A method for online detection of drainage pipes based on low-frequency sound waves
Through the combination of time-frequency image characterization and support vector machine, the problems of noise pollution and insufficient feature extraction in low-frequency sound wave detection are solved, and accurate identification and online diagnosis of drainage pipe blockages and pipeline accessories under complex working conditions are achieved.
Patent Information
- Application Number
- CN202010960502.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-09-14
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing low-frequency acoustic wave detection methods have shortcomings in dealing with noise pollution and extracting features, making it difficult to effectively identify drainage pipe blockages and pipeline accessories under complex working conditions.
Using a combination of time-frequency image characterization and support vector machine, a diagnostic model is established to achieve accurate identification of blockages and pipeline accessories through wavelet packet decomposition, SPWVD time-frequency analysis, grayscale symbiosis matrix and local binary mode feature fusion.
It improves the recognition accuracy of blockages and pipeline accessories, can effectively locate faults under variable working conditions, solves the problems of single time domain signal characteristics and noise interference, and realizes online detection and diagnosis.
Smart Images

Figure BDA0002680376600000032 
Figure BDA0002680376600000034 
Figure BDA0002680376600000062
Abstract
Description
Technical Field
[0001] The invention relates to an online detection method for drainage pipes based on low-frequency sound waves, in particular to a detection method for drainage pipes based on low-frequency sound waves, time-frequency image texture features and support vector machines, belonging to the technical field of pipeline detection. Background Art
[0002] Buried drainage pipes play a vital role in urban and factory infrastructure. By the end of 2018, the length of my country's drainage network exceeded 680,000 kilometers. However, due to unreasonable pipeline design, failure to carry out timely maintenance, the impact of the natural environment, pipeline aging and other factors, pipelines are prone to blockage. Pipeline blockage will not only reduce the transmission efficiency of the fluid in the pipeline and cause waste of resources; but also the presence of blockages will cause changes in the fluid pressure in the pipeline, leading to pipeline leakage and sewage overflow, thereby polluting the environment. Therefore, accurate detection and identification of blockages and pipeline accessories in buried drainage pipes under complex working conditions has important theoretical significance and application value.
[0003] In recent years, the research on pipeline blockage detection and positioning has attracted more and more attention from domestic and foreign researchers, and a large number of solutions have been proposed, such as robot detection, CCTV pipeline endoscopy detection, transient wave and detection, vibration analysis, etc. At present, these detection methods have great limitations in detection efficiency and practicality, and have the disadvantages of cumbersome construction, high cost, and detection susceptible to environmental influences. They cannot effectively solve the detection and identification of blockages and accessories under complex working conditions in buried drainage pipelines. Acoustic detection is a remotely controlled, non-invasive active detection method for pipeline faults. Sound waves can propagate in solid, liquid and gas states, and carry information about the internal operating conditions of the pipeline during the propagation process. The collected pipeline detection sound wave signals are processed and analyzed, and the presence of blockages is determined based on the characteristics of the sound wave signals. The pipeline working conditions are identified. It has the characteristics of being unaffected by the subjective influence of operators, low cost, easy operation, and the ability to detect small diameter pipelines. In addition, low-frequency sound waves have a long transmission distance, making it a very potential pipeline fault detection method.
[0004] In past studies, the processing of low-frequency acoustic wave detection signals mainly focused on time domain processing; however, time domain signals are easily contaminated by noise, and the extracted time domain features are single and cannot fully characterize the fault information. For drainage pipes with changeable and evolving operating conditions, it is difficult to effectively detect and identify faults. Summary of the invention
[0005] The technical problem to be solved by the present invention is to provide an online detection method for drainage pipes based on low-frequency sound waves, which characterizes fault signals through two-dimensional time-frequency images, extracts texture features of the time-frequency images for fusion, and effectively characterizes faults in the presence of background noise. The established drainage pipe diagnostic model can effectively locate and identify blockages and pipe accessories under variable working conditions, solving the above problems.
