CN113343402A - Pipeline corrosion grade evaluation method based on multilayer convolution sparse coding - Google Patents

Pipeline corrosion grade evaluation method based on multilayer convolution sparse coding Download PDF

Info

Publication number
CN113343402A
CN113343402A CN202110768576.2A CN202110768576A CN113343402A CN 113343402 A CN113343402 A CN 113343402A CN 202110768576 A CN202110768576 A CN 202110768576A CN 113343402 A CN113343402 A CN 113343402A
Authority
CN
China
Prior art keywords
layer
sparse coding
multilayer
convolution
sparse
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.)
Granted
Application number
CN202110768576.2A
Other languages
Chinese (zh)
Other versions
CN113343402B (en
Inventor
华佳东
张晗
彭勃
高飞
童彤
梁振霖
林京
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.)
Beihang University
Original Assignee
Beihang 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 Beihang University filed Critical Beihang University
Priority to CN202110768576.2A priority Critical patent/CN113343402B/en
Publication of CN113343402A publication Critical patent/CN113343402A/en
Application granted granted Critical
Publication of CN113343402B publication Critical patent/CN113343402B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/18—Network design, e.g. design based on topological or interconnect aspects of utility systems, piping, heating ventilation air conditioning [HVAC] or cabling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/14—Pipes

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Geometry (AREA)
  • Software Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Physics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Evolutionary Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Hardware Design (AREA)
  • Medical Informatics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
  • Testing Resistance To Weather, Investigating Materials By Mechanical Methods (AREA)

Abstract

The invention provides a pipeline corrosion grade evaluation method based on multilayer convolution sparse coding, which comprises the following steps: s1: installing a first acceleration sensor and a second acceleration sensor on the outer surface of the pipeline, and knocking the concerned area by using a force hammer to obtain an impact response signal; s2: constructing a grade evaluation model based on multilayer convolution sparse coding; s3: acquiring force hammer actual measurement data of a pipeline to be tested, preprocessing the force hammer actual measurement data, inputting a time domain signal serving as a test set into the model trained in the step S2; s4: and giving out a pipeline corrosion grade evaluation result according to an evaluation model based on multilayer convolution sparse coding. The method adopts multilayer convolution sparse coding as a model, utilizes the oscillation attenuation function to extract effective characteristics in a multi-frequency band, can effectively extract weak damage characteristics in a vibration signal under pipeline corrosion, completes automatic corrosion grade evaluation, avoids the dependence of a traditional method on manual threshold selection, and is suitable for online pipeline corrosion monitoring.

