CN109556797B - Pipeline leakage detection and positioning method based on spline local mean decomposition and convolutional neural network - Google Patents
Pipeline leakage detection and positioning method based on spline local mean decomposition and convolutional neural network Download PDFInfo
- Publication number
- CN109556797B CN109556797B CN201811375623.1A CN201811375623A CN109556797B CN 109556797 B CN109556797 B CN 109556797B CN 201811375623 A CN201811375623 A CN 201811375623A CN 109556797 B CN109556797 B CN 109556797B
- Authority
- CN
- China
- Prior art keywords
- leakage
- neural network
- convolutional neural
- signal
- local mean
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M3/00—Investigating fluid-tightness of structures
- G01M3/02—Investigating fluid-tightness of structures by using fluid or vacuum
- G01M3/26—Investigating fluid-tightness of structures by using fluid or vacuum by measuring rate of loss or gain of fluid, e.g. by pressure-responsive devices, by flow detectors
Abstract
The invention discloses a pipeline leakage detection and positioning method based on spline interpolation local mean decomposition and a convolutional neural network, which comprises the following steps: firstly, reducing noise of negative pressure waves generated by leakage by adopting improved sample strip local mean decomposition; on the basis, negative pressure wave signals generated by leakage are converted into image signals to be used as input of a convolutional neural network model, and detection of different leakage sizes is realized through training of the convolutional neural network; and finally, determining the leakage time delay by calculating the generalized correlation function of the pressure signals on the upstream and the downstream of the leakage point, thereby determining the leakage position. Compared with the existing local mean value noise reduction method, the method has higher calculation efficiency, better decomposition precision and more accurate reduction of pressure signals; the method for detecting the leakage based on the convolutional neural network has good accuracy and model generalization capability.
Description
Technical Field
The invention relates to a pipeline leakage detection and positioning method.
Background
The pipeline plays an important role as an infrastructure of a city and a factory, but the pipeline is easy to leak under the influence of factors such as environment, external force, corrosion and the like, the leakage not only causes the waste of resources, but also causes the pollution of the environment, and the noise reduction of a leakage signal is the key of leakage detection.
Currently, there are Empirical Mode Decomposition (EMD) and Local Mean Decomposition (LMD) methods for noise reduction commonly used in pipeline detection and positioning. Empirical mode decomposition (EMF) is a self-adaptive signal time-frequency analysis method, and the signal decomposition has self-adaptive characteristics and decomposes an original signal into superposition of a plurality of eigenmode functions (IMFs). The EMD does not need a basis function, only carries out signal decomposition according to the scale characteristics of the signal, can be used for the decomposition of any type of signal, and has obvious advantages particularly on the decomposition of nonlinear and non-stationary signals, thereby being widely applied to engineering. However, EMD has some disadvantages, such as modal aliasing, over-envelope, under-envelope, and reduced accuracy especially for complex signal decomposition.
The local mean decomposition method is a novel nonlinear and non-steady signal processing method, and decomposes an original signal into Product Functions PFs (Product Functions, PFs) with physical meanings of a plurality of instantaneous frequencies. Compared with the EMD, the LMD has small operation iteration times, can reduce the endpoint effect in the decomposition process, can solve the problems of over-enveloping and under-enveloping, and can save more frequency and enveloping information than the IMF component in the PF component, so that the LMD is more suitable for processing non-stationary and nonlinear signals than the EMD. However, LMD has some disadvantages in practical applications. For example, the LMD calculates local mean functions and envelope estimation functions with low accuracy and boundary distortion. The spline local mean decomposition method can overcome low accuracy and boundary distortion, and the spline local mean decomposition adopts cubic spline interpolation to replace a moving average process to calculate a corresponding local mean function and an envelope estimation function; meanwhile, the boundary distortion in the process of calculating the upper and lower envelopes is overcome by properly expanding the original signal based on the adaptive waveform matching technology. Although SLMD has high accuracy and efficiency, PF components containing information cannot be accurately selected. The PF component is conventionally selected based on frequency, but the noise and leakage signal frequencies in the pipe are unknown and the frequencies vary with operating conditions and environment. The improved spline local mean decomposition determines whether the PF component is noise according to the cross correlation degree between the PF component and the reference signal, and completes signal noise filtering.