[0006] The technical solution of the present invention is: a drainage pipe online detection method based on low-frequency sound waves, comprising the following steps:
[0007] Step 1: Install a pipeline detection device on the drainage pipeline and check whether the detection device is working properly. The detection device includes a computer, a power amplifier, a microphone, a speaker, a filter, a sound card, and a signal transmission line. Use low-frequency sound waves to detect the drainage pipeline section and obtain pipeline section detection data under normal conditions, complex conditions such as those containing repeated blockages and containing tees (pipeline accessories); the computer controls the sound card to generate a low-frequency sound wave X(t) of 0-6000Hz, which is amplified by the power amplifier and transmitted to the drainage pipeline section through the speaker for detection. The microphone collects the echo signal Y(t), which is filtered by the filter and transmitted to the computer.
[0008] Step 2: The acoustic response signal of the pipeline can effectively reduce the interference of random noise and better express the operating condition information inside the pipeline. Therefore, the collected low-frequency sound wave detection signal is denoised, the optimal wavelet packet basis is determined according to the correlation function, and the optimal decomposition layer number n is determined according to the sample entropy. The denoising threshold is adaptively selected based on the "heursure" criterion, the signal is decomposed by wavelet packet, and the sound signal R(t) is subdivided in the low-frequency part and the high-frequency part to obtain the decomposition coefficients of the 2n sub-bands from low frequency to high frequency in the nth layer. The denoised sound signal is reconstructed as f(t).
[0009] Step 3: Obtain the time-frequency image of the pipeline section system, process f(t) using the SPWVD time-frequency analysis method, and obtain the time and frequency distribution image of the acoustic signal;
[0010] Step 4: The time-frequency clustering points in the system time-frequency image obtained in Step 3 are segmented by the Otsu threshold segmentation method. According to the sound propagation speed of 343 m / s, the distance from the time-frequency clustering points to the sound wave emission source of the pipeline section is calculated by d=ct / 2, and the gray-level co-occurrence matrix and local binary patterns (LBP) of the segmented time-frequency clustering points are extracted for feature fusion.
[0011] Step 5: Input the features extracted in Step 4 into the classifier established by support vector machine for training to obtain the SVM diagnostic model. The extracted and fused features are used as training set and test set respectively. The training set is used to train the SVM classification model, and the test set is used to determine the type of time-frequency cluster points and output the recognition results. The corresponding data in the SVM diagnostic model can accurately classify blockages, tees and pipe ends, and realize online detection and diagnosis of drainage pipes.
[0012] In the process of denoising the collected pipeline acoustic wave detection signal in Step 2, the collected sound signal R(t)=Y(t) / X(t), wherein X(t) is the low-frequency acoustic wave signal generated by the sound card and emitted to the drainage pipe section, and Y(t) is the echo signal after the acoustic wave is transmitted and detected in the pipe section.
[0013] In the process of denoising the sound signal in Step 2, an improved method for determining the optimal number of decomposition layers is used. According to the principle that entropy is a representation of signal complexity, the more signal noise there is, the higher the signal complexity is, and the larger the entropy value is. Sample entropy is used as a basis for judgment to determine the optimal number of decomposition layers. The specific process is as follows:
[0014] r(t)=f(t)+n(t)
[0015] Where r(t) is the noisy signal, f(t) is the denoised signal, and n(t) is the noise. r(t) is denoised by wavelet denoising, and the sample entropy value of n(t) is calculated. As the number of decomposition layers increases, the sample entropy value increases until the sample entropy value decreases, which is the optimal decomposition layer number for wavelet denoising.
[0016] In Step 3, SPWVD time-frequency processing is performed to generate a two-dimensional time-frequency image. The formula of the SPWVD time-frequency analysis method is:
[0017]
[0018] After SPWVD time-frequency transformation, the one-dimensional time-domain sound signal is converted into a two-dimensional time-frequency distribution image, enriching the representation information.