Description

Pipeline corrosion grade evaluation method based on multilayer convolution sparse coding
Technical Field
The invention relates to the technical field of nondestructive testing, in particular to a pipeline corrosion grade evaluation method based on multilayer convolution sparse coding.
Background
The corrosion damage of pipelines is one of the main problems faced by industries such as petroleum, buildings and the like. The pipeline inevitably receives external force and environmental change's influence in the in service process, causes ageing problems such as corruption, wearing and tearing, and then arouses the accident. Therefore, developing the research of the nondestructive testing and evaluating method of the pipeline has very important significance for ensuring the safe and reliable service of the pipeline. At present, the nondestructive testing technology for pipeline corrosion mainly comprises ultrasonic testing, magnetic flux leakage testing, ray testing and the like. However, the service condition of the pipeline is complex, the service condition of the conventional method is limited, and the service condition of the pipeline is difficult to effectively reflect. Therefore, it is necessary to develop an economical and efficient corrosion grade evaluation method.
The vibration analysis can reflect the change of the dynamic characteristics of the structure caused by damage, is commonly used for dynamically measuring the structure and acquiring the change or abnormality of vibration information so as to judge the damage degree, and has the advantages of easy extraction, rapidness, convenience, low test cost and the like. The existing signal processing method based on vibration information mainly comprises a time domain analysis method, a frequency domain analysis method, a time-frequency analysis method and the like, which are all applied to damage detection of various structures, and damage information is effectively identified by extracting a proper evaluation index. However, when the experimental signal is complex and the interference is large, the weak damage information is difficult to be sufficiently obtained by manually extracting the index, thereby affecting the accuracy of grade evaluation. The traditional grade evaluation method generally requires a threshold value to be preset manually, the threshold value is usually related to the acquisition scene of the signal, and the threshold value is usually different in different application scenes and often depends on the experience value of an analyst. In order to reduce subjectivity and difference of features depending on manual extraction and reduce influence of manual setting of related parameters as much as possible, an automatic detection method for pipeline corrosion is concerned.
In recent years, deep learning becomes a popular research direction as a feature automatic learning method, data and corresponding labels are input into a network and are trained layer by layer, effective damage features are extracted, network parameters are continuously updated through back propagation, and a stable optimal solution is obtained and then a classification result is output. However, the network working mechanism is difficult to clean, the interpretability is lacked, and the model with perfect performance usually needs a large number of parameters and has huge calculation amount.
Disclosure of Invention
In order to overcome the defects of the conventional method, the invention aims to provide a pipeline corrosion grade evaluation method based on multilayer convolution sparse coding, which not only overcomes the subjective difference that the conventional method needs to manually set a threshold, but also improves the classification precision, reduces the number of parameters, increases the interpretability and can realize the automatic evaluation of the pipeline corrosion grade.
In order to achieve the above object, the present invention provides a pipeline corrosion level evaluation method based on multilayer convolution sparse coding, which comprises the following steps:
s1: installing a first acceleration sensor and a second acceleration sensor on the outer surface of a pipeline, and knocking a concerned area by using a force hammer to obtain an impact response signal, wherein the impact response signal is a time domain signal;
s2: constructing a grade evaluation model based on multilayer convolution sparse coding, and inputting the time domain signal in the step S1 into the model for training as a training set;
an evaluation model based on multilayer convolution sparse coding is constructed, multilayer convolution sparse coding is solved by using a multilayer iteration soft threshold algorithm, and the method can be realized in a frame of a convolution neural network; therefore, firstly, the oscillation attenuation function is used as a first layer of convolution dictionary to obtain a first layer of sparse mapping, meanwhile, other layers of dictionaries are initialized randomly, the following layers of sparse mapping are solved in sequence by using a formula (5), multiple iterations and parameter updating are realized by setting expansion times, the last layer of sparse mapping is obtained and input to full connection, and finally, the three levels are divided; wherein the formula (5) is specifically:
Figure BDA0003152868140000021
wherein the content of the first and second substances,
Figure BDA0003152868140000022
is sparse mapping at the ith layer k +1 iteration,
Figure BDA0003152868140000023
