Method for identifying vibration signal by optical fiber vibration sensor in high-noise environment

Abstract

The invention relates to the field of optical fiber vibration signal processing, and discloses a vibration signal identification method of an optical fiber vibration sensor in a high-noise environment, which comprises the following steps of: S1, preprocessing an optical fiber vibration signal, eliminating a direct current component, performing amplitude normalization and eliminating redundant correlation; s2, separating the preprocessed signals by using an independent component analysis technology, and extracting independent signals of different vibration sources; s3, performing time alignment on the separated signals by using a dynamic time warping algorithm, and calculating an alignment path distance between the signals; and S4, judging whether the vibration signal is abnormal or not according to comparison between the alignment path distance and a preset threshold value. According to the method, independent component analysis and dynamic time warping technologies are combined, and accurate separation and efficient identification of vibration signals and environmental noise are realized through non-Gaussian optimization and time axis dynamic alignment without depending on prior mark data and a data driving model.

Description

Technical Field

[0001] The invention relates to the field of optical fiber vibration signal processing, in particular to a method for optical fiber vibration sensor to identify vibration signals in a high noise environment. Background Art

[0002] Fiber optic vibration sensor technology in high noise environments has been widely used in key areas such as oil and gas pipeline safety monitoring, border protection, and vibration monitoring of important infrastructure. Fiber optic vibration sensors have become an important tool for monitoring vibration signals due to their high sensitivity, strong anti-electromagnetic interference ability, and adaptability to harsh environments. However, in complex noise environments, accurately extracting and identifying target vibration signals still faces many technical challenges.

[0003] Traditional vibration signal recognition methods are mostly based on local feature extraction techniques, such as Fourier transform (FFT) and wavelet transform (DWT). These methods usually identify the vibration source by analyzing the frequency domain or time-frequency domain characteristics of the signal, but in an environment with large noise interference, the characteristics between the signal and the noise are easily confused, resulting in a decrease in recognition accuracy. In addition, these methods usually rely on the extraction of static features and have difficulty in processing the time-varying characteristics of the signal in complex scenes.

[0004] In recent years, data-driven machine learning methods have been gradually applied to the recognition of vibration signals. They rely on the training of a large amount of labeled data and use deep learning models to extract high-dimensional features of signals. However, such methods are highly dependent on labeled data, especially in high-noise environments, where the cost and quality of data annotation are difficult to guarantee. In addition, the generalization ability of machine learning models is limited. When the environment and characteristics of the signal change, the model often needs to be retrained to adapt to the new scenario.

[0005] These limitations of existing technologies lead to the following problems in vibration signal recognition in complex and high-noise environments:

[0006] Insufficient recognition accuracy: Traditional feature extraction methods have poor robustness to noise, and signals are easily masked in a strong noise background.

[0007] High model dependence: Machine learning methods rely on labeled data and training models, and are difficult to quickly adapt to dynamically changing scenarios.

[0008] Lack of adaptability: Existing technologies cannot effectively handle the time alignment problem of vibration signals, resulting in inaccurate classification and detection results.

[0009] Therefore, in order to solve the above problems, there is an urgent need for an efficient and robust vibration signal recognition method. To this end, the present invention proposes a method for vibration signal recognition using an optical fiber vibration sensor in a high noise environment. Summary of the invention

[0010] In view of the deficiencies in the prior art, the present invention provides a method for optical fiber vibration sensors to identify vibration signals in a high-noise environment, thereby solving the problem that optical fiber vibration sensors cannot accurately separate and identify target vibration signals in a high-noise environment.

[0011] To achieve the above object, the present invention is implemented by the following technical scheme: A method for optical fiber vibration sensor to identify vibration signals in a high noise environment comprises the following steps:

[0012] S1. Preprocessing the optical fiber vibration signal, eliminating the DC component, normalizing the amplitude and removing redundant correlation;

[0013] S2. Separate the preprocessed signals using independent component analysis technology to extract independent signals from different vibration sources;

[0014] S3, using a dynamic time warping algorithm to time align the separated signals and calculate the alignment path distance between the signals;

[0015] S4. Determine whether the vibration signal is abnormal based on the comparison between the alignment path distance and the preset threshold value;

[0016] S5. Classify the vibration signal according to the abnormal detection result, and classify the abnormal signal as a specific target vibration source, and output a warning signal at the same time.

[0017] Preferably, the formula for eliminating the DC component in step S1 is as follows:

[0018] x dc-free (t) = x(t) - mean(x(t))

[0019] Among them, x dc-free (t) is the signal after removing the DC component, mean(x(t)) is the mean of the signal;

[0020] The amplitude normalization operation in step S1 standardizes the amplitude range, and the formula is as follows:

[0021]

[0022] Among them, x norm (t) is the normalized signal, min(x) is the minimum value of the signal, and max(x) is the maximum value of the signal;

[0023] In the step S1, a whitening operation is used to remove redundant correlations, and the formula is as follows:

[0024] x w =ED -1 / 2 E T x

[0025] Where E is the eigenvector matrix of the signal covariance matrix, D is the eigenvalue matrix of the signal covariance matrix, x w is the signal after whitening.