[0019] In the step 4, the gray level co-occurrence matrix and LBP features are extracted, wherein the gray level co-occurrence matrix reflects the comprehensive information of the gray level of the time-frequency clustering point on the direction, adjacent interval, and change amplitude. It is the basis for analyzing the local patterns of the image and their arrangement rules. The gray level co-occurrence matrix H(d,i,j,θ) represents the probability of a pixel point with a gray level value i and a gray level value j on the time-frequency image appearing in the direction of θ when the distance is d. First, the gray level matrix of the time-frequency image is normalized:
[0020]
[0021] Then, four texture parameters, namely energy, contrast, entropy and correlation, of the gray-level co-occurrence matrix are extracted. In the invented method, offset parameters of 0°, 45°, 90° and 135° are taken as the gray-level co-occurrence matrix to obtain characteristic indices respectively.
[0022] LBP is an operator used to describe the local texture features of an image. It reflects the relationship between each pixel and the surrounding pixels. It has the characteristics of simple principle, small amount of calculation, strong classification ability and high calculation efficiency. Its calculation formula is as follows:
[0023]
[0024]
[0025] In the formula, (x c ,y c ) is the coordinate of the center pixel, n is the nth pixel in the neighborhood, i p is the gray value of the pixel, i c is the gray value of the center pixel, and S(x) is the sign function.
[0026] The gray-level co-occurrence matrix and the features extracted by LBP are further integrated into P, P = [K1, K2, K3…K n ], and input the fused feature vector P into the support vector machine for training.
[0027] In the Step 5, the fused features are divided into a training set and a test set at a ratio of 7:3. The training set is used to train the classifier, and the test set is used to test the model.
[0028] Step 5.1, input the training set into the classifier established by support vector machine for training, and select radial basis function as the kernel function of the SVM diagnosis model.
[0029] Step 5.2, apply the test set to the SVM model with trained parameters, and output the recognition results of blockages and pipe accessories under different working conditions.
[0030] The beneficial effects of the present invention are: solving the problems of identification of blockages and pipeline accessories under variable working conditions, as well as weak fault information of time domain signals and insufficient feature characterization. The one-dimensional time domain signal is converted into a two-dimensional time-frequency image through time-frequency transformation, enriching the amount of characterization information; according to the propagation speed of sound waves in the pipeline, the time-frequency aggregation points in the time-frequency image of the pipeline system are used to accurately locate blockages and pipeline accessories; the gray-level co-occurrence matrix and LBP texture features are extracted and fused, effectively solving the problem of single feature extraction and improving the recognition accuracy of the classifier; the fault can be effectively located and identified, and used for online detection and diagnosis of pipeline working conditions. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 is a flow chart of the steps of the present invention;
[0032] Figure 2 This is a time domain waveform diagram of the pipeline section detection sound wave with a sampling time of 0.1s in an embodiment of the present invention;
[0033] Figure 3 is a diagram of the steps for determining the optimal number of decomposition layers for denoising the acoustic wave signal of pipeline section detection in an embodiment of the present invention;
[0034] Figure 4 This is a SPWVD time-frequency diagram of a pipe section containing a 55 mm plug, a tee, and a 20 mm plug in the embodiment of the present invention;
[0035] Figure 5 It is a time-frequency diagram of the obstruction and pipeline accessories in the time-frequency diagram of the segmented pipeline segment system in an embodiment of the present invention. DETAILED DESCRIPTION
[0036] The present invention will be further described below in conjunction with the accompanying drawings and specific implementation methods.
[0037] The present invention uses low-frequency sound waves to detect drainage pipe sections and obtains sound wave signal data after pipeline section detection under various complex working conditions. The improved wavelet denoising method is adopted, sample entropy is introduced as the basis for determining the optimal number of decomposition layers, the optimal wavelet basis is determined by using correlation functions, and the denoising threshold is adaptively selected through the "heursure" criterion to achieve denoising of the sound wave detection signal.
[0038] The denoised signal is reconstructed by wavelet coefficients to make the amplitude change of the time domain acoustic wave signal more obvious and improve the signal recognition; the denoised signal is processed by time-frequency transformation to obtain the time-frequency image of the pipeline section system; then, the time-frequency clustering points in the system time-frequency image are segmented to obtain the time-frequency image samples of the blockage and pipeline accessories.