for the ith layer of k iterations of sparse mapping, prox represents the near-end operator,
Figure BDA0003152868140000024
as to the function tgiK is the number of iterations, mu and t are constants,
Figure BDA0003152868140000025
to relate to f and gi-1Gradient mapping operator of DiIs the ith layer of convolution dictionary;
s3: acquiring force hammer actual measurement data of a pipeline to be tested, preprocessing the force hammer actual measurement data, inputting a time domain signal serving as a test set into the model trained in the step S2;
s4: and giving out a pipeline corrosion grade evaluation result according to an evaluation model based on multilayer convolution sparse coding.
Preferably, the step S2 specifically includes the following steps:
s21: matching a time domain signal generated by knocking of a force hammer by using a formula (1) as a first layer of convolution dictionary, and extracting effective corrosion characteristics in a multi-frequency band;
c(t)＝sin(2πft)e-at (1)
wherein t is time, f and a are variables, f is taken within 1-10kHz, and a is taken within 1-5 kHz;
s22: determining each layer of convolutional dictionary by using multilayer convolutional sparse coding, DiIs the ith layer of convolution dictionary, i is a positive integer greater than or equal to 2 as shown in formula (2);
Figure BDA0003152868140000031
wherein x isiIs relative to the convolution dictionary DiI ═ 2,3, …, L; l is the number of layers of multilayer convolution sparse coding, y is an input signal, and s is a base number;
Figure BDA0003152868140000032
represents the square of 2-norm, | ·| non-woven phosphor0Is 0-norm, sLIs the base number of the L-th layer,
Figure BDA0003152868140000033
representing that the 0-norm of each layer of sparse mapping is smaller than the base number of the corresponding layer, namely ensuring the sparsity;
solving the basis pursuit problem of the multilayer convolution sparse coding by adopting a multilayer iteration soft threshold algorithm, and solving to obtain x meeting the conditionsi；
Utilizing a gradient mapping method, as shown in formula (4), and therefore, aiming at the multilayer convolution sparse coding, the ith layer iteration calculation is as shown in formula (5), so as to construct a multilayer convolution sparse coding model; the multi-layer form can be expanded into a multi-layer convolutional neural network by increasing the number of expansions;
Figure BDA0003152868140000034
Figure BDA0003152868140000035
wherein f and g in the upper subscripts represent f (x) and g (x), respectively;
Figure BDA0003152868140000036
for gradient mapping operators on f and g, LcBeing the Lipschitz constant, prox denotes the near-end operator,
Figure BDA0003152868140000037
is about a function
Figure BDA0003152868140000038
The operator of the near-end is used,
Figure BDA0003152868140000039
as to the function tgiThe near-end operator of (a) is,
Figure BDA00031528681400000310
is the gradient of the microprotrusive function, k is the number of iterations,
Figure BDA00031528681400000311
is sparse mapping at the ith layer k +1 iteration,
Figure BDA00031528681400000312
for the ith layer of k iterations of sparse mapping, mu and t are constants,
Figure BDA00031528681400000313
to relate to f and gi-1Gradient mapping operator of DiIs the ith layer of convolution dictionary;
s23: the last layer is processed to obtain xLThe input full connection respectively corresponds to different corrosion degrees, the structure forming the circulating network finishes the modification of the model structure and the debugging of parameters in the training process, and the trained model is stored.
Preferably, the multilayer convolution sparse coding model is two layers; the formation of the multi-layer basis pursuit in the step S22 is expressed as a minimization problem min f (x) + g (x) as shown in formula (3),
Figure BDA00031528681400000314
is a slightly convex function, g1(x)＝λ1||D2x||1g2(x)＝λ2||x||1If the function is not a slightly convex function, the minimization problem is:
Figure BDA00031528681400000315
wherein D is1And D2Respectively representing a first and a second layer of a convolutional dictionary, x2For second-level sparse mapping, λ1And λ2Respectively representing the sparsity of the first layer and the second layer;
the second-level sparse mapping x2Can be expressed as equation (6) with the constants t > 0, μ > 0, and its equivalent convolutional network form can be expressed as equation (7), b1And b2The bias terms for the first and second layers, respectively, and their equivalent convolutional network form can be expressed as equation (7):
Figure BDA0003152868140000041
Figure BDA0003152868140000042
the last layer is processed to obtain xLInputting full connection corresponding to three types of corrosion degrees;
preferably, the multilayer convolution sparse coding model structure is a three-layer basic convolution dictionary, the characteristics of the knocking response signal are included, and the expansion times are set to be 2 times.
Preferably, the first acceleration sensor and the second acceleration sensor are located on both sides of the region of interest of the pipe, respectively.
Preferably, the pipeline corrosion grades are light, medium and heavy grades.
Compared with the prior art, the invention has the following beneficial effects:
(1) the method can effectively extract weak damage characteristics in the vibration signal under the corrosion of the pipeline;
