CN110208375B - Detection method for anchor rod anchoring defect and terminal equipment - Google Patents

Detection method for anchor rod anchoring defect and terminal equipment Download PDF

Info

Publication number
CN110208375B
CN110208375B CN201910511203.XA CN201910511203A CN110208375B CN 110208375 B CN110208375 B CN 110208375B CN 201910511203 A CN201910511203 A CN 201910511203A CN 110208375 B CN110208375 B CN 110208375B
Authority
CN
China
Prior art keywords
layer
neural network
node
signal
anchor rod
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910511203.XA
Other languages
Chinese (zh)
Other versions
CN110208375A (en
Inventor
孙晓云
林童
***
宫世杰
闫志勋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shijiazhuang Tiedao University
Original Assignee
Shijiazhuang Tiedao University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shijiazhuang Tiedao University filed Critical Shijiazhuang Tiedao University
Priority to CN201910511203.XA priority Critical patent/CN110208375B/en
Publication of CN110208375A publication Critical patent/CN110208375A/en
Application granted granted Critical
Publication of CN110208375B publication Critical patent/CN110208375B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/04—Analysing solids
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/22—Details, e.g. general constructional or apparatus details
    • G01N29/24—Probes
    • G01N29/2412—Probes using the magnetostrictive properties of the material to be examined, e.g. electromagnetic acoustic transducers [EMAT]
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
    • G01N29/4481—Neural networks
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
    • G01N29/46—Processing the detected response signal, e.g. electronic circuits specially adapted therefor by spectral analysis, e.g. Fourier analysis or wavelet analysis
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/02—Indexing codes associated with the analysed material
    • G01N2291/023—Solids
    • G01N2291/0234—Metals, e.g. steel
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/02—Indexing codes associated with the analysed material
    • G01N2291/028—Material parameters
    • G01N2291/0289—Internal structure, e.g. defects, grain size, texture
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/04—Wave modes and trajectories
    • G01N2291/044—Internal reflections (echoes), e.g. on walls or defects
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/26—Scanned objects
    • G01N2291/269—Various geometry objects

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Pathology (AREA)
  • Immunology (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Signal Processing (AREA)
  • Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
  • Acoustics & Sound (AREA)
  • Mathematical Physics (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Electromagnetism (AREA)

Abstract

The invention provides a detection method and terminal equipment for anchor rod anchoring defects, wherein the method comprises the following steps: acquiring an echo signal; decomposing the echo signal by using a variational modal decomposition method to obtain intrinsic modal components corresponding to each decomposed layer; denoising the intrinsic mode components corresponding to each layer to obtain a denoised target signal; and inputting the target signal into a target neural network to obtain a defect detection result of anchor rod anchoring. According to the invention, the acquired echo signals are decomposed by using an improved variational modal decomposition method to obtain decomposed intrinsic modal components of each layer, the intrinsic modal components of each layer are subjected to noise reduction, the noise-reduced intrinsic modal components are used as input values of the structural self-organization Elman neural network, and the structural self-organization Elman neural network is adopted to carry out anchor rod anchoring quality detection, so that the anchor rod anchoring defect can be rapidly detected, and the anchor rod anchoring defect detection precision is higher.

