A method for structural health monitoring of rail transit

Abstract

The present invention discloses a method for structural health monitoring of rail transit, belonging to the technical field of structural health monitoring of rail transit. In this method, vibration sensors are symmetrically arranged on both sides of the track to obtain a pair of vibration signals. The cepstrum is extracted from each segment of the vibration signal to obtain the real and imaginary cepstrum data. The cepstrum features of both are respectively extracted and corresponding feature vectors are constructed. Furthermore, the mismatch degree of a pair of vibration signals on the feature vectors is obtained, resulting in the real and imaginary mismatch degrees. The various mismatch degrees are evenly divided into three parts in chronological order and the mean values are extracted to calculate the coefficient of variation. The real and imaginary eigenvalue features of the mismatch degree of each part are extracted, and feature enhancement is performed using the coefficient of variation, finally obtaining the structural health value of the rail transit. This method processes and analyzes the vibration signals through multiple steps, improving the accuracy of structural health monitoring of rail transit.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit structural health monitoring, and particularly to a method for rail transit structural health monitoring. Background Art

[0002] In the modern transportation system, rail transit has become an important means of transportation between cities and regions due to its advantages such as large passenger capacity and high efficiency. However, long-term high-load operation, complex geological conditions, and harsh environmental factors pose challenges to the safety and reliability of rail transit structures. Once health problems occur in rail transit structures, not only will the smooth operation of trains be affected, but also serious safety accidents may be triggered, causing huge economic losses and social impacts.

[0003] Traditional rail transit vibration monitoring uses a single-point sensor layout method. Through simple threshold judgment and spectrum analysis, the health condition of the rail transit structure is determined. Specifically, a fixed vibration amplitude threshold is preset in advance. Once the vibration amplitude monitored by the sensor exceeds this threshold, it is determined that there may be an abnormal situation in the track structure and a warning is issued. At the same time, by performing spectrum analysis on the collected vibration signals, an attempt is made to find the abnormal frequency components existing in the signals, so as to judge whether there are health problems in the track structure.

[0004] However, the operating conditions of rail transit are complex and variable. Different train types, operating speeds, load conditions, etc. will all have different degrees of influence on track vibration. These influences will cause fluctuations in the vibration amplitude, and the abnormal frequency components will also change accordingly. This makes it difficult for the traditional monitoring method based on fixed thresholds and simple spectrum analysis to accurately judge the health condition of the track structure, and there is a problem of low accuracy in rail transit structural health monitoring. Summary of the Invention

[0005] Aiming at the above deficiencies in the prior art, a method for rail transit structural health monitoring provided by the present invention solves the problem of low accuracy in rail transit structural health monitoring existing in the prior art.

[0006] To achieve the above invention purpose, the technical solution adopted by the present invention is: A method for rail transit structural health monitoring, comprising the following steps:

[0007] Vibration sensors are symmetrically arranged on both sides of the track to obtain a pair of vibration signals, wherein the signals collected by symmetrically positioned sensors are a pair;

[0008] For each segment of the vibration signals, cepstrum is extracted to obtain real cepstrum data and imaginary cepstrum data;

[0009] Extract cepstrum features from the real cepstrum data and the imaginary cepstrum data respectively, and construct a real cepstrum feature vector and an imaginary cepstrum feature vector;

[0010] Obtain the mismatch degrees of a pair of vibration signals on the real cepstrum feature vector and the imaginary cepstrum feature vector, and obtain the real mismatch degree and the imaginary mismatch degree;

[0011] In chronological order, evenly divide each real and imaginary mismatch degree into three parts, extract the mean value of each part, and calculate the coefficient of variation of the real and imaginary mismatch degrees;

[0012] Extract the real part eigenvalue and the imaginary part eigenvalue of the real and imaginary mismatch degrees of each part, and perform feature enhancement using the coefficient of variation of the real and imaginary mismatch degrees to obtain the rail transit structure health value.

[0013] Furthermore, the cepstrum features of the real cepstrum data include: the mean of the real cepstrum, the variance of the real cepstrum, and the mean of the key real cepstrum, where the key real part is the real part greater than the mean of the real cepstrum;

[0014] The cepstrum features of the imaginary cepstrum data include: the mean of the imaginary cepstrum, the variance of the imaginary cepstrum, and the mean of the key imaginary cepstrum, where the key imaginary part is the imaginary part greater than the mean of the imaginary cepstrum.