(2) according to the method, multilayer convolution sparse codes are used as a model, effective characteristics in a multi-frequency band are extracted by using an oscillation attenuation function, so that efficient extraction of damage characteristics can be realized, and automatic corrosion grade evaluation is completed;
(3) the method can realize the pipeline corrosion grade evaluation result with higher accuracy, and the parameter quantity is less compared with that of a neural network with the same specification;
(4) the invention avoids the dependence of the traditional method on manual threshold selection and is suitable for online health monitoring.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a graph of vibration signals collected by the embodiment;
FIG. 3 is a schematic diagram of a model structure based on multi-layer convolutional coding according to an embodiment;
FIG. 4 is a diagram illustrating the accuracy of the corrosion rating assessment results obtained by the application model of the embodiment on the test set.
Detailed Description
In order to better understand the technical solution of the present invention, the following detailed description is made with reference to the accompanying drawings and examples. In the drawings, like reference numbers can indicate functionally identical or similar elements. While the various aspects of the embodiments are presented in drawings, the drawings are not necessarily drawn to scale unless specifically indicated.
As shown in fig. 2, a pipeline corrosion level evaluation method based on multilayer convolution sparse coding includes the following steps:
s1: installing a first acceleration sensor and a second acceleration sensor on the outer surface of a pipeline, wherein the first acceleration sensor and the second acceleration sensor are respectively positioned at two sides of a concerned area of the pipeline, knocking the concerned area by using a force hammer to obtain an impact response signal, the impact response signal is a time domain signal, and the impact response signals obtained by the first acceleration sensor and the second acceleration sensor are respectively time domain signalsNumbers are respectively y1、y2。
In this embodiment, nine types of pipelines with lengths of 600mm, 800mm and 1000mm and corrosion degrees of mild, moderate and severe are used to train the multilayer convolution sparse coding model. Respectively adsorbing vibration acceleration transmissions on two sides of a concerned area of the nine types of pipelines, respectively collecting impact response signals under the nine conditions of impact hammer knocking, and recording the time domain signals as time domain signals
Figure BDA0003152868140000051
Where the subscripts 600, 800, 1000 indicate the pipe length and the superscripts 1, 2,3 indicate the level of mild, moderate, and severe corrosion, respectively. The nine time domain signals represent 9 types of signals, and multiple acquisition results in multiple groups of signals, but all belong to the nine types of signals, and fig. 3 shows an example of one time domain signal.
S2: a level estimation model based on multilayer convolutional sparse coding is constructed, and as shown in fig. 4, the time domain signal in step S1 is input to the model as a training set for training.
An evaluation model based on multilayer convolution sparse coding is constructed, multilayer convolution sparse coding is solved by using a multilayer iteration soft threshold algorithm, and the method can be realized in a frame of a convolution neural network. Therefore, firstly, the oscillation attenuation function is used as a first layer of convolution dictionary to obtain a first layer of sparse mapping, meanwhile, other layers of dictionaries are initialized randomly, the following layers of sparse mapping are solved sequentially by using the formula (5), and the formula (5) can also be transformed into a forward propagation process of the convolution neural network due to the equivalent forms of the formula (6) and the formula (7). Multiple iterations and parameter updating are realized by setting the expansion times, the last layer of sparse mapping is obtained and input to full connection, and the three levels are finally divided, and the specific steps are as follows:
s21: matching a time domain signal generated by knocking with a force hammer by using a formula (1) as a first layer of convolution dictionary, and extracting effective corrosion features c (t) in a multiband;
c(t)＝sin(2πft)e-at (1)
wherein t is time, f and a are variables, f takes a value within 1-10kHz, a takes a value within 1-5, and e is a mathematical constant.
The time domain signal generated by the knocking of the hammer is an oscillation attenuation function, so that the design formula (1) takes the oscillation attenuation function as a first layer of convolution dictionary to process a time domain input signal, is used for matching the impact response generated in the experiment, and is favorable for extracting effective corrosion characteristics in a multi-frequency band;
s22: determining each layer of convolutional dictionary by using multilayer convolutional sparse coding, DiIs the ith layer of convolution dictionary, i is a positive integer greater than or equal to 2 as shown in formula (2);
Figure BDA0003152868140000061
wherein x isiIs relative to the convolution dictionary DiI ═ 2,3, …, L; l is the number of layers of multilayer convolution sparse coding, y is an input signal, and s is a base number;
Figure BDA0003152868140000062
represents the square of 2-norm, | ·| non-woven phosphor0Is 0-norm, sLIs the base number of the L-th layer,