Disclosure of Invention
The invention provides a pipeline leakage detection and positioning method based on improved spline local mean decomposition and convolutional neural network, which aims to overcome the defects in the prior art.
In order to more accurately detect the size of the leakage, the invention classifies the pressure images with different leakage apertures by adopting a convolutional neural network model. Firstly, pressure images with different leakage sizes are input as a convolutional neural network model, the network is trained, and the obtained convolutional neural network model can realize leakage detection of different leakage apertures. Compared with other detection methods, the convolutional neural network has higher accuracy and generalization capability
The purpose of the invention is: aiming at the difficulty of noise reduction of negative pressure wave signals generated by leakage, the original negative pressure wave signals are decomposed by adopting spline interpolation local mean, and the spline local mean decomposition has better accuracy and efficiency. In order to select a reasonable PF component, a cross-correlation factor is calculated from the cross-correlation between the reference signal and the PF component. By setting the number of critical correlation factors, the PF component containing leakage information is selected and reconstructed. Meanwhile, the negative pressure wave pressure signal images are classified by adopting a convolutional neural network aiming at the similarity between different positions and different sizes of the noise reduction signals, so that the detection of different leakage sizes is realized.
The technical solution of the invention is as follows: firstly, reducing noise of negative pressure waves generated by leakage by adopting improved sample strip local mean decomposition; on the basis, negative pressure wave signals generated by pipeline leakage are converted into image signals; secondly, detecting different leakage sizes by constructing a convolutional neural network model; and finally, determining leakage time delay by calculating a generalized correlation function of the pressure signals on the upstream and the downstream of the leakage point, so as to determine the leakage position. Compared with the existing local mean value noise reduction method, the method has higher calculation efficiency and better decomposition precision, can more accurately restore the pressure signal, and realizes more accurate positioning precision. In addition, the detection method based on the convolutional neural network has higher accuracy and generalization capability compared with the conventional method.
The pipeline leakage detection and positioning method based on the improved sample local mean decomposition and the convolutional neural network comprises the following specific steps:
Where PF represents the product function, rk(t) represents the residual.
And 2, introducing a leakage point downstream signal (reference signal), calculating the cross correlation between each PF component and the reference signal, and selecting the PF component with the correlation factor larger than the critical constant to reconstruct the signal.
The concrete basis is as follows: coefficient of correlation Ri(τ) may be:
according to the correlation coefficient Ri(τ) calculating a correlation coefficient factor, which is defined by:
PF component more than or equal to 0.75 is selected for signal reconstruction, and signals are reconstructed
Can be expressed as:
and 3, performing wavelet decomposition on the reconstructed signal by using a db4 wavelet, performing 7-layer decomposition, setting coefficients of 1 layer, 7 layer and 8 layer to be 0, and reconstructing to obtain the reconstructed signal.
And 4, converting the reconstructed signal into an image, wherein all image lines have the same color and size, removing any marks which are not related to the pressure, setting the pixel value to be 227 multiplied by 227, and filling the whole image with pressure changes.
And 5, performing the processing in steps 1-4 on 5 groups of leakage pore diameters of 400 × 5-2000 groups of no leakage, 2mm, 3mm, 5mm and 10mm to generate 2000 images with 5 different leakage sizes in total.
And 6, inputting the images into a convolutional neural network for training, and constructing a convolutional neural network detection model.
And 7, processing the required detection signal in the same way. After processing, the data is input into a convolutional neural network to finish detection of different leakage sizes.
And 8, after the convolutional neural network detects the leakage, processing the pressure signal at the downstream of the leakage point by using an improved local mean decomposition method and small noise filtering, wherein the processed signal at the upstream of the leakage point is a reference signal.
And 9, calculating generalized correlation functions of the processed upstream and downstream pressure signals, and determining leakage time delay so as to realize leakage point positioning.
The invention has the advantages that: the method has the advantages that not only is the calculation efficiency higher, but also the decomposition precision is better, and the pressure signal can be restored more accurately; the method for detecting the leakage based on the convolutional neural network has good accuracy and model generalization capability
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
FIG. 1 is a design drawing of an experimental pipeline of the present invention.
Fig. 2 is a flow chart of the present invention.
FIG. 3 is a noise reduction plot of a leak caliber of 3mm at 1000m upstream.