[0015] Furthermore, the specific process of obtaining the real mismatch degree is as follows: calculate the similarity of a pair of vibration signals on the real cepstrum feature vector, and use 1 minus the similarity to obtain the real mismatch degree; or calculate the Euclidean distance of a pair of vibration signals on the real cepstrum feature vector, perform normalization processing on the Euclidean distance, and use 1 minus the normalization result to obtain the real mismatch degree;

[0016] Calculate the similarity of a pair of vibration signals on the imaginary cepstrum feature vector, and use 1 minus the similarity to obtain the imaginary mismatch degree; or calculate the Euclidean distance of a pair of vibration signals on the imaginary cepstrum feature vector, perform normalization processing on the Euclidean distance, and use 1 minus the normalization result to obtain the imaginary mismatch degree.

[0017] Furthermore, the formula for calculating the mismatch degree is:

[0018] or ;

[0019] where θ is the mismatch degree, A1 is the real cepstrum feature vector or the imaginary cepstrum feature vector corresponding to one vibration signal, A2 is the real cepstrum feature vector or the imaginary cepstrum feature vector corresponding to another vibration signal, A 1,i is the i-th element in A1, A 2,iis the i-th element in A2, where i is the element number, and f(A1, A2) is the cosine similarity of A1 and A2.

[0020] Furthermore, the specific process of calculating the coefficient of variation of the real and imaginary part mismatches is as follows:

[0021] In chronological order, the real part mismatch data and the imaginary part mismatch data at all time points are equally divided into three parts corresponding to the starting time period, the middle time period, and the ending time period according to the time length;

[0022] Calculate the mean values of the real part mismatches and the imaginary part mismatches in the starting, middle, and ending time periods respectively;

[0023] Calculate the ratios of the mean values of the real part mismatches in the middle and starting, and the ending and middle time periods in sequence, and add the two ratios to obtain the coefficient of variation of the real part mismatch;

[0024] Calculate the ratios of the mean values of the imaginary part mismatches in the middle and starting, and the ending and middle time periods in sequence, and add the two ratios to obtain the coefficient of variation of the imaginary part mismatch.

[0025] Furthermore, the specific process of obtaining the coefficient of variation of the real part mismatch includes: taking the ratio of the mean value of the real part mismatches in the middle time period to the mean value of the real part mismatches in the starting time period, taking the ratio of the mean value of the real part mismatches in the ending time period to the mean value of the real part mismatches in the middle time period, and adding the two ratios to obtain the coefficient of variation of the real part mismatch;

[0026] The specific process of obtaining the coefficient of variation of the imaginary part mismatch includes: taking the ratio of the mean value of the imaginary part mismatches in the middle time period to the mean value of the imaginary part mismatches in the starting time period, taking the ratio of the mean value of the imaginary part mismatches in the ending time period to the mean value of the imaginary part mismatches in the middle time period, and adding the two ratios to obtain the coefficient of variation of the imaginary part mismatch.

[0027] Furthermore, the specific process of obtaining the health value of the rail transit structure is as follows:

[0028] Screen out the real part mismatches greater than 0.5 in each part and calculate the real part outliers of this part;

[0029] Screen out the imaginary part mismatches greater than 0.5 in each part and calculate the imaginary part outliers of this part;

[0030] Process the real part outliers of each part using the first fully connected layer to obtain real part eigenvalues;

[0031] Process the imaginary part outliers of each part using the second fully connected layer to obtain imaginary part eigenvalues;

[0032] The real - part eigenvalue is feature - enhanced by using the variation coefficient of the real - part mismatch degree to obtain the real - part enhanced eigenvalue;

[0033] The imaginary - part eigenvalue is feature - enhanced by using the variation coefficient of the imaginary - part mismatch degree to obtain the imaginary - part enhanced eigenvalue;

[0034] The third fully - connected layer is used to process the real - part enhanced eigenvalue and the imaginary - part enhanced eigenvalue to obtain the rail transit structure health value.

[0035] Further, the specific process of calculating the real - part outlier of this part includes: screening out the real - part mismatch degrees greater than 0.5 in each part to obtain the number M of real - part mismatch degrees greater than 0.5 re,H and the number M of real - part mismatch degrees less than or equal to 0.5 re,L , subtract M re,H from M re,L , normalize the subtraction result, add 1 to the normalized result to obtain the real - part mismatch factor, take the mean of the real - part mismatch degrees greater than 0.5, and multiply the mean by the real - part mismatch factor to obtain the real - part outlier of this part.