Figure BDA0003152868140000063
and (4) representing that the 0-norm of each layer of sparse mapping is smaller than the base number of the corresponding layer, namely ensuring the sparsity.
The desired goal of equation (2) is: on the premise of satisfying the constraint condition, the error between the input signal and the output is minimum, so as to determine the sparse mapping x of each layeri。
Solving the basis pursuit problem of the multilayer convolution sparse coding by adopting a multilayer iteration soft threshold algorithm, and solving to obtain x meeting the conditionsi。
Taking a two-layer example, a multi-layer basis pursuit is formed as shown in formula (3), which can be expressed as the minimization problem min f (x) + g (x),
Figure BDA0003152868140000064
is a slightly convex function, g1(x)＝λ1||D2x||1g2(x)＝λ2||x||1If the function is not a slightly convex function, the minimization problem is:
Figure BDA0003152868140000065
wherein D is1And D2Respectively representing a first and a second layer of a convolutional dictionary, x2For second-level sparse mapping, λ1And λ2The sparsity of the first layer and the second layer are represented, respectively.
The gradient mapping method is used, as shown in equation (4). Therefore, for the multi-layer convolutional sparse coding, the ith layer iteration calculation is as shown in formula (5), so as to construct the multi-layer convolutional sparse coding model. The multi-layered form can be extended into a multi-layered convolutional neural network by increasing the number of expansions.
Figure BDA0003152868140000066
Figure BDA0003152868140000067
Wherein f and g in the upper subscripts represent f (x) and g (x), respectively;
Figure BDA0003152868140000071
for gradient mapping operators on f and g, LcBeing the Lipschitz constant, prox denotes the near-end operator,
Figure BDA0003152868140000072
to concern about gouty
Figure BDA0003152868140000073
The operator of the near-end is used,
Figure BDA0003152868140000074
as to the function tgiThe near-end operator of (a) is,
Figure BDA0003152868140000075
is the gradient of the microprotrusive function, k is the number of iterations,
Figure BDA0003152868140000076
is sparse mapping at the ith layer k +1 iteration,
Figure BDA0003152868140000077
for the ith layer of k iterations of sparse mapping, mu and t are constants,
Figure BDA0003152868140000078
to relate to f and gi-1The gradient mapping operator. E.g., second-level sparse mapping x2Can be expressed as equation (6) with the constants t > 0 and μ > 0, and its equivalent convolutional network form can be expressed as equation (7) with b as the bias term. The equivalent convolutional network form can be expressed as formula (7), formula (6) and formula (7) are in one-to-one correspondence, and prox corresponds to ReLU and b, and is nested. Compared with the conventional convolutional neural network, the method can reduce a large number of parameters, improve the efficiency and increase the interpretability.
Figure BDA0003152868140000079
Figure BDA00031528681400000710
S23: the last layer is processed to obtain xLThe input full connection corresponds to three types of corrosion degrees respectively, the multilayer convolution sparse coding model structure of the embodiment takes a two-layer basic convolution dictionary as an example, the structure forming a circulating network completes the modification of the model structure and the debugging of parameters in the training process, and the trained model is stored.
The structure of the multilayer convolution sparse coding model can also be preferably a three-layer basic convolution dictionary, the three-layer basic convolution dictionary comprises the characteristics of a knocking response signal, and the expansion times are set to be 2 times.
S3: acquiring force hammer actual measurement data of a pipeline to be tested, preprocessing the force hammer actual measurement data, inputting a time domain signal serving as a test set into the model trained in the step S2;
s4: and giving out a pipeline corrosion grade evaluation result according to an evaluation model based on multilayer convolution sparse coding. In the test process, two groups of signals of the two sensors are classified respectively, then the two groups of signals are mixed to be used as a new data set for classification, and the final result is shown in fig. 4, namely the test sets obtained by the sensor 1, the sensor 2 and the two sensors are evaluated respectively, and the corresponding accuracy is obtained.
It can be found that no matter how long the target pipeline is, the technical scheme can achieve high accuracy rate, and the accuracy rate is over 97 percent, so that the corrosion grade evaluation of the pipeline is further realized. By setting the first layer of the convolution dictionary of the time domain signal input model, fast convergence is realized and interpretability is improved. The multilayer convolution sparse coding model applied by the method has certain generalization and does not need to depend on the selection of an artificial threshold, so that the influence of the traditional method on the robustness of the final result is reduced. The invention has high intelligent degree and is more applicable to engineering practice.
Finally, it should be noted that: the above-mentioned embodiments are only used for illustrating the technical solution 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 or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.

