frequency domain variance feature and a frequency domain discreteness feature based on the vibration signal expression;

[0138] The evaluation module includes a bolt loosening quantification deep confidence network, which obtains the bolt loosening evaluation result based on frequency domain variance features, frequency domain discreteness features, and PLP-MFCC features.

[0139] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. The protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for quantitatively assessing the degree of bolt looseness based on voiceprint features, characterized in that: The method comprises the following steps: S1. Arrange acoustic signal collectors at multiple bolt connection structures of the transmission tower; S2, applying a controllable external excitation to the transmission tower, and obtaining the acoustic signal from the bolt connection structure by the acoustic signal collector; S3, preprocessing the acoustic signal, extracting PLP features and MFCC features from the preprocessed acoustic signal, and fusing the PLP features with the MFCC features to form PLP-MFCC features; S4, converting the acoustic signal into a vibration signal, and obtaining a frequency domain variance feature and a frequency domain discreteness feature based on the vibration signal; S5. Construct and train a deep confidence network for bolt loosening quantification, input frequency domain variance features, frequency domain discrete features, and PLP-MFCC features into the deep confidence network for bolt loosening quantification, and obtain the bolt loosening assessment results.

2. The method for quantitatively evaluating the degree of bolt looseness based on voiceprint features according to claim 1 is characterized in that: The noise elimination method based on wavelet transform and adaptive threshold selection is used to eliminate noise from the acoustic signal to achieve preprocessing of the acoustic signal, which specifically includes: S201, using wavelet transform of multi-resolution analysis to decompose the acoustic signal S(n) into sub-band signals of different frequencies, and setting the number of wavelet decomposition layers; S202, setting an initial soft threshold or hard threshold by the following formula: Where σ is the standard deviation of the noise, and N is the length of the acoustic signal S(n); S203, using the initial threshold to perform soft threshold or hard threshold processing on the wavelet coefficients of each layer, and then perform inverse wavelet transform to form a reconstructed signal; S204, calculating the mean square error between the reconstructed signal and the acoustic signal S(n), and if the mean square error cannot reach the minimum value, continuously iteratively adjusting the threshold to find the optimal soft threshold or the optimal hard threshold that minimizes the mean square error; S205, using the optimized optimal soft threshold and the optimal hard threshold to perform threshold processing on the wavelet coefficients of each layer, remove the noise part, obtain the processed wavelet coefficients, use the processed wavelet coefficients to perform inverse wavelet transform, and reconstruct the clean sound signal S(N).

3. The method for quantitatively assessing the degree of bolt looseness based on voiceprint features according to claim 2 is characterized in that: The optimized optimal soft threshold and optimal hard threshold are used to perform threshold processing on the wavelet coefficients of each layer, including: The wavelet coefficients whose absolute values ​​are less than the optimal soft threshold are set to zero, and the wavelet coefficients whose absolute values ​​are greater than or equal to the optimal soft threshold are subtracted from the threshold to implement soft threshold processing on the wavelet coefficients of each layer; The wavelet coefficients whose absolute values ​​are less than the optimal hard threshold are set to zero, and the wavelet coefficients greater than or equal to the optimal hard threshold remain unchanged, so that the wavelet coefficients of each layer after soft threshold processing are subjected to hard threshold processing.

4. The method for quantitatively assessing bolt looseness based on voiceprint features according to claim 3 is characterized in that: The PLP feature is fused with the MFCC feature to form a PLP-MFCC feature, specifically comprising: fusing the PLP feature and the MFCC feature in series to obtain a series fusion matrix K, performing principal component analysis on the series fusion result to obtain a transformation matrix P, and multiplying the series fusion matrix K by the transformation matrix P to obtain a PLP-MFCC feature.

5. The method for quantitatively assessing bolt looseness based on voiceprint features according to claim 3 is characterized in that: Convert sound signals into vibration signals, including: S401. Measure the stress areas of bolts of different types and construct axial vibration equations of bolts of different types; S402, forming an expression for the force caused by the sound wave on the bolt surface based on the expression of the sound signal; S403, obtaining a homogeneous solution, a special solution, a general solution and a constant of the axial vibration equation of the bolt based on the force expression; S404. Substitute the homogeneous solution, the special solution, the general solution and the constant into the axial vibration equations of the bolts of different types to form a vibration signal expression.

6. The method for quantitatively assessing bolt looseness based on voiceprint features according to claim 5 is characterized in that: Based on the multiple vibration signals, frequency domain variance features and frequency domain discreteness features are obtained, specifically including: S411, numerically differentiating the vibration signal expressions at the multiple bolt connection structures twice respectively to obtain multiple vibration accelerations, and calculating the average values ​​of the multiple vibration accelerations; S412, solving the frequency domain variance of the vibration acceleration based on multiple vibration accelerations; S413, solving the frequency domain variance and average value of the vibration acceleration based on the multiple vibration accelerations, and calculating the frequency domain dispersion, wherein the frequency domain dispersion is calculated by the following formula: C=s / B, wherein B represents the average value, and s represents the frequency domain variance; S414: forming a frequency domain variance feature and a frequency domain dispersion feature based on the frequency domain variance and the average value.

7. The method for quantitatively assessing bolt looseness based on voiceprint features according to claim 6 is characterized in that: Based on the frequency domain variance and average value, the frequency domain variance feature and the frequency domain dispersion feature are formed, including: The frequency domain variance feature is formed by the following formula: Where s1 represents the variance in the current bolt state, and s1 represents the variance when the bolt is in a tightened state; The frequency domain discreteness feature is formed by the following formula: Where C1 represents the frequency domain discreteness of the current bolt state, and C2 represents the frequency domain discreteness of the bolt in the tightened state.

8. The method for quantitatively assessing bolt looseness based on voiceprint features according to claim 7 is characterized in that: The specific steps of constructing and training a deep confidence network for bolt loosening quantification include: using a restricted Boltzmann machine to construct a deep confidence network for bolt loosening quantification, introducing a stochastic gradient descent method to optimize the deep confidence network for bolt loosening quantification, using training set data to train the deep confidence network for bolt loosening quantification, initializing each layer of the network through unsupervised learning, and using the training results as the input of a higher layer of the network, wherein the final output result of the deep confidence network for bolt loosening quantification is expressed as healthy / unhealthy.

9. The method for quantitatively evaluating the degree of bolt looseness based on soundprint features according to any one of claims 1 to 8, characterized in that: The evaluation method is applied to an evaluation system, which includes an exciter, an acoustic signal acquisition device, and a signal processing device. The exciter is used to generate an adjustable external excitation acting on a transmission tower. The acoustic signal acquisition device is connected to the signal processing device. The signal processing device includes a preprocessing module, a first feature extraction module, a second feature extraction module, and an evaluation module. The preprocessing module is configured to remove noise from the acoustic signal using a noise removal method based on wavelet transform and adaptive threshold selection; The first feature extraction module is configured to extract PLP features and MFCC features from the preprocessed acoustic signal, and fuse the PLP features with the MFCC features to form PLP-MFCC features; The second feature extraction module is configured to convert the preprocessed acoustic signal into a vibration signal expression, and obtain a frequency domain variance feature and a frequency domain discreteness feature based on the vibration signal expression; The evaluation module includes a bolt loosening quantification deep confidence network, which obtains the bolt loosening evaluation result based on frequency domain variance features, frequency domain discreteness features, and PLP-MFCC features.

Images