[0026] Preferably, the formula of the independent component analysis technique in step S2 is as follows:

[0027] x=As

[0028] Among them, x is the observed signal vector, A is the unknown mixing matrix, and s is the independent signal source;

[0029] The goal of the independent component analysis technique in step S2 is to solve the separation matrix W = A -1 , so that the separated signal is s'=Wx, where the components of s' are independent of each other.

[0030] Preferably, the independent component analysis in step S2 is achieved by maximizing the non-Gaussianity of the signal, and the non-Gaussianity is measured by the kurtosis index, and the formula is:

[0031]

[0032] Among them, kurt() represents kurtosis, y represents input signal, and E[] represents expectation.

[0033] Preferably, the step S3 specifically includes the following steps:

[0034] S3.1. Define signal time series: define two separated signal time series as s 1 =[s 11 ,s 12 ,...,s 1m ] and s 2 =[s 21 ,s 22 ,...,s 2n ];

[0035] S3.2, calculating local alignment distance: calculating the local Euclidean distance between signal sample points;

[0036] S3.3, using a recursive formula to calculate the cumulative alignment path;

[0037] S3.4. Determine the alignment path: Calculate the optimal alignment path of the signal by backtracking the minimum path, and output the cumulative alignment path distance D(i, j).

[0038] Preferably, the calculation formula of the local Euclidean distance in step S3.2 is d(s 1i ,s 2j )=(s 1i -s 2j ) 2, the recursive formula in step S3.3 is:

[0039] D(i,j)=d(s 1i ,s 2j )+min{D(i-1,j),D(i,j-1),D(i-1,j-1)}

[0040] Where D(i,j) is the signal s 1 and 2 Cumulative alignment distance at the i-th and j-th positions.

[0041] Preferably, the classification in step S5 is based on: using the DTW matching result, matching the abnormal signal with a predefined characteristic signal template, calculating the similarity, and determining the signal category.

[0042] An optical fiber vibration signal monitoring system based on the above method is characterized by comprising:

[0043] Optical fiber sensor, used to collect environmental vibration signals in real time;

[0044] Data acquisition module, used for collecting and digitally processing vibration signals;

[0045] Signal processing module, used to implement signal preprocessing, independent component analysis and dynamic time warping;

[0046] The classification and anomaly detection module is used to classify vibration signals, identify anomalies, and issue early warning signals.

[0047] The invention provides a method for optical fiber vibration sensor to identify vibration signals in a high noise environment.

[0048] It has the following beneficial effects:

[0049] 1. In view of the defect of low recognition accuracy of the prior art in high noise environment, the present invention proposes a method combining independent component analysis and dynamic time warping. Independent component analysis can achieve accurate separation of vibration signals and environmental noise without relying on prior labeled data by optimizing the non-Gaussianity of mixed signals; dynamic time warping overcomes the problem of traditional methods being sensitive to the time position of signals by dynamically aligning the signal time axis. This method shows significant robustness in complex noise environments and greatly improves the reliability of vibration signal recognition. It is particularly suitable for practical application scenarios such as oil and gas pipelines and border protection.

[0050] 2. Unlike machine learning methods that rely on a large amount of labeled data, the present invention does not need to rely on data-driven models, and realizes signal processing and classification through analysis based on physical properties and signal characteristics. Through the combination of signal preprocessing, independent component analysis and dynamic time warping, the system can adapt to complex signal scenes, accurately distinguish vibration source signals from noise signals, and significantly improve the generalization ability of the algorithm. In an environment where data is scarce or labeling is incomplete, the present invention can still work efficiently and meet the actual needs of optical fiber vibration signal recognition in complex scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a structural diagram of the method of the present invention;

[0052] Figure 2 is a flow chart of the method of the present invention;

[0053] Figure 3 It is a framework diagram of the system of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Please see attached Figure 1 and attached Figure 2 The embodiment of the present invention provides a method for identifying vibration signals using an optical fiber vibration sensor in a high noise environment, comprising the following steps:

[0056] S1. Preprocessing the optical fiber vibration signal, eliminating the DC component, normalizing the amplitude and removing redundant correlation;

[0057] S2. Separate the preprocessed signals using independent component analysis technology to extract independent signals from different vibration sources;

[0058] S3, using a dynamic time warping algorithm to time align the separated signals and calculate the alignment path distance between the signals;

[0059] S4. Determine whether the vibration signal is abnormal based on the comparison between the alignment path distance and the preset threshold value;

[0060] S5. Classify the vibration signal according to the abnormal detection result, and classify the abnormal signal as a specific target vibration source, and output a warning signal at the same time.