Description

Detection method for anchor rod anchoring defect and terminal equipment
Technical Field
The invention belongs to the technical field of quality detection, and particularly relates to a method for detecting anchor rod anchoring defects and terminal equipment.
Background
The anchor rod is mostly steel nails in engineering, and the anchor is mostly composed of concrete, and the anchor is fixed outside the anchor rod. The application environment of the anchor rod anchoring is severe, so that the anchoring quality of the anchor rod is damaged, and serious safety accidents can be caused when the anchor rod anchoring is damaged or corroded seriously. How to detect the anchoring internal quality of the anchor rod quickly and accurately is very important.
At present, the use environment of anchor rod anchoring has concealment, so that whether the anchor rod anchoring has defects or not can not be accurately detected, and the detection difficulty of the anchor rod anchoring is high.
Disclosure of Invention
In view of this, the embodiment of the invention provides a method for detecting anchor rod anchoring defects and terminal equipment, so as to solve the problem that the anchor rod anchoring defects cannot be accurately detected at present.
The first aspect of the embodiment of the invention provides a method for detecting anchor rod anchoring defects, which comprises the following steps:
acquiring echo signals, wherein the echo signals are acquired initial information used for detecting the anchor defect of the anchor rod;
decomposing the echo signal by using a variational modal decomposition method to obtain target eigenmode components corresponding to each decomposed layer;
denoising the target eigenmode components corresponding to each layer to obtain a denoised target signal;
and inputting the target signal into a target neural network to obtain a defect detection result of anchor rod anchoring.
A second aspect of an embodiment of the present invention provides a terminal device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the steps of the method for detecting anchor rod anchoring defects are implemented.
A third aspect of embodiments of the present invention provides a computer readable storage medium having a computer program stored thereon, which when executed by a processor, carries out the steps of the method of detecting bolt anchorage defects as described above.
According to the invention, the acquired echo signals are decomposed by using a decomposition method based on improved variation mode to obtain decomposed intrinsic mode components of each layer, the intrinsic mode components of each layer are subjected to noise reduction, the noise-reduced intrinsic mode components are used as input values of a target neural network, the target neural network is adopted for anchor rod anchoring quality detection, structure self-adaptive adjustment can be completed according to different detection conditions, the anchor rod anchoring defect can be rapidly detected, and the anchor rod anchoring defect detection precision is higher.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.
FIG. 1 is a schematic flow chart of a method for detecting anchor rod anchoring defects according to an embodiment of the invention;
FIG. 2 is a schematic diagram of an echo signal provided by one embodiment of the present invention;
FIG. 3 is a schematic diagram illustrating the denoising effect of a wavelet threshold denoising method according to an embodiment of the present invention;
FIG. 4 is an exemplary diagram of an echo signal after a metamorphic modal decomposition method and noise reduction processing according to an embodiment of the present invention;
FIG. 5 is an exemplary diagram of a training process for a constructed neural network provided by one embodiment of the present invention;
FIG. 6 is a diagram of an example of a constructed neural network provided by one embodiment of the present invention;
FIG. 7 is a diagram illustrating an example structure for node splitting provided by an embodiment of the present invention;
fig. 8 is a structural example diagram of an echo signal acquisition device and a bolt anchor according to an embodiment of the invention;
FIG. 9 is a structural illustration of a void defect in the front end of the anchor rod according to an embodiment of the present invention;
FIG. 10 is a structural illustration of a void defect at the rear end of the anchor rod anchoring provided by an embodiment of the present invention;
FIG. 11 is a structural illustration of a dual void defect in anchor rod anchoring provided by an embodiment of the present invention;
FIG. 12 is a schematic structural view of a device for detecting anchor defects according to an embodiment of the present invention;
fig. 13 is a schematic diagram of a terminal device according to an embodiment of the present invention.
Wherein: 1. an anchor rod; 2. anchoring; 3. a permanent magnet; 4. a yoke; 5. a coil; 6. the location of the defect.
Detailed Description
In the following description, for purposes of explanation and not limitation, specific details are set forth, such as particular system structures, techniques, etc. in order to provide a thorough understanding of the embodiments of the invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.
The terms "comprises" and "comprising," as well as any other variations, in the description and claims of this invention and the drawings described above, are intended to mean "including but not limited to," and are intended to cover non-exclusive inclusions. For example, a process, method, or system, article, or apparatus that comprises a list of steps or elements is not limited to only those steps or elements listed, but may alternatively include other steps or elements not listed, or inherent to such process, method, article, or apparatus. Furthermore, the terms "first," "second," and "third," etc. are used to distinguish between different objects and are not used to describe a particular order.
In order to explain the technical means of the present invention, the following description will be given by way of specific examples.
Example 1:
fig. 1 shows a flow chart of an implementation of a method for detecting a defect in anchoring of a rock bolt according to an embodiment of the present invention, and for convenience of description, only the parts related to the embodiment of the present invention are shown, which is detailed as follows:
as shown in fig. 1, a method for detecting a defect in anchoring of a rock bolt provided by an embodiment of the present invention includes:
s101, acquiring an echo signal, wherein the echo signal is acquired initial information used for detecting the anchor rod anchoring defect.
And S102, decomposing the echo signal by using a variational modal decomposition method to obtain target eigenmode components corresponding to each decomposed layer.
S103, denoising the target eigenmode components corresponding to each layer to obtain denoised target signals.
And S104, inputting the target signal into a target neural network to obtain a defect detection result of anchor rod anchoring.
In an embodiment of the present invention, S102 includes:
s201, carrying out initial layering on the echo signals, circularly updating the initial eigenmode component of the layer based on each layer, and obtaining the target eigenmode component of the current layer after the updated eigenmode component meets a first convergence condition.
S202, based on the target eigenmode component of the current layer, calculating the weighted kurtosis value of each layer after the current layering.
And S203, if the weighted kurtosis value of each layer after the current layering meets a first preset condition, stopping layering, and obtaining the final layering number which is the number of layers after the previous layering.
And S204, if the weighted kurtosis value of each layer after the current layering does not meet the first preset condition, continuing to layer until the weighted kurtosis value of each layer after the layering meets the first preset condition.
Wherein the first preset condition is as follows:
Figure BDA0002093547860000044
wherein the content of the first and second substances,
Figure BDA0002093547860000041
sgn () is a sign function to ensure that the phases of the output signal and the original signal are consistent as much as possible; kwIs a weighted kurtosis value; r is an index of C; d is a kurtosis value; μ is the mean of the echo signals f (t); σ is the standard deviation of the echo signal f (t); t is time; t is the length of the echo signal f (T); u. ofk(t) is the target eigenmode component of the kth layer at the current delamination;
Figure BDA0002093547860000042
is uk(t) mean value;
Figure BDA0002093547860000043
is the mean of f (t); zeta is a preset threshold; c is the cross-correlation coefficient between the two signals.
Wherein the target eigenmode component
Figure BDA0002093547860000051
Is composed of
Figure BDA0002093547860000052
Obtaining through inverse Fourier transform:
Figure BDA0002093547860000053
wherein:
Figure BDA0002093547860000054
n is the cycle number; k is the kth layer;
Figure BDA0002093547860000055
to represent
Figure BDA0002093547860000056
The Fourier transform of (1) is obtained by n +1 times of circulation of the kth layer;
Figure BDA0002093547860000057
is the value of the echo signal f (t) after Fourier transformation; alpha is a constant;
Figure BDA0002093547860000058
to represent
Figure BDA0002093547860000059
The Fourier transform of (1) is obtained by n times of circulation of the d-th layer, wherein d is not equal to k; omega is frequency;
Figure BDA00020935478600000510
is the k-th layer and is the modal center frequency at cycle n + 1;
Figure BDA00020935478600000511
representing the Lagrangian λn(t) Fourier transform, obtained for the nth cycle; τ is a time constant; k is the current layering number;
for example, when K is 5, the eigenmode components d of the first layer are calculated to be 2, 3, 4, and 5, respectively. And calculating the intrinsic mode components d of the second layer to be 1, 3, 4 and 5 respectively.
The first convergence condition is as follows:
Figure BDA00020935478600000512
wherein: 10 ∈ ═ 10-7。
In this embodiment, if the number of tiers is K equal to 3, the 3 initial eigenmode components are updated respectively, so that the initial eigenmode components satisfy the first convergence condition, the updating is stopped, the initial eigenmode components satisfying the first convergence condition are output, if the weighted kurtosis value of each tier at this time satisfies the first preset condition, the target signal is divided into 2 tiers, and the 2 initial eigenmode components satisfying the first convergence condition, which are output when K equal to 2, are stored as the target eigenmode components.
