Pipeline detection method and device, computer device and storage medium

Abstract

This application relates to a pipeline inspection method, apparatus, computer equipment, storage medium, and computer program product. The method includes: extracting the power spectrum of an original time-series signal; determining the number of modes required for signal decomposition based on peak distribution; extracting multiple spectral shape features from the power spectrum and determining a penalty factor for signal decomposition based on a comprehensive evaluation of these features; performing variational mode decomposition on the original time-series signal using the number of modes and the penalty factor to obtain decomposed sub-signals; extracting time-frequency domain features of the decomposed sub-signals and dividing them into an effective signal subset and a noise signal subset based on these features; performing noise suppression processing on the decomposed sub-signals in the noise signal subset and merging the processed noise signal subset with the effective signal subset to reconstruct a denoised time-series signal. This method can more accurately extract pipeline structural state features, thereby improving the accuracy and reliability of pipeline defect identification.

Description

Technical Field

[0001] This application relates to the field of nondestructive testing technology, and in particular to a pipeline inspection method, apparatus, computer equipment, storage medium, and computer program product. Background Technology

[0002] With the development of pipeline structural health monitoring and non-destructive testing technologies, variational mode decomposition (VMD) has been widely applied to feature extraction and noise reduction analysis of time-series signals such as pipeline vibration, acoustic waves, and ultrasonic waves due to its advantages in non-stationary signal processing and adaptive frequency band decomposition. In traditional methods, the performance of VMD is highly dependent on the selection of its key parameters—the number of modes K and the penalty factor α. To address this issue, heuristic optimization algorithms (such as particle swarm optimization and genetic algorithms) are often used to adaptively search for VMD parameters. These methods find the optimal solution in the parameter space by defining a fitness function (such as envelope entropy and kurtosis), thereby improving the adaptability of the decomposition effect to the signal.

[0003] However, current heuristic optimization methods still have significant limitations in practical engineering applications: since they are essentially iterative search strategies, each fitness evaluation requires a complete variational mode decomposition, resulting in a heavy overall computational burden; especially when processing long sequences of pipeline detection signals with high sampling rates, the parameter optimization process is extremely time-consuming and cannot meet the needs of rapid on-site analysis and real-time diagnosis. Summary of the Invention

[0004] Therefore, it is necessary to provide a pipeline inspection method, apparatus, computer equipment, computer-readable storage medium, and computer program product to address the aforementioned technical problems.

[0005] Firstly, this application provides a pipeline inspection method. The method includes:

[0006] Obtain the original timing signal of the pipeline to be tested; wherein, the original timing signal is used to characterize the structural state of the pipeline to be tested;

[0007] Extract the power spectrum of the original time-series signal; and extract the peak distribution in the power spectrum, and determine the number of modes required for signal decomposition based on the peak distribution;

[0008] Extract multiple spectral shape features from the power spectrum, and determine the penalty factor for signal decomposition based on a comprehensive evaluation of the multiple spectral shape features;

[0009] Using the number of modes and the penalty factor, variational mode decomposition is performed on the original time-series signal to obtain decomposed sub-signals;

[0010] Extract the time-frequency domain features of the decomposed sub-signals, and divide the decomposed sub-signals into an effective signal subset and a noise signal subset based on the time-frequency domain features;

[0011] The decomposed sub-signals in the noise signal subset are subjected to noise suppression processing, and the processed noise signal subset is merged with the effective signal subset to reconstruct the denoised time sequence signal.

[0012] In one embodiment, extracting the peak distribution in the power spectrum and determining the number of modes required for signal decomposition based on the peak distribution includes:

[0013] Determine the background noise level in the power spectrum; determine the peak threshold based on the background noise level;

[0014] In the power spectrum, data exceeding the peak threshold are matched;

[0015] The number of data points exceeding the peak threshold is determined, and this number of data points is defined as the number of modalities.

[0016] In one embodiment, the step of extracting multiple spectral shape features from the power spectrum and determining a penalty factor for signal decomposition based on a comprehensive evaluation of the multiple spectral shape features includes:

[0017] Feature extraction is performed on the power spectrum to obtain spectral flatness features and spectral centroid features;

[0018] The initial penalty factor is obtained by linearly combining the spectral flatness feature and the spectral centroid feature.

[0019] The initial penalty factor is mapped to a parameter range corresponding to the number of modes to obtain the penalty factor.

[0020] In one embodiment, extracting the time-frequency domain features of the decomposed sub-signal specifically includes:

[0021] Determine the kurtosis of the decomposed sub-signals in the time domain to obtain kurtosis features;

[0022] Based on the high-frequency energy and total energy in the power spectrum, determine the high-frequency energy proportion characteristics of the power spectrum;

[0023] The kurtosis feature is combined with the high-frequency energy ratio feature to form the time-frequency domain feature vector used to distinguish between signal and noise.

[0024] In one embodiment, the method further includes:

[0025] Obtain the first spectral flatness parameter of the decomposed sub-signal;

[0026] Using the number of modes and the penalty factor, the denoised time-series signal is subjected to mode decomposition to obtain the residual signal; the second spectral flatness parameter of the residual signal is determined.

[0027] Based on the first spectral flatness parameter and the second spectral flatness parameter, the denoising parameters of the denoised time-series signal are determined; wherein, the denoising parameters are used to characterize the degree of denoising.

[0028] In one embodiment, dividing the decomposed sub-signal into an effective signal subset and a noise signal subset based on the time-frequency domain characteristics includes:

[0029] The time-frequency domain features are input into a preset classification model to obtain the category identifiers corresponding to the time-frequency domain features; wherein, the classification model is trained based on valid signal and noise signal samples identified in historical data;

[0030] Based on the category identifier, the decomposed sub-signals are divided into a valid signal subset and a noise signal subset.

[0031] Secondly, this application also provides a pipeline inspection device. The device includes:

[0032] The signal acquisition module is used to acquire the original timing signals of the pipeline under test;

[0033] The peak extraction module is used to extract the power spectrum of the original time-series signal; and to extract the peak distribution in the power spectrum, and determine the number of modes required for signal decomposition based on the peak distribution;

[0034] The feature extraction module is used to extract multiple spectral shape features from the power spectrum and determine the penalty factor for signal decomposition based on a comprehensive evaluation of the multiple spectral shape features.

[0035] The signal decomposition module is used to perform variational mode decomposition on the original time-series signal using the number of modes and the penalty factor to obtain decomposed sub-signals;

[0036] The feature extraction module is further configured to extract the time-frequency domain features of the decomposed sub-signals, and divide the decomposed sub-signals into an effective signal subset and a noise signal subset based on the time-frequency domain features;

[0037] The signal reconstruction module is used to perform noise suppression processing on the decomposed sub-signals in the noise signal subset, and merge the processed noise signal subset with the effective signal subset to reconstruct a denoised time-series signal, wherein the denoised time-series signal is used to characterize the structural state of the pipeline to be detected.

[0038] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the pipeline detection method as described in any one of the embodiments of this disclosure.

[0039] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the pipeline inspection method as described in any one of the embodiments of this disclosure.