[0061] Specifically, the core of the method design of the present invention is to accurately separate the target vibration signal and efficiently detect abnormal signals in a complex noise environment. Through a series of interlocking steps, the method can extract key features from the original signal and solve the problem of insufficient robustness of traditional methods in high-noise scenarios.

[0062] In the signal preprocessing stage, the original vibration signal collected by the fiber optic sensor often contains a variety of complex mixed vibrations and noise. Therefore, it is necessary to fully denoise and standardize the signal. First, by eliminating the DC offset in the signal, the fluctuation is centered around the zero value, thereby avoiding the influence of the DC component on the subsequent analysis. Next, the signal amplitude is normalized and the amplitude is adjusted to a uniform range to ensure the comparability between different signals. The redundant correlation between signals is further eliminated through the whitening operation, thereby reducing the computational complexity and enhancing the separability of the signal.

[0063] After preprocessing, the signal is sent to the independent component analysis module. The goal at this time is to separate the mixed signal into independent vibration source signals. Independent component analysis assumes that the observed signal is a linear combination of multiple independent signal sources. The key is to maximize the non-Gaussianity of the signal, optimize the separation matrix, and extract independent vibration sources. The measurement of non-Gaussianity is completed by kurtosis, and the gradient optimization algorithm is used to finally obtain the signal components that meet the independence. This step is very important because it not only achieves signal separation, but also provides an accurate data basis for subsequent time alignment and anomaly detection.

[0064] However, the separated signals may not be aligned on the time axis. Therefore, dynamic time warping technology is introduced to align the time axes of the two signals. By calculating the Euclidean distance between samples and the alignment path generated by the recursive optimization algorithm, the optimal time alignment of the signals can be found. The final alignment distance value is directly used for subsequent abnormal signal judgment.

[0065] The core of anomaly detection is to compare the alignment path distance with the preset threshold. If the alignment distance exceeds the set threshold, the signal is marked as anomaly. To ensure adaptation to diverse scenarios, the threshold setting is usually based on experiments and actual usage environments, and is flexibly adjusted to achieve the best detection effect.

[0066] Finally, the abnormal signal not only needs to be identified, but also needs to be further classified into specific target vibration sources. By matching the separated signal with the predefined template signal, the category of the abnormal signal can be accurately determined, such as pipeline impact, mechanical vibration or environmental noise. The classification results will then be sent to the output module for early warning in a variety of ways, such as buzzer alarm, LCD screen display or remote communication interface to notify relevant personnel. This complete set of closed-loop design ensures efficiency and accuracy from acquisition, processing to output.

[0067] Through the close combination of the above steps, this method can cope with the problem of complex vibration signal recognition in high-noise environments, and is particularly suitable for vibration anomaly detection in oil and gas pipeline safety monitoring, border protection, and important infrastructure.

[0068] The formula for eliminating the DC component in step S1 is as follows:

[0069] x dc-free (t) = x(t) - mean(x(t))

[0070] Among them, x dc-free (t) is the signal after removing the DC component, mean(x(t)) is the mean of the signal;

[0071] The amplitude normalization operation in step S1 standardizes the amplitude range. The formula is as follows:

[0072]

[0073] Among them, x norm (t) is the normalized signal, min(x) is the minimum value of the signal, and max(x) is the maximum value of the signal;

[0074] In step S1, whitening operation is used to remove redundant correlations, and the formula is as follows:

[0075] x w =ED -1 / 2 E T x

[0076] Where E is the eigenvector matrix of the signal covariance matrix, D is the eigenvalue matrix of the signal covariance matrix, x w is the signal after whitening.

[0077] Specifically, the whitening operations involved in step S1, such as eliminating DC components, normalizing amplitudes, and removing redundant correlations, constitute an important part of the preprocessing of optical fiber vibration signals. Their main purpose is to provide high-quality signal input for subsequent signal separation, time alignment, and anomaly detection.

[0078] In this embodiment, the original vibration signal collected by the optical fiber sensor may contain low-frequency offsets, that is, the DC component of the signal. These offsets will interfere with the dynamic characteristics of the signal and subsequent analysis. Therefore, the following formula is used to process the signal to eliminate the DC component:

[0079] x dc-free (t) = x(t) - mean(x(t))

[0080] Among them, x(t) is the collected original optical fiber vibration signal, mean(x(t)) is the mean of the signal, and the calculation formula is:

[0081]

[0082] Among them, x i is the value of a single sampling point of the signal, and N is the total number of sampling points.

[0083] In practical applications, the signal x after removing the DC component dc-free The fluctuation of (t) on the time axis revolves around the zero value. This zeroing operation ensures the signal's reference is clear and effectively avoids errors caused by reference drift. At the same time, this operation reduces the impact of low-frequency offset on subsequent vibration feature extraction and frequency domain analysis, especially in scenes with complex environmental noise.