As an example, setting an initial layering number K equal to 2, and an initial iteration number n equal to 0;
(1) determining the target eigenmode components of the current layer:
defining an eigenmode component function:
performing a Variational Modal Decomposition (VMD) decomposition on the echo signal f (t), and defining an Intrinsic Mode Function (IMF) whose expression is:
Figure BDA0002093547860000061
wherein u isk(t) am-fm signals that are the target eigenmode components,
Figure BDA0002093547860000062
phase of the signal, Ak(t) is the instantaneous amplitude; k is the k-th level of the current number of levels.
A constraint model of intrinsic mode components:
performing Hilbert transform on each layer Intrinsic Mode Function (IMF) and mixing with index
Figure BDA00020935478600000612
The bandwidth of the eigenmode function is estimated by the square norm of the gradient, and a constraint model is obtained:
Figure BDA0002093547860000063
wherein the content of the first and second substances,
Figure BDA0002093547860000064
then means that the partial derivative, u, is calculated for tk(t) is the target eigenmode component, ωk(t) represents all center frequencies, f (t) represents echo signals; k is the current layering number; omegakIs the center frequency; δ (t) is the shock function; h is an imaginary unit.
Determining the target eigenmode components of the current layer:
introducing a Lagrange operator lambda (t) and a secondary penalty factor alpha, wherein the penalty factor alpha is 2000, and solving the problem of Lagrange 'saddle point' by an alternative direction multiplier method, namely alternately updating lambdan+1、
Figure BDA0002093547860000065
And seeking an optimal solution of the constraint variation problem. After Parseval/Plancherel Fourier equidistant transformation, replacing omega with omega to omega-omegakAnd converting the obtained result into a negative frequency interval integral form to obtain a modal component updating formula and a central frequency updating formula, and updating the intrinsic modal components of each layer to obtain the target intrinsic modal component of the current layer.
Figure BDA0002093547860000066
Figure BDA0002093547860000067
Wherein:
Figure BDA0002093547860000068
n is the cycle number; k is the kth layer;
Figure BDA0002093547860000069
to represent
Figure BDA00020935478600000610
The Fourier transform of (1) is obtained by n +1 times of circulation of the kth layer;
Figure BDA00020935478600000611
is the value of the echo signal f (t) after Fourier transformation; alpha is a constant;
Figure BDA0002093547860000071
to represent
Figure BDA0002093547860000072
The Fourier transform of (1) is obtained by n times of circulation of the d-th layer, wherein d is not equal to k; omega is frequency;
Figure BDA0002093547860000073
is the k-th layer and is the modal center frequency at cycle n + 1;
Figure BDA0002093547860000074
representing the Lagrangian λn(t) Fourier transform, obtained for the nth cycle; τ is a time constant; k is the current number of tiers.
Calculate out
Figure BDA0002093547860000075
Then, the eye is obtained by Fourier inverse transformationEigenmode component uk(t)。
Until it meets
Figure BDA0002093547860000076
Accuracy of discrimination e 10-7And ending the loop, outputting K components, otherwise, keeping the iteration when n is n + 1.
(2) Determining the number of decomposition layers:
calculating the kurtosis values of all layers in the current layering process.
Figure BDA0002093547860000077
Wherein D is a kurtosis value; μ is the mean of the echo signals f (t); σ is the standard deviation of the echo signal f (t).
Calculating the cross-correlation coefficient between each layer of intrinsic mode component and the original signal:
Figure BDA0002093547860000078
wherein, T is the length of the echo signal f (T); u. ofk(t) is the target eigenmode component of the kth layer at the current delamination;
Figure BDA0002093547860000079
is uk(t) mean value;
Figure BDA00020935478600000710
is the mean of f (t); ζ is a preset threshold.
According to the Cauchy-Schwarz inequality | C | ≦ 1, the following can be found:
when u iskWhen (t) ═ f (t), C ═ 1; when u iskWhen (t) — f (t), C ═ 1, the phases are 180 degrees apart. When C is present>0 is said to be positively correlated with the two signals, when C<0 is referred to as a negative correlation of the two signals.
Determining the layering number:
calculating the weighted kurtosis value of each layer during current layering, setting a threshold value as 1, stopping the decomposition if the minimum value of the weighted kurtosis of each layer after decomposition is smaller than the threshold value, and taking the value of K as K-1; otherwise, K is equal to K +1, and the decomposition is continued until the stop condition is met.
Kw＝sgn(C)D|C|r＜ζ；
Wherein, KwIs a weighted kurtosis value; r is an exponent of C, a positive real number is taken, indexes of an input signal and an output signal are adjusted, and r is usually 1; d is a kurtosis value; ζ is a preset threshold value of 1.
As a verification, as shown in fig. 2, a set of data is extracted, weighted kurtosis values of intrinsic mode components (IMFs) of respective layers are calculated as shown in table 1, and finally, when K is 6, a component with a weighted kurtosis value smaller than 1 occurs, so K is 5.
TABLE 1 IMF component weighted kurtosis values for each layer
Figure BDA0002093547860000081
In an embodiment of the present invention, S103 includes:
and denoising the target eigenmode components of each layer by using a wavelet threshold method to obtain denoised target signals.
By way of example: selecting a wavelet basis as sym4 and determining the number of decomposition layers as 5 for the decomposed target intrinsic mode components (IMF) of each layer, and performing wavelet transformation on the target intrinsic mode components of each layer to obtain a low-frequency wavelet coefficient and a high-frequency wavelet coefficient.
Secondly, selecting a maximum and minimum threshold, and applying a hard threshold function to perform threshold processing on the decomposed high-frequency wavelet coefficient.
And thirdly, performing wavelet reconstruction on the low-frequency wavelet coefficient and the processed high-frequency wavelet coefficient to obtain a signal subjected to noise reduction on each layer of intrinsic mode component. And recombining the processed target intrinsic mode components (IMF) of each layer to finish noise reduction.
As a verification, the noise reduction comparison results are shown in table 2 below:
TABLE 2 comparison of noise reduction results
Figure BDA0002093547860000082
As can be seen from Table 2, the method of the present invention has a good noise reduction effect. The data is subjected to noise reduction, and the comparison results are shown in fig. 2, fig. 3 and fig. 4, and it can be seen from the waveforms in fig. 2 that a large amount of noise signals exist in the original signals, and the characteristics of the reflected signals are difficult to judge. It can be seen from fig. 3 that the wavelet de-noising waveform is smooth, but the bottom echo is lost.
As shown in fig. 5, in the embodiment of the present invention, before S104, the method further includes:
s1101, obtaining the constructed neural network, as shown in FIG. 6.
S1102, a sample set is obtained, and variation modal decomposition and noise reduction processing are carried out on samples in the sample set to obtain a sample signal set.
And S1103, inputting a group of sample signals in the sample signal set into the constructed neural network to obtain the output of the hidden layer in the training process.
And S1104, based on the output of the hidden layer, deleting the nodes meeting the second preset condition, splitting the nodes meeting the third preset condition, and obtaining the neural network after the training.
And S1105, determining whether the neural network after the training meets a second convergence condition.
And if the second convergence condition is met, the neural network after the training is the target neural network.
If the second convergence condition is not met, updating the weight and the bias value of the neural network after the training through an Elman neural network weight updating formula; and selecting a group of untrained sample signals from the sample signal set, inputting the untrained sample signals into the neural network updated at this time, and carrying out next training until the trained neural network meets a second convergence condition.
In the present embodiment, S1101 includes:
and performing three-layer wavelet packet decomposition on the noise-reduced target signal, namely the noise-reduced target eigenmode component, expressing the decomposition result according to an energy mode (namely a wavelet packet energy spectrum), extracting the energy of each frequency band, normalizing, and forming an anchor rod defect characteristic value with the dimension of 8 as the input of the defect identification of the network.
By way of example: the anchor rod anchoring quality signal after the noise reduction of the J-th group in 560 groups randomly extracted is FJ(t) for FJ(t) for the mth signal of the kth layer after three-layer wavelet packet decomposition
Figure BDA0002093547860000091
It is shown that,
Figure BDA0002093547860000092
the length of (a) is lambda, and the wavelet packet decomposition energy formula is as follows:
Figure BDA0002093547860000093
wherein the content of the first and second substances,
Figure BDA0002093547860000094
the decomposition energy of the mth signal in the jth layer of the jth group; k represents the number of decomposition layers, k is 3; m represents the position of the subband, i.e. the mth signal of the third layer; j represents the anchor rod anchoring quality signal after J group noise reduction. From the principle of conservation of energy, it is known that:
Figure BDA0002093547860000101
wherein, E [ F ]J(t)]And (4) the total energy of the anchor rod anchoring quality signals subjected to the J-th group noise reduction.
Normalization to construct a defect eigenvector matrix:
Figure BDA0002093547860000102
e (m) is a defect feature vector matrix.
Then we can get the anchor defect eigenvalue matrix with dimension 8 as the input of the network.