[0040] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the pipeline detection method as described in any of the embodiments of this disclosure.

[0041] The aforementioned pipeline inspection methods, devices, computer equipment, storage media, and computer program products acquire the original time-series signals of the pipeline, adaptively determine the number of modes and penalty factors required for variational mode decomposition, accurately decompose the original signals, and then filter effective and noise signals through time-frequency domain features. After suppressing the noise signals, they are merged and reconstructed with the effective signals, achieving high-quality noise reduction of pipeline vibration signals and providing a more reliable data foundation for subsequent pipeline defect identification and assessment. This method can dynamically adjust the decomposition parameters according to the characteristics of the signal itself, effectively avoiding mode aliasing or over-decomposition problems, significantly improving the accuracy and robustness of pipeline structural state information extraction, and is particularly suitable for pipeline inspection scenarios under complex working conditions. Attached Figure Description

[0042] Figure 1 This is a diagram illustrating the application environment of a pipeline inspection method in one embodiment;

[0043] Figure 2 This is a flowchart illustrating a pipeline inspection method in one embodiment;

[0044] Figure 3 This is a schematic diagram of the AI ​​model training curve before data processing in one embodiment;

[0045] Figure 4 This is a schematic diagram of the AI ​​model validation curve before data processing in one embodiment;

[0046] Figure 5 This is a bar chart illustrating the accuracy and recall of an AI model before data processing in one embodiment.

[0047] Figure 6 This is a schematic diagram of the AI ​​model training curve after data processing in one embodiment;

[0048] Figure 7 This is a schematic diagram of the AI ​​model validation curve after data processing in one embodiment;

[0049] Figure 8 This is a bar chart illustrating the accuracy and recall of an AI model after data processing in one embodiment.

[0050] Figure 9 This is a schematic diagram illustrating the average distribution of the root mean square values ​​in one embodiment;

[0051] Figure 10 This is a schematic diagram of the average signal-to-noise ratio of the root mean square value in one embodiment;

[0052] Figure 11 This is a schematic diagram illustrating the average correlation between adjacent channels in one embodiment;

[0053] Figure 12 This is a structural block diagram of a pipeline inspection device in one embodiment;

[0054] Figure 13 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0056] The pipeline inspection method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. In this application environment, terminal 102 can be used to collect the original time-series signal of the pipeline under test and upload it to server 104. Server 104 is configured to execute the method in this embodiment: extract the power spectrum of the signal, determine the number of modes required for signal decomposition based on its peak distribution; simultaneously extract multiple spectral shape features from the power spectrum, and determine the decomposition penalty factor after comprehensive evaluation; perform variational mode decomposition on the original signal based on the number of modes and the penalty factor to obtain several decomposed sub-signals; extract the time-frequency domain features of each sub-signal, and divide them into an effective signal subset and a noise signal subset; suppress the noise signal subset, and then reconstruct it with the effective signal subset to obtain a denoised time-series signal, which can be used for further analysis of the pipeline's structural state. After processing, server 104 can return the reconstructed signal and analysis results to terminal 102 or archive them to the data storage system. The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle systems. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0057] In one embodiment, such as Figure 2 As shown, a pipeline inspection method is provided, including the following steps:

[0058] Step S200: Obtain the original timing signal of the pipeline to be tested.

[0059] The original time-series signal can be used to characterize the structural state of the pipeline under test. For example, it can be acquired by sensors (such as accelerometers, acoustic sensors, ultrasonic sensors, etc.) installed on or inside the pipeline. It can be vibration signals, sound wave signals, or ultrasonic echo signals generated by the pipeline during operation. These signals contain information such as whether there are defects in the pipeline (such as corrosion, cracks, deformation, etc.) and the degree of defects.

[0060] In one exemplary embodiment, if the pipeline to be tested is a main urban water supply pipeline, piezoelectric accelerometers can be installed at equal intervals on the outer wall of the pipeline to collect the original vibration time sequence signal generated by the pipeline under the impact of water flow and its own vibration. The sampling frequency can be set according to the pipeline diameter and the frequency range of possible defect characteristics. For example, for a DN1000mm pipeline, a sampling frequency of 50kHz can be used, and the continuous sampling time is not less than 10 seconds to ensure that sufficient structural state information is included.

[0061] Step S202: Extract the power spectrum of the original time-series signal; and extract the peak distribution in the power spectrum, and determine the number of modes required for signal decomposition based on the peak distribution.

[0062] After acquiring the original time-series signal, its time-domain signal can be converted into a power spectrum in the frequency domain using Fourier transform. This power spectrum clearly reflects the energy distribution of different frequency components in the signal. Peak distribution in the power spectrum is extracted. Specifically, the background noise level in the power spectrum is determined by statistical analysis of the lower-energy frequency bands in the power spectrum, for example, by calculating the average energy value in this region using a sliding window method as an estimate of the background noise. Based on this background noise level, a reasonable peak threshold is set, typically several times the background noise level (e.g., 3 or 5 times the standard deviation), to ensure effective differentiation between real signal peaks and spurious peaks caused by noise. All data points exceeding this peak threshold are searched throughout the power spectrum; each such data point represents a potential signal mode component. Finally, the number of these data points exceeding the peak threshold is determined as the number of modes K required for variational mode decomposition. For example, if five peaks significantly higher than the background noise threshold are detected in the power spectrum, the number of modes K is set to 5, allowing variational mode decomposition to specifically decompose the original signal into five sub-signals with different center frequencies.

[0063] In one exemplary embodiment, the power spectrum calculation can employ the Welch method, segmenting the original time-series signal and applying a window, such as a Hamming window, with a window length of 1024 points and an overlap rate of 50%, to achieve a balance between frequency resolution and statistical accuracy. Assuming that after power spectrum calculation, the resulting power spectrum exhibits significant peaks at 100Hz, 300Hz, 500Hz, 800Hz, and 1200Hz, and the energy of these peaks is all higher than four times the background noise level (a set peak threshold), then the determined number of modes K is 5.

[0064] In one exemplary embodiment, determining the number of modes may include determining the background noise level in the power spectrum; determining a peak threshold based on the background noise level; matching data in the power spectrum that are greater than the peak threshold; determining the number of data that are greater than the peak threshold, and defining the number of data as the number of modes. Specifically, the background noise level may be determined by statistically analyzing low-energy regions in the power spectrum (e.g., the frequency bands with the lowest energy values ​​in the bottom 30%), calculating their mean and standard deviation, and using the mean plus a multiple of the standard deviation (e.g., three times) as an initial background noise estimate; if there are isolated high-energy points (possibly noise spikes) in this region, smoothing is performed using median filtering (window size of 5 frequency points) before recalculation. The peak threshold can be dynamically adjusted according to the signal-to-noise ratio (SNR). For example, when the SNR is higher than 20 dB, the threshold is set to 3 times the background noise level; when the SNR is between 10 and 20 dB, the threshold is set to 2.5 times the background noise level; when the SNR is lower than 10 dB, the threshold is set to 2 times the background noise level, to avoid missing valid peaks in low SNR scenarios. When matching peak values ​​in the power spectrum, an adaptive peak search algorithm can be used: calculate the first derivative of the power spectrum to find all points where the derivative is zero; combine the second derivative to determine its concavity and convexity, and filter out local maxima; compare the energy values ​​of these maxima with the peak threshold, and count the number of maxima greater than the threshold. This number is the final determined number of modes K. For example, in a vibration signal from a pipeline containing two corrosion defects, after processing, the power spectrum shows three significant peaks at 150Hz, 420Hz, and 680Hz (all higher than the threshold). In this case, the number of modes K is determined to be 3, ensuring that variational mode decomposition can accurately separate these three defect-related characteristic frequency components.