FIG. 4 is a graph showing the noise reduction result of the downstream with a leak caliber of 10mm at a distance of 1000m from the upstream.
FIG. 5 is a graph showing the noise reduction result of the downstream with a leak caliber of 5mm at a distance of 1000m from the upstream.
FIG. 6 is a graph showing the noise reduction result of the downstream with a leak caliber of 3mm at a distance of 1000m from the upstream.
FIG. 7 is a graph showing the noise reduction result of the downstream with a leak caliber of 2mm at a distance of 1000m from the upstream.
Fig. 8 is a structural diagram of a convolutional neural network.
FIGS. 9 a-9 e are graphs of training and test inputs at 0, 2, 3, 5, 10mm upstream 1000m respectively.
FIGS. 9 f-9 j are graphs of training and test inputs at 0, 2, 3, 5, 10mm upstream 500m respectively.
Detailed Description
The following detailed description of the embodiments of the present invention will be provided with reference to the drawings and examples, so that how to apply the technical means to solve the technical problems and achieve the technical effects can be fully understood and implemented.
Negative pressure waves generated by leaks with different sizes at different positions are collected, and the specific operating conditions are as follows: the medium is water, the length L of the pipe is 1510m, the inner diameter is 0.05m, the roughness of the relative wall in the pipe is 0.025, the difference between the upstream and the downstream is 120m, the propagation speed of the negative pressure wave is 1000m/s, and the temperature is 20 ℃. To simulate leakage at different locations and different apertures, flow ball valves were installed 500m and 1010m upstream, with the apertures of the ball valves being 0mm, 2mm, 3mm, 5mm and 10mm, respectively. The upstream and downstream pressure measuring points are node1 and node2, the sampling frequency of a pressure gauge is 100HZ, the duration is 40s, a pipeline valve is opened within 2s, and a leakage valve is opened within 20 s.
The pressure of the leakage at different positions and sizes is collected for 400 groups respectively. Firstly, SLMD decomposition is carried out on a pressure signal which is 1000m away from the upstream, the cross correlation degree of each decomposed PF component and a downstream signal (reference signal) is calculated, and the PF component with the cross correlation degree larger than 0.75 is selected for signal reconstruction. The reconstructed signal is wavelet decomposed using a db4 wavelet, 7-layer decomposition, setting the coefficients of 1, 7, and 8 layers to 0, and then reconstructed to obtain the reconstructed signal. The reconstructed signal is converted into an image in which all image lines are the same color, size, and any markers not related to pressure are erased, the pixel value is set to 227 x 227, and pressure changes fill the entire image. The above processing was performed on 5 leak pore diameters of no leak, 2mm, 3mm, 5mm and 10mm in total of 400 × 5 to 2000 sets of data, and a total of 2000 images of 5 different leak sizes were generated. The above images were divided into training and testing data sets at an 8:2 ratio and tested for accuracy.
In practical situations, the leakage position has randomness, and the convolutional neural network model has good generalization capability. In order to detect the generalization capability of the training model, the upstream pressure signal 500m away from the upstream is processed by the same processing method, and the 1000m trained convolutional neural network model is used for carrying out classification detection on the images. The specific accuracy is shown in table 1 below:
table 1: leak detection accuracy meter
And when the convolutional neural network detects that leakage occurs, processing a downstream signal by the same noise reduction method, wherein the reference signal is an upstream signal after noise reduction, and other parameters are consistent with the upstream noise reduction parameters. The time delay delta t of the leakage signal is calculated by the generalized cross-correlation function of the upstream and the downstream, and the leakage position L is obtainedxCalculated according to the following formula.
Lx＝(L+v×Δt)/2
LxThe distance between the leakage point and the upstream position, the length of the L pipe, the propagation speed of the v negative pressure wave and the delta v time difference. Specifically, 20 sets of data at 1000m are randomly extracted for calculation, and the positioning error is obtained from the average value of the absolute values of the errors, and the specific results are shown in table 2 below:
table 2: leak detection error table
The embodiments described in this specification are merely illustrative of implementations of the inventive concept and the scope of the present invention should not be considered limited to the specific forms set forth in the embodiments but rather by the equivalents thereof as may occur to those skilled in the art upon consideration of the present inventive concept.