[0084] In this embodiment, the signal amplitude normalization process is intended to unify the signal amplitudes at different sampling points into a standardized range, thereby enhancing the comparability of the signals and avoiding calculation errors caused by amplitude differences. The specific formula is:

[0085]

[0086] Where x(t) is the signal from which the DC component has been removed, min(x) and max(x) are the minimum and maximum values ​​of the signal in the current sampling period, respectively. norm (t) is the normalized signal, and its amplitude range is usually between [0,1] or [-1,1].

[0087] The significance of normalization processing is to eliminate the interference caused by the difference in amplitude of the signal itself or the difference in sensitivity of the sampling device, so that all signals can be calculated on a unified amplitude scale. This not only improves the calculation accuracy of subsequent processing, but also reduces the problem of pseudo-feature extraction caused by amplitude imbalance. Especially when the fiber optic sensor collects long-distance complex signals, this processing method can well compensate for the deviation caused by amplitude changes.

[0088] In this embodiment, in order to further reduce the redundant components in the signal and improve the signal independence, a whitening operation is used to process the signal. The signal x after whitening w The formula is:

[0089] x w =ED -1 / 2 E T x

[0090] Where E is the eigenvector matrix of the signal covariance matrix, D is the eigenvalue matrix of the signal covariance matrix, and D -1 / 2is the inverse square root of the eigenvalue matrix, x is the signal after amplitude normalization, x w is the signal after whitening.

[0091] The specific implementation steps of the whitening operation include the following:

[0092] Compute the signal covariance matrix:

[0093] The covariance of the normalized signal x is calculated as follows:

[0094]

[0095] Where N is the total number of sampling points and μ is the signal mean.

[0096] Perform eigendecomposition:

[0097] Perform eigendecomposition on the signal covariance matrix Cov(x) to obtain the eigenvalue matrix D and eigenvector matrix E, which satisfy:

[0098] Cov(x)=EDE T

[0099] Signal whitening processing:

[0100] According to the whitening formula x w =ED -1 / 2 E T x, linearly transform the signal to eliminate redundant correlations.

[0101] As a result of the whitening operation, the covariance matrix of the signal is normalized to the identity matrix, namely: Cov(x w )=I, where I is the unit matrix. This indicates that the original correlation in the signal has been eliminated and the signals are independent. For the subsequent independent component analysis operation, this whitening process is particularly important because it directly affects the accuracy of signal source separation.

[0102] The formula for the independent component analysis technique in step S2 is as follows:

[0103] x=As

[0104] Among them, x is the observed signal vector, A is the unknown mixing matrix, and s is the independent signal source;

[0105] The goal of the independent component analysis technique in step S2 is to solve the separation matrix W = A -1 , so that the separated signal is s'=Wx, where the components of s' are independent of each other.

[0106] In the S2 step, independent component analysis is achieved by maximizing the non-Gaussianity of the signal. The non-Gaussianity is measured by the kurtosis index, and its formula is:

[0107]

[0108] Among them, kurt() represents kurtosis, y represents input signal, and E[] represents expectation.

[0109] Specifically, independent component analysis (ICA) is an important step in the fiber vibration signal identification method, which aims to separate independent vibration source signals from complex observation signals. By assuming that the observation signal is a linear mixture of independent signal sources, independent component analysis uses an optimization algorithm to solve the separation matrix to ensure that the separated signal sources meet statistical independence. At the same time, independent component analysis achieves signal separation by maximizing the non-Gaussianity of the signal (measured by the kurtosis index).

[0110] The model assumptions of independent component analysis are:

[0111] In this embodiment, independent component analysis models the observed signal as a linear mixture of independent signal sources. Let the observed signal be x and the independent signal source be s, and the relationship between the two can be expressed as: x = As

[0112] in:

[0113] x is the observation signal vector, which contains n-dimensional observation signals;

[0114] s is an independent signal source vector, containing m independent signal sources (usually n = m);

[0115] A is an unknown n×m dimensional mixing matrix, which describes the linear mixing relationship from independent signal sources to observed signals.

[0116] The goal of independent component analysis is to separate independent signal sources from the observed signal, that is, to solve the separation matrix W so that the separated signal vector is:

[0117] s′=Wx,W=A -1

[0118] Among them, the components s' of the separated signal s' i are statistically independent of each other.

[0119] In this embodiment, the core goal of independent component analysis is to ensure that the separated signal components s' have statistical independence. Since it is difficult to measure independence directly, independent component analysis uses a method of maximizing the non-Gaussianity of the signal to indirectly achieve independence.

[0120] According to the central limit theorem, the non-Gaussianity of the mixed signal is lower than that of its source signals. The separation of independent signals can be achieved indirectly by maximizing the non-Gaussianity of the separated signals.

[0121] Non-Gaussianity measure: Kurtosis

[0122] In this embodiment, non-Gaussianity is measured by kurtosis, and the calculation formula is as follows:

[0123]

[0124] Where y is the input signal, Represents the fourth-order moment of the signal, which is used to describe the "peakedness" of the distribution. Represents the second-order moment of a signal, which is used to measure the energy of the signal.