[0065] Step S204: Extract multiple spectral shape features from the power spectrum, and determine the penalty factor for signal decomposition based on a comprehensive evaluation of the multiple spectral shape features.

[0066] Among them, the spectral shape feature can be used to characterize the overall distribution characteristics of the power spectrum. Through comprehensive evaluation, the penalty factor can be adaptively adjusted to optimize the effect of variational mode decomposition. Specifically, feature extraction is performed on the power spectrum to obtain spectral flatness and spectral centroid features. The spectral flatness feature can be achieved by calculating the flatness index of the power spectrum, defined as the ratio of the geometric mean to the arithmetic mean of the power spectrum. The closer the value is to 1, the flatter the spectrum, and the more noise components may be in the signal; conversely, the smaller the value, the steeper the spectrum, and the more concentrated the signal energy is at a specific frequency. The spectral centroid feature is a frequency-weighted average, with the weight being the power spectral density at each frequency point. It reflects the concentrated location of signal energy in the frequency domain. A spectral centroid biased towards the high-frequency band may indicate that the signal contains more high-frequency noise or impulse components. The initial penalty factor is obtained by linearly combining the spectral flatness feature and the spectral centroid feature. For example, an initial penalty factor α0 can be set as α0 = w1 * F + w2 * C, where F is the normalized spectral flatness index, C is the normalized spectral centroid, and w1 and w2 are preset weighting coefficients, which can be determined based on historical data or experience, such as w1 = 0.6 and w2 = 0.4, to balance the influence of flatness and centroid on the penalty factor. The initial penalty factor is mapped to a parameter range corresponding to the number of modes to obtain the penalty factor. Different numbers of modes K correspond to different reasonable penalty factor ranges. For example, when K = 3, the recommended range for the penalty factor is [1000, 3000]; when K = 5, the recommended range is [2000, 5000]. The initial penalty factor α0 is adjusted to this range through linear or nonlinear mapping (such as mapping based on the Sigmoid function) to obtain the final penalty factor α used for variational mode decomposition. For example, if the initial penalty factor α0 is 0.5 and the parameter range corresponding to the number of modes K=4 is [1500, 4000], it can be mapped to 2750 by the formula α=1500+(4000-1500)*α0, ensuring that the penalty factor matches the number of modes and improving the decomposition accuracy.

[0067] Step S206: Using the number of modes and the penalty factor, perform variational mode decomposition on the original time series signal to obtain decomposed sub-signals.

[0068] Variational Mode Decomposition (VMD) can be an adaptive signal decomposition method. Specifically, it can involve constructing and solving a constrained variational problem to decompose the original signal into K modal components (i.e., decomposed sub-signals) with center frequencies and finite bandwidths. In this step, the number of modes K determined in step S202 and the penalty factor α determined in step S204 are used as key input parameters for the VMD algorithm. The VMD algorithm initializes the center frequency and bandwidth of each modal component, and then continuously updates each modal component through alternating iterative optimization to minimize the sum of the estimated bandwidths of all modal components, while ensuring that the sum of the decomposed modal components can reconstruct the original signal. During the iteration process, the penalty factor α mainly affects the sparsity and bandwidth control of the signal. A larger α value will make the bandwidth of the modal components narrower, which is beneficial for separating signal components with similar frequencies, but may lead to over-decomposition; a smaller α value may make the bandwidth of the modal components wider, resulting in mode aliasing. The α value dynamically determined based on the spectral shape characteristics in step S204 can work in conjunction with the number of modes K to effectively avoid the aforementioned problems and ensure that each decomposed sub-signal has a clear physical meaning and independent frequency characteristics. For example, for an original time-series signal containing 5 modes, using a determined penalty factor, the VMD algorithm will decompose the original signal into 5 single-component sub-signals. Each sub-signal corresponds to a component with a different center frequency in the original signal. These components may correspond to normal vibration of the pipeline, vibration caused by a specific defect, and environmental noise, etc.

[0069] Step S208: Extract the time-frequency domain features of the decomposed sub-signals, and divide the decomposed sub-signals into an effective signal subset and a noise signal subset based on the time-frequency domain features.

[0070] Time-frequency domain features can be used to characterize the properties of the decomposed sub-signals in the time and frequency dimensions, providing a multi-dimensional basis for distinguishing effective signals from noise signals. Specifically, for each decomposed sub-signal, its time-domain and frequency-domain features can be extracted. Time-domain features may include, but are not limited to: the signal's peak factor (the ratio of peak value to root mean square value, used to characterize the impulse characteristics in the signal), kurtosis (the fourth central moment, sensitive to abnormal impulses in the signal), waveform factor (the ratio of root mean square value to absolute mean value, reflecting changes in the signal waveform), and impulse index (the ratio of peak value to absolute mean value, highlighting the impulse components in the signal). Frequency-domain features may include, but are not limited to: the centroid frequency of the signal power spectrum (reflecting the main frequency location of energy concentration), mean square frequency (the weighted average of the squares of the frequencies, characterizing the dispersion of the frequency distribution), and frequency band energy proportion (for example, calculating the proportion of the sub-signal's energy within a preset pipeline defect characteristic frequency band to the total energy). The extracted time-frequency domain features are normalized to eliminate the influence of dimensions. Then, the decomposed sub-signals are classified using preset classification rules or machine learning models. For example, a kurtosis threshold can be set. When the kurtosis value of a sub-signal exceeds this threshold, it is determined that it may contain characteristics of impact defects and is initially classified as a subset of valid signals. Simultaneously, considering the frequency band energy proportion, if the energy proportion of the sub-signal within the defect characteristic frequency band is higher than a set proportion (e.g., 30%), it is further confirmed as a valid signal. For sub-signals with low time-domain characteristics (e.g., peak factor, kurtosis) and a very small energy proportion within the characteristic frequency band, they are classified as a subset of noise signals. In a specific example, a decomposed sub-signal has a kurtosis value of 5.2 (with a set threshold of 3.5) and an energy proportion of 45% within the 200-800Hz range (the known characteristic frequency range of pipe corrosion defects), thus this sub-signal is classified as a subset of valid signals; while another decomposed sub-signal has a kurtosis value of 1.8 and an energy proportion of only 8% within the same frequency band, thus it is classified as a subset of noise signals.