[0039] The gray-level co-occurrence matrix and local binary pattern (LBP) texture features of the blockage and system time-frequency image are extracted and fused to obtain classification features. The support vector machine (SVM) can better solve the advantages of small sample, nonlinear, and high-dimensional feature pattern recognition, and establish a mapping relationship between the fused time-frequency image features and the blockage and pipeline accessories, so as to realize the identification of blockages and accessories in the pipeline section under variable working conditions.
[0040] The problem of difficulty in identifying blockages and pipe fittings at different positions and degrees in drainage pipes under variable working conditions was solved, and a new method for drainage pipe object detection based on the combination of time-frequency images and support vector machines was determined.
[0041] like Figure 1 As shown, a drainage pipe online detection method based on low-frequency sound waves includes the following steps:
[0042] Step 1: Install the pipeline detection device on the drainage pipe and check whether the computer, power amplifier, microphone, speaker, filter, sound card and signal transmission line in the detection device are working properly. Use low-frequency sound waves to detect the drainage pipe section and obtain the pipeline section detection data under normal conditions, complex conditions such as those with repeated blockages and those with tees (pipe accessories); the computer controls the sound card to generate 0-6000Hz low-frequency sound waves X(t), which are amplified by the power amplifier and transmitted to the drainage pipe section through the speaker for detection. The microphone collects the echo signal Y(t) as shown in Figure 2 As shown, it is filtered and then transmitted to the computer.
[0043] Step 2: The acoustic response signal of the pipeline can effectively reduce the interference of random noise and better express the operating condition information inside the pipeline. Therefore, the collected low-frequency sound wave detection signal is denoised, the optimal wavelet packet basis is determined according to the correlation function, and the optimal decomposition layer number n is determined according to the sample entropy. The denoising threshold is adaptively selected based on the "heursure" criterion, the signal is decomposed by wavelet packet, and the sound signal R(t) is subdivided in the low-frequency part and the high-frequency part to obtain the decomposition coefficients of the 2n sub-bands from low frequency to high frequency in the nth layer. The denoised sound signal is reconstructed as f(t).
[0044] Step 3: Obtain the time-frequency image of the pipeline section system, process f(t) using the SPWVD time-frequency analysis method, and obtain the time and frequency distribution image of the acoustic signal;
[0045] Step 4: The time-frequency clustering points in the system time-frequency image obtained in Step 3 are segmented by the Otsu threshold segmentation method. According to the sound propagation speed of 343 m / s, the distance from the time-frequency clustering points to the sound wave emission source of the pipeline section is calculated by d=ct / 2, and the gray-level co-occurrence matrix and local binary patterns (LBP) of the segmented time-frequency clustering points are extracted for feature fusion.
[0046] Step 5: Input the features extracted in Step 4 into the classifier established by support vector machine for training to obtain the SVM diagnostic model. The extracted and fused features are used as training set and test set respectively. The training set is used to train the SVM classification model, and the test set is used to determine the type of time-frequency cluster points and output the recognition results. The corresponding data in the SVM diagnostic model can accurately classify blockages, tees and pipe ends, and realize online detection and diagnosis of drainage pipes.
[0047] In the process of denoising the collected pipeline acoustic wave detection signal in Step 2, the collected sound signal R(t)=Y(t) / X(t), wherein X(t) is the low-frequency acoustic wave signal generated by the sound card and emitted to the drainage pipe section, and Y(t) is the echo signal after the acoustic wave is transmitted and detected in the pipe section.
[0048] In the process of removing noise from the sound signal in Step 2, if Figure 3 As shown in the figure, an improved method for determining the optimal number of decomposition layers is adopted. According to the principle that entropy is a representation of signal complexity, the more signal noise there is, the higher the signal complexity is, and the larger the entropy value is. Sample entropy is used as the basis for judgment to determine the optimal number of decomposition layers. The specific process is as follows:
[0049] r(t)=f(t)+n(t)
[0050] Where r(t) is the noisy signal, f(t) is the denoised signal, and n(t) is the noise. r(t) is denoised by wavelet denoising, and the sample entropy value of n(t) is calculated. As the number of decomposition layers increases, the sample entropy value increases until the sample entropy value decreases, which is the optimal decomposition layer number for wavelet denoising.
