A real-time monitoring method for the health status of high-speed rail sound barriers based on DAS

Abstract

The present invention discloses a real-time monitoring method for the health status of a high-speed rail sound barrier based on DAS, comprising: laying a sensing optical cable on the high-speed rail sound barrier; using a DAS system connected to the sensing optical cable to sense multiple groups of sound barrier plate vibration signals caused by the high-speed rail passing, and dividing the sensed multiple groups of sound barrier plate vibration signals into normal group data and fault group data; preprocessing the normal group data and the fault group data; extracting corresponding multi-domain features based on statistical characteristics from the normal group data and the fault group data; real-time acquisition of the sound barrier plate vibration signal, extracting its corresponding multi-domain features; using a K-nearest neighbor classification algorithm based on Mahalanobis distance, identifying and classifying the multi-domain features of the extracted real-time vibration signal. The present invention can effectively use DAS technology to monitor the vibration of the high-speed rail sound barrier under the excitation of train wind pressure, and judge the health status of the sound barrier.

Description

Technical Field

[0001] The present invention belongs to the technical field of distributed optical fiber sensing, and in particular relates to a real-time monitoring method for the health status of a high-speed railway sound barrier based on DAS, which is used to monitor the vibration of the high-speed railway sound barrier under the excitation of train wind pressure by using DAS technology, and to judge the health status of the sound barrier. Background Art

[0002] The rapid development of high-speed railways has also brought about noise problems along the railways, which has a huge impact on the lives of residents along the lines. At present, building sound barriers on both sides of the high-speed railway is the main measure to solve the noise problem. Metal plug-in sound barriers are one of the commonly used sound barriers. However, the pulsating force generated by the high-speed train will cause the sound-absorbing plugs to loosen, posing a threat to the normal operation of the high-speed railway and the safety of passengers. The rubber parts connected between each sound-absorbing plug of the sound barrier can absorb the pulsating force, thereby extending the service life of the sound barrier. However, when the rubber parts are aged or damaged, the sound-absorbing board may fall off under the repeated excitation of the pulsating force generated by the high-speed train, posing a major safety hazard. Therefore, after the line is opened for operation, the steel components of the sound barrier should be comprehensively checked at least once a month.

[0003] At present, the health inspection of sound barriers is basically carried out manually. The manual inspection method requires the use of a jack to lift the sound barrier, and the inspectors drill out to conduct visual inspection. This inspection method is not only inefficient, but also has relatively high uncertainty and labor costs. At the same time, most high-speed rail lines are located on bridges, which also poses a certain risk to the life safety of sound barrier security inspectors. In contrast, the video monitoring method uses a visual acquisition module to regularly inspect the sound barriers along the train line, which can reduce labor costs and risks to security inspectors. However, this method is easily affected by bad weather such as rain and fog, and the recognition accuracy is not high.

[0004] Compared with the video acquisition method, the distributed optical fiber strain detection method has the advantages of real-time and not being restricted by weather and environment. By detecting the Brillouin frequency shift, the strain parameters can be continuously measured in space, so as to determine the state of the sound barrier according to the size of the strain. However, this method has strict requirements on the layout of the optical fiber, and the optical fiber needs to be pre-stretched to determine the strain coefficient. This method is sensitive to both strain and temperature changes, so the difference in ambient temperature in space may cause strain measurement errors, thereby reducing the recognition accuracy.