[0071] Step S210: Perform noise suppression processing on the decomposed sub-signals in the noise signal subset, and merge the processed noise signal subset with the effective signal subset to reconstruct a noise-reduced time-series signal; wherein, the original time-series signal is used to characterize the structural state of the pipeline to be detected.

[0072] The decomposed sub-signals within the noise signal subset can be subjected to noise suppression processing, specifically using threshold-based wavelet denoising methods or adaptive filtering algorithms. For example, for wavelet denoising, a suitable wavelet basis function (such as the db4 wavelet) and decomposition level (such as 3 levels) can be selected to perform wavelet decomposition on the noise sub-signals, obtaining wavelet coefficients for different frequency bands. Then, a threshold is determined based on the noise level estimate (such as the standard deviation of the high-frequency coefficients obtained from the decomposition), and soft thresholding is performed on the high-frequency wavelet coefficients (i.e., setting coefficients with absolute values ​​less than the threshold to zero and subtracting the threshold from coefficients greater than the threshold) to suppress noise components. After processing, an inverse wavelet transform is performed on the thresholded wavelet coefficients to obtain the denoised noise sub-signals. By linearly superimposing all the noise-suppressed sub-signals with the decomposed sub-signals in the effective signal subset, the denoised time-series signal can be reconstructed. Compared to the original time-series signal, this reconstructed signal effectively removes most irrelevant noise interference and retains effective feature information related to the pipeline structure state, thus facilitating the subsequent accurate identification and assessment of pipeline defects. For example, the original vibration signal is mixed with background noise from equipment operation and environmental interference. After the above decomposition, classification, denoising and reconstruction steps, the resulting denoised time sequence signal can more clearly highlight the characteristic vibration frequency and impact pulse caused by pipeline corrosion or cracks, laying a solid foundation for subsequent defect location and severity analysis.

[0073] The aforementioned pipeline inspection method acquires the original time-series signal of the pipeline, adaptively determines the number of modes and penalty factor required for variational mode decomposition, accurately decomposes the original signal, and then filters the effective signal and noise signal through time-frequency domain features. After suppressing the noise signal, it is merged and reconstructed with the effective signal, achieving high-quality noise reduction of the pipeline vibration signal and providing a more reliable data foundation for subsequent pipeline defect identification and assessment. This method can dynamically adjust the decomposition parameters according to the signal's own characteristics, effectively avoiding mode aliasing or over-decomposition problems, significantly improving the accuracy and robustness of pipeline structural state information extraction, and is particularly suitable for pipeline inspection scenarios under complex working conditions.

[0074] In one embodiment, extracting the peak distribution in the power spectrum and determining the number of modes required for signal decomposition based on the peak distribution includes:

[0075] Determine the background noise level in the power spectrum; determine the peak threshold based on the background noise level.

[0076] In the power spectrum, data greater than the peak threshold are matched.

[0077] The number of data points exceeding the peak threshold is determined, and this number of data points is defined as the number of modalities.

[0078] The background noise level can be determined by statistically analyzing the low-energy regions of the power spectrum (e.g., the bottom 30% of the frequency bands). The mean and standard deviation are calculated, and the mean plus a factor of the standard deviation (e.g., three times) is used as the initial background noise estimate. If isolated high-energy points (potentially noise spikes) exist within this region, median filtering (window size of 5 frequency points) is applied for smoothing before recalculation. The peak threshold can be dynamically adjusted based on the signal-to-noise ratio (SNR). For example, when the SNR is higher than 20 dB, the threshold is set to 3 times the background noise level; when the SNR is between 10 and 20 dB, the threshold is set to 2.5 times the background noise level; and when the SNR is lower than 10 dB, the threshold is set to 2 times the background noise level to avoid missing valid peaks in low SNR scenarios. When matching peak values ​​in the power spectrum, an adaptive peak search algorithm can be used: calculate the first derivative of the power spectrum to find all points where the derivative is zero; combine the second derivative to determine its concavity and convexity, and filter out local maxima; compare the energy values ​​of these maxima with the peak threshold, and count the number of maxima greater than the threshold. This number is the final determined number of modes K. For example, in a vibration signal from a pipeline containing two corrosion defects, after processing, the power spectrum shows three significant peaks at 150Hz, 420Hz, and 680Hz (all higher than the threshold). In this case, the number of modes K is determined to be 3, ensuring that variational mode decomposition can accurately separate these three defect-related characteristic frequency components.

[0079] In one exemplary embodiment, the background noise level can be determined by statistically analyzing low-energy regions in the power spectrum (e.g., the frequency bands with the lowest energy values ​​in the bottom 30%), calculating their mean and standard deviation, and using the mean plus a multiple of the standard deviation (e.g., three times) as the initial background noise estimate. If isolated high-energy points (potentially noise spikes) exist within this region, they are smoothed using median filtering (with a window size of 5 frequency points) before recalculation. The peak threshold can be dynamically adjusted based on the signal-to-noise ratio (SNR). For example, when the SNR is higher than 20 dB, the threshold is set to 3 times the background noise level; when the SNR is between 10 and 20 dB, the threshold is set to 2.5 times the background noise level; and when the SNR is lower than 10 dB, the threshold is set to 2 times the background noise level to avoid missing valid peaks in low SNR scenarios. When matching peak values ​​in the power spectrum, an adaptive peak search algorithm can be used: calculate the first derivative of the power spectrum to find all points where the derivative is zero; combine the second derivative to determine its concavity and convexity, and filter out local maxima; compare the energy values ​​of these maxima with the peak threshold, and count the number of maxima greater than the threshold. This number is the final determined number of modes K. For example, in a vibration signal from a pipeline containing two corrosion defects, after processing, the power spectrum shows three significant peaks at 150Hz, 420Hz, and 680Hz (all higher than the threshold). In this case, the number of modes K is determined to be 3, ensuring that variational mode decomposition can accurately separate these three defect-related characteristic frequency components.

[0080] In this embodiment, by accurately estimating the background noise level of the power spectrum and dynamically adjusting the peak threshold, combined with an adaptive peak search algorithm to determine the number of modes, intelligent and data-driven setting of the number of modes in the variational mode decomposition parameters is achieved. This approach avoids the subjectivity and limitations of traditional empirical setting of the number of modes, and can adaptively match the optimal number of decomposed modes according to the actual spectral characteristics of different pipeline vibration signals. This ensures that each decomposed modal component corresponds to a real characteristic frequency component in the signal, providing a scientific and reliable foundation for subsequent signal denoising and feature extraction.

[0081] In one embodiment, extracting multiple spectral shape features from the power spectrum and determining a penalty factor for signal decomposition based on a comprehensive evaluation of the multiple spectral shape features includes:

[0082] Feature extraction is performed on the power spectrum to obtain spectral flatness features and spectral centroid features;

[0083] The initial penalty factor is obtained by linearly combining the spectral flatness feature and the spectral centroid feature.

[0084] The initial penalty factor is mapped to a parameter range corresponding to the number of modes to obtain the penalty factor.