For ElmaThe n neural network parameters are initialized, the structure and parameters of the network are set as follows, the number of input layer nodes is set to be 8, the number of hidden layer initial nodes is set to be 18, the number of output layer nodes is set to be 4, the network pruning threshold value is c equal to 1.70, the increasing threshold value d equal to 1.60, the network learning rate is 0.08, and the hidden layer bias beta is set to be1Output layer bias β 0.12And a is the self-feedback factor a of the receiving layer is 0.08, and each layer of the network is weighted to be a random number smaller than 1.
In the embodiment of the present invention, the sample set in S1102 is:
640 groups of data are respectively collected for the four types of defect anchor rods, after the denoising operation is completed, three-layer wavelet packet decomposition is carried out on the data, 8-dimensional defect signal characteristic values are extracted, 560 groups of characteristic value data sets which are randomly extracted are used as training samples of the constructed neural network, and the constructed neural network is trained.
In an embodiment of the present invention, S1103 includes:
obtaining parameters such as output values of each layer of the network, including
x(p)＝f(W1xc(p)+W2(u(p-1))+β1)；
xc(p)＝x(p-1)+a·xc(p-1)；
y(p)＝g(W3x(p)+β2)；
Wherein the content of the first and second substances,
Figure BDA0002093547860000103
Figure BDA0002093547860000111
wherein f (x) is a hidden layer activation function, g (x) is an output layer activation function, W1For the connection of the weight matrix between the hidden layer and the anchor layer, W2For connecting the weight matrix between the hidden layer and the input layer, W3For the connection of the weight matrix between the output layer and the hidden layer, x (p) for the pth output of the hidden layer, y (p) for the networkOutput of p-th group, xc(p) a pth group output for the receiving layer; u (p-1) is the p-1 group (previous group) input of the network, namely the p-1 group characteristic data of the anchor defect characteristic signal data set; x (p-1) is the p-1 group (previous group) output of the hidden layer, xc(p-1) output for the p-1 group of receiving layers; beta is a1Outputting a bias for the hidden layer; beta is a2Outputting a bias for the output layer; a is a self-feedback factor of the receiving layer, and the value a is 0.1; x is the number ofh inAn input that is a hidden layer; x is the number ofo inIs an input to the output layer.
In an embodiment of the present invention, S1104 includes:
s11041, based on the output of each hidden layer, obtaining the contribution degree of each node of each hidden layer, the pruning threshold value of the current neural network and the growth threshold value of the current neural network.
S11042, comparing the contribution degree of each node of each layer of hidden layer with the pruning threshold value, and deleting the hidden layer node with the contribution degree smaller than the pruning threshold value and the carrying layer node corresponding to the hidden layer node.
S11043, comparing the contribution degree of each node of each layer of hidden layer with the increase threshold, splitting the hidden layer node with the contribution degree larger than the increase threshold and the carrying layer node corresponding to the hidden layer node, and obtaining the neural network after the training.
In the embodiment of the present invention, the contribution degree of each node of the hidden layer is:
Figure BDA0002093547860000112
wherein, SConjContribution of node j which is a hidden layer;
Figure BDA0002093547860000113
connecting weight values between a node j of the hidden layer and a node i of the output layer; z is a radical ofj(p) a pth set of output values for node j of the hidden layer; s is the s-th group of data of the feature data set; n is the total group number of the characteristic data set; m is the number of nodes of the output layer；
The pruning threshold of the current neural network is:
Figure BDA0002093547860000114
wherein the content of the first and second substances,
Figure BDA0002093547860000121
SACon is the average contribution of the hidden layer; pth is the pruning threshold of the current neural network; n is the number of nodes of the hidden layer; c is a pruning constant, a constant greater than 1;
the growth threshold of the current neural network is:
Gth＝d·SACon；
wherein, Gth is the growth threshold of the current neural network; d is a growth constant, a constant greater than 1.
As shown in fig. 6, in an embodiment of the present invention, S11042 includes:
according to
Figure BDA0002093547860000122
Figure BDA0002093547860000123
Figure BDA0002093547860000124
Figure BDA0002093547860000125
Wherein j is deleted hidden layer node, jcIn order for a bearer node to be deleted,
Figure BDA0002093547860000126
for the connection weights between the hidden layer j node and the input layer q node,
Figure BDA0002093547860000127
for the output layer i node and the hidden layer j node connection weight,
Figure BDA0002093547860000128
is the connection weight, w, between the node of the hidden layer j and the node of the receiving layer lz,jc 1For hidden layer z nodes and socket layer jcAnd connecting the weights among the nodes.
And deleting hidden layer nodes with contribution degrees smaller than the pruning threshold of the nodes and carrying layer nodes corresponding to the hidden layer nodes.
As shown in fig. 7, in an embodiment of the present invention, S11043 includes: according to
Figure BDA0002093547860000129
Figure BDA00020935478600001210
b is a random number between 0 and 1 and is 1-a a;
wherein, the hidden layer j' and the k nodes are the nodes after the hidden layer node j is split, wjq 2As a connection weight, w, between the hidden layer j node and the input layer q node before splittingj′q 2For the connection weight, w, between the hidden layer j' node and the input layer q node after the network growthkq 2In order to newly increase the connection weight between the k node of the hidden layer and the q node of the input layer, a and b are node splitting coefficients.
Figure BDA0002093547860000131
1,2 … n and l ≠ jc′,kc(ii) a n is the number of nodes of the bearing layer;
Figure BDA0002093547860000132
1,2 … n and l ≠ jc′,kc(ii) a n is the number of nodes of the bearing layer;
Figure BDA0002093547860000133
z is 1,2 … n; n is the number of hidden layer nodes;
Figure BDA0002093547860000134
z is 1,2 … n; n is the number of hidden layer nodes;
wherein, the receiving layer jc′And kcThe node is a bearing layer node jcNode after splitting, wjl 1The connection weight, w, between the hidden layer j node and the carrying layer l node before the network structure is increasedj′l 1The connection weight value w between the hidden layer j' node and the receiving layer l node after the network node is increasedkl 1The connection weight between the node k of the hidden layer and the node l of the bearing layer is increased; w is az,jc 1Implicit layer z nodes and socket layer j before network growthcConnection weight between nodes, wz,jc′ 1For growing implicit layer node z and carry layer node jc′Is connected with weight value wz,kc 1For growing implicit layer node z and bearer layer node kcThe weight value is connected between the two nodes.
After the network node is split and increased, the splitting change implies that the nodes j' and k of the layer output as,
Figure BDA0002093547860000135
Figure BDA0002093547860000136
Figure BDA0002093547860000137
Figure BDA0002093547860000138
m is the number of output layer nodes;
Figure BDA0002093547860000139
m is the number of output layer nodes;
wherein x isj′(p) p-th set of outputs, x, for a new split node j' of the post-split hidden layerk(p) p-th set of outputs, y, for a new split node k of the post-split hidden layeriIs the output of the output layer node i, yi expIs the desired output of the output layer node i, eiError, x, of network training output layer node ikFor the output of the hidden layer node k, xj'output as hidden layer node j', wik 3For the connection weight between the output layer node i after splitting and the new splitting node k of the hidden layer,
Figure BDA00020935478600001310
the connection weight value of the output layer node i after splitting and the new splitting node j' of the hidden layer is obtained;
Figure BDA00020935478600001311
the connection weight of the output layer node i and the hidden layer node j before splitting; w is aj′l 1The connection weight of a new splitting node j' of the hidden layer after splitting and the receiving layer l is obtained;
Figure BDA0002093547860000141
the connection weight of a new splitting node k of the hidden layer after splitting and the receiving layer l is set;
Figure BDA0002093547860000142
connecting weights between the nodes j' of the hidden layer and the nodes q of the input layer after splitting;
Figure BDA0002093547860000143
connecting weights between the nodes k of the hidden layer and the nodes q of the input layer after splitting; x is the number ofc,l(p) as the p-th group output of the bearer node l；uq(p-1) is the p-1 th group input to the network.
And obtaining the neural network after the training by using the hidden layer nodes with the contribution degree of the split nodes larger than the increase threshold value and the carrying layer nodes corresponding to the hidden layer nodes.
In an embodiment of the present invention, S1105 includes:
calculating a network error:
Figure BDA0002093547860000144
wherein MSE is the network error; m is the total number of elements in the network output matrix in one iteration, yiFor the actual output of the network output layer node i,
Figure BDA0002093547860000145
expecting an output for a network output layer node i; and m is the number of nodes of the network output layer.
And judging whether the network error is smaller than the cycle jump-out error, finishing network cycle when a jump-out condition, namely a second convergence condition, is met, and storing the network structure and other parameters.
MSE < ψ, where ψ is the cycle out error.
And when the error does not meet the cycle ending condition, updating the weight value and the offset value of the network according to the following formula.
Figure BDA0002093547860000146
Figure BDA0002093547860000147
Figure BDA0002093547860000148
Figure BDA0002093547860000149
Figure BDA00020935478600001410
Figure BDA00020935478600001411
Wherein
g'(x)＝1；
f'j(x)＝fj(x)(1-fj(x))；
Figure BDA0002093547860000151
Wherein, Δ wij3 is the update quantity of the connection weight between the node of the output layer i and the node of the hidden layer j,
Figure BDA0002093547860000152
for the connection weight value updating quantity between the j node of the hidden layer of the network and the q node of the input layer,
