A method for predicting depth based on a sound wave vibration type pipeline detector
By constructing a sample set and using a supervised neural network to process the characteristics of acoustic signals, the problem of acoustic vibration-type pipeline detectors being unable to accurately estimate depth was solved, achieving efficient gas pipeline depth prediction and reducing construction risks and costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CCCC SECOND HIGHWAY CONSULTANTS CO LTD
- Filing Date
- 2022-12-09
- Publication Date
- 2026-06-05
AI Technical Summary
Existing acoustic vibration-type pipeline detectors cannot accurately estimate the burial depth of gas pipelines, and ground-penetrating radar methods have limitations in detecting depth in small-diameter pipelines and complex geological conditions, making it difficult to distinguish multiple pipelines, which increases labor and cost.
By constructing a sample set, a supervised neural network is used to process the frequency, power, receiver position, and vibration signal data of the signal transmitter. Combined with empirical mode decomposition and Fourier transform, a database of signal features and pipeline depth is established, and a supervised neural network is trained to predict depth.
It has achieved an accuracy rate of over 85% in predicting the depth of gas pipelines, reducing construction risks and costs, and providing a scientific basis for pipeline network information.
Smart Images
Figure CN115860103B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of acoustic detection technology, and more specifically to a depth prediction method based on an acoustic vibration-type pipeline detector. Background Technology
[0002] In recent years, the rapid pace of urbanization has led to frequent accidents involving gas pipelines being severed or burst during third-party construction work, causing significant harm to people's lives and property and resulting in extremely negative social impacts. Therefore, the detection of non-metallic gas pipelines has become a hot topic for many geophysical exploration technicians. Currently, the main methods for detecting and locating underground gas pipelines both domestically and internationally include ground-penetrating radar (GPR), tracer line methods, magnetic gradient methods, and inertial gyroscope methods. While these methods have achieved certain results, they all have corresponding constraints. For example, GPR requires a relatively flat surface and an environment free from strong electrical or magnetic interference, and it cannot distinguish between multiple pipelines running side-by-side. Pipeline detectors based on electromagnetic induction are only suitable for PE gas pipelines with complete tracer lines. While magnetic gradient and inertial gyroscope methods are very accurate in pinpointing pipeline locations, these methods involve cumbersome preparation, long construction periods, and high costs, and are typically used as verification tools to evaluate the effectiveness of other methods. Inertial gyroscope positioning requires cutting both ends of the pipeline, which is very costly and unsuitable for large-scale detection and location work.
[0003] The application of acoustic vibration detection method has effectively solved the aforementioned problems. Compared with traditional methods such as ground-penetrating radar, APL acoustic detection, and probing, this method has advantages such as accuracy, efficiency, and low cost. It has become an indispensable tool for domestic gas companies and municipal pipeline detection units in conducting PE gas pipeline detection.
[0004] Basic principle: A specific frequency sound wave signal is emitted into the gas pipeline through the sound wave oscillator of the signal transmitter. The signal propagates directionally along the pipeline to the far end. At the same time, the sound wave signal propagates directionally in the pressurized gas in the pipeline and propagates through the soil of the pipe wall to the ground. At this time, the receiver on the ground captures the sound wave signal, determines the vertical disturbance range of the pipeline by the strong volume area received, and then determines the horizontal orientation of the pipeline based on the vibration intensity characteristics.
[0005] Currently, none of the instruments of this type have depth prediction capabilities, and they generally rely on ground-penetrating radar (GPR) to assist in depth determination. However, GPR has limitations in detecting small-diameter pipes and pipes located in saline-alkali soil, shale layers, or clay layers. Sometimes, multiple pipes, such as telecommunications, gas, electricity, water supply, and drainage pipes, are laid side by side or vertically on a single road, and distinguishing these pipelines becomes a challenge for GPR detection. Furthermore, GPR can only perform profile detection in pipeline detection and cannot continuously track pipelines. Different frequency antennas must be used for different depths, which undoubtedly increases the workload of personnel and the cost of equipment operation.