[0005] The patent with the publication number CN113866684A discloses a distribution transformer fault diagnosis method based on hybrid sampling and cost sensitivity, including: collecting vibration data of the transformer under different states, and extracting features to form a new sample set; hybrid sampling of the new sample set, using Tomek Links to clean the new sample set generated after oversampling, and dividing it into a training set and a test set; introducing a cost factor, constructing a cost-sensitive multi-classification SVM, and using the training set to train the multi-classification SVM; using the test set to verify the classification effect of the trained cost-sensitive multi-classification SVM classifier, and applying the verified classifier to the fault diagnosis of the distribution transformer. The invention can reliably complete the transformer state detection, overcome the adverse effects of unbalanced data sets on the classifier, improve the separability of positive and abnormal samples, effectively solve the problem of data overlap between categories, and improve the classifier's recognition ability and classification accuracy for transformer fault samples. The patent with publication number CN112084885A discloses a rolling bearing performance evaluation method based on multi-domain feature extraction and related compensation distance. The method extracts the time domain, frequency domain and time-frequency domain features of the collected vibration signal data and forms a multi-domain feature matrix; the extracted feature matrix is ​​subjected to data dimension reduction by the Laplace score algorithm to screen out the features containing the most fault information; the related distance between the fault signal and the normal signal is calculated, and the compensation distance between the fault signals is calculated at the same time. The sum of the related distance and the compensation distance is the related compensation distance; the performance degradation of the rolling bearing can be judged by the size of the related compensation distance. The evaluation index obtained by the performance degradation evaluation method proposed in the invention can monitor the performance degradation trend of the rolling bearing in real time and detect early faults in time.

[0006] However, these traditional feature extraction methods only extract time domain and frequency domain features, and cannot reflect the complexity of the energy distribution of the vibration signal in the frequency domain and the complex state of the time series; the Euclidean distance calculation method in the traditional method cannot avoid the interference of different feature dimensions and the correlation between different features. Therefore, the combination of traditional multi-domain feature extraction and classification methods cannot be directly applied to the health status measurement of high-speed rail sound barrier panels. Summary of the invention

[0007] Technical problem to be solved: The present invention proposes a real-time monitoring method for the health status of high-speed railway sound barriers based on DAS, which can solve the problem of real-time measurement of the health status of high-speed railway sound barrier panels.

[0008] Technical solution:

[0009] A real-time monitoring method for the health status of a high-speed railway sound barrier based on DAS, the real-time monitoring method comprising the following steps:

[0010] S1, laying the sensor optical cable on the high-speed rail sound barrier, so that the sensor optical cable is laid on the sound barrier panels at different heights in each span of the sound barrier, and the sensor optical cable is tightly fixed in the middle of each sound barrier panel;

[0011] S2, using the DAS system connected to the sensor optical cable to sense multiple groups of sound barrier plate vibration signals caused by the high-speed rail passing by, and dividing the sensed multiple groups of sound barrier plate vibration signals into normal group data and fault group data;

[0012] S3, preprocessing the normal group data and the fault group data; extracting corresponding multi-domain features based on statistical characteristics from the normal group data and the fault group data;

[0013] S4, collects the vibration signal of the sound barrier board in real time and extracts its corresponding multi-domain features; uses the K nearest neighbor classification algorithm based on Mahalanobis distance to identify and classify the multi-domain features of the extracted real-time vibration signal: calculates the Mahalanobis distance between the multi-domain features of the real-time perceived sound barrier board vibration signal and the multi-domain features of the known normal group data and fault group data, and takes the sound barrier board sample category with the smallest Mahalanobis distance as the health status category of the real-time sound barrier board vibration signal.

[0014] Furthermore, in step S1, the sensor optical cable is laid in an inverted V-shape on the high-speed railway sound barrier board.

[0015] Furthermore, an optical cable fixing clamp is used to fix the sensor optical cable in the middle of each sound barrier panel.

[0016] Furthermore, in step S2, the DAS system is placed on the platform and connected to the sensor optical cables on the sound barrier boards of different lines through an optical switch. The input end of the optical switch is connected to the DAS system, and the N output ends of the optical switch are connected one-to-one to the sensor optical fibers of the sound barrier boards of the N lines. The line selection state of the optical switch is adjusted to control the DAS system to separately collect the vibration signals on the sound barrier boards of different lines.