[0036] Further, the specific process of calculating the imaginary - part outlier of this part includes: screening out the imaginary - part mismatch degrees greater than 0.5 in each part to obtain the number M of imaginary - part mismatch degrees greater than 0.5 im,H and the number M of imaginary - part mismatch degrees less than or equal to 0.5 im,L , subtract M im,H from M im,L , normalize the subtraction result, add 1 to the normalized result to obtain the imaginary - part mismatch factor, take the mean of the imaginary - part mismatch degrees greater than 0.5, and multiply the mean by the imaginary - part mismatch factor to obtain the imaginary - part outlier of this part.

[0037] Further, multiply the variation coefficient of the real - part mismatch degree by the real - part eigenvalue to obtain the real - part enhanced eigenvalue;

[0038] Multiply the variation coefficient of the imaginary - part mismatch degree by the imaginary - part eigenvalue to obtain the imaginary - part enhanced eigenvalue.

[0039] The beneficial effects of the present invention are:

[0040] 1. In the present invention, vibration sensors are symmetrically arranged on both sides of the track to obtain a pair of vibration signals. This symmetric sensor arrangement can acquire vibration information on both sides of the track simultaneously compared with the traditional single-point arrangement. The vibration conditions on both sides of the track can corroborate and complement each other. When local damage or abnormality occurs on one side of the track, the signal difference at the symmetric position can be more prominently reflected, thereby capturing the vibration characteristics of the track more comprehensively and accurately, avoiding missing key information due to the limitations of single-point monitoring, and helping to more precisely judge the health status of the track structure.

[0041] 2. The present invention extracts the cepstrum from the vibration signals and further obtains the real and imaginary cepstrum data, and then constructs real and imaginary cepstrum feature vectors respectively to reflect the cepstrum distribution of the real and imaginary parts.

[0042] 3. The present invention obtains the mismatching degrees of a pair of vibration signals on the real and imaginary cepstrum feature vectors to obtain the real mismatching degree and the imaginary mismatching degree, quantifying the difference in cepstrum features of the vibration signals at symmetric positions on both sides of the track. Under normal circumstances, the vibration signals at symmetric positions on both sides of the track have similarity in cepstrum features. When an abnormality occurs in the track structure on one side, the vibration response will change significantly, resulting in an increase in the mismatching degree. Using the mismatching degree as a judgment index can more sensitively identify the abnormal conditions of the track structure. Compared with the traditional threshold judgment based on a single signal only, it can more accurately distinguish the normal signal features from the fault signal features, reducing the possibility of false judgment and missed judgment.

[0043] 4. The present invention evenly divides each real and imaginary mismatching degree into three parts in chronological order, extracts the mean value for each part and calculates the coefficient of variation. It can conduct trend analysis on the health status of the track structure. When the coefficient of variation gradually increases, it means that the mismatching degree of the track structure is increasing.

[0044] 5. The present invention extracts the real and imaginary feature values of each part of the real and imaginary mismatching degrees, and uses the coefficient of variation of the real and imaginary mismatching degrees for feature enhancement, finally obtaining the health value of the rail transit structure. The coefficient of variation reflects the degree of change of the mismatching degree. Using it for feature enhancement can make the features related to the health status of the track structure more prominent and obvious, improving the monitoring accuracy of the health value of the rail transit structure. Description of the Drawings

[0045] Figure 1 It is a flowchart of a method for monitoring the health of a rail transit structure.

[0046] Figure 2 It is a schematic diagram of the arrangement. Detailed Embodiment

[0047] The specific embodiments of the present invention will be described below to facilitate those skilled in the art of this technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art of this technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are within the scope of protection.

[0048] As Figure 1 shown, a method for structural health monitoring of rail transit includes the following steps:

[0049] Vibration sensors are symmetrically arranged on both sides of the track to obtain a pair of vibration signals. Among them, the signals collected by the sensors with symmetrical positions are a pair.

[0050] For each segment of the vibration signal, cepstrum is extracted to obtain real part cepstrum data and imaginary part cepstrum data.

[0051] Cepstrum features are respectively extracted from the real part cepstrum data and the imaginary part cepstrum data to construct a real part cepstrum feature vector and an imaginary part cepstrum feature vector.

[0052] The mismatch degrees of a pair of vibration signals on the real part cepstrum feature vector and the imaginary part cepstrum feature vector are obtained to obtain a real part mismatch degree and an imaginary part mismatch degree.

[0053] In chronological order, each real part and imaginary part mismatch degree is evenly divided into three parts, the mean value is extracted for each part, and the variation coefficients of the real part and imaginary part mismatch degrees are calculated.

[0054] The real part eigenvalue and the imaginary part eigenvalue of the real part and imaginary part mismatch degrees of each part are extracted, and feature enhancement is performed using the variation coefficients of the real part and imaginary part mismatch degrees to obtain the structural health value of the rail transit.

