Industrial equipment pipeline leakage vibration voiceprint detection method and system

Abstract

The invention belongs to the technical field of industrial equipment pipeline detection, and discloses an industrial equipment pipeline leakage vibration voiceprint detection method and system, and the method comprises the steps: installing a vibration sensor on a pipeline, transmitting a collected voiceprint to a central control system, carrying out NRBO-FMD processing, extracting six general feature indexes of the voiceprint, and calculating a weight vector through a CVITW-CV method. The method comprises the following steps: extracting three chaotic feature indexes of voiceprints at the same time, constructing a VFMF theory to obtain a cross membership matrix, obtaining a fuzzy weighted feature vector, calculating a weight vector and the membership matrix through an LSH-FSVM, outputting a fuzzy probability, and detecting a pipeline leakage degree. The leakage degree of the pipeline can be effectively detected, and the method is suitable for scenes with complex working conditions.

Description

Technical Field

[0001] This invention belongs to the field of industrial equipment pipeline inspection technology, specifically relating to a method and system for detecting vibration and acoustic signatures of leaks in industrial equipment pipelines. Background Technology

[0002] Leaks in industrial pipelines are often difficult to detect in their early stages due to the small size of the seepage pores and the insignificant changes in flow rate. If such hidden dangers are not repaired in a timely manner, they not only exacerbate water loss and soil pollution but may also induce secondary disasters such as land subsidence, threatening regional ecological security and infrastructure stability. More seriously, the reverse osmosis of pollutants caused by leaks can directly threaten drinking water quality, exacerbating public health risks. Therefore, building a precise and efficient pipeline leak early warning system is of great value for urban infrastructure operation and maintenance and the protection of residents' health.

[0003] From a fluid mechanics perspective, when a pipeline ruptures, the pressure difference between the inside and outside of the pipe wall causes severe pressure pulsations at the leak point. The dynamic conversion of kinetic and potential energy in this area excites mechanical vibration waves at specific frequencies, and their spectral characteristics are strongly correlated with the degree of pipeline leakage. However, in actual operating conditions, the signal-to-noise ratio of vibration waveforms decreases significantly due to the coupling effects of environmental noise interference, fluctuations in medium flow velocity, and pipe aging, leading to deviations in leak feature extraction. Furthermore, existing detection technologies generally face the dual contradiction of high cost of high-precision equipment and insufficient adaptability to complex scenarios. Developing a detection solution with both early identification capabilities and anti-interference performance within a limited budget has become a core challenge for the industry. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method and system for detecting vibration and acoustic signatures of pipeline leaks in industrial equipment. The system provides accurate identification results and can effectively detect pipeline leaks, making it suitable for scenarios with complex working conditions.

[0005] The present invention discloses a vibration acoustic sound detection method for pipeline leakage in industrial equipment, comprising the following steps: Step 1: Collect vibration sound samples from the pipeline leak as the basis for subsequent detection; Step 2: Perform NRBO-FMD processing on the collected original vibration acoustic text. Use NRBO to optimize the filter length and decomposition order of FMD. Substitute the optimized index into FMD to decompose the original acoustic text to obtain each intrinsic mode function (IMFs). Reconstruct the vibration acoustic text based on the correlation to remove the influence of irrelevant acoustic text. Step 3: Extract key feature indicators based on the reconstructed vibration soundprint, and use the Convolutional Variational Inference Time Warping (CVITW)-Cross-Validation (CV) method to calculate the weight vector of each indicator and quantify its contribution to the detection of different leakage levels. Step 4: Extract chaotic feature indices from the reconstructed vibration acoustic signature and construct a chaotic feature vector; Step 5: Propose a variant fuzzy membership function (VFMF) theory to construct a membership matrix, convert the uncertainty of different chaotic and non-chaotic feature indicators into membership degree information, and construct a chaotic fuzzy cross membership degree matrix to describe the probability distribution characteristics of indicators under different leakage levels. Step 6: Combining the weight vector, membership matrix, and chaotic feature vector, the leakage degree of the pipeline is calculated using Local Sensitive Hash-Fuzzy Support Vector Machine (LSH-FSVM). The order of the largest element in the result vector is taken as the current leakage degree of the pipeline, thus realizing real-time detection of the pipeline leakage degree.

[0006] Furthermore, in step 2, the vibration sound of pipeline leakage is processed by NRBO-FMD. NRBO is used to determine the filter length and decomposition order in FMD for the original vibration time domain sound, and the number of IMFs is determined by the center frequency of the components.

[0007] Furthermore, step 2, voiceprint reconstruction, specifically involves: The formula for calculating the correlation coefficient is as follows: , in, This is the original vibration voiceprint. The decomposition mode is represented by l, which represents the number of IMFs. Indicates variance calculation; when Correlation coefficient with original voiceprint When the value exceeds a threshold, it is retained; all others are discarded. The remaining IMFs are then reconstructed to obtain the processed voiceprint. : .

[0008] Furthermore, in step 3, the reconstructed voiceprint... Extract the general feature indicators of each reconstructed voiceprint, and set... This is the acoustic waveform of a non-destructive pipeline vibration, i.e., the reference acoustic waveform. For real-time detection of vibration acoustic signatures; the six general characteristic indicators are: , in, The Time Frequency Coherence Index (TFCI) measures the overall coherence between the reference and detected voiceprints in the time and frequency domains. It uses wavelet coherence calculation to comprehensively reflect the coupling relationship between non-stationary voiceprints. , are the wavelet transforms of the reference voiceprint and the detected voiceprint, respectively, and t and f are the time and frequency domain variables, respectively; , in, The Time Varying Energy Entropy (TVEE) is an indicator used to detect whether the energy distribution of the acoustic signature changes abruptly or becomes discrete due to leakage. For voiceprint in the first Energy probability distribution within a window Divide the time window into total numbers; , in, The Spectral Entropy Difference (SED) is an index that measures the difference in spectral complexity between the baseline acoustic signature and the detected acoustic signature, revealing the degree to which the leaked acoustic signature disturbs the system's spectral structure. , These are the spectral entropy values ​​of the reference voiceprint and the detected voiceprint, respectively. , The reference voiceprint and the detected voiceprint are respectively in the 1st... Energy probability distribution within a window; , in, It is the Wavelet Energy Divergence (WED), which reflects the energy characteristic perturbation of the leakage vibration sound signature at different time scales. , The first and second parts of the reference voiceprint and the detected voiceprint are respectively Layer wavelet energy, Indicates the total number of wavelet decomposition layers; , in, The nonlinear similarity coefficient (NCC) is used to compare the structural similarity between the detected acoustic pattern and the reference acoustic pattern in the reconstructed phase space, thereby capturing the characteristics of nonlinear leakage vibration acoustic patterns. , The reference voiceprint and the detected voiceprint in the reconstructed phase space A trajectory point, , These are the mean values ​​of the trajectory points of the reference acoustic signature and the detected acoustic signature in the reconstructed phase space, respectively. A phase space trajectory point; , in, Multi-Scale Energy-Phase Deviation (MS-EPD) extracts the structural deviation of acoustic signatures at different time scales from both energy and instantaneous phase perspectives, and is sensitive to the temporal and phase abrupt changes in acoustic signatures caused by pipeline leaks. For scale serial numbers, there are a total of One scale; , These represent the energy of the reference voiceprint and the detected voiceprint at different scales; Indicates the proportion of wavelet energy; This represents the weighting factor for energy and phase deviation; Indicates the total duration of the voiceprint. , These represent the detected voiceprint and the reference voiceprint at different scales. Below, the instantaneous phase value at time t.

[0009] Furthermore, in step 3, the CVITW-CV method is used to calculate the weight vector of each general indicator, quantifying the contribution of different indicators to the detection of leakage levels, specifically as follows: Step 3-1: Calculate the CVITW distance matrix; For feature indicators For i=1,2,…,6, calculate the CVITW distance matrix among all samples. This measures the similarity of the temporal changes of this characteristic under different leakage levels; , Where Q is the total number of samples and ; Let the sample , The following iterative relationship exists: , in, As an intermediate inference term, R and S represent samples , The number of elements contained in each. Indicates the i-th feature in the sample and CVITW distance between Align the weights for the convolution kernel. This is the variational inference error term; Step 3-2: Feature discrimination score; For matrix Calculate the distance between samples with the same level of leakage. and the distance between samples with different leakage levels As shown in the following formula: , in, , Indicates the first , Number of samples under each level of leakage; At this point, a certain characteristic index is defined. The discrimination score is The higher the discrimination score, the stronger the ability of the feature to distinguish the degree of leakage; Step 3-3: Assign weights; Based on the discrimination score of different feature indicators Initialize weights ; The original discrimination score for the j-th feature indicator. At this point, the weighted CVITW distance is defined as... ; Steps 3-4: Cross-validation optimization and weight normalization; Five-fold cross-validation is used, and weights are adjusted via gradient descent. The normalized weights are obtained as follows: ; The original weights represent the importance of the j-th feature; the final weight vector is obtained. .

[0010] Furthermore, in step 4, the reconstructed vibration acoustic signature is... Extract various chaotic feature indicators, including: , in, The Lyapunov Exponent (LE) is used because the vibration sound pattern after a valve leak may show a rapid divergence in trajectory, leading to... Increase; among them, This represents the small distance between the two trajectories in the initial phase space. This represents the actual distance between the two trajectories after time t. , in, Correlation Dimension Because valve leakage may complicate the spatial trajectory of the vibration acoustic signature. Increase In distance The probability of finding a pair of phase space trajectory points within the space. , The first The and the first One reconstructed vector, For Heaviside step function; , in, Kolmogorov Entropy This indicator measures the unpredictability of vibration and acoustic signatures after a system valve leaks. It will increase; Embedding dimension of phase space, Using distance as the scale, For time delay, For the embedding dimension is The lower distance is less than The probability of the trajectory point pairs.

[0011] Furthermore, step 5 specifically includes: Step 5-1: Establish the theory of Variational Fuzzy Membership Function (VFMF); Assuming the f-th leakage level, the first sample of each... The general characteristic indicators are All general characteristic indicators constitute the basic feature vector. ,have: , Assume that the g-th chaotic feature index of each sample under the f-th leakage level is All chaotic feature indicators constitute the basic feature vector. ,have The cross-feature matrix is ​​obtained as shown in the following formula: , To calculate whether a sample belongs to the f-th leakage level, a variogram fuzzy membership function is established. : , in, This represents the membership degree corresponding to each feature. For the first The standard deviation of each feature Represents each eigenvalue, Indicates the first under the same leakage level The feature values ​​corresponding to each sample For the first The mean of the samples; Step 5-2: Construct the membership matrix: .

[0012] Furthermore, step 6 specifically includes: Step 6-1: Weighting of fuzzy features; By combining weights and membership degrees, a fuzzy weighted feature vector is generated. ,in: , in, For the first A fuzzy weighted feature, For cross-membership weights, Weights for chaotic features; , , These represent the f-th leakage level encoding values ​​under voiceprint features, cross-membership degree, and chaotic features, respectively. Step 6-2: Fuzzy membership assignment; Assign fuzzy membership degree to the k-th training sample , This indicates that the sample belongs to the label. Confidence level: , in, express Its real label The overall membership degree; when The closer the value is to 1, the more reliable the sample label is; Step 6-3: Fuzzy constraint optimization; For the f-th level of leakage, the original optimized state of LSH-FSVM is as follows: , in, , These represent the weights and bias coefficients of LSH-FSVM, respectively. The Gaussian kernel function; C is the kernel index, which aims to control the rate of decay of sample similarity; C is the regularization index, which balances the classification margin with the cost of misclassification. It is a random projection hash function; It is an LSH random matrix; This is the LSH offset; Set the LSH quantization step size; These are slack variables, used to adjust for misclassification penalties of unreliable samples; The decision values ​​for U, representing the true label at the level of leakage f, are as follows: , in, The fuzzy weighted feature mean; Step 6-4: Fuzzy probability output; Map each decision value to the probability space: , in, and The scaling index for the f-th leakage level is obtained by maximizing the log-likelihood function. and As shown in the following formula: , in, Custom label values; The predicted probability that feature sample k belongs to leakage level f; the scaling index is updated iteratively using Newton's method. and As shown in the following formula: , Where L is the log-likelihood function. and H represents the scaling index at the nth iteration, and H is the Hessian matrix. The gradient vector is used; the index is gradually adjusted using Newton's iteration method. and This maximizes the likelihood function L, mapping the LSH-FSVM decision values ​​to fuzzy probability outputs for each level of leakage, as shown below: , The state with the highest probability is selected as the final classification result.

[0013] This invention also provides an industrial equipment pipeline leakage vibration and acoustic fingerprint detection system, comprising a pipeline leakage vibration and acoustic fingerprint detection system, an on-site monitoring system, and a core control system, wherein: The pipeline leakage vibration and acoustic sound detection system uses vibration sensors installed on the leaking pipeline to collect the vibration and acoustic sound of the pipeline leakage, preprocess it, and then transmit the processed vibration and acoustic sound to the local control system. The field monitoring system transmits the vibration and acoustic signals received from the leak pipeline detection system to the core control system via wireless communication. The core control system processes the received acoustic signatures to achieve real-time leak detection in pipelines.

[0014] Furthermore, the core control system includes an NRBO-FMD vibration acoustic signature processing section, a section for extracting mixed indicators of vibration acoustic signature characteristics of leaking pipelines, a VFMF-cross membership matrix section, and an LSH-FSVM detection section, among which: The NRBO-FMD vibration acoustic text processing section uses NRBO to optimize FMD indicators and complete the preprocessing of acoustic text. The hybrid index extraction section for vibration acoustic signatures of leaking pipelines extracts multiple general and chaotic feature indices from the preprocessed vibration acoustic signatures. The general feature indices are weighted using CVITW-CV to quantify the contribution of different indices to leak detection. The chaotic feature indices directly constitute the chaotic feature vector. The VFMF-cross membership matrix part is based on VFMF function theory. It constructs a cross membership matrix, integrates general feature information and chaotic feature information, and transforms different indicators into membership information to express the probabilistic characteristics of feature indicators under different leakage levels. The LSH-FSVM detection part takes a fuzzy weighted feature vector as input, which is obtained by combining a general index weight vector, a chaotic fuzzy cross-membership matrix, and a chaotic feature vector. The degree of leakage in the pipeline can be determined through fuzzy inference using LSH-FSVM, enabling real-time detection of pipeline leakage.

[0015] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor, when executing the computer program, implements the vibration and acoustic sound detection method for pipeline leakage in industrial equipment according to the present invention.

[0016] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the vibration and acoustic sound detection method for pipeline leakage in industrial equipment as described in the present invention.

[0017] The beneficial effects of this invention are as follows: 1) This invention simultaneously extracts the general features and chaotic features of pipeline leakage vibration soundprints. The two features are fused once using the proposed variant fuzzy membership function (VFMF) theory to obtain the cross membership matrix. When constructing the fuzzy weighted feature vector, the general feature weight vector, the cross membership matrix, and the non-fuzzy chaotic features are fused a second time, which can more fully capture the features of pipeline leakage vibration soundprints. 2) This invention proposes a Convolutional Variational Inference Time Warping (CVITW)-Cross-Validation (CV) method to calculate the weight vector of general features. Through multiple rounds of cross-validation, time series with different leakage levels are aligned, and the weight vector of each modality parameter is dynamically adjusted. 3) This invention is based on LSH-FSVM, combines fuzzy weighted feature vectors to quantify the contribution probability of each feature to the leakage degree, and introduces a local sensitive hash factor to dynamically adjust the classification boundary. It fully integrates historical data. As the detection data continues to expand and the database continues to enrich, this fuzzy detection method has high robustness. Attached Figure Description

[0018] Figure 1This is a schematic diagram of the system described in this invention; Figure 2 This is a schematic diagram of the vibration acoustic texture processing module; Figure 3 This is a schematic diagram of the field monitoring system structure; Figure 4 This is a flowchart of the method described in this invention; Figure 5 This is a schematic diagram of the specific algorithm of the method described in this invention; Figure 6 This is a schematic diagram of the original pipeline leakage vibration acoustic waveform. Figure 7 This is a schematic diagram of the vibration sound signature of a pipeline leak after NRBO-FMD treatment; Figure 8 This is a schematic diagram showing the discrimination scores of six characteristic parameters for pipelines under four different leakage levels; Figure 9 It is the weighted feature vector of each indicator of the pipeline under four different leakage levels; Figure 10 This is a probability density function graph of nine indicators for leakage level 1; Figure 11 This is a hash distribution diagram under chaotic and non-chaotic environments with a leakage level of 1; Figure 12 These are the results of LSH-FSVM fuzzy computation for four different leakage levels. Detailed Implementation

[0019] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0020] The system described in this invention consists of three parts: a pipeline leakage vibration and acoustic detection system, a field monitoring system, and a core control system. The overall structure is as follows: Figure 1 As shown.

[0021] The pipeline leakage vibration acoustic fingerprint detection system consists of a vibration sensor and an acoustic fingerprint processing module. The vibration sensor is mounted on the outer wall and converts the abnormal mechanical vibration caused by pipeline leakage into an electroacoustic fingerprint through the piezoelectric effect. The acoustic fingerprint processing module consists of a voltage amplification circuit, an acoustic fingerprint filtering circuit, a microprocessor, an acoustic fingerprint transceiver module, and a power supply module. The voltage amplification circuit performs gain compensation on weak vibration acoustic fingerprints to match their amplitude to the input range of the subsequent filtering circuit. The acoustic fingerprint filtering circuit first removes noise from the amplified acoustic fingerprints and then sends them to the microprocessor. The microprocessor unit is an STM32, which transmits the denoised acoustic fingerprints to the local control system via an RS232 bus. The system uses a battery pack to provide a stable operating voltage.

[0022] To achieve accurate detection of pipeline leaks, a real-time vibration acoustic signature acquisition system needs to be constructed. When a pipeline leaks, a local high-pressure gradient field is formed in the region with a significant pressure difference between the inside and outside of the pipe and a thinner pipe wall. Within this region, fluid flow induces the energy conversion of kinetic and pressure energy, exciting mechanical vibrations and generating stress waves at the leak point. This structural vibration induced by the leak propagates through the pipe wall medium as acoustic energy and is ultimately captured by vibration sensors placed on the pipe surface. The acquired vibration acoustic signatures are processed using NRBO-FMD. General and chaotic feature indicators for both pipeline health and leakage states are extracted. The weight vector of the general feature indicators is calculated using CVITW-CV. A cross-membership matrix is ​​constructed by combining the chaotic and general feature indicators using VFMF theory. Then, a fuzzy weighted feature vector is constructed by integrating various indicators, and an LSH-FSVM detection model is built based on fuzzy theory to achieve effective detection of the pipeline leak degree.

[0023] The vibration sensor employs a flexible magnetic base structure, achieving reliable adhesion to pipe surfaces of different materials through a bottom hot-melt adhesive fixing process. It utilizes an MA21 piezoelectric sensor. Chip selection must meet the target frequency band coverage requirements and possess high sensitivity response characteristics. Given the presence of high-frequency background noise above 5kHz in the experimental environment, while typical leakage characteristic frequencies are concentrated below 2kHz, a sensing element with optimized low-frequency sensitivity is prioritized. Furthermore, the chip design must meet environmental adaptability indicators such as IP67 waterproof rating and 120dB dynamic range. The sensor achieves mechanical impedance matching with the pipe surface through non-contact magnetic coupling, effectively eliminating acoustic attenuation problems caused by poor contact. In this example, the ADXL362 accelerometer chip is selected, with a frequency response range of 0.005-2000Hz and a sensitivity of 2V / g.

[0024] Vibration sensors are deployed on the outer wall of the leaking pipe and collect data at a certain frequency. Their start and stop are controlled by the central control system.

[0025] Voiceprint processing module, such as Figure 2As shown. To address issues such as interference within the pipeline, the system achieves voiceprint enhancement through dual processing of pre-amplification and dynamic filtering. The voiceprint conditioning circuit uses Maxim Integrated's Max44280 low-noise operational amplifier to construct the differential amplification unit. This chip has a 130dB open-loop gain, and impedance matching of 50Ω-1MΩ is achieved through optimized PCB layout. A 40dB fixed-gain amplification circuit is designed with ±15V dual power supply. The filtering module integrates the Max7410 general-purpose filter chip, and achieves bandpass filtering through an external PWM voiceprint adjustment capacitor array. It can dynamically adjust the passband range of 0.1-2kHz to suppress ambient noise above 5kHz, and uses adaptive voltage adjustment technology to ensure amplitude-frequency response stability within the range of -40℃ to +85℃. The main control unit uses an STM32 processor, combined with 16MB of on-chip Flash and 2MB of SRAM to achieve real-time voiceprint acquisition and edge computing. The system consumes only 120mA of current during data acquisition. The digital voiceprint after analog-to-digital conversion is transmitted to the core control system.

[0026] The working mode of the pipeline leakage vibration and acoustic detection system, the specific steps are as follows: 1) The core control system sends vibration acoustic signature acquisition commands to the field monitoring system; 2) The field monitoring system sends instructions to the vibration acoustic signature detection system; 3) The pipeline leakage vibration and acoustic signature detection system performs preprocessing and transmits the data to the on-site monitoring system; 4) The on-site monitoring system transmits the received voiceprints to the core control system.

[0027] The architecture of the field monitoring system is as follows: Figure 3 As shown, its core function is to process abnormal acoustic fingerprint data output by the regional vibration detection nodes and achieve bidirectional command interaction with the core control platform via a wireless link. The system uses an STM32 microcontroller as its core and employs time-division multiplexing technology to connect multiple vibration sensor acoustic fingerprints to different I / O channels for serialized acquisition. To address the need for continuous acquisition of pipe segment vibration acoustic fingerprints, the system expands with a large-capacity storage module: a high-speed data channel is established with the STM32 via the SDIO protocol, and the FAT32 file system is used to implement time-division sequential storage of multi-channel acoustic fingerprints, with each vibration sensing channel corresponding to a unique physical storage address mapping.

[0028] To optimize wireless communication performance, the system uses a cable management system to externalize the GPRS communication module to an open area with good acoustic signature. The wireless transmission unit, built on the SIM800A communication chip, supports TCP / IP protocol stack for transparent data transmission. The communication module interacts with the STM32 microcontroller via a USART interface, and a built-in watchdog mechanism ensures the reliability of the communication link.

[0029] The core control system is responsible for aggregating vibration data and equipment numbers from each detection node, and monitoring equipment status through a heartbeat mechanism. The platform supports remote start / stop control: control codes are sent via downlink commands, and the local control system executes the corresponding operation upon receiving the start / stop audible signals.

[0030] like Figure 4 and Figure 5 As shown, the vibration and acoustic sound detection method for industrial equipment pipeline leakage according to the present invention includes the following steps: Step 1: Collect vibration acoustic samples of pipeline leaks; Step 2: Perform NRBO-FMD processing on the collected original vibration acoustic text. Use NRBO to optimize the filter length and decomposition order of FMD. Substitute the optimized index into FMD to decompose the original acoustic text to obtain each intrinsic mode function (IMFs). Reconstruct the vibration acoustic text based on the correlation. Step 3: Extract key feature indicators based on the reconstructed vibration acoustic signature, and use the CVITW-CV method to calculate the weight vector of each indicator to quantify its contribution to the detection of different leakage levels. Step 4: Extract chaotic feature indices from the reconstructed vibration acoustic signature and construct a chaotic feature vector; Step 5: Based on VFMF theory, the uncertainty of different chaotic and non-chaotic characteristic indicators is converted into membership information, and a chaotic fuzzy cross-membership matrix is ​​constructed to describe the probability distribution characteristics of indicators under different leakage levels. Step 6: Combining the weight vector, membership matrix, and chaotic feature vector, the leakage degree of the pipeline is calculated using LSH-FSVM fuzzy calculation. The order of the largest element in the result vector is taken as the current leakage degree of the pipeline, realizing real-time detection of the pipeline leakage degree.

[0031] Figure 6 and Figure 7 The vibration acoustic waveforms before and after NRBO-FMD treatment under different leakage levels are presented. Figure 6 (a), (b), (c), and (d) represent the vibration acoustic waveforms of the original pipeline under leakage levels 1, 2, 3, and 4, respectively. Figure 7 (a), (b), (c), and (d) represent the pipeline leakage vibration acoustic waveforms under leakage levels 1, 2, 3, and 4, respectively, after processing with NRBO-FMD. After processing by the NRBO-FMD algorithm, the high-frequency noise components in the original acoustic waveform are effectively separated into higher-order IMF components, achieving noise suppression and feature enhancement of the original vibration acoustic waveform.

[0032] Figure 8The pipeline was shown under four different leakage conditions. , , , , , The discrimination scores of six general characteristic indicators. Among them, Figure 8 (a), (b), (c), and (d) correspond to the discrimination scores of the six characteristic parameters under the conditions of leakage level 1, leakage level 2, leakage level 3, and leakage level 4, respectively.

[0033] Four leakage levels (circular holes with a leakage notch diameter of 2 cm) are defined, and the CVITW distance matrix for leakage level 1 is shown in Table 1:

[0034] Table 1. CVITW distance matrix under leakage level 1

[0035] A smaller CVITW distance indicates higher temporal similarity. The weight vector is given below: , parameter It contributes the most to leak detection.

[0036] Figure 9 (a), (b), (c), and (d) respectively demonstrate the indicators for leakage levels 1, 2, 3, and 4. , The weighted feature vector of the model is shown in the figure. This figure allows for a visual analysis of the differences in the weights of each parameter at different leakage levels, reflecting the contribution of each parameter to the detection of leakage levels. For example, the fluctuations in the weight values ​​of each parameter at different leakage levels reflect the differences in the detection of different leakage levels by the parameters.

[0037] Next, the output results of the VFMF function vector under leakage level 1 are given in Table 2.

[0038] Table 2. VFMF function vector output results under leakage level 1

[0039] Combining the row vectors yields the cross-membership matrix, as shown below: , The general feature index weight vector, cross membership matrix, and chaotic feature index are fuzzily weighted, and the probability vector output by LSH-FSVM is: , The probability vector maximum value of 0.293 appears in the first position, indicating that the leakage level is most likely to be 29.3%, which is consistent with the actual pipeline leakage level.

[0040] Figure 10 The output probability density distribution for a leakage level of 1 is shown. TSNE-1 and TSNE-2 are two two-dimensional coordinate dimensions obtained after processing with the t-distribution random neighbor embedding algorithm, corresponding to the first and second principal dimensions after dimensionality reduction, respectively. By combining these two dimensions, the similarity and clustering of high-dimensional data can be visualized on a plane. Elliptical regions represent probability components, and the darkest colored regions correspond to the locations of the maximum probability density of the distribution mean. The analysis results show that the probability distributions of different leakage indicators maintain a relatively concentrated distribution pattern under the same leakage level. The probability density distribution under leakage level 1 maintains a high degree of morphological similarity.

[0041] Figure 11 The hash distribution diagrams for chaotic and non-chaotic environments at leakage level 1 are shown. Most indicators exhibit a bimodal distribution, indicating that hash values ​​have relatively concentrated areas in different state intervals of chaotic and non-chaotic conditions. Indicator 5 shows a single peak with a wide peak area; Indicator 8 is also single-peaked, reflecting that its distribution is more concentrated in a certain state interval, and this indicator is not effective for fuzzy identification of leakage conditions.

[0042] Figure 12 Images (a), (b), (c), and (d) respectively demonstrate the fuzzy identification results for four leakage levels (1-4) determined by LSH-FSVM. The fuzzy detection vectors corresponding to each leakage level exhibit feature shifts as the leakage level increases, while the index value corresponding to the peak value consistently matches the actual leakage level. This correspondence verifies the effectiveness of the proposed method in multi-condition identification. The consistency between the maximum element number of the detection result and the actual leakage level indicates that the algorithm has reliable discriminative ability.

[0043] This embodiment also provides a computer device applicable to the vibration and acoustic sound detection method for pipeline leakage in industrial equipment, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the vibration and acoustic sound detection method for pipeline leakage in industrial equipment as proposed in the above embodiment.

[0044] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. 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 stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0045] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the vibration and acoustic signature detection method for pipeline leakage in industrial equipment as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0046] The above description is merely a preferred embodiment of the present invention and is not intended to further limit the present invention. All equivalent changes made based on the description and drawings of the present invention are within the protection scope of the present invention.

Claims

1. A method for detecting a leak in a pipe of an industrial plant by vibration acoustic fingerprinting, characterized in that, The method comprises the following steps: Step 1, collecting pipeline leakage vibration sound print; Step 2, processing the collected original vibration sound print by NRBO-FMD, using NRBO to optimize the filter length and decomposition order of FMD, substituting the optimized indicators into FMD, decomposing the original sound print to obtain each intrinsic mode function IMF, and reconstructing the vibration sound print according to correlation; Step 3, extracting key feature indicators from the reconstructed vibration sound print, using the CVITW-CV method to calculate the weight vector of each indicator, and quantifying the contribution of the indicator to the detection of different leakage degrees; Step 4, extracting chaotic feature indicators from the reconstructed vibration sound print, and constructing a chaotic feature vector; Step 5, based on the VFMF theory, converting the uncertainty of different chaotic feature indicators and non-chaotic feature indicators into membership information, constructing a chaotic fuzzy cross-membership matrix to describe the probability distribution characteristics of the indicators under different leakage degrees; Step 6, comprehensively considering the weight vector, membership matrix and chaotic feature vector, and calculating the leakage degree of the pipeline by LSH-FSVM fuzzy calculation, taking the order of the maximum element in the result vector as the current pipeline leakage degree to realize real-time detection of the pipeline leakage degree.

2. The method of claim 1, wherein, In step 3, the general feature indicators of each reconstructed voiceprint are extracted The lossless pipeline vibration voiceprint, i.e. the reference voiceprint; The vibration voiceprint detected in real time; 6 general feature indicators are:​ ， wherein, is a time-frequency coherence index, , are wavelet transforms of the reference voiceprint and the detected voiceprint, respectively, t and f are time and frequency domain variables, respectively. ， wherein, is a time-varying energy entropy, is an energy probability distribution of the voiceprint in the th window, is a total number of time window divisions; ， wherein, is a spectral entropy difference, , are spectral entropies of the reference voiceprint and the detection voiceprint, respectively, , are energy probability distributions of the reference voiceprint and the detection voiceprint in the first window, respectively. ， wherein, is the wavelet energy dispersion, , are the first layer wavelet energies of the reference voiceprint and the detection voiceprint, respectively, denotes the total number of wavelet decomposition layers.​ ， in, For nonlinear similarity coefficients, , The reference voiceprint and the detected voiceprint in the reconstructed phase space A trajectory point, , These are the mean values ​​of the trajectory points of the reference acoustic signature and the detected acoustic signature in the reconstructed phase space, respectively. A phase space trajectory point; ， wherein, is a multi-scale energy-phase deviation; is a scale index, and there are scales; , respectively represent the energy of the reference voiceprint and the detection voiceprint at the scale; represents the wavelet energy proportion; represents the energy and phase deviation weight factor; represents the total duration of the voiceprint, , respectively represent the instantaneous phase value of the detection voiceprint and the reference voiceprint at the scale at time t.

3. The method of claim 2, wherein, In step 3, the CVITW-CV method is used to calculate the weight vector of each general indicator, and the contribution of different indicators to the detection of leakage degree is quantified, which is specifically as follows: Step 3-1, calculating the CVITW distance matrix; For the feature indicator The CVITW distance matrix between all samples is calculated , measuring the similarity of the timing variation of the feature under each leakage level; ， where Q is the total number of samples and ; let the sample , , have the following iterative relationship: ， wherein, is an intermediate inference term, R, S represent samples 、 the number of elements contained in each, represents the CVITW distance between the i-th feature in the sample and , is a convolution kernel alignment weight, is a variational inference error term; Step 3-2, feature discrimination score; On a matrix The distance between samples of the same leakage degree And the distance between samples of different leakage degrees As shown in the following formula: ， wherein , represents the number of samples under the first , leakage degree. At this time, the discrimination score of a certain feature index is defined as The greater the discrimination score, the stronger the discrimination ability of the feature to the leakage degree. Step 3-3, assigning weights; According to the discrimination score of different characteristic indexes , initialize the weight ; is the original discrimination score of the jth characteristic index, At this time, the weighted CVITW distance is defined as ; Step 3-4, CV cross-validation optimization and weight normalization; Using 5-fold cross-validation, weights are adjusted by gradient descent , and the normalized weights are ; are the original weights, indicating the importance of the jth feature; and the final weight vector is .

4. The method of claim 3, wherein, In Step 4, each chaotic feature index is extracted from the reconstructed vibration voiceprint including: ， wherein, is the Lyapunov exponent; is the infinitesimal distance between two trajectories in the initial phase space; is the actual distance between two trajectories after time t; ， wherein is the correlation dimension; is the probability of finding a pair of phase space trajectory points within a distance ; , are the first and second reconstruction vectors, respectively; is the Heaviside step function;​ ， where, is the Kolmogorov entropy; is the phase space embedding dimension, is the distance scale, is the time delay, is the probability of trajectory points pairs with a distance less than is the embedding dimension, is the probability of trajectory points pairs with a distance less than 5. The method of claim 4, wherein, Step 5 is specifically as follows: Step 5-1, establishing VFMF theory; Assume that the fth general characteristic index of each sample under the fth leakage degree is , all general characteristic indexes constitute a basic characteristic vector , and there are: ​ ， Assuming that the gth chaotic characteristic index of each sample under the fth leakage degree is , all chaotic characteristic indexes constitute a basic feature vector , there are , and a cross feature matrix is obtained, as shown in the following formula: ， To calculate the sample belongs to the fth degree of leakage, the variation fuzzy membership function is established : ， wherein, represents the membership degree corresponding to each feature, is the standard deviation of the th feature, represents each feature value, represents the feature value corresponding to the th sample under the same leakage degree, is the mean value of the th sample; Step 5-2, constructing a membership matrix: 。 6. The method of claim 5, wherein, Step 6 is specifically as follows: Step 6-1, fuzzy feature weighting; combining the weights and the membership degrees to generate a fuzzy weighted feature vector wherein: ， wherein, is the fth fuzzy weighted feature, is the fth fuzzy weighted feature, is the cross membership weight, is the chaotic feature weight; , , respectively represent the fth leakage degree encoding value under the voiceprint feature, cross membership, chaotic feature. Step 6-2, fuzzy membership assignment; Assigning fuzzy membership to the kth training sample , , reflecting the confidence that the sample belongs to the label ​ ， wherein, denotes the true label of its integrated membership degree, denotes the maximum integrated membership degree under the fth leakage degree; when the closer to 1, the more reliable the sample label is; Step 6-3, fuzzy constraint optimization; For the fth leakage degree, the original optimization state of LSH-FSVM is as follows: ， wherein, , respectively represent the weight and bias coefficient of LSH-FSVM; is a Gaussian kernel function; is a kernel index; C is a regularization index; is a random projection hash function; is an LSH random matrix; is an LSH offset; is an LSH quantization step; is a relaxation variable; is the real label under the leakage degree f, and the decision value of U is as follows: ， wherein, is the fuzzy weighted feature mean; Step 6-4, fuzzy probability output; Map each decision value to the probability space: ， wherein and is a scaling index for the fth leakage level, solved by maximizing the log-likelihood function and as shown in the following equation: ， wherein, is a custom label value; is a predicted probability that feature sample k belongs to leakage level f; update the scaling indicator by Newton's method iteration and as shown in the following equation: ， where L is the log-likelihood function, and denotes the scaling index at the n-th iteration, H is the Hessian matrix, is the gradient vector; through the Newton iteration method, the index is gradually adjusted and to maximize the likelihood function L, and the decision value of the LSH-FSVM is mapped to the fuzzy probability output of each leakage degree, as follows: ， Select the state with the maximum probability as the final classification result.

7. An industrial plant piping leak vibration acoustic print detection system for implementing the method of any one of claims 1-6, characterized by, The system comprises a pipeline leakage vibration sound print detection system, a field monitoring system and a core control system, wherein: The pipeline leakage vibration sound print detection system collects the vibration sound print of the pipeline leakage by installing a vibration sensor on the leakage pipeline, pre-processes the vibration sound print, and transmits the pre-processed vibration sound print to the field monitoring system; The field monitoring system receives the vibration sound print from the pipeline leakage detection system and transmits it to the core control system; The core control system pre-processes the received vibration sound print to realize real-time leakage detection of the leakage pipeline.

8. The industrial equipment piping leak vibration acoustic print detection system of claim 7, wherein, The core control system comprises an NRBO-FMD vibration sound print processing part, a leakage pipeline vibration sound print feature mixed indicator extraction part, a CVITW-CV weight vector part, a VFMF-cross membership matrix part and an LSH-FSVM detection part; The NRBO-FMD vibration sound print processing part uses NRBO to optimize the FMD indicators to complete the preprocessing of the sound print; The leakage pipeline vibration soundprint feature mixed index extraction part is used for extracting a plurality of general feature indexes and chaotic feature indexes of the preprocessed vibration soundprint, the general feature indexes are calculated by CVITW-CV weight vectors, and the contribution degrees of different indexes to the leakage state detection are quantified. The chaotic feature indexes directly constitute a chaotic feature vector; The VFMF-cross membership matrix part is based on the VFMF function theory, constructs a cross membership matrix, fuses general feature information and chaotic feature information, converts different indexes into membership degree information, and expresses the probability characteristics of the feature indexes under different leakage degrees; The LSH-FSVM detection part inputs a fuzzy weighted feature vector, the vector is obtained by synthesizing a general index weight vector, a chaotic fuzzy cross membership matrix and a chaotic feature vector. The leakage degree of the pipeline can be obtained by LSH-FSVM fuzzy reasoning, and the real-time detection of the pipeline leakage degree is realized. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to realize the method in any one of claims 1-6.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the method in any one of claims 1-6.

Images