Claims (6)

1. A pipeline corrosion grade assessment method based on multilayer convolution sparse coding is characterized by comprising the following steps: which comprises the following steps:
s1: installing a first acceleration sensor and a second acceleration sensor on the outer surface of a pipeline, and knocking a concerned area by using a force hammer to obtain an impact response signal, wherein the impact response signal is a time domain signal;
s2: constructing a grade evaluation model based on multilayer convolution sparse coding, and inputting the time domain signal in the step S1 into the model for training as a training set;
an evaluation model based on multilayer convolution sparse coding is constructed, multilayer convolution sparse coding is solved by using a multilayer iteration soft threshold algorithm, and the method can be realized in a frame of a convolution neural network; therefore, firstly, the oscillation attenuation function is used as a first layer of convolution dictionary to obtain a first layer of sparse mapping, meanwhile, other layers of dictionaries are initialized randomly, the following layers of sparse mapping are solved in sequence by using a formula (5), multiple iterations and parameter updating are realized by setting expansion times, the last layer of sparse mapping is obtained and input to full connection, and finally, the three levels are divided; wherein the formula (5) is specifically:
Figure FDA0003152868130000011
wherein the content of the first and second substances,
Figure FDA0003152868130000012
is sparse mapping at the ith layer k +1 iteration,
Figure FDA0003152868130000013
for the ith layer of k iterations of sparse mapping,
Figure FDA0003152868130000014
for the i-1 st layer k iterations of sparse mapping, prox represents the near-end operator,
Figure FDA0003152868130000015
as to the function tgiK is the number of iterations, mu and t are constants,
Figure FDA0003152868130000016
to relate to f and gi-1Gradient mapping operator of DiIs the ith layer of convolution dictionary;
s3: acquiring force hammer actual measurement data of a pipeline to be tested, preprocessing the force hammer actual measurement data, inputting a time domain signal serving as a test set into the model trained in the step S2;
s4: and giving out a pipeline corrosion grade evaluation result according to an evaluation model based on multilayer convolution sparse coding.
2. The pipeline corrosion grade evaluation method based on multilayer convolution sparse coding according to claim 1, characterized by comprising the following steps: the step S2 specifically includes the following steps:
s21: matching a time domain signal generated by knocking of a force hammer by using a formula (1) as a first layer of convolution dictionary, and extracting effective corrosion characteristics in a multi-frequency band;
c(t)＝sin(2πft)e-at (1)
wherein t is time, f and a are variables, f is taken within 1-10kHz, and a is taken within 1-5 kHz;
s22: determining each layer of convolutional dictionary by using multilayer convolutional sparse coding, DiIs the ith layer of convolution dictionary, i is a positive integer greater than or equal to 2 as shown in formula (2);
Figure FDA0003152868130000017
Figure FDA0003152868130000018
wherein x isiIs relative to the convolution dictionary DiI ═ 2,3, …, L; l is the number of layers, x, of the multilayer convolutional sparse codingLSparse mapping of an L-th layer is carried out, y is an input signal, and s is a base number; d(1，L)＝D1…DL，
Figure FDA0003152868130000021
Represents the square of the 2-norm, | |)0Is 0-norm, sLIs the base number of the L-th layer,
Figure FDA0003152868130000022
representing that the 0-norm of each layer of sparse mapping is smaller than the base number of the corresponding layer, namely ensuring the sparsity;
solving the basis pursuit problem of the multilayer convolution sparse coding by adopting a multilayer iteration soft threshold algorithm, and solving to obtain x meeting the conditionsi；