[0055] As Figure 2 shown, one vibration sensor is arranged on each side of the track, and another vibration sensor is arranged on the other side of the track. Figure 2 This is only a schematic diagram. In actual arrangement, the vibration sensors are all arranged on the side of the track and in the non-contact part between the track and the train to avoid the influence of train operation on them.

[0056] During the train running process, mechanical vibrations will be propagated on the track. Under the conditions of ideal working conditions and less external interference, the distance that the vibration sensors can significantly detect vibration signals usually can reach dozens of meters. However, this detection distance will be affected by various factors such as track materials and train types.

[0057] The vibration signals collected by the present invention are the vibration signals generated during the process of the train traveling towards the vibration sensors.

[0058] In this embodiment, the collected vibration signals are divided into multiple segments, and the cepstrum is extracted from each segment of the signal. For example, for a 10-minute long vibration signal, it is equally divided into 10 segments or 20 segments, and a sampling frequency of 2000 Hz is set for each segment of the signal.

[0059] The cepstrum features of the real part cepstrum data include: the mean of the real part cepstrum, the variance of the real part cepstrum, and the mean of the key real part cepstrum, where the key real part is the real part greater than the mean of the real part cepstrum;

[0060] The cepstrum features of the imaginary part cepstrum data include: the mean of the imaginary part cepstrum, the variance of the imaginary part cepstrum, and the mean of the key imaginary part cepstrum, where the key imaginary part is the imaginary part greater than the mean of the imaginary part cepstrum.

[0061] When calculating the cepstrum features of the present invention, the real part and the imaginary part are considered separately, and the mean, variance, and the mean of the key part are extracted. The mean reflects the overall level of the cepstrum, the variance measures the degree of dispersion of the cepstrum data. After the cepstrum transformation, the larger cepstrum components correspond to the important periodic or harmonic information of the signal. By taking the part above the mean, the noise interference can be reduced, and more attention can be paid to the main structural features of the signal.

[0062] The cepstrum includes: the real part and the imaginary part. The present invention processes the real part and the imaginary part of the cepstrum separately. The statistical features of the real part cepstrum can reveal the power spectrum information of the signal, such as the energy distribution of the fundamental wave and the harmonic waves. The statistical features of the imaginary part cepstrum can be used for phase structure analysis. For example, in vibration signal analysis, the phase information can provide important information about the signal source.

[0063] In this embodiment, the specific process of obtaining the real part mismatch degree is as follows: calculate the similarity of a pair of vibration signals on the real part cepstrum feature vector, and use 1 minus the similarity to obtain the real part mismatch degree; or calculate the Euclidean distance of a pair of vibration signals on the real part cepstrum feature vector, and perform normalization processing on the Euclidean distance, and use 1 minus the normalization result to obtain the real part mismatch degree;

[0064] Calculate the similarity of a pair of vibration signals on the imaginary part cepstrum feature vector, and use 1 minus the similarity to obtain the imaginary part mismatch degree; or calculate the Euclidean distance of a pair of vibration signals on the imaginary part cepstrum feature vector, and perform normalization processing on the Euclidean distance, and use 1 minus the normalization result to obtain the imaginary part mismatch degree.

[0065] In this embodiment, the formula for calculating the mismatch degree is:

[0066] or ;

[0067] Where, θ is the mismatch degree, A1 is the real cepstrum feature vector or the imaginary cepstrum feature vector corresponding to one vibration signal, A2 is the real cepstrum feature vector or the imaginary cepstrum feature vector corresponding to another vibration signal, A 1,i is the i-th element in A1, A 2,i is the i-th element in A2, i is the element number, and f(A1, A2) is to calculate the cosine similarity of A1 and A2. When calculating the real part mismatch degree, A1 is the real cepstrum feature vector corresponding to one vibration signal, and A2 is the real cepstrum feature vector corresponding to another vibration signal; when calculating the imaginary part mismatch degree, A1 is the imaginary cepstrum feature vector corresponding to one vibration signal, and A2 is the imaginary cepstrum feature vector corresponding to another vibration signal.

[0068] The present invention calculates the similarity of a pair of vibration signals on the real and imaginary cepstrum feature vectors, which can directly reflect the similarity degree of the two signal features. Under normal operating conditions, the vibration signals at symmetric positions on both sides of the track have a high similarity. When the track structure is abnormal, such as track wear, ballast bed diseases, etc., the characteristics of the vibration signals on both sides will change and the similarity will decrease. By subtracting the similarity from 1 to obtain the mismatch degree, the mismatch degree will increase correspondingly, which can clearly reflect the abnormal changes of the track structure, so as to detect the occurrence of faults more accurately.