[0006] Clearly, using ground-penetrating radar to determine depth is technically challenging, costly, and prone to misjudgments. Therefore, an alternative approach is needed to determine the depth of gas pipelines.
[0007] Existing acoustic vibration-based pipeline detection methods rely on experience and analysis of acoustic wave intensity variation characteristics to determine the pipeline's planar position. However, this method cannot accurately estimate the pipeline's burial depth. To address the problem of acoustic vibration-based pipeline detection instruments being unable to determine the burial depth, this invention proposes a method based on acoustic wave attenuation characteristics. Vibration signal data is picked up directly above a known pipeline and subjected to empirical mode decomposition. Then, Fourier transform processing is performed on each intrinsic mode component. By establishing a diagnostic curve showing the frequency of the signal source, the distance between the acoustic signal pickup point and the signal source, the amplitude of the intrinsic mode components changing with frequency, and a historical feature database of the pipeline's burial depth, the vibration signal data of the test point is then matched with this feature database after the above processing to obtain the reference burial depth of the pipeline. Summary of the Invention
[0008] The purpose of this invention is to overcome the shortcomings of the prior art and propose a depth prediction method based on an acoustic vibration-type pipeline detector.
[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:
[0010] A depth prediction method based on an acoustic vibration-type pipe detector includes the following steps:
[0011] Step 1: Construct samples [g, f, p, l, a] i f i The actual pipe depth h corresponding to the sample is used as the label of the sample. All samples constitute a sample set, where g represents the pipe laying environment, f is the frequency of the transmitted signal from the signal transmitter, p is the transmission power of the signal transmitter, l is the distance between the location of the receiver where the sound wave amplitude is strongest and the signal transmitter, and a... i The component amplitude, f, corresponding to the vibration signal data at the point of strongest sound wave amplitude. i The dominant frequency component of the vibration signal data at the point where the sound wave amplitude is strongest;
[0012] Step 2: Build a supervised neural network and determine the loss function of the supervised neural network;
[0013] Step 3: Train the supervised neural network based on the samples;
[0014] Step 4: For gas pipelines with unknown depths, record the corresponding pipeline laying environment g', the transmitter transmission frequency f', the transmitter transmission power p', the distance l' between the location of the strongest sound wave amplitude and the transmitter, and the component amplitude a' corresponding to the vibration signal data at the location of the strongest sound wave amplitude. i The dominant frequency f' of the vibration signal data corresponding to the point of strongest sound wave amplitude. i , and [g', f', p', l', a' i , f' i The input is fed into the supervised neural network trained in step 3, and the supervised neural network outputs the predicted pipeline depth.
[0015] As described above, step 1 includes the following steps:
[0016] Step 1.1: After the signal transmitter transmits a signal of a set frequency f to the gas pipeline, the signal receiver of the acoustic vibration pipeline detector directly above the pipeline determines the point of strongest acoustic amplitude and records the vibration signal data of the point of strongest acoustic amplitude. The vibration signal data is then denoised to obtain denoised vibration signal data x(t). The distance l between the location of the signal receiver at the point of strongest acoustic amplitude and the signal transmitter is recorded. The actual depth h of the pipeline, the pipeline laying environment g, the transmission frequency f of the signal transmitter, and the transmission power p of the signal transmitter are also recorded.
[0017] Step 1.2: Perform empirical mode decomposition on the denoised vibration signal data x(t) to obtain the intrinsic mode components C of the denoised vibration signal data x(t). i (t); i∈{1~m}, where m is the number of intrinsic modal components in the decomposition;
[0018] Step 1.3: Extract each intrinsic mode component c of the denoised vibration signal data x(t). i Perform a Fourier transform on (t) to obtain the amplitudes a of each component. i Component main frequency f i ;
[0019] Step 1.4: Change the pipeline laying environment g. Implement steps 1.1-1.3 to obtain the intrinsic modal components of the denoised vibration signal of different pipeline laying environments after empirical mode decomposition, as well as the amplitude and dominant frequency of each component.