[0017] Furthermore, in step S2, the process of sensing the vibration signals of multiple groups of sound barrier panels caused by the high-speed train passing by includes the following steps:

[0018] S21, the light source module of the DAS system emits direct current light, which is pulse modulated by an acousto-optic modulator, pulse amplified by a constant-power erbium-doped fiber amplifier, and filtered to generate a narrow-linewidth laser pulse signal, which is injected into a sensor optical cable laid on the sound barrier board of the line where a train is about to pass, and generates a Rayleigh backscattered light signal in the sensor optical cable. When the train passes by, the Rayleigh backscattered light signal contains the vibration information of the sound barrier under the excitation of wind pressure;

[0019] S22, beat the Rayleigh backscattered light signal with the local oscillator light signal, output an electrical signal through photoelectric conversion, convert the electrical signal into a digital signal, perform orthogonal phase demodulation on it, and obtain the amplitude and phase of the Rayleigh backscattered light signal corresponding to each position along the sensor optical cable distribution on the high-speed railway sound barrier, as follows:

[0020] S23, the obtained digital signal is processed and divided into two digital signals, and the two digital signals are multiplied by the sine signal sinΔωt and the cosine signal cosΔωt respectively to obtain an in-phase signal I and an orthogonal signal Q. L = c × t / 2, and the amplitude E(t) and phase of the Rayleigh backscattered light signal corresponding to each position along the sensor optical cable distribution on the high-speed rail sound barrier are obtained Among them, Δω is the frequency shift introduced by the DC light through the optical pulse modulation module, t is the time variable, k is a constant, L is the distribution position of the sensor optical cable, and c is the speed of light;

[0021] S24, divide the vibration signal into fixed-length windows, perform spectrum analysis on each vibration signal, calculate its power spectrum density, compare the power spectrum density in different windows, distinguish the signals of the wide-spectrum impulse part and the free resonance part, avoid the time period covered by the wide-spectrum impulse generated by the sound barrier board due to train excitation, and only extract the signal of the free resonance part for analysis and processing for subsequent feature recognition and classification.

[0022] Furthermore, in step S3, the process of preprocessing the normal group data and the fault group data includes the following steps:

[0023] The normal group data and the fault group data are filtered by using a Chebyshev I-type digital filter to remove noise below 20 Hz and above 45 Hz.

[0024] The filtered data were normalized using the Z-score method to center the data to a mean of 0 and scale it to a standard deviation of 1.

[0025] Furthermore, in step S3, the process of extracting corresponding multi-domain features based on statistical characteristics from the normal group data and the fault group data includes the following steps:

[0026] S31, calculating the energy value of the vibration signal at each point at the calibrated position corresponding to each group of sound barrier panels, and smoothing the energy value using a moving average filter;

[0027] S32, performing a peak search operation on the smoothed energy value to find the maximum peak point and the peak-valley point, find the peak-valley values ​​on the left and right of the maximum peak, and set the peak-valley values ​​on the left and right as the frame interval corresponding to the maximum peak;

[0028] S33, extracting six characteristic values ​​located in the time domain, frequency domain and information domain respectively for each obtained frame interval; among which, the maximum value, variance and root mean square value are extracted from the time domain; the frequency standard deviation is extracted from the frequency domain; and the power spectrum entropy and singular spectrum entropy characteristics are extracted from the information domain.

[0029] Furthermore, in step S4, the process of using the Mahalanobis distance-based K-nearest neighbor classification algorithm to identify and classify the multi-domain features of the extracted real-time vibration signal includes the following steps:

[0030] The six eigenvalues ​​of the vibration signal sample are combined into a vector X=(x1, x2, x3, x4, x5, x6), where x1 is the maximum value of the vibration signal sample, x2 is the variance of the vibration signal sample, x3 is the root mean square value of the vibration signal sample, x4 is the frequency standard deviation of the vibration signal sample, x5 is the power spectrum entropy of the vibration signal sample, and x6 is the singular spectrum entropy of the vibration signal sample; the mean vector μ of the known normal plate vibration signal population is calculated. n =(μ n1 , μ n2 , μ n3 , μ n4 , μ n5 , μ n6 ) and the covariance matrix ∑ n , where μ n1 is the mean of the maximum values ​​of the normal plate vibration signal, μ n2 is the mean of the variance of the normal plate vibration signal population, μ n3 is the mean of the root mean square value of the normal plate vibration signal, μ n4 is the mean of the frequency standard deviation of the normal plate vibration signal population, μ n5 is the mean value of the power spectrum entropy of the normal plate vibration signal, μ n6 is the mean of the singular spectral entropy of the normal plate vibration signal population: the mean vector μ of the known faulty plate vibration signal population is calculated f =(μ f1 , μ f2 , μf3, μ f4 , μ f5 , μ f6 ) and the covariance matrix ∑ f , where μ f1 is the mean of the maximum values ​​of the faulty board vibration signal, μ f2 is the mean of the variance of the faulty plate vibration signal population, μ f3 is the mean of the root mean square value of the faulty plate vibration signal, μ f4 is the mean of the frequency standard deviation of the faulty plate vibration signal, μ f5 is the mean value of the power spectrum entropy of the faulty plate vibration signal, μ f6is the mean of the singular spectral entropy of the faulty plate vibration signal;

[0031] The Mahalanobis distance between the real-time vibration signal sample and the known normal plate vibration signal population is calculated. The Mahalanobis distance between the real-time vibration signal sample and the known faulty plate vibration signal population is calculated.

[0032] The category of the sound barrier panel sample with the smallest Mahalanobis distance with the real-time vibration signal sample is taken as the category of the sound barrier panel corresponding to the vibration signal position. If it belongs to the category of the faulty panel, an alarm is triggered.

[0033] Beneficial effects:

[0034] The present invention uses the characteristic values ​​of the collected vibration signal in the time domain, frequency domain and information domain to determine whether the characteristics of the sound barrier board belong to the category of the faulty board, thereby determining whether the sound barrier board is faulty, providing safety protection for the high-speed rail sound barrier, and the cost is low; compared with optical fiber stress sensing, the present invention does not need to consider the prestressing setting requirements when laying the optical cable, and will not cause strain measurement errors due to ambient temperature, and has the advantages of convenient laying and low cost. The present invention can also be used as a method for detecting whether the sound barrier is qualified when the sound barrier is delivered; and during long-term operation, the health status of the sound barrier is detected in real time so that it can be repaired in time, providing important guarantees for the safe operation of the high-speed rail. Compared with the traditional feature extraction method, in addition to the time domain and frequency domain, the present invention also extracts power spectrum entropy and singular spectrum entropy features from the information domain. Compared with the time-frequency domain features, it can reflect the complexity of the energy distribution of the vibration signal in the frequency domain and the complex state of the time series; in the K nearest neighbor classification, the Mahalanobis distance is calculated. Compared with the Euclidean distance of the traditional method, the influence of different feature dimensions can be avoided, and the interference of the correlation between features can be eliminated, making the classification more practical. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of the inverted V-shaped layout of the metal plug-in sound barrier structure and the optical cable.

[0036] Figure 2 It is a schematic diagram of multiple line monitoring.

[0037] Figure 3 It is a flow chart of signal processing.

[0038] Figure 4 It is a schematic diagram that distinguishes forced vibration from free resonance. DETAILED DESCRIPTION

[0039] The following examples will enable those skilled in the art to more fully understand the present invention, but are not intended to limit the present invention in any way.

[0040] See also Figure 3 The present invention mentions a real-time monitoring method for the health status of a high-speed railway sound barrier based on DAS. A sensor optical cable is laid on the high-speed railway sound barrier in an inverted V-shape. A single DAS device located at a main station is connected to the sensor optical cables on sound barriers of different lines through an optical switch. The sensor optical cable signals are collected in time-sharing manner, and the collected vibration signals are preprocessed. After multi-domain feature extraction in the time domain, frequency domain and information domain, the features are classified and identified using a K-nearest neighbor classification algorithm based on the Mahalanobis distance, thereby determining whether the sound barrier board where the feature column is located belongs to the faulty board category.