[0085] The spectral flatness feature can be characterized by the Spectral Flatness Measure (SFM), which is defined as the ratio of the geometric mean to the arithmetic mean of the power spectrum. The formula is SFM = exp(∑(log(P(f_i))) / N) / (∑P(f_i) / N), where P(f_i) is the power spectrum value at frequency point f_i, and N is the total number of frequency points. A value closer to 1 indicates a flatter spectrum and potentially more noise in the signal; a smaller value indicates a steeper spectrum and potentially more significant peaks. The spectral centroid feature is represented by the Spectral Centroid Frequency (SC), calculated as SC = ∑(f_i*P(f_i)) / ∑P(f_i). It reflects the concentration of signal energy on the frequency axis; a higher centroid indicates richer high-frequency components. After obtaining the spectral flatness index and spectral centroid, they need to be normalized. For example, the spectral flatness index and the spectral centroid should be normalized to the [0,1] interval (the upper and lower limits can be set according to the common frequency range of the pipeline vibration signal, such as 0-2000Hz). Then, according to the degree of influence of the two on the penalty factor, the corresponding weighting coefficients w1 and w2 are set, and a linear combination is performed to obtain the initial penalty factor α0, that is, α0=w1*SFM_norm + w2*SC_norm, where SFM_norm is the normalized spectral flatness index and SC_norm is the normalized spectral centroid.

[0086] In one exemplary embodiment, the weighting coefficients w1 and w2 can be set to 0.6 and 0.4 respectively to highlight the influence of spectral flatness on the penalty factor. For example, when the spectral flatness index SFM of a certain pipeline vibration signal is 0.3 (normalized SFM_norm=0.3), and the spectral centroid SC is 800Hz (if the frequency range is set to 0-2000Hz, then normalized SC_norm=800 / 2000=0.4), then the initial penalty factor α0=0.6*0.3+0.4*0.4=0.18+0.16=0.34. Subsequently, according to the number of modes K determined in step S203, the initial penalty factor α0 is mapped to the corresponding parameter range. For example, when K=3, the preset parameter range is [2000, 5000]. Using the linear mapping formula α=α_min+(α0*(α_max-α_min)), where α_min=2000 and α_max=5000, we get α=2000+(0.34*(5000-2000))=2000+1020=3020, meaning the final determined penalty factor is 3020. If the number of modes K=5, the corresponding parameter range is [3000, 7000]. Then α=3000+(0.34*(7000-3000))=3000+1360=4360. This method of dynamically adjusting the penalty factor range based on the number of modes allows for better coordination between the penalty factor and the number of modes, avoiding over-decomposition or mode aliasing caused by an increase in the number of modes. For example, when a signal contains many modal components (with a large K value), a larger penalty factor is usually needed to constrain the bandwidth of each mode and prevent mutual interference between components. This can be achieved by shifting the parameter range upward as a whole.

[0087] In this embodiment, the initial penalty factor is obtained by extracting the spectral flatness and spectral centroid features of the power spectrum and linearly combining them. This initial factor is then mapped to the corresponding parameter range based on the number of modes, enabling dynamic adjustment of the penalty factor. This method of determining the penalty factor based on spectral shape features allows it to be adapted to the frequency distribution characteristics and the number of decomposed modes of the signal. This avoids the over-decomposition or mode aliasing problems that may occur in complex signal decomposition with a fixed penalty factor, further improving the accuracy and stability of variational mode decomposition and providing a reliable guarantee for subsequent effective signal processing.

[0088] In one embodiment, extracting the time-frequency domain features of the decomposed sub-signals specifically includes:

[0089] The steepness of the decomposed sub-signal in the time domain is determined to obtain the kurtosis feature.

[0090] Based on the high-frequency energy and total energy in the power spectrum, the high-frequency energy ratio characteristics of the power spectrum are determined.

[0091] The kurtosis feature is combined with the high-frequency energy ratio feature to form the time-frequency domain feature vector used to distinguish between signal and noise.

[0092] The kurtosis feature is determined by calculating the kurtosis value of the time-domain sequence of the decomposed sub-signals, and its calculation formula is Kurtosis=E[(x-μ)]. 4 ] / (σ 4 ), where E[·] represents the expected value, μ is the signal mean, and σ is the signal standard deviation. Kurtosis is used to measure the steepness of the signal probability density function. For signals containing impact components (such as vibration impact caused by pipe defects), the kurtosis value is usually large; while the kurtosis value of a stable noise signal is close to 3 (the kurtosis value of a normal distribution signal is 3). To calculate the high-frequency energy proportion characteristic, it is first necessary to determine the high-frequency range in the signal power spectrum. This range can be set according to the characteristic frequencies generated by common defects in the pipe to be detected (such as corrosion and cracks). For example, for metal pipes, the high-frequency range can be set to 500-2000Hz. Then, the sum of the power spectrum energy of all frequency points in this high-frequency range is calculated and divided by the total power spectrum energy of the entire frequency range to obtain the high-frequency energy proportion. For example, if the total energy of the power spectrum of the decomposed sub-signal in the 500-2000Hz frequency band is 850 and the total energy is 2000, then the high-frequency energy proportion is 850 / 2000=0.425 (i.e., 42.5%). The calculated kurtosis value and high-frequency energy ratio are used as two dimensions to form a time-frequency domain feature vector, for example (Kurtosis=6.8, HighFrequencyRatio=0.38). This vector can comprehensively characterize the characteristics of the decomposed sub-signal from the perspectives of time-domain impulse characteristics and frequency-domain energy distribution, providing a quantitative basis for the subsequent classification of effective signals and noise signals.

[0093] In one exemplary embodiment, when the decomposed sub-signal has a time-domain kurtosis of 7.2 and a high-frequency energy proportion of 0.45, its feature vector (7.2, 0.45) can be clearly distinguished from the noise sub-signal with a kurtosis of 2.9 and a high-frequency energy proportion of 0.12. This feature combination method can effectively capture the impact of pipeline defect signals in the time domain and the energy concentration characteristics in the frequency domain, providing highly discriminative input features for subsequent signal classification models based on support vector machines or decision trees.

[0094] In this embodiment, by combining kurtosis features and high-frequency energy proportion features, a time-frequency domain feature vector that can comprehensively reflect the impulse characteristics and frequency distribution characteristics of the decomposed sub-signals is constructed. This enables accurate differentiation between effective signals and noise signals.

[0095] In one embodiment, the method further includes:

[0096] Obtain the first spectral flatness parameter of the decomposed sub-signal.

[0097] Using the number of modes and the penalty factor, the denoised time-series signal is subjected to mode decomposition to obtain the residual signal; the second spectral flatness parameter of the residual signal is determined.

[0098] Based on the first spectral flatness parameter and the second spectral flatness parameter, the denoising parameters of the denoised time-series signal are determined; wherein, the denoising parameters are used to characterize the degree of denoising.