[0069] The Euclidean distance measures the distance between two feature vectors in space, which can intuitively reflect the difference degree of signal features. Normalizing the Euclidean distance eliminates the influence of different data scales on the results, making the calculation of the mismatch degree more objective and accurate. By subtracting the normalized result from 1 to obtain the mismatch degree, it can clearly reflect the difference situation of the vibration signal features.

[0070] In this embodiment, the specific process of calculating the variation coefficients of the real part and the imaginary part mismatch degrees is as follows:

[0071] According to the time sequence, divide all the real part mismatch degree data and the imaginary part mismatch degree data at all time points into three parts corresponding to the starting time period, the middle time period, and the ending time period according to the time length;

[0072] Calculate the mean values of each real part mismatch degree and each imaginary part mismatch degree in the starting, middle, and ending time periods respectively;

[0073] Calculate the ratios of the mean values of the real part mismatch degrees between the middle and the starting, and between the ending and the middle time periods in sequence, and add the two ratios to obtain the variation coefficient of the real part mismatch degree;

[0074] Calculate the ratios of the mean values of the imaginary part mismatch degrees between the middle and the starting, and between the ending and the middle time periods in sequence, and add the two ratios to obtain the variation coefficient of the imaginary part mismatch degree.

[0075] In this embodiment, the specific process of obtaining the variation coefficient of the real - part mismatch degree includes: taking the ratio of the mean value of the real - part mismatch degrees in the middle time period to the mean value of the real - part mismatch degrees in the starting time period, taking the ratio of the mean value of the real - part mismatch degrees in the ending time period to the mean value of the real - part mismatch degrees in the middle time period, and adding the two ratios to obtain the variation coefficient of the real - part mismatch degree;

[0076] The specific process of obtaining the variation coefficient of the imaginary - part mismatch degree includes: taking the ratio of the mean value of the imaginary - part mismatch degrees in the middle time period to the mean value of the imaginary - part mismatch degrees in the starting time period, taking the ratio of the mean value of the imaginary - part mismatch degrees in the ending time period to the mean value of the imaginary - part mismatch degrees in the middle time period, and adding the two ratios to obtain the variation coefficient of the imaginary - part mismatch degree.

[0077] The present invention equally divides the mismatch - degree data into three time periods according to time and calculates the variation coefficient, which can clearly capture the dynamic changes of the real - part and imaginary - part mismatch degrees over time. For example, in the track monitoring scenario, if the variation coefficient of the real - part mismatch degree gradually increases, it indicates that the difference in the real - part characteristics of the vibration signals on both sides of the track is continuously expanding, predicting that new problems have emerged in the track structure or the existing problems are deteriorating.

[0078] In this embodiment, another implementation method for calculating the variation coefficients of the real - part and imaginary - part mismatch degrees is given. Arrange the real - part or imaginary - part mismatch - degree data of all time points in chronological order, denoted as y k , and the corresponding time points are denoted as t k , k = 1, 2, …, K. Use the least - squares method to perform linear fitting on the data to obtain the fitting straight - line equation y = at + b. Where a is the slope of the fitting straight line and b is the intercept. To obtain the variation coefficient, three specific time points t1, t2, t3 (selected at equal intervals, for example) can be selected, and calculate the corresponding fitting values y1 = at1 + b, y2 = at2 + b, y3 = at3 + b at these three time points respectively. Then the variation coefficient C=(y2 - y1) / y1+(y3 - y2) / y2, k is a positive integer, and K is the data volume.

[0079] In this embodiment, the specific process of obtaining the health value of the rail transit structure is as follows:

[0080] In each part, screen out the real - part mismatch degrees greater than 0.5 and calculate the real - part outliers of this part;

[0081] In each part, screen out the imaginary - part mismatch degrees greater than 0.5 and calculate the imaginary - part outliers of this part;

[0082] Use the first fully - connected layer to process the real - part outliers of each part to obtain the real - part feature values;

[0083] The imaginary part outliers of each part are processed by a second fully-connected layer to obtain imaginary part eigenvalue;

[0084] The real part eigenvalues are feature-enhanced by using the variation coefficient of the real part mismatch degree to obtain real part enhanced eigenvalues;

[0085] The imaginary part eigenvalues are feature-enhanced by using the variation coefficient of the imaginary part mismatch degree to obtain imaginary part enhanced eigenvalues;

[0086] The real part enhanced eigenvalues and the imaginary part enhanced eigenvalues are processed by a third fully-connected layer to obtain the rail transit structure health value.