[0041] Specifically, the real-time monitoring method comprises the following steps:

[0042] S1, lay the sensor optical cable on the high-speed rail sound barrier, so that the sensor optical cable is laid on the sound barrier boards at different heights in each span of the sound barrier, and the sensor optical cable is tightly fixed in the middle of each sound barrier board.

[0043] like Figure 1 As shown, the sensor optical cable is laid in an inverted V shape on the high-speed railway sound barrier, ensuring that the sensor optical cable is laid on the sound barrier boards at different heights in each span of the sound barrier. In order to make the sensor optical cable close to the sound barrier board, a special optical cable fixing clip is used to fix the sensor optical cable in the middle of each sound barrier board.

[0044] S2 uses a DAS system connected to a sensor optical cable to sense multiple groups of vibration signals of sound barrier panels caused by the passage of high-speed rail, and divides the sensed multiple groups of vibration signals of sound barrier panels into normal group data and fault group data.

[0045] A single DAS device can realize time-sharing monitoring of the health status of sound barriers on different lines. When there are multiple railway lines at the main station, the single DAS device located at the main station can be connected to the sensor optical cables on the sound barriers of different lines through an optical switch. On the line where a train is about to pass, a narrow pulse laser signal is injected into the sensor optical cable of the line through an optical switch to monitor the vibration information generated by the sound barrier after the train is excited; after the sound barrier monitoring of the line is completed, the optical switch of the DAS device is switched to other lines to realize the function of polling monitoring of all lines.

[0046] The connection between the DAS system of the main site and the sensor optical cables of each line is as follows Figure 2 As shown in the figure, a single DAS device located at the main site is connected to the sensor optical cables on the sound barriers of different lines through an optical switch. On the line where a train is about to pass, a narrow pulse laser signal is injected into the sensor optical cable of the line through the optical switch to collect the sensor optical cable signal of the line. When the train passes by, the signal contains the vibration information of the sound barrier under the excitation of wind pressure.

[0047] The light source module of the DAS system first emits a direct current light with a wavelength of 1550nm. The direct current light is pulse modulated by an acousto-optic modulator, amplified by a constant-power erbium-doped fiber amplifier, and filtered to generate a narrow-linewidth laser pulse signal. The laser pulse signal is injected into the sensor optical cable laid on the sound barrier of the line where the train is about to pass, and a Rayleigh backscattered light signal is generated in the sensor optical cable. When the train passes, the signal contains the vibration information of the sound barrier under the excitation of wind pressure. The Rayleigh backscattered light signal is beat with the local oscillator light signal, and an electrical signal is output through photoelectric conversion. The electrical signal is converted into a digital signal, and orthogonal phase demodulation is performed on it to obtain the amplitude and phase of the Rayleigh backscattered light signal corresponding to each position along the sensor optical cable distribution on the high-speed rail sound barrier, as follows:

[0048] The obtained digital signal is processed and divided into two digital signals. The two digital signals are multiplied by the sine signal sinΔωt and the cosine signal cosΔωt to obtain the in-phase signal I and the orthogonal signal Q. L = c × t / 2, and the amplitude E(t) and phase of the Rayleigh backscattered light signal corresponding to each position along the sensor optical cable distribution on the high-speed rail sound barrier are obtained Among them, Δω is the frequency shift introduced by the DC light passing through the optical pulse modulation module, t is the time variable, k is a constant, L is the distribution position of the sensor optical cable, and c is the speed of light.

[0049] This embodiment uses the DAS system to collect vibration information generated by the sound barrier, avoids the time period covered by the wide spectrum impulse generated by the sound barrier due to train excitation, and only extracts the signal of the free resonance part for analysis and processing for subsequent feature recognition and classification. The extraction method is: divide the signal into fixed-length windows, perform spectrum analysis on each signal segment, calculate its power spectrum density, and compare the power spectrum density in different windows to distinguish the signal of the wide spectrum impulse part from the signal of the free resonance part, such as Figure 4 As shown, forced vibration and free resonance can be distinguished directly. As a general strategy, the extraction can be started 50 milliseconds after the end of the wide spectrum impulse part.