[0020] Step 1.5: Under different laying environment conditions (g), transmitter transmission frequency (f), and transmitter transmission power (p), determine the distance l between the location of the strongest sound wave amplitude and the transmitter, and the inherent mode component C corresponding to the denoised vibration signal data x(t). i The amplitude of the component a of (t) i and main frequency f i Construct samples [g, f, p, l, a] i f i The actual pipe depth h corresponding to the sample is used as the label of the sample, and the samples together constitute the sample set.
[0021] As mentioned above, the laying environment g includes hardened pavement and dirt pavement, wherein hardened pavement includes cement pavement, asphalt pavement and sidewalks, and dirt pavement includes hard soil and sandy grass.
[0022] As mentioned above, a supervised neural network consists of one input layer, three hidden layers, and one output layer. The loss function of a supervised neural network is...
[0023]
[0024] in, Let h be the predicted pipe depth output by the supervised neural network after the sample is input, h be the actual pipe depth corresponding to the sample, and n be the total number of samples.
[0025] As described in step 3 above, the initial hyperparameters for supervised neural network training are set as follows: the initial bias vectors of the input layer, each hidden layer, and the final output layer are all zero; the number of iterations is 2000; the iterative algorithm used is the adaptive moment estimation algorithm; and the learning rate is 1e-5.
[0026] Compared with the prior art, the present invention has the following advantages:
[0027] By establishing a sample containing the pipeline laying environment, the frequency of the signal transmitter's transmitted signal, the transmission power of the signal transmitter, the distance between the signal receiver and the signal transmitter at the location of the strongest sound wave amplitude, the component amplitude of the vibration signal data corresponding to the location of the strongest sound wave amplitude, and the vibration signal data at the location of the strongest sound wave amplitude, and using the actual pipeline depth as the sample label, a supervised neural network is trained to predict the pipeline burial depth. The depth prediction accuracy reaches over 85%, providing a scientific basis for pipeline network informatization and greatly reducing the construction risks in urban areas. Attached Figure Description
[0028] Figure 1 This is a schematic diagram of a supervised neural network. Detailed Implementation
[0029] To facilitate understanding and implementation of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to embodiments. It should be understood that the embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.
[0030] The basis of this invention: Sound source distance perception is the distance between the listener and the sound source; therefore, sound source distance perception is also called sound source distance localization. When humans or animals hear a sound, in addition to sensing the approximate location of the sound, they can also sense the distance of the sound source. To accurately perceive the depth of a sound source, one must be familiar with the sound field environment and the timbre of the sound source to estimate and measure its distance. This demonstrates that sound source distance perception is acquired and can be trained. Based on this, this invention proposes a depth prediction method based on a sound wave vibration-type pipe detector, comprising the following steps:
[0031] Step 1: Construct samples [g, f, p, l, a] i f i The actual pipe depth h corresponding to the sample is used as the label of the sample. All samples constitute a sample set, where g represents the pipe laying environment, f is the frequency of the transmitted signal from the signal transmitter, p is the transmission power of the signal transmitter, l is the distance between the location of the receiver where the sound wave amplitude is strongest and the signal transmitter, and a... i The component amplitude, f, corresponding to the vibration signal data at the point of strongest sound wave amplitude. i The dominant frequency component of the vibration signal data at the point where the sound wave amplitude is strongest;
[0032] Step 1.1: After the signal transmitter transmits a signal of a set frequency f to the gas pipeline, the signal receiver of the acoustic vibration pipeline detector picks up the acoustic vibration signal above the pipeline at the known spatial location (planar location and burial depth) directly above the pipeline. The point with the strongest acoustic amplitude above the pipeline is found and the vibration signal data of the point with the strongest acoustic amplitude for 3 to 5 seconds is recorded. The signal is then subjected to bandpass filtering to obtain the denoised vibration signal data x(t). The distance l between the location of the signal receiver at the point with the strongest acoustic amplitude and the signal transmitter is recorded. The actual depth h of the pipeline, the pipeline laying environment g, the transmission frequency f of the signal transmitter, and the transmission power p of the signal transmitter are also recorded.