Utilizing a gradient mapping method, as shown in formula (4), and therefore, aiming at the multilayer convolution sparse coding, the ith layer iteration calculation is as shown in formula (5), so as to construct a multilayer convolution sparse coding model; the multi-layer form can be expanded into a multi-layer convolutional neural network by increasing the number of expansions;
Figure FDA0003152868130000023
Figure FDA0003152868130000024
wherein f and g in the upper subscripts represent f (x) and g (x), respectively;
Figure FDA0003152868130000025
for gradient mapping operators on f and g, LcBeing the Lipschitz constant, prox denotes the near-end operator,
Figure FDA0003152868130000026
is about a function
Figure FDA0003152868130000027
The operator of the near-end is used,
Figure FDA0003152868130000028
as to the function tgiThe near-end operator of (a) is,
Figure FDA0003152868130000029
is the gradient of the microprotrusive function, k is the number of iterations,
Figure FDA00031528681300000210
is sparse mapping at the ith layer k +1 iteration,
Figure FDA00031528681300000211
for the ith layer of k iterations of sparse mapping, mu and t are constants,
Figure FDA00031528681300000212
to relate to f and gi-1Gradient mapping operator of DiIn order to form the i-th layer convolution dictionary,
s23: the last layer is processed to obtain xLThe input full connection respectively corresponds to different corrosion degrees, the structure forming the circulating network finishes the modification of the model structure and the debugging of parameters in the training process, and the trained model is stored.
3. The pipeline corrosion level evaluation method based on multilayer convolution sparse coding according to claim 2, characterized in that:
the multilayer convolution sparse coding model is divided into two layers;
the formation of the multi-layer basis pursuit in the step S22 is expressed as a minimization problem minf (x) + g (x) as shown in formula (3),
Figure FDA00031528681300000213
is a slightly convex function, g1(x)＝λ1||D2x||1g2(x)＝λ2||x||1If the function is not a slightly convex function, the minimization problem is:
Figure FDA00031528681300000214
wherein D is1And D2Respectively representing a first layer and a second layer of convolutional dictionaries,x2for second-level sparse mapping, λ1And λ2Respectively representing the sparsity of the first layer and the second layer;
the second-level sparse mapping x2Can be expressed as formula (6), where the constant t is>0，μ>0, the equivalent convolutional network form of which can be expressed as formula (7), b1And b2The bias terms for the first and second layers, respectively, and their equivalent convolutional network form can be expressed as equation (7):
Figure FDA0003152868130000031
Figure FDA0003152868130000032
the last layer is processed to obtain xLThe input full connection corresponds to three types of corrosion degrees respectively.
4. The pipeline corrosion level evaluation method based on multilayer convolution sparse coding according to claim 2, characterized in that:
the multilayer convolution sparse coding model structure is a three-layer basic convolution dictionary, contains the characteristics of knocking response signals, and the expansion times are set to be 2 times.
5. The pipeline corrosion grade evaluation method based on multilayer convolution sparse coding according to claim 1, characterized by comprising the following steps:
the first acceleration sensor and the second acceleration sensor are respectively positioned on two sides of a region of interest of the pipeline.
6. The pipeline corrosion grade evaluation method based on multilayer convolution sparse coding according to claim 1, characterized by comprising the following steps:
the corrosion grade of the pipeline is light, moderate and severe.
CN202110768576.2A 2021-07-07 2021-07-07 Pipeline corrosion grade evaluation method based on multilayer convolution sparse coding Active CN113343402B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110768576.2A CN113343402B (en) 2021-07-07 2021-07-07 Pipeline corrosion grade evaluation method based on multilayer convolution sparse coding