[0050] S3, preprocessing the normal group data and the fault group data; extracting corresponding multi-domain features based on statistical characteristics from the normal group data and the fault group data.

[0051] The extracted signal was preprocessed, specifically, a Chebyshev I-based digital filter was used to filter out noise below 20 Hz and above 45 Hz. The filtered data was normalized using the “Z-score” method, centering the data to a mean of 0 and scaling it to a standard deviation of 1.

[0052] Calculate the signal energy value of each point at the calibrated position corresponding to each group of sound barriers, use a moving average filter to smooth the energy value, perform peak search on the energy value, find the maximum peak point and valley point, and finally have a valley value on the left and right of the maximum peak. Set the left and right valleys as frame intervals.

[0053] Six eigenvalues ​​are extracted in the time domain, frequency domain and information domain for each frame interval. The extracted time domain eigenvalues ​​are the maximum value, variance and root mean square value, and the extracted frequency domain eigenvalues ​​are the frequency standard deviation. In addition, in order to overcome the problem that the time-frequency domain features cannot reflect the complexity of the energy distribution of the vibration signal in the frequency domain and the complex state of the time series, the power spectrum entropy and singular spectrum entropy are extracted in the information domain. A total of 6 eigenvalues ​​are extracted.

[0054] S4, collects the vibration signal of the sound barrier board in real time and extracts its corresponding multi-domain features; uses the K nearest neighbor classification algorithm based on Mahalanobis distance to identify and classify the multi-domain features of the extracted real-time vibration signal: calculates the Mahalanobis distance between the multi-domain features of the real-time perceived sound barrier board vibration signal and the multi-domain features of the known normal group data and fault group data, and takes the sound barrier board sample category with the smallest Mahalanobis distance as the health status category of the real-time sound barrier board vibration signal.

[0055] In order to avoid the problems of Euclidean distance calculation in traditional classification methods being affected by different feature dimensions and correlation interference between features, the K nearest neighbor classification algorithm based on Mahalanobis distance is used to identify and classify the extracted features: the Mahalanobis distances between the vibration signal samples and the known normal and faulty panels are calculated respectively. The category of the sound barrier sample with the smallest Mahalanobis distance to the sample, that is, the category of the sound barrier panel corresponding to the vibration signal position, is the category of the sound barrier panel. If it belongs to the faulty panel category, an alarm is triggered.

[0056] Specifically, the six eigenvalues ​​of the vibration signal sample form a vector X = (x1, x2, x3, x4, x5, x6); the mean vector μ of the known normal plate vibration signal population n =(μ n1 , μ n2 , μ n3 , μ n4 , μ n5 , μ n6 ), the covariance matrix is ​​∑ n ; The known mean vector μ of the faulty plate vibration signal population f =(μ f1 , μ f2 , μ f3 , μ f4 , μ f5 , μ f6 ), the covariance matrix is ​​∑ f; Mahalanobis distance between the vibration signal sample and the known normal plate vibration signal population Mahalanobis distance between the vibration signal sample and the known faulty board vibration signal population The category of the sound barrier sample with the smallest Mahalanobis distance to the sample, that is, the category of the sound barrier board corresponding to the vibration signal position, will be alarmed if it belongs to the category of faulty board.

[0057] After the sound barrier monitoring of the line is completed, the optical switch of the DAS equipment is switched to other lines, and a narrow linewidth laser pulse signal is injected into the sensor optical cable on the sound barrier of other lines where trains are about to pass, and the above steps are repeated to realize the function of polling monitoring of all lines.