Figure BDA0002093547860000153
updating the connection weight between the node of the hidden layer j and the node of the receiving layer l; x is the number ofc,l(p) is the output of the p-th group of l nodes of the bearing layer, and E is the reverse transfer error function of the network; eta is the learning rate of the neural network, fj' (x) is the hidden layer activation function fj(x) The partial derivative of the output at the hidden layer j node, g' (x), is the derivative of the output layer activation function; Δ w is a weight update quantity matrix of the network;
Figure BDA0002093547860000154
representing the partial derivative of the network reverse transfer error function to the weight matrix;
Figure BDA0002093547860000155
weight update for output layer node iA factor; x is the number ofj(p) a pth group output for hidden layer node j; m is the number of nodes of the output layer; n is the number of hidden layer nodes;
Figure BDA0002093547860000156
updating a factor for the weight value of the hidden layer node j; u. ofq(P-1) output layer group P-1 inputs; r is the number of input layer nodes;
Figure BDA0002093547860000157
for the pth group of outputs x to the hidden layer node jj(p) determining a weight value related to a connection between a hidden layer node j and a bearer layer node l
Figure BDA0002093547860000158
Partial derivatives of (a);
Figure BDA0002093547860000159
a p-th group of expected outputs for a network output layer node i; y isi(p) p-th group of actual outputs for network output layer node i; g' (. cndot.) is the derivative of the output layer activation function on the output layer output; f. ofj' (. -) is the derivative of the hidden layer activation function on the output of the hidden layer node j; y is an output matrix of the network in the one-time iteration process of the network; y isexpAn expected output matrix of the network for one iteration process of the network; t is transposing the matrix.
By way of verification, the detection result of the target neural network (namely the structural self-organizing Elman neural network) of the invention on the anchor rod anchoring defect, the detection result of the pruning-growth BP neural network defect and the detection result of the fixed structural Elman neural network defect are shown in table 3.
TABLE 3 results of Elman neural network detection of self-organized and fixed structures
Figure BDA00020935478600001510
Figure BDA0002093547860000161
As can be seen from table 3, the structural self-organizing Elman neural network can complete structural self-organizing adjustment during training, and has higher defect detection accuracy compared with the fixed structural Elman neural network and the pruning-growing BP neural network. The structural self-organization Elman neural network is adopted, the network structure can be adjusted in a self-adaptive mode according to different detection conditions, the method is high in adaptability to anchor rod anchoring quality detection and high in detection and identification accuracy, and the method is suitable for anchor rod anchoring nondestructive detection in complex engineering environments.
As shown in fig. 8, in the embodiment of the present invention, the echo signal is initial information acquired by a preset echo signal acquisition device and used for detecting the anchor defect of the anchor rod;
wherein, echo signal acquisition device includes:
signal generating means, magnetic field generating means and coil 5; the coil 5 is wound on the anchor rod 1, the anchor 2 is arranged outside the anchor rod 1, the magnetic field generating device is arranged on the anchor rod, the coil is arranged in the magnetic field generating device, and the signal generating device is electrically connected with the coil;
and exciting signals sent by the signal generating device enable the anchor rod to excite ultrasonic guided waves under the action of the magnetic field generating device, and the ultrasonic guided waves reflected by the anchoring of the anchor rod are echo signals.
In this embodiment, the echo signal acquisition device further includes a power amplifier and a duplexer, the power amplifier is respectively connected to the signal generation device and the duplexer, and the duplexer is connected to the coil 5.
In this embodiment, the magnetic field generating device includes permanent magnet 3 and yoke 4, and permanent magnet 3 is in even symmetric distribution with stock 1 as the axis.
The signal generating device transmits an excitation signal to the signal amplifier, the signal amplifier transmits the amplified excitation signal to the duplexer, the duplexer transmits the amplified excitation signal to the coil, and under the action of a magnetic field generated by the magnetic field generating device, a magnetostrictive effect is generated in the detected anchor rod, so that ultrasonic guided waves are excited in the anchor rod; the signal returned in the anchoring of the anchor rod enables the coil to generate induced voltage under the action of the magnetic field, namely the echo signal.
In this embodiment, the signal generating device is Tektronix AFG3052, the signal amplifier is AE TECRON 7224, and the duplexer is RITEC RDX-EM2DIPLEX PRE-AMPLIFIER.
In this embodiment, the echo signal acquisition device is a self-excited self-retracting device based on a magnetostrictive mechanism.
As shown in fig. 9 to 11, in the present embodiment, the anchor defect mainly includes: a front-end void defect, a back-end void defect and a double-void defect, the front-end void defect being a defect having a defect position 6 at the front end.
It should be understood that, the sequence numbers of the steps in the foregoing embodiments do not imply an execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.
Example 2:
as shown in fig. 12, an embodiment of the present invention provides a device 100 for detecting anchor defect for performing the method steps in the embodiment corresponding to fig. 1, wherein the device 100 for detecting anchor defect is connected to a duplexer, and comprises:
the signal acquisition module 110 is configured to acquire an echo signal, where the echo signal is acquired initial information used for detecting the anchor defect of the anchor rod;
the first calculation module 120 is configured to decompose the echo signal by using a variational modal decomposition method to obtain a target eigenmode component corresponding to each decomposed layer;
the second calculating module 130 is configured to perform noise reduction processing on the target eigenmode components corresponding to each layer to obtain a noise-reduced target signal;
and the detection module 140 is configured to input the target signal into a target neural network to obtain a defect detection result of anchor rod anchoring.
In an embodiment of the present invention, the first calculation module 120 includes:
the first calculation unit is used for carrying out initial layering on the echo signals, circularly updating the initial eigen-modal component of the layer based on each layer, and obtaining the target eigen-modal component of the current layer after the updated eigen-modal component meets a first convergence condition;
a second calculating unit, configured to calculate a weighted kurtosis value of each layer after the current layering based on the target eigenmode component of the current layer;
the first judgment unit is used for stopping layering if the weighted kurtosis value of each layer after the current layering meets a first preset condition, and obtaining the final layering number as the number of layers after the previous layering;
a second judging unit, configured to continue layering until the weighted kurtosis values of the layered layers satisfy a first preset condition if the weighted kurtosis values of the layered layers do not satisfy the first preset condition;
wherein the first preset condition is as follows:
Kw＝sgn(C)D|C|r<ζ；
wherein the content of the first and second substances,
Figure BDA0002093547860000181
sgn () is a sign function to ensure that the phases of the output signal and the original signal are consistent as much as possible; kwIs a weighted kurtosis value; r is an index of C; d is a kurtosis value; μ is the mean of the echo signals f (t); σ is the standard deviation of the echo signal f (t); t is time; t is the length of the echo signal f (T); u. ofk(t) is the target eigenmode component of the kth layer at the current delamination;
Figure BDA0002093547860000182
is uk(t) mean value;
Figure BDA00020935478600001814
is the mean of f (t); zeta is a preset threshold; c is the cross-correlation coefficient between the two signals.
Wherein the target eigenmode component
Figure BDA0002093547860000183
Is composed of
Figure BDA0002093547860000184
Obtaining through inverse Fourier transform:
Figure BDA0002093547860000185
wherein:
Figure BDA0002093547860000186
n is the cycle number; k is the kth layer;
Figure BDA0002093547860000187
to represent
Figure BDA0002093547860000188
The Fourier transform of (1) is obtained by n +1 times of circulation of the kth layer;
Figure BDA0002093547860000189
is the value of the echo signal f (t) after Fourier transformation; alpha is a constant;
Figure BDA00020935478600001810
to represent
Figure BDA00020935478600001811
The Fourier transform of (1) is obtained by n times of circulation of the d-th layer, wherein d is not equal to k; omega is frequency;
Figure BDA00020935478600001812
is the k-th layer and is the modal center frequency at cycle n + 1;
Figure BDA00020935478600001813
representing the Lagrangian λn(t) Fourier transform, obtained for the nth cycle; τ is a time constant; k is the current number of tiers.
The first convergence condition is as follows:
Figure BDA0002093547860000191
wherein: 10 ∈ ═ 10-7。
In an embodiment of the present invention, the second calculation module 130 includes:
and denoising the target eigenmode components of each layer by using a wavelet threshold method to obtain denoised target signals.
In an embodiment of the present invention, the connection with the detection module 140 further includes:
the network building module is used for obtaining a built neural network;
the system comprises a sample acquisition module, a data processing module and a data processing module, wherein the sample acquisition module is used for acquiring a sample set, and carrying out variation modal decomposition and noise reduction processing on samples in the sample set to obtain a sample signal set;