[0087] In the present invention, all the real part mismatch degrees and the imaginary part mismatch degrees are divided into three parts, namely, the starting time period, the middle time period and the ending time period. Therefore, the first fully-connected layer processes the real part outliers of the starting time period, the middle time period and the ending time period, and the second fully-connected layer processes the imaginary part outliers of the starting time period, the middle time period and the ending time period.

[0088] In the present invention, the outliers are calculated by screening out the real part and the imaginary part mismatch degrees greater than 0.5, which can focus on the data with large deviation from the normal range and highlight the possible structural problems. The outliers of the real part and the imaginary part are calculated separately in the present invention, which can capture the abnormal information in the vibration signal from two different dimensions. In the present invention, the fully-connected layer is used to process the real part and the imaginary part outliers to obtain eigenvalues, and the variation coefficients of the real part and the imaginary part mismatch degrees are used to enhance the corresponding eigenvalues, which can further highlight the correlation between the eigenvalues and the structural health state. The variation coefficient reflects the change trend of the mismatch degree in different time periods. Combining it with the eigenvalues can make the eigenvalues more comprehensively reflect the dynamic changes of the structural health state, enhance the expression ability of the features, and improve the accuracy of the evaluation of the structural health condition. In the present invention, the real part enhanced eigenvalues and the imaginary part enhanced eigenvalues are processed by a third fully-connected layer to obtain the rail transit structure health value, realizing the fusion of the real part and the imaginary part information. The feature information extracted from different angles in the present invention comprehensively considers the structural health conditions reflected by the real part and the imaginary part, can more comprehensively and accurately evaluate the overall health state of the rail transit structure, and avoids the one-sidedness of relying only on the information of a single dimension for evaluation.

[0089] In this embodiment, another implementation scheme for obtaining the health value of the rail transit structure is given: in each part, arrange the real part mismatch degrees into a matrix to obtain the real part mismatch degree matrix, and arrange the imaginary part mismatch degrees into a matrix in each part to obtain the imaginary part mismatch degree matrix. Process the real part mismatch degree matrix and the imaginary part mismatch degree matrix respectively through principal component analysis (PCA) to obtain the real part principal components and the imaginary part principal components. Multiply the variation coefficient of the real part mismatch degree by the real part principal components, multiply the variation coefficient of the imaginary part mismatch degree by the imaginary part principal components, and input the multiplied features into the support vector regression (SVR) model. SVR is a machine learning algorithm for regression analysis, which fits the data by finding the optimal hyperplane to predict the health value of the rail transit structure. When training the SVR model, historical real part outliers, imaginary part outliers, variation coefficients, and corresponding known health value data can be used for training to determine the parameters of the model, and then predict the health value of new data.

[0090] In this embodiment, the real part outlier is equal to the mean of the real part mismatch degrees greater than 0.5. In a more optimal scheme, the specific process of calculating the real part outlier of this part includes: screening out the real part mismatch degrees greater than 0.5 in each part to obtain the number M of the real part mismatch degrees greater than 0.5 re,H and the number M of the real part mismatch degrees less than or equal to 0.5 re,L , subtract M re,H from M re,L , normalize the subtraction result, add the normalized result to 1 to obtain the real part mismatch factor, take the mean of the real part mismatch degrees greater than 0.5, and multiply the mean by the real part mismatch factor to obtain the real part outlier of this part.

[0091] In this embodiment, the formula for calculating the real part outlier of this part is: ;

[0092] where is the real part outlier, ε re is the mean of the real part mismatch degrees greater than 0.5, M re,H is the number of the real part mismatch degrees greater than 0.5, M re,L is the number of the real part mismatch degrees less than or equal to 0.5;

[0093] In this embodiment, the imaginary part outlier is equal to the mean of the imaginary part mismatch degrees greater than 0.5. In a more optimal scheme, the specific process of calculating the imaginary part outlier of this part includes: screening out the imaginary part mismatch degrees greater than 0.5 in each part to obtain the number M of the imaginary part mismatch degrees greater than 0.5 im,H and the number M of the imaginary part mismatch degrees less than or equal to 0.5 im,L , subtract M im,HSubtract from M im,L and normalize the subtraction result, then add 1 to the normalized result to obtain the imaginary part mismatch factor. Take the average of the imaginary part mismatch degrees greater than 0.5, and multiply the average by the imaginary part mismatch factor to obtain the imaginary part outlier of this part.

[0094] In this embodiment, the formula for calculating the imaginary part outlier of this part is: ;

[0095] where is the imaginary part outlier, ε im is the average of the imaginary part mismatch degrees greater than 0.5, M im,H is the number of imaginary part mismatch degrees greater than 0.5, M im,L is the number of imaginary part mismatch degrees less than or equal to 0.5.