[0125] The physical meaning of kurtosis is to measure the deviation of signal distribution from Gaussian distribution:

[0126] When kurt(y)>0, the distribution of the signal is sharper than the Gaussian distribution, showing a "peak" characteristic;

[0127] When kurt(y)<0, the distribution of the signal is flatter than the Gaussian distribution and exhibits a "broad peak" characteristic.

[0128] In implementation, ICA optimizes the separation matrix W so that the kurtosis value of the output signal is as far away from 0 as possible, thereby maximizing the non-Gaussianity of the signal.

[0129] Optimization process: solving the separation matrix

[0130] In this embodiment, in order to achieve signal separation, it is necessary to iteratively optimize the separation matrix W, and the steps are as follows:

[0131] 1. Initial setting: Initialize the separation matrix W, which is usually taken as the unit matrix to ensure the stability of the initial solution.

[0132] 2. Non-Gaussian objective function: Take kurt(y) as the optimization objective function. The goal is to maximize kurt(W x ), where W x is the separated signal.

[0133] 3. Gradient optimization: Use the gradient descent method to optimize the separation matrix W\mathbf{W}W. The update formula is as follows:

[0134]

[0135] Among them, α is the learning rate, is the gradient of the objective function, indicating the adjustment direction of the separation matrix.

[0136] 4. Signal orthogonalization: In order to avoid the correlation between separated signals, the separation matrix needs to be orthogonalized after each iteration to ensure that the separated signals are orthogonal to each other. The orthogonalization formula is:

[0137] W←(WW T )-1 / 2 W

[0138] 5. Convergence judgment: When the updated value of the separation matrix W\mathbf{W}W is less than the set threshold, or the change of the objective function is less than a certain set value, the iteration is terminated and the final separation matrix is ​​output.

[0139] Separation signal results:

[0140] After optimization, the separation signal s ′ =Wx satisfies the following characteristics:

[0141] Statistical independence: There is minimal statistical dependence between the components of the separated signal.

[0142] Non-Gaussianity: The distribution of the separated signal deviates significantly from the Gaussian distribution and exhibits sharp or broad peak characteristics.

[0143] Physical meaning: The separated signal is approximately equivalent to the original independent signal source s and has a clear physical meaning (for example, corresponding to different vibration sources).

[0144] The S3 step specifically includes the following steps:

[0145] S3.1. Define signal time series: define two separated signal time series as s 1 =[s 11 ,s 12 ,...,s 1m ] and s 2 =[s 21 ,s 22 ,...,s 2n ];

[0146] S3.2, calculating local alignment distance: calculating the local Euclidean distance between signal sample points;

[0147] S3.3, using a recursive formula to calculate the cumulative alignment path;

[0148] S3.4. Determine the alignment path: Calculate the optimal alignment path of the signal by backtracking the minimum path, and output the cumulative alignment path distance D(i, j).

[0149] The calculation formula of the local Euclidean distance in step S3.2 is d(s 1i ,s 2j )=(s 1i -s 2j ) 2 , the recursive formula in step S3.3 is:

[0150] D(i,j)=d(s 1i ,s 2j)+min{D(i-1,j),D(i,j-1),D(i-1,j-1)}

[0151] Where D(i,j) is the signal s 1 and 2 Cumulative alignment distance at the i-th and j-th positions.

[0152] Specifically, Dynamic Time Warping (DTW) technology is an important step in this method to align the time series of two separated signals. In order to solve the time offset problem that may exist in the separated signals, DTW realizes the alignment of the signals on the time axis by calculating the local alignment distance, recursive cumulative path and optimal path. This technology strictly follows the core idea of ​​the technical solution, based on the recursive calculation of time series and path optimization, and provides important support for the quantification of similarity between signals.

[0153] In this embodiment, in order to time align the two separated signals, the signals are first defined as discrete time series, which are recorded as:

[0154] s 1 =[s 11 ,s 12 ,...,s 1m ] and s 2 =[s 21 ,s 22 ,...,s 2n ]

[0155] Where: s 1 and 2 They are two separate signals; s 1i and 2i Respectively represent the signal s 1 and 2 The vibration signal values ​​at the i-th and j-th time points; m and n are the lengths of the two signals respectively.

[0156] This time series definition ensures a clear representation of the signal in the time domain, providing a basis for subsequent calculation of local alignment distance and path planning.

[0157] In this embodiment, in order to quantify the similarity of signals at a single time point, the local Euclidean distance is used as the metric for two signal segments. The calculation formula is:

[0158] d(s 1i ,s 2j )=(s 1i -s 2j ) 2

[0159] Where: d(s 1i ,s2j ) represents the signal s 1 and 2 The local alignment distance between the i-th and j-th time points, s 1i and 2j are the vibration values ​​of the two signals at the corresponding time points.