the first output module is used for inputting a group of sample signals in the sample signal set into the constructed neural network to obtain the output of the hidden layer in the training process;
the second output module is used for deleting the nodes meeting the second preset condition based on the output of the hidden layer, splitting the nodes meeting the third preset condition and obtaining the neural network after the training;
the judging module is used for determining whether the neural network after the training meets a second convergence condition;
if the second convergence condition is met, the neural network after the training is the target neural network;
if the second convergence condition is not met, updating the weight and the bias value of the neural network after the training through an Elman neural network weight updating formula; and selecting a group of untrained sample signals from the sample signal set, inputting the untrained sample signals into the neural network updated at this time, and carrying out next training until the trained neural network meets a second convergence condition.
In an embodiment of the present invention, the second output module includes:
the third calculation unit is used for obtaining the contribution degree of each node of each hidden layer, the pruning threshold value of the current neural network and the growth threshold value of the current neural network based on the output of each hidden layer;
a node deleting unit, configured to compare the contribution degree of each node of each hidden layer with the pruning threshold, and delete a hidden layer node whose contribution degree is smaller than the pruning threshold and a carrying layer node corresponding to the hidden layer node;
and the splitting node unit is used for comparing the contribution degree of each node of each layer of hidden layer with the increase threshold, and obtaining the neural network after the training, wherein the contribution degree of the splitting node is greater than the hidden layer node of the increase threshold and the carrying layer node corresponding to the hidden layer node.
In the embodiment of the present invention, the contribution degree of each node of the hidden layer is:
Figure BDA0002093547860000201
wherein, SConjContribution of node j which is a hidden layer;
Figure BDA0002093547860000202
connecting weight values between a node j of the hidden layer and a node i of the output layer; z is a radical ofj(p) a pth set of output values for node j of the hidden layer; s is the s-th group of data of the feature data set; n is the total group number of the characteristic data set; m is the number of nodes of the output layer;
the pruning threshold of the current neural network is:
Figure BDA0002093547860000203
wherein the content of the first and second substances,
Figure BDA0002093547860000204
SACon is the average contribution of the hidden layer; pth is the pruning threshold of the current neural network; n is the number of nodes of the hidden layer; c is a pruning constant;
the growth threshold of the current neural network is:
Gth＝d·SACon；
wherein, Gth is the growth threshold of the current neural network; d is a growth constant.
In an embodiment of the present invention, the second convergence condition is:
MSE<ψ；
wherein the content of the first and second substances,
Figure BDA0002093547860000205
MSE is the network error; m is the total number of elements of the output matrix after one iteration of the network is finished, yiFor the network output layer node i actual output, yi expExpecting an output for a network output layer node i; m is the number of output layer nodes; psi is the cycle slip-out error.
In the embodiment of the invention, the echo signal is initial information which is acquired by a preset echo signal acquisition device and used for detecting the anchor defect of the anchor rod;
wherein, echo signal acquisition device includes:
a signal generating device, a magnetic field generating device and a coil; the coil is wound on the anchor rod, the magnetic field generating device is arranged on the anchor rod, the coil is arranged in the magnetic field generating device, and the signal generating device is electrically connected with the coil;
and exciting signals sent by the signal generating device enable the anchor rod to excite ultrasonic guided waves under the action of the magnetic field generating device, and the ultrasonic guided waves reflected by the anchoring of the anchor rod are echo signals.
It is clearly understood by those skilled in the art that, for convenience and brevity of description, the above-mentioned division of the functional modules is merely used as an example, in practical applications, the above-mentioned function distribution may be performed by different functional modules according to needs, that is, the internal structure of the device for detecting an anchoring defect of the anchor rod is divided into different functional modules, so as to perform all or part of the above-mentioned functions. Each functional module in the embodiments may be integrated in one processing unit, or each unit may exist alone physically, or two or more units are integrated in one unit, and the integrated module may be implemented in a form of hardware, or in a form of software functional unit. In addition, specific names of the functional modules are only used for distinguishing one functional module from another, and are not used for limiting the protection scope of the application. The specific working process of the module in the device for detecting anchor rod anchoring defects may refer to the corresponding process in the foregoing method embodiment 1, and is not described herein again.
Example 3:
fig. 13 is a schematic diagram of a terminal device according to an embodiment of the present invention. As shown in fig. 13, the terminal device 136 of this embodiment includes: a processor 1360, a memory 1361, and computer programs 1362 stored in the memory 1361 and executable on the processor 1360. The processor 1360, when executing the computer program 1362, implements steps in the embodiments as described in embodiment 1, such as steps S101 through S104 shown in fig. 1. Alternatively, the processor 1360, when executing the computer program 1362, implements the functions of the modules/units in the system embodiments as described in embodiment 2, such as the functions of the modules 110 to 140 shown in fig. 12.
The terminal device 136 refers to a terminal with data processing capability, and includes but is not limited to a computer, a workstation, a server, and even some Smart phones, palmtop computers, tablet computers, Personal Digital Assistants (PDAs), Smart televisions (Smart TVs), and the like with excellent performance. The terminal device is generally installed with an operating system, including but not limited to: windows operating system, LINUX operating system, Android (Android) operating system, Symbian operating system, Windows mobile operating system, and iOS operating system, among others. While specific examples of terminal equipment 136 are listed above in detail, those skilled in the art will appreciate that terminal equipment is not limited to the listed examples.
The terminal devices may include, but are not limited to, a processor 1360, a memory 1361. Those skilled in the art will appreciate that fig. 13 is merely an example of the terminal device 136 and does not constitute a limitation of the terminal device 136 and may include more or less components than those shown, or combine certain components, or different components, for example, the terminal device 136 may also include an input-output device, a network access device, a bus, etc.
The Processor 1360 may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components, etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The storage 1361 may be an internal storage unit of the terminal device 136, such as a hard disk or a memory of the terminal device 136. The memory 1361 may also be an external storage device of the terminal device 136, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like provided on the terminal device 136. Further, the memory 1361 may also include both an internal storage unit and an external storage device of the terminal device 136. The memory 1361 is used for storing the computer programs and other programs and data required by the terminal device 136. The memory 1361 may also be used to temporarily store data that has been output or is to be output.
Example 4:
an embodiment of the present invention further provides a computer-readable storage medium, in which a computer program is stored, and the computer program, when executed by a processor, implements the steps in the embodiments described in embodiment 1, for example, step S101 to step S104 shown in fig. 1. Alternatively, the computer program, when executed by a processor, implements the functions of the respective modules/units in the respective system embodiments as described in embodiment 2, for example, the functions of the modules 110 to 140 shown in fig. 12.
The computer program may be stored in a computer readable storage medium, which when executed by a processor, may implement the steps of the various method embodiments described above. Wherein the computer program comprises computer program code, which may be in the form of source code, object code, an executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, usb disk, removable hard disk, magnetic disk, optical disk, computer Memory, Read-Only Memory (ROM), Random Access Memory (RAM), electrical carrier wave signals, telecommunications signals, software distribution medium, and the like.
In the above embodiments, the description of each embodiment has a respective emphasis, and embodiments 1 to 4 may be combined arbitrarily, and a new embodiment formed by combining is also within the scope of the present application. For parts which are not described or illustrated in a certain embodiment, reference may be made to the description of other embodiments.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
In the embodiments provided in the present invention, it should be understood that the disclosed terminal device and method may be implemented in other ways. For example, the above-described system/terminal device embodiments are merely illustrative, and for example, the division of the modules or units is only one logical division, and there may be other divisions when actually implemented, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The above-mentioned embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present invention, and are intended to be included within the scope of the present invention.