[0096] In the present invention, since the range of the mismatch degree is 0 to 1, when it is close to 0, the cepstrum similarity of the two vibration sensors is high, and when it is close to 1, the cepstrum similarity of the two vibration sensors is low. Therefore, the middle value is taken as the threshold, which is not limited to the value selected in the present invention and can be adjusted according to needs.

[0097] The present invention calculates the difference between the number of real part (or imaginary part) mismatch degrees greater than 0.5 and the number of mismatch degrees less than or equal to 0.5, and performs normalization processing to obtain the mismatch factor. This method can effectively reflect the proportion of abnormal data in the overall data. When the number of mismatch degrees greater than 0.5 is relatively large, the mismatch factor will be large, so that the finally calculated outlier will also be large, highlighting the influence degree of the abnormal data. For example, in a certain part of the data, if most of the real part mismatch degrees are greater than 0.5, then the real part mismatch factor will be significantly greater than 1, and the real part outlier obtained by multiplying with the average value will also increase significantly, intuitively reflecting that the abnormal degree of this part of the data is relatively high.

[0098] In this embodiment, multiply the variation coefficient of the real part mismatch degree by the real part eigenvalue to obtain the real part enhanced eigenvalue;

[0099] Multiply the variation coefficient of the imaginary part mismatch degree by the imaginary part eigenvalue to obtain the imaginary part enhanced eigenvalue.

[0100] The first fully connected layer, the second fully connected layer, and the third fully connected layer can be trained using historical real part outliers, imaginary part outliers, variation coefficients, and corresponding known health value data (the training process is implemented using the existing gradient descent method), determine the parameters of the fully connected layer, and then predict the health value of new data.

[0101] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for health monitoring of rail transit structures, characterized in that, The steps include: Vibration sensors are symmetrically arranged on both side tracks to obtain a pair of vibration signals. Among them, the signals collected by the sensors with symmetrical positions are a pair. For each segment of the vibration signals, cepstrum is extracted to obtain real part cepstrum data and imaginary part cepstrum data. Cepstrum features are respectively extracted from the real part cepstrum data and the imaginary part cepstrum data to construct a real part cepstrum feature vector and an imaginary part cepstrum feature vector. The mismatching degrees of a pair of vibration signals on the real part cepstrum feature vector and the imaginary part cepstrum feature vector are obtained to get a real part mismatching degree and an imaginary part mismatching degree. In chronological order, the real part and imaginary part mismatching degrees are respectively evenly divided into three parts. The mean value is extracted for each part, and the variation coefficients of the real part and imaginary part mismatching degrees are calculated. The real part eigenvalues and imaginary part eigenvalues of the real part and imaginary part mismatching degrees of each part are extracted, and feature enhancement is performed using the variation coefficients of the real part and imaginary part mismatching degrees to obtain the rail transit structure health value. The specific process of obtaining the real part mismatching degree is as follows: calculate the similarity of a pair of vibration signals on the real part cepstrum feature vector, and use 1 minus the similarity to obtain the real part mismatching degree; or calculate the Euclidean distance of a pair of vibration signals on the real part cepstrum feature vector, perform normalization processing on the Euclidean distance, and use 1 minus the normalization result to obtain the real part mismatching degree. Calculate the similarity of a pair of vibration signals on the imaginary part cepstrum feature vector, and use 1 minus the similarity to obtain the imaginary part mismatching degree; or calculate the Euclidean distance of a pair of vibration signals on the imaginary part cepstrum feature vector, perform normalization processing on the Euclidean distance, and use 1 minus the normalization result to obtain the imaginary part mismatching degree. The specific process of calculating the variation coefficients of the real part and imaginary part mismatching degrees is as follows: In chronological order, the real part mismatching degree data and the imaginary part mismatching degree data at all time points are respectively equally divided into three parts corresponding to the starting time period, the middle time period, and the ending time period according to the time length. The mean values of the real part mismatching degrees and the imaginary part mismatching degrees in the starting, middle, and ending time periods are respectively calculated. Calculate the ratios of the mean value of the real part mismatching degree in the middle period to that in the starting period and the ratio of the mean value of the real part mismatching degree in the ending period to that in the middle period in turn, and add the two ratios to obtain the variation coefficient of the real part mismatching degree. Calculate the ratios of the mean value of the imaginary part mismatching degree in the middle period to that in the starting period and the ratio of the mean value of the imaginary part mismatching degree in the ending period to that in the middle period in turn, and add the two ratios to obtain the variation coefficient of the imaginary part mismatching degree.