[0160] The local Euclidean distance measures the difference between two signals at a single time point and provides basic data for the cumulative calculation of subsequent recursive paths.

[0161] In this embodiment, in order to construct a global alignment path from local alignment distances, a recursive formula is used to calculate the cumulative alignment path. The specific formula is:

[0162] D(i,j)=d(s 1i ,s 2j )+min{D(i-1,j),D(i,j-1),D(i-1,j-1)}

[0163] Where: D(i,j) is the signal s 1 and 2 The cumulative alignment distance between the i-th and j-th time points, d(s 1i ,s 2j ) is the local Euclidean distance at the current time point;

[0164] min{·} is to select the path with the smallest cumulative distance from the three paths:

[0165] D(i-1,j) means the current path extends from the position (i-1,j);

[0166] D(i,j-1) means the current path extends from position (i,j-1);

[0167] D(i-1,j-1) means that the current path extends from the position (i-1,j-1) (diagonal path).

[0168] The recursive formula takes the starting point of the time series as the initial condition:

[0169] D(0,00=0,

[0170] During the recursive calculation process, the path matrix is ​​gradually filled, and finally the global alignment path distance matrix of the time series can be obtained.

[0171] In this embodiment, in order to extract the optimal alignment path of two signal segments from the cumulative path distance matrix, a backtracking method is used to perform path optimization. The backtracking process is as follows:

[0172] Starting from the end point (m,n) of the path matrix, select the previous step with the smallest cumulative distance in the path matrix:

[0173] (i,j)←argmin{D(i-1,j),D(i,j-1),D(i-1,j-1)}

[0174] Repeat the above process until you reach the starting point (0,0).

[0175] Output the optimal alignment path and the cumulative path distance D(m,n).

[0176] The optimal alignment path reflects the alignment relationship between the two signals on the time axis, and the cumulative path distance D(m,n) is the quantified result of the global similarity of the two signals:

[0177] The smaller D(m,n), the higher the overall similarity of the two signals;

[0178] The larger the D(m,n), the more significant the difference between the two signals.

[0179] Dynamic time warping technology solves the problem of the offset of separated signals on the time axis in this technical solution, and provides a high-precision alignment and similarity measurement method for anomaly detection. The local Euclidean distance calculation, recursive path solution, and path optimization in the steps are strictly implemented based on mathematical models, which conform to the logic of the technical solution and ensure the rigor and robustness of the alignment process. Through the backtracking of the path matrix and the cumulative distance, this embodiment finally realizes the optimal alignment path calculation of the two signal segments, laying a solid technical foundation for subsequent anomaly detection.

[0180] The classification basis in step S5 is: using the DTW matching result, matching the abnormal signal with the predefined characteristic signal template, calculating the similarity, and determining the signal category.

[0181] Specifically, classification is a key step in vibration signal processing, which aims to further analyze and identify the detected abnormal signals and classify them as specific target vibration sources. In step S5, the abnormal signals are matched with the predefined feature signal templates through the dynamic time warping (DTW) matching results, and the signal category is determined based on the similarity. This method relies on the alignment path distance output by DTW to ensure the accuracy and robustness of the classification results.

[0182] Classification basis: Dynamic time warping matching results

[0183] In this embodiment, the classification of abnormal signals mainly depends on the DTW matching results. The specific classification basis is:

[0184] Matching principle: Match the abnormal signal with multiple predefined target feature signal templates T KPerform one-by-one matching, and the template set is defined as:

[0185] {T 1 ,T 2 ,…,T K}

[0186] Among them, each template T K Represents the typical signal characteristics of a specific target vibration source.

[0187] Similarity measure:

[0188] Use DTW to calculate the anomaly signal s abn The alignment path distance D between each feature template signal k , the formula is as follows:

[0189] D k =DTW(s abn ,T k )

[0190] Here, DTW(·,·) represents the dynamic time warping matching operation between signals.

[0191] Category judgment: Based on the similarity (i.e., alignment path distance), the abnormal signal is classified into the template category that is closest to it. The specific judgment rules are:

[0192]

[0193] in, is the classification result of abnormal signal, indicating the difference between abnormal signal and The feature template is closest.

[0194] In this embodiment, the classification implementation includes predefined template generation, abnormal signal matching and final category output.

[0195] 1. Pre-definition of characteristic signal templates: According to the actual application scenario, pre-define several target characteristic signal templates. For example, in the pipeline safety monitoring scenario, common templates include:

[0196] External force impact signal template: generated by the vibration signal of the pipeline when it is hit by external force through experimental simulation;

[0197] Mechanical vibration signal template: simulates the vibration signal generated by external construction machinery on the pipeline;

[0198] Environmental background noise template: collects environmental background noise signals without abnormalities.

[0199] Template signals can be generated through cluster analysis of historical data or manually selected through expert experience.

[0200] 2. Matching of abnormal signals with templates: Use DTW to match abnormal signals with all template signals to obtain the alignment path distance set D k The matching steps include:

[0201] The abnormal signal and the template signal are time-aligned through step S3;

[0202] Calculate the cumulative alignment path distance D k To quantify the similarity between signals.