Claims (8)

1. A method for detecting anchor defects of a rock bolt is characterized by comprising the following steps:
acquiring echo signals, wherein the echo signals are acquired initial information used for detecting the anchor defect of the anchor rod;
decomposing the echo signal by using a variational modal decomposition method to obtain target eigenmode components corresponding to each decomposed layer;
denoising the target eigenmode components corresponding to each layer to obtain a denoised target signal;
inputting the target signal into a target neural network to obtain a defect detection result of anchor rod anchoring;
the decomposing the echo signal by using the variational modal decomposition method to obtain the intrinsic modal components corresponding to each decomposed layer comprises the following steps:
carrying out initial layering on the echo signals, circularly updating the initial eigenmode component of the layer based on each layer, and obtaining the target eigenmode component of the current layer after the updated eigenmode component meets a first convergence condition; calculating the weighted kurtosis value of each layer after the current layering based on the target eigenmode component of the current layer; if the weighted kurtosis value of each layer after the current layering meets a first preset condition, stopping layering, and obtaining the final layering number which is the number of layers after the previous layering; if the weighted kurtosis value of each layer after the current layering does not meet the first preset condition, continuing to layer until the weighted kurtosis value of each layer after the layering meets the first preset condition;
before inputting the target signal into the target neural network, the method further comprises:
acquiring a constructed neural network; acquiring a sample set, and performing variational modal decomposition and noise reduction processing on samples in the sample set to obtain a sample signal set;
inputting a group of sample signals in the sample signal set into the constructed neural network to obtain the output of a hidden layer in the training process;
based on the output of the hidden layer, deleting the nodes meeting a second preset condition, splitting the nodes meeting a third preset condition, and obtaining the neural network after the training; the method specifically comprises the following steps: based on the output of each hidden layer, obtaining the contribution degree of each node of each hidden layer, the pruning threshold value of the current neural network and the growth threshold value of the current neural network; comparing the contribution degree of each node of each hidden layer with the pruning threshold, and deleting the hidden layer nodes with the contribution degree smaller than the pruning threshold and the carrying layer nodes corresponding to the hidden layer nodes; comparing the contribution degree of each node of each hidden layer with the increase threshold, and acquiring the neural network after the training, wherein the contribution degree of the split node is greater than the hidden layer node of the increase threshold and the carrying layer node corresponding to the hidden layer node;
specifically, the contribution degree of each node of the hidden layer is as follows:
Figure FDA0003335113810000021
wherein, SConjContribution of node j which is a hidden layer; w is aij 3Connecting weight values between a node j of the hidden layer and a node i of the output layer; z is a radical ofj(p) is the pth set of output values for node j of the hidden layer; s is the s-th group of data of the feature data set; n is the total group number of the characteristic data set; m is the number of nodes of the output layer;
the pruning threshold of the current neural network is:
Figure FDA0003335113810000022
wherein the content of the first and second substances,
Figure FDA0003335113810000023
SACon is the average contribution of the hidden layer; pth is the pruning threshold of the current neural network; n is the number of nodes of the hidden layer; c is a pruning constant;
the growth threshold of the current neural network is:
Gth＝d·SACon；
wherein, Gth is the growth threshold of the current neural network; d is a growth constant.
2. The method of detecting defects in anchoring of a rock bolt according to claim 1,
the target eigenmode component
Figure FDA0003335113810000024
Is composed of
Figure FDA0003335113810000025
Obtaining through inverse Fourier transform:
Figure FDA0003335113810000026
wherein:
Figure FDA0003335113810000031
n is the cycle number; k is the kth layer;
Figure FDA0003335113810000032
to represent
Figure FDA0003335113810000033
The Fourier transform of (1) is obtained by n +1 times of circulation of the kth layer;
Figure FDA0003335113810000034
is the value of the echo signal f (t) after Fourier transformation; alpha is a constant;
Figure FDA0003335113810000035
to represent
Figure FDA0003335113810000036
The Fourier transform of (1) is obtained by n times of circulation of the d-th layer, wherein d is not equal to k; omega is frequency;
Figure FDA0003335113810000037
is the k-th layer and is the modal center frequency at cycle n + 1;
Figure FDA0003335113810000038
representing the Lagrangian λn(t) Fourier transform, obtained for the nth cycle; τ is a time constant; k is the current layering number;
the first convergence condition is as follows:
Figure FDA0003335113810000039
wherein: 10 ∈ ═ 10-7；
The first preset condition is as follows:
Kw＝sgn(C)D|C|r<ζ；
wherein the content of the first and second substances,
Figure FDA00033351138100000310
sgn () is a sign function to ensure that the phases of the output signal and the original signal are consistent as much as possible; kwIs a weighted kurtosis value; r is an index of C; d is a kurtosis value; μ is the mean of the echo signals f (t); sigmaIs the standard deviation of the echo signal f (t); t is time; t is the length of the echo signal f (T); u. ofk(t) is the target eigenmode component of the kth layer at the current delamination;
Figure FDA00033351138100000311
is uk(t) mean value;
Figure FDA00033351138100000312
is the mean of f (t); zeta is a preset threshold; c is the cross-correlation coefficient between the two signals.
3. The method for detecting defects in anchoring of a rock bolt according to claim 1, wherein the step of denoising the intrinsic mode components corresponding to each layer to obtain a denoised target signal comprises:
and denoising the intrinsic mode components of each layer by using a wavelet threshold method to obtain a denoised target signal.
4. The method of detecting defects in anchoring of a rock bolt according to claim 1, further comprising, prior to inputting said target signal into a target neural network:
determining whether the neural network after the training meets a second convergence condition;
if the second convergence condition is met, the neural network after the training is the target neural network;
if the second convergence condition is not met, updating the weight and the bias value of the neural network after the training through an Elman neural network weight updating formula; and selecting a group of untrained sample signals from the sample signal set, inputting the untrained sample signals into the neural network updated at this time, and carrying out next training until the trained neural network meets a second convergence condition.
5. The method of detecting defects in anchoring of a rock bolt according to claim 4, wherein said second convergence condition is:
MSE<ψ；
wherein the content of the first and second substances,
Figure FDA0003335113810000041
MSE is the network error; m is the total number of elements in the network output matrix in one iteration, yiFor the network output layer node i actual output, yi expExpecting an output for a network output layer node i; m is the number of output layer nodes; psi is the cycle slip-out error.
6. The method for detecting the anchor rod anchoring defect of any one of claims 1 to 5, wherein the echo signal is initial information which is acquired by a preset echo signal acquisition device and is used for detecting the anchor rod anchoring defect;
wherein, echo signal acquisition device includes:
a signal generating device, a magnetic field generating device and a coil; the coil is wound on the anchor rod, the magnetic field generating device is arranged on the anchor rod, the coil is arranged in the magnetic field generating device, and the signal generating device is electrically connected with the coil;
and exciting signals sent by the signal generating device enable the anchor rod to excite ultrasonic guided waves under the action of the magnetic field generating device, and the ultrasonic guided waves reflected by the anchoring of the anchor rod are echo signals.
7. Terminal device, characterized in that it comprises a memory, a processor and a computer program stored in said memory and executable on said processor, said processor implementing the steps of the method of detection of bolting defects according to any of claims 1 to 6 when executing said computer program.
8. A computer-readable storage medium, characterized in that it stores a computer program which, when executed by a processor, carries out the steps of the method of detecting a bolt anchorage defect according to any one of claims 1 to 6.
CN201910511203.XA 2019-06-13 2019-06-13 Detection method for anchor rod anchoring defect and terminal equipment Active CN110208375B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910511203.XA CN110208375B (en) 2019-06-13 2019-06-13 Detection method for anchor rod anchoring defect and terminal equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910511203.XA CN110208375B (en) 2019-06-13 2019-06-13 Detection method for anchor rod anchoring defect and terminal equipment