[0033] Step 1.2: Perform Empirical Mode Decomposition (EMD) on the denoised vibration signal data x(t). EMD decomposes a complex signal into a finite number of intrinsic mode components (IMFs). Each IMF contains local characteristic signals of the denoised vibration signal data x(t) at different time scales. In EMD, the denoised vibration signal data x(t) is decomposed into intrinsic mode components and residual terms, i.e.: c i(t) represents the intrinsic mode components under different decompositions, r m (t) is the residual term, and m is the number of intrinsic modal components.
[0034]
[0035] The various intrinsic modal components C of the denoised vibration signal data x(t) are obtained. i (t); i∈{1~m}, m is the number of intrinsic mode components to be decomposed (since the frequency of the signal source is between 200-600Hz, only the instantaneous frequency of the intrinsic mode components needs to be analyzed, which is between 200-600Hz).
[0036] Step 1.3: Extract each intrinsic mode component c of the denoised vibration signal data x(t). i Perform a Fourier transform on (t) to obtain the amplitudes a of each component. i Component main frequency f i ;
[0037] Step 1.4: Change the pipeline laying environment g. Implement steps 1.1-1.3 to obtain the intrinsic mode components of the denoised vibration signals from different pipeline laying environments after empirical mode decomposition, as well as the amplitude and dominant frequency of each component. The laying environment above the pipeline is generally divided into hardened road surfaces (cement road surface, asphalt road surface, sidewalk) and soil road surfaces (hard soil, sandy grass), etc. The set frequency f of the signal transmitted by the signal transmitter is known, and the distance l between the signal pickup point (the point with the strongest sound wave amplitude directly above the pipeline) and the signal transmitter is known.
[0038] Step 1.5: Under different laying environment conditions (g), transmitter transmission frequency (f), and transmitter transmission power (p), determine the distance l between the location of the strongest sound wave amplitude and the transmitter, and the inherent mode component C corresponding to the denoised vibration signal data x(t). i The amplitude of the component a of (t) i and main frequency f i Construct samples [g, f, p, l, a] i f i The actual pipe depth h corresponding to the sample is used as the label of the sample, and the samples together constitute the sample set.
[0039] Step 2: Establish a supervised neural network
[0040] like Figure 1As shown, a supervised neural network consists of one input layer, three hidden layers, and one output layer; the activation function of the hidden layers is the Rectified Linear Unit (ReLU) function. The activation function of the output layer is the Softmax function; the training process of a supervised neural network is the process of minimizing a predefined loss function to determine the parameters of the neural network.
[0041] The loss function of a supervised neural network is
[0042]
[0043] in, The output of the supervised neural network is the predicted pipe depth after the sample is input, h is the actual pipe depth corresponding to the sample, and n is the total number of samples. Learning stops when the loss function reaches its minimum value or is less than a set threshold.
[0044] Step 3: Supervised Neural Network Training
[0045] To obtain a deep learning neural network model with high accuracy and generalization ability, the order of samples in the sample set prepared in step 1 is first shuffled. The sample set contains 50,000 samples. Then, 40,000 samples are randomly selected from the sample set as the training dataset, and the remaining 10,000 samples are used as the validation dataset. A 10-fold cross-validation method is used to train the supervised neural network, minimizing the loss function or reducing it to a set threshold to obtain the optimal mapping between the input samples and the pipeline prediction depth. The validation dataset is used to evaluate the supervised neural network during training and to fine-tune its hyperparameters to improve its generalization ability.
[0046] In this embodiment, the initial hyperparameters for supervised neural network training are set as follows: the initial bias vectors for the input layer, each hidden layer, and the final output layer are all zero, and the number of epochs is 2000. The iterative algorithm used is the Adam adaptive moment estimation algorithm, with a learning rate of 1e-5.