2. The method for monitoring the structural health of rail transit according to claim 1, characterized in that The cepstrum features of the real part cepstrum data include: the mean value of the real part cepstrum, the variance of the real part cepstrum, and the mean value of the key real part cepstrum, where the key real part is the real part greater than the mean value of the real part cepstrum. The cepstrum features of the imaginary part cepstrum data include: the mean value of the imaginary part cepstrum, the variance of the imaginary part cepstrum, and the mean value of the key imaginary part cepstrum, where the key imaginary part is the imaginary part greater than the mean value of the imaginary part cepstrum.

3. The method for monitoring the structural health of rail transit according to claim 1, wherein The formula for calculating the mismatching degree is: or ; Where θ is the mismatch degree, A1 is the real cepstrum feature vector or the imaginary cepstrum feature vector corresponding to one vibration signal, A2 is the real cepstrum feature vector or the imaginary cepstrum feature vector corresponding to another vibration signal, A 1,i is the i-th element in A1, A 2,i is the i-th element in A2, i is the element number, and f(A1, A2) is the cosine similarity of A1 and A2.

4. The method for monitoring the structural health of rail transit according to claim 1, wherein, The specific process of obtaining the variation coefficient of the real part mismatch degree includes: taking the ratio of the mean value of each real part mismatch degree in the middle time period to the mean value of each real part mismatch degree in the starting time period, taking the ratio of the mean value of each real part mismatch degree in the ending time period to the mean value of each real part mismatch degree in the middle time period, adding the two ratios to obtain the variation coefficient of the real part mismatch degree; The specific process of obtaining the variation coefficient of the imaginary part mismatch degree includes: taking the ratio of the mean value of each imaginary part mismatch degree in the middle time period to the mean value of each imaginary part mismatch degree in the starting time period, taking the ratio of the mean value of each imaginary part mismatch degree in the ending time period to the mean value of each imaginary part mismatch degree in the middle time period, adding the two ratios to obtain the variation coefficient of the imaginary part mismatch degree.

5. The method for monitoring the structural health of rail transit according to claim 1, characterized in that, The specific process of obtaining the rail transit structure health value is as follows: In each part, filter out the real part mismatch degrees greater than 0.5 and calculate the real part outliers of this part; In each part, filter out the imaginary part mismatch degrees greater than 0.5 and calculate the imaginary part outliers of this part; Use the first fully connected layer to process the real part outliers of each part to obtain the real part feature values; Use the second fully connected layer to process the imaginary part outliers of each part to obtain the imaginary part feature values; Use the variation coefficient of the real part mismatch degree to perform feature enhancement on the real part feature values to obtain the real part enhanced feature values; Use the variation coefficient of the imaginary part mismatch degree to perform feature enhancement on the imaginary part feature values to obtain the imaginary part enhanced feature values; Use the third fully connected layer to process the real part enhanced feature values and the imaginary part enhanced feature values to obtain the rail transit structure health value.

6. The method for structural health monitoring of rail transit according to claim 5, wherein The specific process of calculating the real part outlier of this part includes: screening out the real part mismatch degrees greater than 0.5 in each part to obtain the number M of real part mismatch degrees greater than 0.5 re,H and the number M of real part mismatch degrees less than or equal to 0.5 re,L , subtracting M re,H from M re,L , normalizing the subtraction result, adding the normalization result to 1 to obtain the real part mismatch factor, taking the mean of the real part mismatch degrees greater than 0.5, and multiplying the mean by the real part mismatch factor to obtain the real part outlier of this part.

7. The method for structural health monitoring of rail transit according to claim 5, characterized in that, The specific process of calculating the imaginary part outliers of this part includes: screening out the imaginary part mismatch degrees greater than 0.5 in each part to obtain the number M of the imaginary part mismatch degrees greater than 0.5 im,H and the number M of the imaginary part mismatch degrees less than or equal to 0.5 im,L , subtracting M im,H from M im,L , normalizing the subtraction result, adding the normalization result to 1 to obtain the imaginary part mismatch factor, taking the mean of the imaginary part mismatch degrees greater than 0.5, and multiplying the mean by the imaginary part mismatch factor to obtain the imaginary part outliers of this part.

8. The method for monitoring the structural health of rail transit according to claim 5, wherein Multiply the variation coefficient of the real part mismatch degree by the real part feature values to obtain the real part enhanced feature values; Multiply the variation coefficient of the imaginary part mismatch degree by the imaginary part feature values to obtain the imaginary part enhanced feature values.

Images