Claims (2)
1. A pipeline leakage detection and positioning method based on spline interpolation and a convolutional neural network comprises the following steps:
step 1, decomposing an upstream pressure signal through spline local mean decomposition to obtain a product function PFs;
step 2, introducing a leakage point downstream pressure wave signal, namely a reference signal, calculating the cross correlation degree between each PF component and the reference signal, and selecting the PF component containing leakage information to reconstruct by setting the number of critical correlation factors; the improvement of the spline local mean decomposition is as follows: introducing a reference signal, and determining whether the PF component contains leakage information by using the cross correlation between the reference signal and the PF so as to realize noise filtering; the method comprises the following specific steps:
(1) for non-linear, non-stationary signals downstream pressure xu(t), the spline local mean decomposition result can be expressed as:
where k represents the number of PF components, rk(t) represents the residual;
(2) each PF and rk(t) and downstream signal xd(t) performing a correlation analysis;
(3) calculating a correlation coefficient factor:
(4) selecting a threshold value suitable for the PF correlation factor00.75, when not less than0Considering the PF component as a leakage signal; when in use<0The PF component is considered not to be a leakage signal; selecting a PF according to the correlation coefficient factor, and reconstructing the selected PF signal;
(5) selecting PFi signalReconstructing, reconstructing signals
Can be expressed as:
step 3, performing noise reduction processing on the reconstructed signal through wavelet decomposition;
step 4, converting the noise reduction signals obtained in the step 3 into image signals, wherein the colors and the sizes of all images and lines are the same, removing any marks irrelevant to pressure, setting pixel values, and filling the whole image with the pressure negative pressure wave change process;
step 5, processing 5 groups of leakage pore diameters of 400 × 5 to 2000 groups of no leakage, 2mm, 3mm, 5mm and 10mm in total in steps 1-4 to generate 2000 images in total;
step 6, inputting the images into a convolutional neural network for model training and verification, and constructing a convolutional neural network detection model;
step 7, the convolutional neural network model obtained by training can realize detection of different leakage sizes;
step 8, processing the downstream pressure signal of the leakage point by using an improved local mean decomposition method and wavelet noise filtering, and using the upstream signal of the leakage point as a reference signal;
and 9, calculating the generalized correlation function of the reconstructed upstream and downstream pressure signals of the leakage point, and determining the leakage time delay so as to realize the positioning of the leakage point.
2. The spline interpolation and convolutional neural network-based pipeline leakage detection and location method of claim 1, wherein in steps 6 and 7, the convolutional neural network is introduced to the detection of pipeline leakage; by utilizing the advantages of the convolutional neural network in processing 2-D image data, training of one leakage point and detection of the sizes of a plurality of local leaks are realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811375623.1A CN109556797B (en) | 2018-11-19 | 2018-11-19 | Pipeline leakage detection and positioning method based on spline local mean decomposition and convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811375623.1A CN109556797B (en) | 2018-11-19 | 2018-11-19 | Pipeline leakage detection and positioning method based on spline local mean decomposition and convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109556797A CN109556797A (en) | 2019-04-02 |
| CN109556797B true CN109556797B (en) | 2020-10-09 |
Family
ID=65866766
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811375623.1A Active CN109556797B (en) | 2018-11-19 | 2018-11-19 | Pipeline leakage detection and positioning method based on spline local mean decomposition and convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109556797B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110440148B (en) * | 2019-07-15 | 2021-07-13 | 清华大学 | Method, device and system for classifying and identifying leakage acoustic signals |
| CN114486128A (en) * | 2022-02-18 | 2022-05-13 | 汉威科技集团股份有限公司 | Gas leakage detection device and method integrating ultrasonic wave and passive infrared imaging |
| US20230332976A1 (en) * | 2022-04-14 | 2023-10-19 | Professional Flexible Technologies Inc. | Systems and methods for improved pipeline leak detection |