[0099] The first spectral flatness parameter is the spectral flatness exponent of the decomposed sub-signal, calculated in the same way as the aforementioned spectral flatness exponent: SFM1 = exp(∑(log(P1(f_i))) / N) / (∑P1(f_i) / N), where P1(f_i) is the power spectral density of the decomposed sub-signal at frequency point f_i. The residual signal refers to the remaining portion of the denoised time-series signal that cannot be explained by the effective modal components after modal decomposition. Its second spectral flatness parameter, SFM2, is calculated as SFM2 = exp(∑(log(P2(f_i))) / N) / (∑P2(f_i) / N), where P2(f_i) is the power spectral density of the residual signal at frequency point f_i. The denoising parameter can be defined as the ratio of the first spectral flatness parameter to the second spectral flatness parameter, i.e., denoising parameter R = SFM1 / SFM2. When the noise reduction effect is good, the spectral flatness of the decomposed sub-signal (effective signal) is low (SFM1 is small, the spectrum is steep, and the characteristics are obvious), while the spectral flatness of the residual signal (mainly noise) is high (SFM2 is large, and the spectrum is close to the flat characteristics of noise). In this case, the R value is small. Conversely, if the noise reduction is insufficient, the residual signal still contains a lot of effective components, so SFM2 is small and the R value is large. For example, if the SFM1 of the decomposed sub-signal is 0.25 and the SFM2 of the residual signal is 0.75, then the noise reduction parameter R = 0.25 / 0.75 ≈ 0.33, indicating that the separation effect between the effective signal and noise is good. If SFM1 is 0.4 and SFM2 is 0.5, then R = 0.8, indicating that there may still be some effective signal energy in the residual, and the degree of noise reduction needs to be improved. Through this noise reduction parameter, the noise reduction effect can be quantitatively evaluated, and a basis can be provided for whether the decomposition parameters (such as the number of modes or the penalty factor) need to be adjusted in the future, thereby forming a closed-loop optimization mechanism to further ensure the reliability of pipeline vibration signal processing.

[0100] In one exemplary embodiment, when the noise reduction parameter R is greater than a preset threshold (e.g., 0.6), the system can automatically trigger a re-evaluation process for the number of modes K. This involves increasing the number of potential modes by expanding the peak search frequency band or lowering the peak threshold, then recalculating the penalty factor and performing mode decomposition until the noise reduction parameter R falls back below the threshold. For example, in the initial decomposition of a pipeline vibration signal, K=3, and R=0.72 (greater than 0.6). After the system starts optimization, a sub-significant peak at 850Hz (not detected under the original threshold) is re-identified in the power spectrum. K is adjusted to 4, corresponding to an upward shift in the penalty factor parameter range. After re-decomposition, the SFM2 of the residual signal increases to 0.8, and R=0.3 / 0.8=0.37, meeting the noise reduction requirements. This closed-loop optimization mechanism based on the spectral flatness ratio achieves adaptive adjustment of the pipeline vibration signal noise reduction process, avoids the lag of manual intervention, and significantly improves the robustness of signal processing under complex operating conditions.

[0101] In this embodiment, the impact of noise reduction processing on the signal spectral characteristics is quantitatively evaluated by comparing the first spectral flatness parameter of the decomposed sub-signal with the second spectral flatness parameter of the residual signal. This enables dynamic monitoring and feedback of the noise reduction effect. It improves the closed-loop control capability of signal processing in pipeline inspection, ensuring that the denoised time-series signal retains key defect feature information while significantly suppressing noise interference.

[0102] In one embodiment, dividing the decomposed sub-signal into an effective signal subset and a noise signal subset based on the time-frequency domain characteristics includes:

[0103] The time-frequency domain features are input into a preset classification model to obtain the category identifiers corresponding to the time-frequency domain features; wherein, the classification model is trained based on valid signal and noise signal samples identified in historical data;

[0104] Based on the category identifier, the decomposed sub-signals are divided into a valid signal subset and a noise signal subset.

[0105] The pre-defined classification model can employ Support Vector Machine (SVM), Random Forest, or deep learning models (such as Convolutional Neural Networks). During model training, a large number of vibration signal samples under different pipeline operating conditions need to be collected. These samples are manually or through expert annotation, labeled as "effective signals" (containing defect features) or "noise signals" (mainly background interference). The time-frequency domain feature vectors of each sample (such as kurtosis and high-frequency energy proportion features) are extracted as input, and the corresponding category label is output to construct the training dataset. Taking SVM as an example, by selecting an appropriate kernel function (such as radial basis function) and penalty parameters, the training dataset is trained to establish a nonlinear mapping relationship between the time-frequency domain feature vectors and the category label. In the model application phase, the time-frequency domain feature vectors of the decomposed sub-signals are input into the trained classification model. The model outputs the corresponding category label (e.g., "effective" or "noise") based on the learned mapping relationship, thereby achieving automatic classification of the decomposed sub-signals. For example, if a sub-signal has a kurtosis of 6.5 and a high-frequency energy proportion of 0.4, and this feature vector (6.5, 0.4) is input into a trained SVM model, the model outputs a category label of "valid signal," thus classifying the sub-signal into the valid signal subset. If another sub-signal has a kurtosis of 3.2 and a high-frequency energy proportion of 0.1, the model outputs "noise signal," and it is classified into the noise signal subset. In this embodiment, the time-frequency domain features of the sub-signals are automatically identified and classified using a preset classification model, achieving intelligent differentiation between valid and noise signals. This avoids the limitations of traditional threshold classification and can adapt to the signal characteristics of different pipe types, defect modes, and operating conditions, providing accurate input data for subsequent signal reconstruction and defect identification.

[0106] In one exemplary embodiment, data augmentation techniques can be introduced during the training phase, such as temporal stretching, noise addition, and kurtosis fine-tuning of the original vibration signal samples to generate diverse training samples. Simultaneously, cross-validation methods (such as 5-fold cross-validation) are used to optimize model hyperparameters. For example, grid search is used to determine the optimal kernel function parameters (such as the gamma value of the radial basis function) and penalty coefficient C of the SVM to avoid overfitting. After model deployment, a periodic update mechanism can be set up. When new pipeline inspection data (such as novel defect signals or noise samples under different operating conditions) is collected, the classification model is incrementally trained to continuously adapt to signal changes in real-world applications. For example, in the early stages of operation of a city's water supply pipeline inspection system, the classification model's accuracy in identifying "corrosion perforation" defect signals was 89%. After incremental training with 200 newly collected corrosion defect samples from the past three months, the accuracy increased to 95%, and the ability to identify novel defects such as "loose pipe joints" was also enhanced. This classification model construction method, which combines data augmentation, cross-validation, and incremental training, ensures the accuracy and robustness of the distinction between effective and noisy signals, laying a solid foundation for the accurate identification of pipeline defects.

[0107] In this embodiment, by inputting the time-frequency domain feature vector into a preset classification model, the automatic classification of the decomposed sub-signals is achieved by utilizing the model's classification capability of the feature vector. This realizes intelligent and automated division of the decomposed sub-signals, avoiding the subjectivity and uncertainty of traditional manual experience judgment, thereby enabling more objective and efficient screening of effective signals containing pipeline defect features.