[0051] In Step 3, SPWVD time-frequency processing is performed to generate a two-dimensional time-frequency image. The formula of the SPWVD time-frequency analysis method is:
[0052]
[0053] like Figure 4 The two-dimensional time-frequency image shown shows the change of frequency over time. The distribution of sound wave energy intuitively and fully characterizes the operation status inside the pipeline section, and adaptively segment the time-frequency aggregation points in the time-frequency image of the pipeline section system through the Otsu threshold segmentation method. After the SPWVD time-frequency transformation, the one-dimensional time domain sound signal is converted into a two-dimensional time-frequency distribution image, which enriches the representation information.
[0054] In the Step 4, the time-frequency cluster points in the time-frequency image of the pipeline segment system are segmented to obtain the time-frequency image samples of the blockage and pipeline accessories, such as Figure 5As shown. Then the gray level co-occurrence matrix of the segmented time-frequency image is extracted and fused with the LBP texture features. The gray level co-occurrence matrix reflects the comprehensive information of the gray level of the time-frequency clustering points on the direction, adjacent intervals, and change amplitude. It is the basis for analyzing the local patterns of the image and their arrangement rules. The gray level co-occurrence matrix H (d, i, j, θ) represents the probability of a pixel point with a gray value i and a gray value j on the time-frequency image appearing in the direction of θ when the distance is d. First, the gray level matrix of the time-frequency image is normalized:
[0055]
[0056] Then, four texture parameters, namely energy, contrast, entropy and correlation, of the gray-level co-occurrence matrix are extracted. In the invented method, offset parameters of 0°, 45°, 90° and 135° are taken as the gray-level co-occurrence matrix to obtain characteristic indices respectively.
[0057] LBP is an operator used to describe the local texture features of an image. It reflects the relationship between each pixel and the surrounding pixels. It has the characteristics of simple principle, small amount of calculation, strong classification ability and high calculation efficiency. Its calculation formula is as follows:
[0058]
[0059]
[0060] In the formula, (x c ,y c ) is the coordinate of the center pixel, n is the nth pixel in the neighborhood, i p is the gray value of the pixel, i c is the gray value of the center pixel, and S(x) is the sign function.
[0061] The gray-level co-occurrence matrix and the features extracted by LBP are further integrated into P, P = [K1, K2, K3…K n ], and input the fused feature vector P into the support vector machine for training.
[0062] In the Step 5, the fused features are divided into a training set and a test set at a ratio of 7:3. The training set is used to train the classifier, and the test set is used to test the model.
[0063] Step 5.1, input the training set into the classifier established by support vector machine for training, and select radial basis function as the kernel function of the SVM diagnosis model.
[0064] Step 5.2, apply the test set to the SVM model with trained parameters, and output the recognition results of blockages and pipe accessories under different working conditions.
[0065] The recognition results show that the recognition rate of 55mm blockage under variable working conditions is 91.48%, the recognition rate of 20mm blockage is 90.14%, the recognition rate of tee is 89.35%, and the recognition rate of the end of the pipe section is 92.16%.
[0066] Table 1.
[0067] Objects in the pipe Recognition rate 20mm plug 91.48% 20mm plug 90.14% Tee pieces 89.35% Pipe segment end 92.16%
[0068] Table 1: Object recognition results of SVM diagnosis model under different working conditions
[0069] Finally, the experimenters tested the remaining pipeline sections with unknown conditions and repeated the above testing process.
[0070] The specific implementation modes of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above implementation modes, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.
Claims
1. A drainage pipe online detection method based on low-frequency sound waves, characterized in that: Step 1: Install a pipeline detection device on the drainage pipeline and check whether the detection device is working properly; use low-frequency sound waves to detect the drainage pipeline section and obtain the detection data of the normal pipeline section, the pipeline section containing repeated blockages and the pipeline section containing tees; wherein, the computer controls the sound card to generate low-frequency sound waves X(t), which are amplified by the power amplifier and then transmitted to the drainage pipeline section through the speaker for detection, and the microphone collects the echo signal Y(t), which is filtered by the filter and then transmitted to the computer; Step 2: De-noise the collected low-frequency sound wave detection signal, determine the optimal wavelet packet basis according to the correlation function, and determine the optimal decomposition layer n according to the sample entropy. Adaptively select the denoising threshold based on the "heursure" criterion, decompose the signal by wavelet packet, subdivide the sound signal R(t) in the low-frequency part and the high-frequency part, and obtain the decomposition coefficients of 2n sub-bands from low frequency to high frequency in the nth layer. Reconstruct the denoised sound signal into f(t); Step 3: Obtain the time-frequency image of the pipeline section system, wherein f(t) is processed by the SPWVD time-frequency analysis method to obtain the time and frequency distribution image of the acoustic signal; Step 4: Segment the time-frequency clustering points in the system time-frequency image obtained in Step 3 by using the Otsu threshold segmentation method, calculate the distance from the time-frequency clustering points to the acoustic wave emission source of the pipeline section by d=ct / 2, and extract the gray-level co-occurrence matrix and local binary pattern LBP of the segmented time-frequency clustering points for feature fusion; Step 5: Input the features extracted in Step 4 into the classifier established by support vector machine for training to obtain the SVM diagnostic model.
2. The method for online detection of drainage pipes based on low-frequency sound waves according to claim 1 is characterized in that: The sound signal R(t) in Step 2 is R(t)=Y(t) / X(t), wherein X(t) is the low-frequency sound wave signal generated by the sound card and transmitted to the drainage pipe section, and Y(t) is the echo signal after the sound wave is transmitted and detected in the pipe section.
3. The method for online detection of drainage pipes based on low-frequency sound waves according to claim 1 is characterized in that: In the process of denoising the sound signal in Step 2, sample entropy is used as a basis for judgment to determine the optimal number of decomposition layers. The specific process is as follows: r(t)=f(t)+n(t) Where r(t) is the noisy signal, f(t) is the denoised signal, and n(t) is the noise. r(t) is denoised by wavelet denoising, and the sample entropy value of n(t) is calculated. As the number of decomposition layers increases, the sample entropy value increases until the sample entropy value decreases, which is the optimal decomposition layer number for wavelet denoising.
4. The method for online detection of drainage pipes based on low-frequency sound waves according to claim 1 is characterized in that The formula of the SPWVD time-frequency analysis method in Step 3 is:
5. The method for online detection of drainage pipes based on low-frequency sound waves according to claim 1 is characterized in that: In the Step 4, the time-frequency clustering points in the time-frequency image of the pipeline segment system are segmented to obtain the time-frequency image samples of the blockage and pipeline accessories, and then the gray-level co-occurrence matrix and LBP texture features of the segmented time-frequency image are extracted for fusion. The gray-level co-occurrence matrix H (d, i, j, θ) represents the probability of a pixel point with a gray value i and a gray value j on the time-frequency image appearing in the direction of θ when the distance is d. First, the gray-level matrix of the time-frequency image is normalized: Then, the four texture parameters of the gray-level co-occurrence matrix, namely, energy, contrast, entropy and correlation, are extracted as the gray-level co-occurrence matrix, and the characteristic indices are obtained respectively; The calculation formula of LBP is as follows: In the formula, (x c ,y c ) is the coordinate of the center pixel, n is the nth pixel in the neighborhood, i p is the gray value of the pixel, i c is the gray value of the center pixel, S(x) is the sign function; The gray-level co-occurrence matrix and the features extracted by LBP are further integrated into P, P = [K1, K2, K3...K n ], and input the fused feature vector P into the support vector machine for training.