[0203] 3. Classification output: according to The template signal with the highest similarity (i.e., the smallest alignment distance) is selected as the classification result of the abnormal signal.

[0204] Output signal category information, such as "external force impact abnormality" or "mechanical vibration abnormality".

[0205] Classification optimization and robustness design:

[0206] Multi-template matching: set multiple template signals for each target category To cover more possible signal changes. The final classification result is determined by the following rules:

[0207]

[0208] Among them, D kj It represents the alignment path distance between the abnormal signal and the jth template of the kth category.

[0209] Distance threshold filtering: Set the similarity threshold T k when When , it is determined that the abnormal signal does not belong to this category to avoid misclassification.

[0210] Classification probability output: The distance is converted to category probability by normalizing the alignment path distance:

[0211]

[0212] The final classification result is output based on the maximum probability and the classification confidence is provided.

[0213] Please refer to the attached Figure 3 , an optical fiber vibration signal monitoring system based on a method for identifying vibration signals using an optical fiber vibration sensor in a high noise environment, comprising:

[0214] Optical fiber sensor, used to collect environmental vibration signals in real time;

[0215] Data acquisition module, used for collecting and digitally processing vibration signals;

[0216] Signal processing module, used to implement signal preprocessing, independent component analysis and dynamic time warping;

[0217] The classification and anomaly detection module is used to classify vibration signals, identify anomalies, and issue early warning signals.

[0218] Experimental content: Verification experiment of vibration signal recognition method using optical fiber vibration sensor in high noise environment

[0219] Purpose:

[0220] Verify the performance of the optical fiber vibration signal recognition method proposed in the present invention in a high noise environment, evaluate its effectiveness and robustness in signal separation, time alignment, anomaly detection and classification, and analyze the applicability and accuracy of the method.

[0221] Experimental design:

[0222] 1. Experimental environment

[0223] Simulate pipeline monitoring scenarios, and set the ambient noise intensity to ≥80dB.

[0224] Arrange high noise sources (such as mechanical construction equipment) around the pipeline and apply vibration sources to simulate the target signal.

[0225] 2. Experimental Equipment

[0226] Fiber optic sensor: installed at two locations on the pipeline to collect environmental vibration signals.

[0227] Data acquisition module: STM32H743ZIT6 microprocessor, sampling rate is set to 1MHz.

[0228] Signal processing module: The PC platform runs algorithms including signal preprocessing, independent component analysis and dynamic time warping.

[0229] Vibration source equipment:

[0230] External force impact vibration source.

[0231] Slight vibration source.

[0232] High frequency mechanical noise simulation equipment.

[0233] Data recording device: used to record experimental results for subsequent analysis.

[0234] 3. Experimental subjects

[0235] External force impact signal.

[0236] Slight vibration signal.

[0237] High-frequency mechanical construction noise.

[0238] Experimental steps:

[0239] Step 1: Signal Acquisition

[0240] Optical fiber sensors are placed at two locations in the pipeline to collect vibration signals in real time.

[0241] The signal was digitally recorded using a data acquisition module at a sampling rate of 1 MHz and saved as a time series.

[0242] Step 2: Signal preprocessing

[0243] The signal is subjected to DC component removal, amplitude normalization and whitening operations to reduce redundant information and improve signal separation accuracy.

[0244] Step 3: Signal Separation

[0245] Independent component analysis (ICA) is used to separate the mixed signals and extract independent vibration signal sources.

[0246] Step 4: Time Alignment

[0247] Dynamic time warping (DTW) is used to calculate the optimal alignment path between the separated signal and the predefined template signal, and the cumulative alignment path distance is output.

[0248] Step 5: Anomaly Detection and Classification

[0249] Based on the alignment distance and similarity of DTW, the signal is classified as external impact, slight vibration or mechanical noise.

[0250] Determine abnormal signals and output detection results.

[0251] Experimental results:

[0252] In the experiment, three types of signals were collected and processed: external force impact signals, slight vibration signals, and mechanical construction noise. The following are the specific results of signal separation, time alignment, classification, and anomaly detection:

[0253] Experimental items External impact signal Slight vibration signal Mechanical construction noise Separation signal correlation coefficient 0.96 0.94 0.93 DTW alignment path distance 0.45 0.67 0.81 Classification accuracy 98％ 95％ 90％ Anomaly detection sensitivity high high middle Response time (seconds) 0.8 0.9 1.1

[0254] Experimental summary:

[0255] High-precision signal separation: Through independent component analysis (ICA), the experiment achieved high-precision separation of mixed vibration signals. The correlation coefficients of the separated signals exceeded 0.93, which can effectively extract the target vibration source signal while suppressing environmental noise interference.