[0108] In one exemplary embodiment, the pipeline inspection method may be implemented in the following manner, specifically including:

[0109] Data Input and Parameter Initialization: Let the input be X={x1,x2,…,xn}, where xi∈Rn is the sampled value of the n channels at mileage i (unit: millimeters); preset initialization hyperparameters related to adaptive parameter estimation; wherein, the preset initialization hyperparameters related to adaptive parameter estimation may include: preset range of decomposition mode number: Klimits∈[Kmin,Kmax], used to limit the search interval of the initial mode number in VMD; preset range of penalty factor: α∈[αmin, αmax], used to limit the value range of the bandwidth adjustment parameter in VMD; residual spectral flatness threshold (target_res_flat): specifies the target level of spectral flatness of the residual signal that should be achieved during parameter estimation; minimum spectral flatness growth threshold (min_gain): used to determine the minimum increase in spectral flatness when the mode number K increases, as a criterion for whether to continue iteration; maximum number of parameter fine-tuning times (max_extra_k): controls the maximum number of small-range trials allowed based on the initial K value, to avoid computational redundancy caused by excessive search.

[0110] DC component removal from the signal: Extract the original data sequence of the input signal channel by channel to ensure independent processing of the channel signal; in order to improve the accuracy of subsequent frequency domain analysis and mode decomposition, the input signal needs to be processed to remove the DC component.

[0111] Peak analysis and adaptive estimation of initial parameter K: Power spectrum calculation and preprocessing, performing frequency domain transformation and smoothing operations on the original signal to provide basic power spectrum information for subsequent peak analysis; Peak intensity threshold estimation, adaptively estimating the peak significance intensity threshold based on robust statistical methods (such as median and MAD) to improve the robustness and generalization of peak identification; Peak spacing constraint and identification rule setting, setting minimum peak spacing and selective filtering rules to avoid false peak and overlapping peak identification and ensure the accuracy of peak identification; Noise low estimation and peak significance identification, estimating local background noise in the spectrum and judging the significance of the true peak based on it, eliminating low-confidence perturbations; Peak search and initial K estimation, combining the spectral structure and identification rules to complete the peak search and derive the optimal initial mode decomposition number K, providing reference parameters for subsequent VMD.

[0112] Spectral morphology analysis and adaptive estimation of penalty factor α: Spectral feature extraction and normalization: Spectral morphology indicators such as spectral flatness, bandwidth concentration, peak sharpness (Q value), spectral centroid, and energy spread are extracted and normalized to a standard scale; Comprehensive feature score construction: A spectral morphology score is constructed by combining multiple normalized features in a weighted linear combination to reflect the structural complexity of the signal and the decomposition requirements; Parameter mapping and range limitation: The morphology score is mapped to a preset penalty factor range [αmin, αmax] to obtain an initial penalty factor α value, which serves as an important control parameter for subsequent VMD decomposition; Analysis and interpretation and logical control: When the signal spectrum exhibits high concentration, high Q value, and sharp energy aggregation, the system will assign a larger α; conversely, when the spectral structure tends to be flat and divergent, a smaller α is used to maintain decomposition stability.

[0113] Adaptive Mode Decomposition and Feature-Driven Denoising: Adaptive VMD operation and residual evaluation: Based on initial parameters K and α, VMD decomposition is performed to obtain the initial mode sequence; the decomposition quality is evaluated by residual spectral flatness and mode overlap, and feedback correction is performed accordingly; Parameter adaptive fine-tuning mechanism: If the decomposition result does not meet the flatness and overlap conditions, the K value is dynamically adjusted according to optimization criteria (e.g., the flatness decrease is not sufficient when the mode increases) until the optimal decomposition criteria are met or the maximum number of iterations is reached; Modal feature extraction and clustering classification: Modal features (DFA exponent, spectral centroid, spectral sharpness, kurtosis, etc.) are extracted and classified into signal modes and noise modes by K-Means clustering, providing a basis for subsequent denoising processing; Noise mode suppression and signal reconstruction: Wavelet threshold denoising is performed on the noise modes to preserve the original form of the signal modes, and then the processed modes are reconstructed to generate the final denoised signal output;

[0114] Processing Result Evaluation: Multi-dimensional statistical evaluation is conducted, quantitatively assessing the denoised signal from multiple dimensions including time domain, frequency domain, spatial domain, and signal-to-noise ratio (SNR). Key statistical features are extracted, including root mean square (RMS), peak-to-peak value, SNR, spectral centroid, high-frequency energy proportion, mean correlation coefficient of adjacent channels, and abnormal channel proportion, to comprehensively measure the processing effect and signal quality improvement. Intelligent Prediction Effect Verification: The processed signal is input into a pre-trained AI diagnostic model to perform automatic identification and classification tasks of pipeline structures and defect types, covering standard pipeline sections, pipeline components, typical defect morphologies, and comprehensive discrimination effect evaluation of composite structure and defect combination scenarios. Visual Result Display: The differences in signal characteristics in the time and frequency domains before and after processing are visualized, including time-domain waveforms, modal decomposition diagrams, and feature parameter trend diagrams, to intuitively show the denoising effect, feature evolution, and classification boundaries, providing visual support for technical analysis and decision-making. Processing Efficiency Domain Time Optimization Evaluation: The processing time of samples is compared with heuristic optimization and grid search methods. The extracted key statistical features include root mean square (RMS), peak-to-peak value, signal-to-noise ratio, spectral centroid, high-frequency energy proportion, mean correlation coefficient of adjacent channels, and abnormal channel proportion, used to comprehensively measure the processing effect and signal quality improvement; the signal processing results can be used as input to train the AI ​​model and evaluate its recognition performance under various working conditions; specifically, the AI ​​model training before data processing can be as follows: Figure 3 , Figure 4 , Figure 5 As shown, where, Figure 3 and Figure 4 These represent the model training curve and the model validation curve, respectively. Figure 5 A bar chart showing model accuracy and recall; the training status of the AI ​​model after data processing can be seen as follows. Figure 6 , Figure 7 , Figure 8 As shown, where, Figure 3 and Figure 7 These represent the model training curve and the model validation curve, respectively. Figure 8 This is a bar chart showing the model's accuracy and recall; specifically, the average distribution of the root mean square values ​​can be represented as follows: Figure 9 As shown, the average signal-to-noise ratio of the root mean square value can be expressed as follows: Figure 10 As shown, the average correlation between adjacent channels can be expressed as follows: Figure 11 As shown, etc.

[0115] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0116] Based on the same inventive concept, this application also provides a pipeline inspection device for implementing the pipeline inspection method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more pipeline inspection device embodiments provided below can be found in the limitations of the pipeline inspection method described above, and will not be repeated here.

[0117] In one embodiment, such as Figure 12 As shown, a pipeline inspection device 120 is provided, including: a signal acquisition module 121, a peak extraction module 123, a feature extraction module 125, a signal decomposition module 127, and a signal reconstruction module 129, wherein:

[0118] The signal acquisition module is used to acquire the original timing signals of the pipeline under test;

[0119] The peak extraction module is used to extract the power spectrum of the original time-series signal; and to extract the peak distribution in the power spectrum, and determine the number of modes required for signal decomposition based on the peak distribution;

[0120] The feature extraction module is used to extract multiple spectral shape features from the power spectrum and determine the penalty factor for signal decomposition based on a comprehensive evaluation of the multiple spectral shape features.