Claims

1. A real-time monitoring method for the health status of high-speed rail sound barriers based on DAS, characterized in that: The real-time monitoring method comprises the following steps: S1, laying the sensor optical cable on the high-speed rail sound barrier, so that the sensor optical cable is laid on the sound barrier panels at different heights in each span of the sound barrier, and the sensor optical cable is tightly fixed in the middle of each sound barrier panel; S2, using the DAS system connected to the sensor optical cable to sense multiple groups of sound barrier plate vibration signals caused by the high-speed rail passing by, and dividing the sensed multiple groups of sound barrier plate vibration signals into normal group data and fault group data; S3, preprocessing the normal group data and the fault group data; extracting corresponding multi-domain features based on statistical characteristics from the normal group data and the fault group data; S4, collect the vibration signal of the sound barrier board in real time, and extract its corresponding multi-domain features; use the K nearest neighbor classification algorithm based on Mahalanobis distance to identify and classify the multi-domain features of the extracted real-time vibration signal: calculate the Mahalanobis distance between the multi-domain features of the real-time perceived sound barrier board vibration signal and the multi-domain features of the known normal group data and fault group data, and take the sound barrier board sample category with the smallest Mahalanobis distance as the health status category of the real-time sound barrier board vibration signal; In step S2, the process of sensing the vibration signals of multiple groups of sound barrier panels caused by the high-speed train passing by includes the following steps: S21, the light source module of the DAS system emits direct current light, which is pulse modulated by an acousto-optic modulator, pulse amplified by a constant-power erbium-doped fiber amplifier, and filtered to generate a narrow-linewidth laser pulse signal, which is injected into a sensor optical cable laid on the sound barrier board of the line where a train is about to pass, and generates a Rayleigh backscattered light signal in the sensor optical cable. When the train passes by, the Rayleigh backscattered light signal contains the vibration information of the sound barrier under the excitation of wind pressure; S22, beat the Rayleigh backscattered light signal with the local oscillator light signal, output an electrical signal through photoelectric conversion, convert the electrical signal into a digital signal, perform orthogonal phase demodulation on it, and obtain the amplitude and phase of the Rayleigh backscattered light signal corresponding to each position along the sensor optical cable distribution on the high-speed railway sound barrier, as follows: S23, the obtained digital signal is processed and divided into two digital signals, and the two digital signals are multiplied by the sine signal sinΔωt and the cosine signal cosΔωt respectively to obtain an in-phase signal I and an orthogonal signal Q. L = c × t / 2, and the amplitude E(t) and phase of the Rayleigh backscattered light signal corresponding to each position along the sensor optical cable distribution on the high-speed rail sound barrier are obtained Among them, Δω is the frequency shift introduced by the DC light through the optical pulse modulation module, t is the time variable, k is a constant, L is the distribution position of the sensor optical cable, and c is the speed of light; S24, divide the vibration signal into fixed-length windows, perform spectrum analysis on each vibration signal, calculate its power spectrum density, compare the power spectrum density in different windows, distinguish the signals of the wide-spectrum impulse part and the free resonance part, avoid the time period covered by the wide-spectrum impulse generated by the sound barrier board due to train excitation, and only extract the signal of the free resonance part for analysis and processing for subsequent feature recognition and classification.

2. The real-time monitoring method for the health status of high-speed railway sound barriers based on DAS according to claim 1 is characterized in that: In step S1, the sensor optical cable is laid in an inverted V shape on the high-speed railway sound barrier board.

3. The real-time monitoring method for the health status of high-speed railway sound barriers based on DAS according to claim 1 is characterized in that: The sensing optical cable is fixed in the middle of each sound barrier panel using an optical cable fixing clamp.

4. The real-time monitoring method for the health status of high-speed railway sound barriers based on DAS according to claim 1 is characterized in that: In step S2, the DAS system is placed on the platform and connected to the sensor optical cables on the sound barrier boards of different lines through an optical switch. The input end of the optical switch is connected to the DAS system, and the N output ends of the optical switch are connected one-to-one to the sensor optical fibers of the sound barrier boards of the N lines. The line selection state of the optical switch is adjusted to control the DAS system to separately collect the vibration signals on the sound barrier boards of different lines.