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110768576.2A CN113343402B (en) 2021-07-07 2021-07-07 Pipeline corrosion grade evaluation method based on multilayer convolution sparse coding

Publications (2)

Publication Number Publication Date
CN113343402A true CN113343402A (en) 2021-09-03
CN113343402B CN113343402B (en) 2022-05-24

Family

ID=77482910

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110768576.2A Active CN113343402B (en) 2021-07-07 2021-07-07 Pipeline corrosion grade evaluation method based on multilayer convolution sparse coding

Country Status (1)

Country Link
CN (1) CN113343402B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116011183A (en) * 2022-12-08 2023-04-25 中国石油大学(北京) In-service oil and gas pipeline detection method, device, equipment and storage medium
CN116608419A (en) * 2023-07-20 2023-08-18 山东特检科技有限公司 Pipeline fatigue failure risk assessment method combined with vibration monitoring

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105843780A (en) * 2016-04-11 2016-08-10 西安交通大学 Sparse deconvolution method for impact load identification of mechanical structure
CN105912854A (en) * 2016-04-11 2016-08-31 西安交通大学 Sparse representation method for dynamic load identification of mechanical structure
US20170183107A1 (en) * 2014-04-02 2017-06-29 Sikorsky Aircraft Corporation System and method for health assessment of aircraft structure

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170183107A1 (en) * 2014-04-02 2017-06-29 Sikorsky Aircraft Corporation System and method for health assessment of aircraft structure
CN105843780A (en) * 2016-04-11 2016-08-10 西安交通大学 Sparse deconvolution method for impact load identification of mechanical structure
CN105912854A (en) * 2016-04-11 2016-08-31 西安交通大学 Sparse representation method for dynamic load identification of mechanical structure

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
余路等: "基于改进稀疏编码的微弱振动信号特征提取算法", 《仪器仪表学报》 *
李继猛等: "基于自适应随机共振和稀疏编码收缩算法的齿轮故障诊断方法", 《中国机械工程》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116011183A (en) * 2022-12-08 2023-04-25 中国石油大学(北京) In-service oil and gas pipeline detection method, device, equipment and storage medium
CN116011183B (en) * 2022-12-08 2023-09-15 中国石油大学(北京) In-service oil and gas pipeline detection method, device, equipment and storage medium
CN116608419A (en) * 2023-07-20 2023-08-18 山东特检科技有限公司 Pipeline fatigue failure risk assessment method combined with vibration monitoring
CN116608419B (en) * 2023-07-20 2023-11-03 山东特检科技有限公司 Pipeline fatigue failure risk assessment method combined with vibration monitoring

Also Published As

Publication number Publication date
CN113343402B (en) 2022-05-24

Similar Documents

Publication Publication Date Title
CN109783906B (en) Intelligent analysis system and method for detecting magnetic flux leakage data in pipeline
CN112232400B (en) Stainless steel weld ultrasonic defect detection method based on depth feature fusion
CN113343402B (en) Pipeline corrosion grade evaluation method based on multilayer convolution sparse coding
CN112258496A (en) Underground drainage pipeline disease segmentation method based on full convolution neural network
CN111734961B (en) Natural gas pipeline leakage detection method
CN113239504B (en) Pipeline corrosion defect prediction method based on optimized neural network
CN111853555B (en) Water supply pipe network blind leakage identification method based on dynamic process
CN110470729B (en) Eddy current-based nondestructive testing method for oil field casing pipe defects
CN105546352A (en) Natural gas pipeline tiny leakage detection method based on sound signals
CN103343885B (en) Pipeline Magnetic Flux Leakage Inspection online data compression method
CN115824519B (en) Comprehensive diagnosis method for valve leakage faults based on multi-sensor information fusion
CN115906949B (en) Petroleum pipeline fault diagnosis method and system, storage medium and petroleum pipeline fault diagnosis equipment
CN115791969A (en) Jacket underwater crack detection system and method based on acoustic emission signals
CN109632942B (en) Inversion method of pipeline defect size based on ensemble learning
CN114091320B (en) Method and device for predicting corrosion failure time of natural gas pipeline
CN116415182A (en) Arc additive low-carbon steel fatigue crack prediction method based on causality and schematic injection force
CN108195932B (en) Ultrasonic guided wave quantitative assessment method for aircraft pipeline damage
CN115935241B (en) Multi-parameter mutually-fused pipe cleaner real-time positioning method and device
Liu et al. A simple machine learning based framework for processing the inline inspection data of subsea pipelines
CN117668623B (en) Multi-sensor cross-domain fault diagnosis method for leakage of ship pipeline valve
CN116701948B (en) Pipeline fault diagnosis method and system, storage medium and pipeline fault diagnosis equipment
CN117540277B (en) Lost circulation early warning method based on WGAN-GP-TabNet algorithm
CN113899809B (en) In-pipeline detector positioning method based on CNN classification and RNN prediction
Zhang et al. Diagnosis and Recognition of Pipeline Damage Defects Based on Neural Network Algorithm
CN116757091A (en) Prediction model for residual life of external corrosion of buried pipeline

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