[0121] The signal decomposition module is used to perform variational mode decomposition on the original time-series signal using the number of modes and the penalty factor to obtain decomposed sub-signals;

[0122] The feature extraction module is further configured to extract the time-frequency domain features of the decomposed sub-signals, and divide the decomposed sub-signals into an effective signal subset and a noise signal subset based on the time-frequency domain features;

[0123] The signal reconstruction module is used to perform noise suppression processing on the decomposed sub-signals in the noise signal subset, and merge the processed noise signal subset with the effective signal subset to reconstruct a denoised time-series signal, wherein the denoised time-series signal is used to characterize the structural state of the pipeline to be detected.

[0124] In one embodiment, the peak extraction module is further configured to:

[0125] Determine the background noise level in the power spectrum; determine the peak threshold based on the background noise level;

[0126] In the power spectrum, data exceeding the peak threshold are matched;

[0127] The number of data points exceeding the peak threshold is determined, and this number of data points is defined as the number of modalities.

[0128] In one embodiment, the feature extraction module is further configured to:

[0129] Feature extraction is performed on the power spectrum to obtain spectral flatness features and spectral centroid features;

[0130] The initial penalty factor is obtained by linearly combining the spectral flatness feature and the spectral centroid feature.

[0131] The initial penalty factor is mapped to a parameter range corresponding to the number of modes to obtain the penalty factor.

[0132] In one embodiment, the feature extraction module is further configured to:

[0133] Determine the kurtosis of the decomposed sub-signals in the time domain to obtain kurtosis features;

[0134] Based on the high-frequency energy and total energy in the power spectrum, determine the high-frequency energy proportion characteristics of the power spectrum;

[0135] The kurtosis feature is combined with the high-frequency energy ratio feature to form the time-frequency domain feature vector used to distinguish between signal and noise.

[0136] In one embodiment, the device further includes:

[0137] The parameter acquisition module is used to acquire the first spectral flatness parameter of the decomposed sub-signal;

[0138] The parameter acquisition module is further configured to perform mode decomposition on the denoised time-series signal using the number of modes and the penalty factor to obtain the residual signal; and determine the second spectral flatness parameter of the residual signal.

[0139] The signal denoising module is used to determine the denoising parameters of the denoised time-series signal based on the first spectral flatness parameter and the second spectral flatness parameter; wherein the denoising parameters are used to characterize the degree of denoising.

[0140] Each module in the aforementioned pipeline inspection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0141] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 13 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a pipe drain detection method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0142] Those skilled in the art will understand that Figure 13 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0143] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0144] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0145] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0146] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A pipeline inspection method, characterized in that, The method includes: Obtain the original timing signal of the pipeline to be tested; Extract the power spectrum of the original time-series signal; and extract the peak distribution in the power spectrum, and determine the number of modes required for signal decomposition based on the peak distribution; Extract multiple spectral shape features from the power spectrum, and determine the penalty factor for signal decomposition based on a comprehensive evaluation of the multiple spectral shape features; Using the number of modes and the penalty factor, variational mode decomposition is performed on the original time-series signal to obtain decomposed sub-signals; Extract the time-frequency domain features of the decomposed sub-signals, and divide the decomposed sub-signals into an effective signal subset and a noise signal subset based on the time-frequency domain features; The decomposed sub-signals in the noise signal subset are subjected to noise suppression processing, and the processed noise signal subset is merged with the effective signal subset to reconstruct a noise-reduced time-series signal; wherein, the original time-series signal is used to characterize the structural state of the pipeline to be detected.

2. The method according to claim 1, characterized in that, The step of extracting the peak distribution in the power spectrum and determining the number of modes required for signal decomposition based on the peak distribution includes: Determine the background noise level in the power spectrum; determine the peak threshold based on the background noise level; In the power spectrum, data exceeding the peak threshold are matched; The number of data points exceeding the peak threshold is determined, and this number of data points is defined as the number of modalities.

3. The method according to claim 1, characterized in that, The step of extracting multiple spectral shape features from the power spectrum and determining a penalty factor for signal decomposition based on a comprehensive evaluation of the multiple spectral shape features includes: Feature extraction is performed on the power spectrum to obtain spectral flatness features and spectral centroid features; The initial penalty factor is obtained by linearly combining the spectral flatness feature and the spectral centroid feature. The initial penalty factor is mapped to a parameter range corresponding to the number of modes to obtain the penalty factor.

4. The method according to claim 1, characterized in that, The extraction of time-frequency domain features of the decomposed sub-signals includes: Determine the kurtosis of the decomposed sub-signals in the time domain to obtain kurtosis features; Based on the high-frequency energy and total energy in the power spectrum, determine the high-frequency energy proportion characteristics of the power spectrum; The kurtosis feature is combined with the high-frequency energy ratio feature to form the time-frequency domain feature vector used to distinguish between signal and noise.

5. The method according to claim 1, characterized in that, The method further includes: Obtain the first spectral flatness parameter of the decomposed sub-signal; Using the number of modes and the penalty factor, the denoised time-series signal is subjected to mode decomposition to obtain the residual signal; the second spectral flatness parameter of the residual signal is determined. Based on the first spectral flatness parameter and the second spectral flatness parameter, the denoising parameters of the denoised time-series signal are determined; wherein, the denoising parameters are used to characterize the degree of denoising.

6. The method according to claim 1, characterized in that, The step of dividing the decomposed sub-signal into an effective signal subset and a noise signal subset based on the time-frequency domain characteristics includes: The time-frequency domain features are input into a preset classification model to obtain the category identifiers corresponding to the time-frequency domain features; wherein, the classification model is trained based on valid signal and noise signal samples identified in historical data; Based on the category identifier, the decomposed sub-signals are divided into a valid signal subset and a noise signal subset.

7. A pipeline inspection device, characterized in that, The device includes: The signal acquisition module is used to acquire the original timing signals of the pipeline under test; The peak extraction module is used to extract the power spectrum of the original time-series signal; and to extract the peak distribution in the power spectrum, and determine the number of modes required for signal decomposition based on the peak distribution; The feature extraction module is used to extract multiple spectral shape features from the power spectrum and determine the penalty factor for signal decomposition based on a comprehensive evaluation of the multiple spectral shape features. The signal decomposition module is used to perform variational mode decomposition on the original time-series signal using the number of modes and the penalty factor to obtain decomposed sub-signals; The feature extraction module is further configured to extract the time-frequency domain features of the decomposed sub-signals, and divide the decomposed sub-signals into an effective signal subset and a noise signal subset based on the time-frequency domain features; The signal reconstruction module is used to perform noise suppression processing on the decomposed sub-signals in the noise signal subset, and merge the processed noise signal subset with the effective signal subset to reconstruct a denoised time-series signal, wherein the denoised time-series signal is used to characterize the structural state of the pipeline to be detected.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images