[0256] Efficient time alignment: Dynamic time warping (DTW) performs well in signal time alignment, and the alignment path distance reflects the degree of similarity of the signals. The alignment path distance of the external impact signal is the smallest (0.45), indicating that its time characteristics match the predefined template most closely.

[0257] Accurate classification and anomaly detection: The DTW-based classification algorithm achieves a classification accuracy rate of up to 98%. At the same time, the anomaly detection sensitivity is high and can promptly distinguish external force impact, slight vibration and mechanical construction noise signals.

[0258] Real-time and robustness: In the experiment, the signal processing and detection response time of the system were both within 1.1 seconds, which has high real-time performance. The classification and detection results remain stable under high noise background, which proves the robustness of this method.

[0259] This experiment verifies the effectiveness and applicability of the vibration signal recognition method using optical fiber vibration sensors in high noise environments, providing strong support for its application in practical scenarios.

[0260] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying vibration signals using an optical fiber vibration sensor in a high noise environment, characterized in that: The following steps are involved: S1. Preprocessing the optical fiber vibration signal, eliminating the DC component, normalizing the amplitude and removing redundant correlation; S2. Separate the preprocessed signals using independent component analysis technology to extract independent signals from different vibration sources; S3, using a dynamic time warping algorithm to time align the separated signals and calculate the alignment path distance between the signals; S4. Determine whether the vibration signal is abnormal based on the comparison between the alignment path distance and the preset threshold value; S5. Classify the vibration signal according to the abnormal detection result, and classify the abnormal signal as a specific target vibration source, and output a warning signal at the same time.

2. The method for vibration signal recognition by an optical fiber vibration sensor in a high noise environment according to claim 1, characterized in that: The formula for eliminating the DC component in the S1 step is as follows: x dc-free (t)=x(t)-mean(x(t)) Among them, x dc-free (t) is the signal after removing the DC component, mean(x(t)) is the mean of the signal; The amplitude normalization operation in step S1 standardizes the amplitude range, and the formula is as follows: Among them, x norm (t) is the normalized signal, min(x) is the minimum value of the signal, and max(x) is the maximum value of the signal; In the step S1, a whitening operation is used to remove redundant correlations, and the formula is as follows: x w =ED -1 / 2 AND T x Where E is the eigenvector matrix of the signal covariance matrix, D is the eigenvalue matrix of the signal covariance matrix, x w is the signal after whitening.

3. The method for vibration signal recognition by an optical fiber vibration sensor in a high noise environment according to claim 1, characterized in that: The formula of the independent component analysis technique in the S2 step is as follows: x=As Among them, x is the observed signal vector, A is the unknown mixing matrix, and s is the independent signal source; The goal of the independent component analysis technique in step S2 is to solve the separation matrix W = A -1 , so that the separated signal is s'=Wx, where the components of s' are independent of each other.

4. The method for vibration signal recognition by an optical fiber vibration sensor in a high noise environment according to claim 1, characterized in that: The independent component analysis in the S2 step is achieved by maximizing the non-Gaussianity of the signal, which is measured by the kurtosis index, and its formula is: Among them, kurt() represents kurtosis, y represents input signal, and E[] represents expectation.

5. The method for vibration signal recognition by optical fiber vibration sensor in a high noise environment according to claim 1, characterized in that: The S3 step specifically includes the following steps: S3.

1. Define signal time series: define two separated signal time series as s1 = [s 11 ,s 12 ,...,s 1m ] and s2=[s 21 ,s 22 ,...,s 2n ]; S3.2, calculating local alignment distance: calculating the local Euclidean distance between signal sample points; S3.3, using a recursive formula to calculate the cumulative alignment path; S3.

4. Determine the alignment path: Calculate the optimal alignment path of the signal by backtracking the minimum path, and output the cumulative alignment path distance D(i, j).

6. The method for vibration signal recognition by an optical fiber vibration sensor in a high noise environment according to claim 5, characterized in that: The calculation formula of the local Euclidean distance in step S3.2 is d(s 1i ,s 2j )=(s 1i -s 2j ) 2 , the recursive formula in step S3.3 is: D(i,j)=d(s 1i ,s 2j )+min{D(i-1,j),D(i,j-1),D(i-1,j-1)} Where D(i,j) is the cumulative alignment distance between signals s1 and s2 at the i-th and j-th positions.

7. The method for vibration signal recognition by an optical fiber vibration sensor in a high noise environment according to claim 1, characterized in that: The classification in step S5 is based on: using the DTW matching result, matching the abnormal signal with the predefined characteristic signal template, calculating the similarity, and determining the signal category.

8. An optical fiber vibration signal monitoring system based on the method of claim 1, characterized in that: include: Optical fiber sensor, used to collect environmental vibration signals in real time; Data acquisition module, used for collecting and digitally processing vibration signals; Signal processing module, used to implement signal preprocessing, independent component analysis and dynamic time warping; The classification and anomaly detection module is used to classify vibration signals, identify anomalies, and issue early warning signals.

Images