5. The real-time monitoring method for the health status of high-speed railway sound barriers based on DAS according to claim 1 is characterized in that: In step S3, the process of preprocessing the normal group data and the fault group data includes the following steps: The normal group data and the fault group data are filtered by using a Chebyshev I-type digital filter to remove noise below 20 Hz and above 45 Hz. The filtered data were normalized using the Z-score method to center the data to a mean of 0 and scale it to a standard deviation of 1.

6. The method for real-time monitoring of the health status of high-speed railway sound barriers based on DAS according to claim 1 is characterized in that: In step S3, the process of extracting corresponding multi-domain features based on statistical characteristics from the normal group data and the fault group data includes the following steps: S31, calculating the energy value of the vibration signal at each point at the calibrated position corresponding to each group of sound barrier panels, and smoothing the energy value using a moving average filter; S32, performing a peak search operation on the smoothed energy value to find the maximum peak point and the peak-valley point, find the peak-valley values ​​on the left and right of the maximum peak, and set the peak-valley values ​​on the left and right as the frame interval corresponding to the maximum peak; S33, extracting six characteristic values ​​located in the time domain, frequency domain and information domain respectively for each obtained frame interval; among which, the maximum value, variance and root mean square value are extracted from the time domain; the frequency standard deviation is extracted from the frequency domain; and the power spectrum entropy and singular spectrum entropy characteristics are extracted from the information domain.

7. The real-time monitoring method for the health status of high-speed railway sound barriers based on DAS according to claim 1 is characterized in that: In step S4, the process of using the K nearest neighbor classification algorithm based on Mahalanobis distance to identify and classify the multi-domain features of the extracted real-time vibration signal includes the following steps: The six eigenvalues ​​of the vibration signal sample are combined into a vector X = (x1, x2, x3, x4, x5, x6), where x1 is the maximum value of the vibration signal sample, x2 is the variance of the vibration signal sample, x3 is the root mean square value of the vibration signal sample, x4 is the frequency standard deviation of the vibration signal sample, x5 is the power spectrum entropy of the vibration signal sample, and x6 is the singular spectrum entropy of the vibration signal sample; the mean vector μ of the known normal plate vibration signal population is calculated n =(μ n1 ,μ n2 ,μ n3 ,μ n4 ,μ n5 ,μ n6 ) and the covariance matrix Σ n , where μ n1 is the mean of the maximum values ​​of the normal plate vibration signal, μ n2 is the mean of the variance of the normal plate vibration signal population, μ n3 is the mean of the root mean square value of the normal plate vibration signal, μ n4 is the mean of the frequency standard deviation of the normal plate vibration signal population, μ n5 is the mean value of the power spectrum entropy of the normal plate vibration signal, μ n6 is the mean of the singular spectral entropy of the normal plate vibration signal population; the mean vector μ of the known faulty plate vibration signal population is calculated f =(μ f1 ,μ f2 ,μ f3 ,μ f4 ,μ f5 ,μ f6 ) and the covariance matrix Σ f , where μ f1 is the mean of the maximum values ​​of the faulty board vibration signal, μ f2 is the mean of the variance of the faulty plate vibration signal population, μ f3 is the mean of the root mean square value of the faulty plate vibration signal, μ f4 is the mean of the frequency standard deviation of the faulty plate vibration signal, μ fs is the mean value of the power spectrum entropy of the faulty plate vibration signal, μ f6 is the mean of the singular spectral entropy of the faulty plate vibration signal; The Mahalanobis distance between the real-time vibration signal sample and the known normal plate vibration signal population is calculated. The Mahalanobis distance between the real-time vibration signal sample and the known faulty plate vibration signal population is calculated. The category of the sound barrier panel sample with the smallest Mahalanobis distance with the real-time vibration signal sample is taken as the category of the sound barrier panel corresponding to the vibration signal position. If it belongs to the category of the faulty panel, an alarm is triggered.

Images