Publications (2)

Publication Number Publication Date
CN110208375A CN110208375A (en) 2019-09-06
CN110208375B true CN110208375B (en) 2021-12-14

Family

ID=67792519

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910511203.XA Active CN110208375B (en) 2019-06-13 2019-06-13 Detection method for anchor rod anchoring defect and terminal equipment

Country Status (1)

Country Link
CN (1) CN110208375B (en)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110826598B (en) * 2019-10-16 2022-09-02 四川大学 Anchor rod anchoring compactness grade evaluation method based on convolutional neural network
CN110749650B (en) * 2019-10-16 2021-11-02 四川大学 Anchor rod anchoring compactness grade evaluation method based on support vector machine
CN110849968B (en) * 2019-11-05 2021-12-17 东南大学 Crane main beam damage acoustic emission nondestructive detection method based on self-adaptive optimization VMD
CN111723701B (en) * 2020-06-08 2022-05-20 西安交通大学 Underwater target identification method
CN114354969A (en) * 2021-09-23 2022-04-15 山东开泰抛丸机械股份有限公司 Speed measuring device for testing shot projection speed of shot blasting machine
CN114200016B (en) * 2021-10-18 2024-04-16 中国科学院武汉岩土力学研究所 Double-channel nondestructive testing method and related equipment for rock anchor rod
CN117091648A (en) * 2023-07-27 2023-11-21 石家庄铁道大学 Air-ground integrated construction ecological environment monitoring device and visual processing method

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10139376B2 (en) * 2016-03-31 2018-11-27 General Electric Company System for sensing and locating delamination
CN106053602A (en) * 2016-05-04 2016-10-26 石家庄铁道大学 A self-closed rock bolt nondestructive testing method based on a magnetostrictive effect
CN106778809A (en) * 2016-11-23 2017-05-31 安徽理工大学 A kind of blind source separation method based on improvement chicken group's algorithm
CN108845352B (en) * 2018-06-27 2019-09-27 吉林大学 Desert Denoising of Seismic Data method based on VMD approximate entropy and multi-layer perception (MLP)
CN109446928B (en) * 2018-10-10 2021-04-02 南京航空航天大学 Signal noise reduction method based on variational modal decomposition and minimum mean square error adaptive filter
CN109633431B (en) * 2019-01-09 2021-07-09 国网江苏省电力有限公司电力科学研究院 On-load tap-changer fault identification method based on vibration signal feature extraction

Also Published As

Publication number Publication date
CN110208375A (en) 2019-09-06

Similar Documents

Publication Publication Date Title
CN110208375B (en) Detection method for anchor rod anchoring defect and terminal equipment
KR101958674B1 (en) Actually-measured marine environment data assimilation method based on sequence recursive filtering three-dimensional variation
CN110865357B (en) Laser radar echo signal noise reduction method based on parameter optimization VMD
CN111580151B (en) SSNet model-based earthquake event time-of-arrival identification method
CN112213771B (en) Seismic wave impedance inversion method and device
Perelli et al. Best basis compressive sensing of guided waves in structural health monitoring
CN110174658B (en) Direction-of-arrival estimation method based on rank-dimension reduction model and matrix completion
CN113687433A (en) Bi-LSTM-based magnetotelluric signal denoising method and system
CN110071706A (en) Core pulse signal rising edge restoration methods, device and digitlization nuclear instrument
CN115280182A (en) Phase-based method for ultrasound examination
CN102915735A (en) Noise-containing speech signal reconstruction method and noise-containing speech signal device based on compressed sensing
CN110954860B (en) DOA and polarization parameter estimation method
CN104361192A (en) Sparse representation adaptive reconstruction method under compressed sensing analysis model
Rebillat et al. Damage localization in composite plates using canonical polyadic decomposition of Lamb wave difference signals tensor
Zadoianchuk et al. Dynamics of solutions of a class of second-order autonomous evolution inclusions
CN109558819B (en) Depth network lightweight method for remote sensing image target detection
Hu et al. Direct and inverse problems for electromagnetic scattering by a doubly periodic structure with a partially coated dielectric
CN108983160B (en) Compressed sensing radar multi-target parameter estimation method under low signal-to-noise ratio condition
Bagwe et al. Automated Radar Signal Analysis Based on Deep Learning
CN113447987B (en) Method, device and equipment for determining formation fluid saturation
CN113919388A (en) Electromechanical equipment fault diagnosis method and device integrating signal frequency spectrum amplitude modulation and deep learning
CN114630207A (en) Multi-sensing-node perception data collection method based on noise reduction self-encoder
CN110045362B (en) RNN-based radar target track identification method and device
CN108668234B (en) WIFI-based reverse car searching method for parking lot
CN110518588B (en) Filtering method, device, equipment and power electronic device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant