A pipeline leak detection method based on acoustic emission feature space and state estimation

Abstract

The application discloses a pipeline leakage detection method based on acoustic emission characteristic space and state estimation, comprising offline analysis and online detection of pipeline leakage detection, and relates to the technical field of pipeline leakage detection; the application combines Kalman filter and outlier removal technology to estimate the real state in the acoustic emission signal characteristic space, and realizes leakage from an unknown class to a known class; meanwhile, the application provides an efficient calculation method for pipeline leakage detection, which does not directly remove noise from the acoustic emission signal in the signal space, but removes noise from the signal in the characteristic space by using simple calculation, so that the calculation efficiency is improved, and the real-time performance of leakage detection is ensured; compared with the prior art, the method has higher leakage detection accuracy, and can realize real-time leakage detection.

Description

Technical Field

[0001] This invention relates to the field of pipeline leak detection technology, specifically a pipeline leak detection method based on acoustic emission feature space and state estimation. Background Technology

[0002] Pipelines are a core component of fuel transportation networks and the process industry. Despite their stringent design and installation, leaks can still occur due to material fatigue and corrosion, potentially causing personal injury, death, and environmental pollution. Therefore, early leak detection (LD) has been proposed for pipelines to mitigate potentially devastating consequences. However, existing methods have limitations, such as identifying leaks by detecting pipeline flow velocity, pressure, and flow balance. Furthermore, pipeline detection instruments are not sensitive to subtle changes in flow velocity and pressure caused by minute leaks, resulting in inaccurate identification of even minor leaks. Therefore, new leak detection technologies have emerged, among which acoustic emission, as a non-destructive and non-contact detection technique, shows significant promise for early leak detection in pipelines.

[0003] Existing pipe leak detection (LD) methods based on acoustic emission signals typically denoise the original signal directly in the signal space, extract features from the denoised signal, and finally classify the normal / leaking state using a classifier trained on an offline dataset. Their complex computational structure can limit their real-time applications, especially when they require analyzing large amounts of data. These methods may be ineffective in real-world pipe LDs, where the AE signal can be prone to constant fluctuations. Furthermore, real-world pipe networks often consist of many pipes with various sizes, shapes, materials, and operating modes, potentially requiring the simultaneous deployment of multiple acoustic emission sensors. Processing data acquired by these sensors concurrently would require enormous communication and computational resources. Additionally, some denoising techniques may fail to effectively remove unwanted components from the acquired AE signal of a working pipe because these signals are affected by flow turbulence and are therefore non-stationary. Moreover, the offline datasets used to train LD classifiers are often limited in scope, resulting in models that are insufficient to detect leaks in real-world pipes.

[0004] To address these issues, this invention provides a pipeline leakage detection method based on acoustic emission feature space and state estimation. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a pipeline leakage detection method based on acoustic emission feature space and state estimation, thus solving the aforementioned problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a pipeline leakage detection method based on acoustic emission feature space and state estimation, comprising the following steps:

[0007] Step 1: Offline analysis of pipeline leak detection;

[0008] The offline analysis of pipeline leak detection includes the following steps:

[0009] Step S1: Acquisition of acoustic emission signals;

[0010] Step S2: Segment the acquired acoustic emission signal into frames;

[0011] Step S3: Extract features from each frame of acoustic emission signal;

[0012] Step S4: Select the optimal feature set from the extracted features;

[0013] Step 2: Online detection of pipeline leaks;

[0014] The online detection of pipeline leaks includes the following steps:

[0015] Step a: Acquire acoustic emission signals from acoustic emission sensors installed on the pipeline;

[0016] Step b: Segment the acquired acoustic emission signal and extract the optimal features from the optimal feature set selected in step one;

[0017] Step c: Model the pipeline state based on the selected optimal features, and perform state estimation and outlier removal;

[0018] Step d: Use state estimation to measure the normalized distance between the unknown state and the known state, and calculate a threshold based on the given false alarm probability according to the Neyman-Pearson theorem to determine the state of the pipeline.

[0019] Preferably, in step S2, the acoustic emission signal is segmented using a rectangular window function, and the signal is segmented into frames every 1024 samples.

[0020] Preferably, the features extracted in step S3 include short-term energy consumption, root mean square, average amplitude, spectral peak value, spectral diffusion, kurtosis, skewness, entropy, and spectral centroid, and the extracted features form a feature pool.

[0021] Preferably, in step S4, the features extracted in step S3 are scored based on the Fisher Discriminant Ratio (FDR) to select the optimal feature set. The formula for calculating the FDR is as follows:

[0022]

[0023] Among them, a i It is the mean of the i-th type of feature. The variance of the i-th feature, i = 0, 1, corresponds to the normal and leaking states of the pipeline, respectively.

[0024] Preferably, the pipeline state modeling formula in step c is:

[0025]

[0026] Where, x k Let z represent the (n×1) pipe state vector. k w represents the characteristic selected by the (m×1) observation vector. k Indicates process noise, v k Indicates the measurement noise, Φ k and H k These are an (n×n) state transition matrix and an (m×n) measurement matrix, respectively, where k is the discrete exponent and w k and v k The expected function is:

[0027]

[0028]

[0029] Where T is the matrix transpose.

[0030] Preferably, in step c, state estimation is performed on the model using a linear Kalman filter, and the estimation process uses the measurement noise covariance R. k And error covariance P k Calculate the Kalman gain K:

[0031]

[0032] in, and Representing the predicted state and prediction error respectively, expressed by the Kalman gain K and the observed value z. k Update state estimation Error covariance P k The calculation formula is:

[0033]

[0034] Among them, e k To estimate the error, the initial value during the state estimation process is... and initial value and The calculation formula is:

[0035]

[0036] Among them, z is,i = -M, -M+1, …, -1 are M prior measurement values, and R0 are the initial error estimate and its covariance matrix, respectively.

[0037] Preferably, in said step c, the outlier is represented by the likelihood function l, and its calculation formula is:

[0038]

[0039] where, ε k and S k are the innovation and innovation covariance, respectively, and their calculation formulas are:

[0040]

[0041] Preferably, in said step d, the calculation formula of the normalized distance is:

[0042]

[0043] where, is the state estimation after Kalman filtering and outlier removal, μ0 and Σ0 are the known state expectation vector and covariance matrix, respectively. The known state is set as ω0, the unknown state is set as ω1, and the vector of the known state ω0 is set as the vector of the unknown state ω1 is set as where a0 < a1, the given false alarm probability P FA = P(ω1|ω0), and the maximum likelihood ratio detection probability P given by the following formula D = P(ω1|ω1):

[0044]

[0045] where, y is d 2 the observation value of , α is the threshold, and its calculation formula is:

[0046] P FA = ∫ {y:L(y>α} p(y|ω0)dy (11)

[0047] Substituting p(y|ω0) for p(y|ω1), and L(y) after simplification is:

[0048]

[0049] where, β = [σ 2 / (a1-a0)]lnα+(a1+a0) / 2, thus obtaining:

[0050]

[0051] Where Φ(·) is the complementary cumulative distribution function.

[0052] Beneficial effects

[0053] This invention provides a pipeline leak detection method based on acoustic emission feature space and state estimation. Compared with existing technologies, it has the following advantages:

[0054] This invention proposes a novel, non-destructive, and non-invasive pipeline leak detection method for early detection of minute leaks in pipelines under low operating pressure and low flow rate conditions, which may be undetectable by traditional methods. This method combines Kalman filtering and outlier removal techniques to estimate the true state in the acoustic emission signal feature space, and identifies leaks from an unknown class to a known class. Furthermore, this invention presents an efficient computational method for pipeline leak detection. Instead of directly denoising the acoustic emission signal in the signal space (which typically involves complex calculations), this method uses simple computation to denoise the signal in the feature space, thereby improving computational efficiency and ensuring real-time leak detection. Compared to existing methods, this method achieves higher leak detection accuracy and enables real-time leak detection. Attached Figure Description

[0055] Figure 1 This is an overall flowchart of the pipeline leakage detection method of the present invention;

[0056] Figure 2 This is a state estimation diagram in an embodiment of the present invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Example:

[0059] Please see Figure 1-2 A pipeline leakage detection method based on acoustic emission feature space and state estimation includes the following steps:

[0060] Step 1: Offline analysis of pipeline leak detection;

[0061] Offline analysis of pipeline leak detection includes the following steps:

[0062] Step S1: Acquisition of acoustic emission signals;

[0063] Step S2: Segment the acquired acoustic emission signal into frames;

[0064] Step S3: Extract features from each frame of acoustic emission signal;

[0065] Step S4: Select the optimal feature set from the extracted features;

[0066] Step 2: Online detection of pipeline leaks;

[0067] Online pipeline leak detection includes the following steps:

[0068] Step a: Acquire acoustic emission signals from acoustic emission sensors installed on the pipeline;

[0069] Step b: Segment the acquired acoustic emission signal and extract the optimal features from the optimal feature set selected in step one;

[0070] Step c: Model the pipeline state based on the selected optimal features, and perform state estimation and outlier removal;

[0071] Step d: Use state estimation to measure the normalized distance between the unknown state and the known state, and calculate a threshold based on the given false alarm probability according to the Neyman-Pearson theorem to determine the state of the pipeline.

[0072] The flow chart of the pipeline leakage detection method proposed in this invention is as follows: Figure 1 As shown, the method comprises two stages: offline analysis and online detection for pipeline leak detection. Offline analysis includes the acquisition, segmentation, feature extraction, and feature selection of acoustic emission signals to determine the optimal feature set for leak detection. Signals are acquired using acoustic emission sensors for different scenarios, then the acquired signals are segmented into frames, and features are extracted from each frame. To remove redundant and irrelevant features from the feature pool, feature selection is performed to select the optimal feature set, which represents the most distinctive features of a leak in the pipeline. Once the offline analysis stage determines the optimal features for leak detection, the proposed online detection stage can be invoked to detect leaks in the pipeline in real time. In the online detection stage, acoustic emission signals are acquired from sensors installed on pipelines that may be leaking. These acoustic emission signals are first segmented, and then features are extracted, at which point only the optimal features identified in the offline analysis stage are extracted. Subsequently, Kalman filtering is used to estimate the true state of the system in the feature space to determine the health status of the pipeline. Finally, state estimation is used to measure the normalized distance between the unknown state and the known state, and a threshold is calculated based on a given false alarm probability according to the Neyman-Pearson theorem to determine the state of the pipeline, i.e., normal or leaking. Figure 1In this context, PFA stands for False Alarm Probability, ρ for Significance Level, PL for Pipeline, AES for Acoustic Emission Sensor, SS for Signal Segmentation, FE for Feature Extraction, FS for Feature Selection, SEOR for State Estimation and Outlier Suppression, LD for Leak Detection, DB for Database, and N / L for Normal / Leaking.

[0073] In step S2, the acoustic emission signal is segmented using a rectangular window function, and each 1024 samples are segmented into frames.

[0074] The characteristics of non-stationary signals change over time, requiring analysis at different time intervals. This can be achieved by segmenting the signal into sequentially overlapping frames and then extracting features from these frames or segments, which helps mitigate the impact of non-stationarity on feature extraction. Due to the influence of various factors such as flow dynamics and environmental noise, the acoustic emission signals from fluid pipelines are non-stationary. Therefore, signal segmentation is a crucial step before feature extraction in this invention. The acoustic emission signal is segmented using a rectangular window function, with each 1024 samples being segmented into frames.

[0075] The features extracted in step S3 include short-term energy consumption, root mean square, average amplitude, spectral peak value, spectral diffusion, kurtosis, skewness, entropy, and spectral centroid. The extracted features form a feature pool.

[0076] After segmenting the acoustic emission signal, features shown in Table 1 are extracted from each segment to form a feature pool. These features represent the different properties of the signal segments in the time and frequency domains. Furthermore, redundant and irrelevant features need to be discarded, as these may not carry the most discriminative information for the leakage syndrome and could compromise the leakage detection accuracy of the proposed method. To eliminate redundant features, individual components are scored based on the Fisher discriminant ratio (FDR).

[0077] In step S4, the features extracted in step S3 are scored based on the Fisher Discriminant Ratio (FDR), and the optimal feature set is selected. The formula for calculating FDR is:

[0078]

[0079] Among them, a i It is the mean of the i-th type of feature. The variance of the i-th feature, i = 0, 1, corresponds to the normal and leaking states of the pipeline, respectively. The FDR (Fixed Difference Ratio) quantifies the separability of a single feature; a higher FDR score indicates a better feature, while a lower FDR score indicates a worse feature. Features in the feature pool are ranked according to their FDR scores, and the best feature is selected for state estimation and outlier rejection (SEOR).

[0080] Table 1

[0081]

[0082] Where, x' n s (n = 0, 1, ..., N-1) are N samples of the input signal segment, X is the short-time spectral amplitude, and μ and σ are the mean and variance of x, respectively.

[0083] The formula for pipeline state modeling in step c is:

[0084]

[0085] Where, x k Let z represent the (n×1) pipe state vector. k w represents the characteristic selected by the (m×1) observation vector. k Indicates process noise, v k Indicates the measurement noise, Φ k and H k These are an (n×n) state transition matrix and an (m×n) measurement matrix, respectively, where k is the discrete exponent and w k and v k The expected function is:

[0086]

[0087]

[0088] Where T is the matrix transpose.

[0089] Measurement noise v k This includes all signal sources unrelated to leakage, including inherent sources such as turbulent flow or external sources such as vibrations caused by nearby machinery, vehicles, and environmental factors like rain and wind. External noise sources are unrelated to pipeline operation and should be controlled during sensor installation. Therefore, the noise v in the model given by equation (2) k This mainly represents the inherent noise source, which only exists during pipeline operation. According to equation (2), if the pipeline is in normal condition, then assume x k z is the zero vector k Change to v k Therefore, the measurement noise v under normal conditions k This approach, assuming the noise source remains consistent before and after the leak, is applied to overall pipeline health monitoring. This assumption is reasonable because leaks typically begin in a negligible aperture, resulting in relatively small leaks that may not significantly impact the entire system. Therefore, minute changes can be ignored. However, this assumption is incorrect if low-quality sensors are used, as they are insensitive to minute changes and thus cannot detect leaks immediately. In this case, due to the leak and the measurement noise v... kThe flow rate may vary considerably, and the noise determined under normal conditions may need to be re-evaluated correctly to determine the leakage status.

[0090] In step c, a linear Kalman filter is used to perform state estimation on the model. Figure 2 The state estimation diagram is given (where I is the matrix unit), and the estimation process uses the measurement noise covariance R. k And error covariance P k Calculate the Kalman gain K:

[0091]

[0092] in, and Representing the predicted state and prediction error respectively, expressed by the Kalman gain K and the observed value z. k Update state estimation Error covariance P k The calculation formula is:

[0093]

[0094] Among them, e k To estimate the error, the initial value during the state estimation process is... and initial value and The calculation formula is:

[0095]

[0096] Among them, z i s,i=-M,-M+1,…,-1 are the prior measurements of M. R0 and R0 represent the initial error estimate and its covariance matrix, respectively. In practice, pipelines typically operate in noisy environments, and the acoustic emission signals obtained from the pipeline fluctuate continuously, inevitably generating outliers in the feature pool, which need to be removed.

[0097] In step c, outliers are represented by the likelihood function l, which is calculated using the following formula:

[0098]

[0099] Where, ε k and S k These are innovation value and innovation covariance, respectively, and their calculation formulas are as follows:

[0100]

[0101] From equation (8), it can be seen that ε k ~N(0, Sk), therefore we can use ξk Transform This is a matrix containing elements with identical distribution, ξ k ~ N(0,1). If ξ is are identically distributed, then the random variable has a central chi-square probability distribution with v degrees of freedom, where υ=dim(v k )=m, which is the dimension of the observation vector. The distribution is expressed by the following formula:

[0102]

[0103] where Γ(u) is defined as:

[0104]

[0105] Therefore, for a given significance level ρ, the threshold γ can be determined as:

[0106]

[0107] where P[.] represents probability, indicates an outlier, omits the measurement update, and if l≥γ, continues to perform state estimation using the previously predicted state.

[0108] Online detection of pipeline leakage detection:

[0109] The known state is set as ω0, which is a normal state, and the unknown state is set as ω1, which is a leakage state, then the calculation formula for the normalized distance in step d is:

[0110]

[0111] Wherein, is the state estimation after Kalman filtering and outlier removal, μ0 and Σ0 are the expected vector and covariance matrix of the known state respectively, and the vector of the known state ω0 is set as The vector of the unknown state ω1 is set as where a0<a1, the given false alarm probability P FA =P(ω1|ω0), the maximum likelihood ratio detection probability P given by the following formula D =P(ω1|ω1):

[0112]

[0113] wherein y is the observation value of d 2 and α is the threshold, and its calculation formula is:

[0114] P FA =∫ {y:L(y>α}p(y|ω0)dy (11)

[0115] Replace p(y|ω1) with p(y|ω0). and L(y) can be simplified to:

[0116]

[0117] Where β=[σ 2 Therefore, we get:

[0118]

[0119] Where Φ(·) is the complementary cumulative distribution function.

[0120] The threshold β of a given false alarm probability PFA is determined by equation (13), and then the leak is detected by equation (12).

[0121] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0122] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0123] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A pipeline leakage detection method based on acoustic emission feature space and state estimation, characterized in that, Includes the following steps: Step 1: Offline analysis of pipeline leak detection; The offline analysis of pipeline leak detection includes the following steps: Step S1: Acquisition of acoustic emission signals; Step S2: Segment the acquired acoustic emission signal into frames; Step S3: Extract features from each frame of acoustic emission signal; Step S4: Select the optimal feature set from the extracted features; In step S4, the features extracted in step S3 are scored based on the Fisher Discriminant Ratio (FDR) to select the optimal feature set. The formula for calculating the FDR is as follows: （1） Among them, a i It is the mean of the i-th type of feature. The variance of the i-th type of feature, These correspond to the normal and leaking states of the pipeline, respectively. Step 2: Online detection of pipeline leaks; The online detection of pipeline leaks includes the following steps: Step a: Acquire acoustic emission signals from acoustic emission sensors installed on the pipeline; Step b: Segment the acquired acoustic emission signal and extract the optimal features from the optimal feature set selected in step one; Step c: Model the pipeline state based on the selected optimal features, and perform state estimation and outlier removal; Step d: Use state estimation to measure the normalized distance between the unknown state and the known state, and calculate a threshold based on the given false alarm probability according to the Neyman-Pearson theorem to determine the state of the pipeline.

2. The pipeline leakage detection method based on acoustic emission feature space and state estimation according to claim 1, characterized in that: In step S2, the acoustic emission signal is segmented using a rectangular window function, and each 1024 samples are segmented into frames.

3. The pipeline leakage detection method based on acoustic emission feature space and state estimation according to claim 2, characterized in that: The features extracted in step S3 include short-term energy consumption, root mean square, average amplitude, spectral peak value, spectral diffusion, kurtosis, skewness, entropy, and spectral centroid. The extracted features form a feature pool.

4. The pipeline leakage detection method based on acoustic emission feature space and state estimation according to claim 1, characterized in that: The pipeline state modeling formula in step c is: （2） Where, x k Let z represent the (n×1) pipe state vector. k This represents the characteristic selected by the (m×1) observation vector. Indicates process noise. Indicates measurement noise. and H k These are an (n×n) state transition matrix and an (m×n) measurement matrix, respectively, where k is the discrete exponent. and The expected function is: （3） Where T is the matrix transpose.

5. The pipeline leakage detection method based on acoustic emission feature space and state estimation according to claim 4, characterized in that: In step c, state estimation is performed on the model using a linear Kalman filter. The estimation process uses the measurement noise covariance R. k And error covariance P k Calculate the Kalman gain K: （4） in, and Representing the predicted state and prediction error respectively, expressed by the Kalman gain K and the observed value z. k Update state estimation Error covariance P k The calculation formula is: （5） in, To estimate the error, the initial value during the state estimation process is... and initial value and The calculation formula is: （6） in, s, i = −M, −M + 1,…,−1 are the prior measurements of M. = - R0 and R0 are the initial error estimate and its covariance matrix, respectively.

6. The pipeline leakage detection method based on acoustic emission feature space and state estimation according to claim 5, characterized in that: In step c, outliers are represented by the likelihood function l, and its calculation formula is as follows: （7） in, and S k These are innovation value and innovation covariance, respectively, and their calculation formulas are as follows: （8）。 7. The pipeline leakage detection method based on acoustic emission feature space and state estimation according to claim 6, characterized in that: The formula for calculating the normalized distance in step d is: （9） in, For Kalman filtering and state estimation after removing outliers, and These are the expected vector and covariance matrix of the known state, respectively. The known state is set as follows: The unknown state is set to Known state Let the vector be d ~ N Unknown state Let the vector be d ~ N ,in Given the false alarm probability = P(ω1|ω0), the maximum likelihood ratio detection probability P is given by the following formula. D = P(ω1|ω1): （10） in, for The observed values, The threshold value is calculated using the following formula: （11） use Alternative , and , After simplification, it becomes: or y> (12) in, Therefore, we get: （13） in, (·) represents the complementary cumulative distribution function.

Images