[0047] Step 4: Pipeline Depth Prediction
[0048] After a signal transmitter emits a signal of a set frequency f' into a gas pipeline at an unknown depth, an acoustic vibration pipeline detector locates the point of strongest acoustic amplitude above the pipeline and records vibration signal data for 3-5 seconds at that point. The vibration signal data is then bandpass filtered to obtain denoised vibration signal data x'(t). The distance l' between the receiver and the transmitter, the transmitter's transmission frequency f', the transmitter's transmission power p', and the installation environment g' are recorded. Modal decomposition is then performed on the denoised vibration signal data to obtain the individual intrinsic mode components C'. i The component amplitude a' corresponding to (t) i 、 main frequency f' i ,
[0049] [g', f', p', l', a'] i , f' i The input is fed into the supervised neural network trained in step 3, and the supervised neural network outputs the predicted pipeline depth.
[0050] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.
Claims
1. A depth prediction method based on an acoustic vibration-type pipe detector, characterized in that, Includes the following steps: Step 1: Constructing Samples The actual depth of the pipe corresponding to the sample As labels for the samples, the individual samples constitute a sample set, where... For the environment of pipeline laying, The frequency of the transmitted signal from the signal transmitter. This refers to the transmission power of the signal transmitter. The distance between the location of the signal receiver at the point of strongest sound wave amplitude and the signal transmitter. The component amplitude corresponding to the vibration signal data at the point of strongest sound wave amplitude. The dominant frequency component of the vibration signal data at the point where the sound wave amplitude is strongest; Step 2: Build a supervised neural network and determine the loss function of the supervised neural network; Step 3: Train the supervised neural network based on the samples; Step 4: For gas pipelines with unknown depths, record the corresponding pipeline laying environment. The transmission frequency of the signal transmitter Transmitting power of the signal transmitter The distance between the point of strongest sound wave amplitude and the signal transmitter The component amplitude corresponding to the vibration signal data at the point of strongest sound wave amplitude. The dominant frequency component corresponding to the vibration signal data at the point of strongest sound wave amplitude. ,Will[ , , , , , The input is fed into the supervised neural network trained in step 3. The supervised neural network outputs the predicted pipeline depth. Step 1 includes the following steps: Step 1.1: The signal transmitter transmits a set frequency to the gas pipeline. After receiving the signal, the point of strongest acoustic wave amplitude is determined directly above the pipeline using the signal receiver of the acoustic vibration pipeline detector, and the vibration signal data at the point of strongest acoustic wave amplitude is recorded. The vibration signal data is then denoised to obtain denoised vibration signal data. Record the distance between the location of the signal receiver at the point of strongest sound wave amplitude and the signal transmitter. Record the true depth of the pipe Pipeline laying environment The transmission frequency of the signal transmitter and the transmission power of the signal transmitter ; Step 1.2: Denoise the vibration signal data Empirical mode decomposition is performed to obtain denoised vibration signal data. Each intrinsic mode component ; , The number of intrinsic modal components in the decomposition; Step 1.3: Denoising vibration signal data Each intrinsic mode component Perform a Fourier transform to obtain the amplitude of each component. Component main frequency ; Step 1.4: Change the pipeline laying environment Implement steps 1.1-1.3 to obtain the intrinsic modal components of the denoised vibration signals in different pipeline laying environments after empirical mode decomposition, as well as the amplitude and dominant frequency of each component. Step 1.5: Statistics on different laying environments The transmission frequency of the signal transmitter Transmitting power of the signal transmitter Under these conditions, the distance between the location of the strongest sound wave amplitude and the signal transmitter Denoising vibration signal data Corresponding intrinsic mode components component amplitude and clock speed ; Constructing samples The actual depth of the pipe corresponding to the sample As labels for the samples, the individual samples constitute the sample set. The laying environment This includes paved roads and dirt roads. Paved roads include cement roads, asphalt roads, and sidewalks, while dirt roads include hard soil and sandy grass. The supervised neural network includes an input layer, three hidden layers, and an output layer. The loss function of a supervised neural network is : in, The pipe prediction depth is the output of the supervised neural network after the sample is input. This represents the actual depth of the pipe corresponding to the sample. The total number of samples, In step 3, the initial hyperparameters for supervised neural network training are set as follows: the initial bias vectors of the input layer, each hidden layer, and the final output layer are all zero; the number of iteration rounds is 2000; the iterative algorithm used is the adaptive moment estimation algorithm; and the learning rate is 1e-5.