| CN115824519B (en) * | 2023-02-14 | 2023-05-12 | 中国海洋大学 | Comprehensive diagnosis method for valve leakage faults based on multi-sensor information fusion |
| CN115993268B (en) * | 2023-03-22 | 2023-06-02 | 福建天甫电子材料有限公司 | Sampling system and method for anhydrous hydrogen fluoride |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103822097B (en) * | 2012-11-16 | 2016-11-16 | 国际商业机器公司 | The method and device of the velocity of wave of suction wave is estimated in fluid-transporting tubing |
| CN103216731B (en) * | 2013-05-08 | 2015-02-04 | 西安理工大学 | Pipeline leakage judgment method and system thereof |
| CN104390781A (en) * | 2014-11-26 | 2015-03-04 | 中国矿业大学 | Gear fault diagnosis method based on LMD and BP neural network |
| CN105424057B (en) * | 2015-10-30 | 2019-04-09 | 东南大学 | A kind of optical fibre gyro analysis of vibration signal method based on improvement LMD |
| US10386262B2 (en) * | 2016-01-28 | 2019-08-20 | Utopus Insights, Inc. | Leak identification in a mass transport network |
| CN107228282B (en) * | 2017-07-06 | 2019-05-07 | 东北石油大学 | A kind of gas pipeline leakage localization method and device |
| CN108052989A (en) * | 2018-02-07 | 2018-05-18 | 深圳市唯特视科技有限公司 | A kind of image classification method based on Spline convolution neutral net |
| CN108444653B (en) * | 2018-03-09 | 2019-11-12 | 北京理工大学 | A kind of face formation ultrasonic gas leak point positioning detection device and method |
| CN108488638B (en) * | 2018-03-28 | 2019-08-20 | 东北大学 | Line leakage system and method based on sound wave suction wave hybrid monitoring |
| CN108664901B (en) * | 2018-04-20 | 2021-04-13 | 三峡大学 | Improved LMD-based micro-grid power quality disturbance signal detection method |
-
2018
- 2018-11-19 CN CN201811375623.1A patent/CN109556797B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109556797A (en) | 2019-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109556797B (en) | Pipeline leakage detection and positioning method based on spline local mean decomposition and convolutional neural network | |
| Zhou et al. | Leak detection and location based on ISLMD and CNN in a pipeline | |
| CN109827082B (en) | Pipeline multi-point leakage accurate positioning method | |
| CN110671613A (en) | Fluid pipeline leakage signal time delay estimation method based on improved empirical wavelet transform | |
| CN110345392B (en) | Positioning method and device for oil pipeline leakage point | |
| CN111503527B (en) | Fluid pipeline leakage positioning method based on self-adaptive multivariate variational modal decomposition | |
| CN110388570A (en) | A kind of self-adaptation noise reduction method based on VMD and its application in water pipeline leakage locating | |
| CN103488906A (en) | Valve inner leakage defect type recognition and inner leakage rate calculation method | |
| CN111339701B (en) | Pipeline leakage characteristic Godunov simulation method based on Brunone dynamic friction resistance | |
| CN110645483B (en) | Urban buried pipeline early leakage diagnosis method based on spectrum analysis | |
| CN107356677B (en) | Ultrasonic nondestructive testing method based on travel time tomography and reverse time migration imaging | |
| Feng et al. | Fast reconstruction of defect profiles from magnetic flux leakage measurements using a RBFNN based error adjustment methodology | |
| CN109492708A (en) | Missing data interpolating method is detected in a kind of pipe leakage based on LS-KNN | |
| CN114397475A (en) | Water flow velocity measuring method suitable for ultrasonic water meter | |
| CN111365624A (en) | Intelligent terminal and method for detecting leakage of brine transportation pipeline | |
| CN111182705A (en) | Time-varying plasma diagnosis method and diagnosis system based on automatic encoder | |
| CN105674065A (en) | Variational mode decomposition-based method for locating leakage point of pipeline by acoustic emission | |
| CN104456089A (en) | Fluid pipeline multi-leakage-point locating method | |
| CN109630908B (en) | Pipeline leakage positioning method capable of reducing noise for multiple times | |
| CN115031906A (en) | Pipeline leakage on-line monitoring method based on infrasonic wave | |
| CN113111923A (en) | Water supply network leakage detection and positioning method based on one-dimensional migration learning convolutional neural network integrated model | |
| CN113253248B (en) | Small sample vertical array target distance estimation method based on transfer learning | |
| CN107389184A (en) | A kind of method of two-dimensional space window processing acoustical signal error | |
| CN108107437B (en) | Marine environment monitoring method by using normal wave coupling interference | |
| WO2019056121A1 (en) | Methods for detecting pipeline weakening |
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 |