Electronic fingerprint-based rail damage monitoring method and system

Abstract

The present application belongs to the technical field of track detection, and provides a rail damage monitoring method and system based on electronic fingerprints. The method comprises: generating a first amplitude-frequency response curve of the rail based on an electrical signal collected by a measuring end located at one end of the rail; performing noise suppression and feature enhancement processing on the first amplitude-frequency response curve to obtain a second amplitude-frequency response curve, extracting a feature matrix from the second amplitude-frequency response curve, and taking the feature matrix as an electronic fingerprint of the rail section; calculating a feature difference degree of the electronic fingerprint and a preset electronic fingerprint, and calculating a damage index based on the feature difference degree. The present application realizes full-band coverage detection of rail damage, effectively suppresses environmental noise, enhances weak damage features, and improves detection sensitivity and accuracy.

Description

Technical Field

[0001] This invention relates to the field of track inspection technology, and more specifically, to a method and system for monitoring rail damage based on electronic fingerprints. Background Technology

[0002] As the core infrastructure of rail transit, the service condition of rails is directly related to the safety of train operation. However, under the influence of multiple factors such as long-term load, environmental erosion and material fatigue, rails are prone to damage such as cracks, wear and peeling. Traditional rail damage detection methods mainly rely on manual inspection, ultrasonic testing and eddy current testing, but these methods have significant limitations: (1) Manual inspection: low efficiency, high labor intensity, and significantly affected by the experience and subjective judgment of the inspectors, making it difficult to achieve real-time and comprehensive monitoring of rail damage. (2) Ultrasonic testing: requires high operational skills from the inspectors, and the test results are greatly affected by factors such as the effect of the coupling agent and the surface condition of the rail, especially in areas with complex geometric structures (such as welds, frogs, etc.). (3) Eddy current testing: limited by the skin effect, it has insufficient ability to detect deep damage, and is easily interfered with by metal surface oxide scale, oil stains, etc., resulting in a high false detection rate.

[0003] With the development of rail transit towards higher speeds and heavier loads, higher demands are placed on the real-time performance, accuracy, and intelligence of rail damage monitoring. The traditional detection methods mentioned above are no longer sufficient to meet the needs of modern rail transit operation and maintenance, necessitating the development of an efficient and reliable rail damage monitoring solution. Summary of the Invention

[0004] In response, the present invention provides a rail damage monitoring method, system, electronic device, computer storage medium, and computer program product based on electronic fingerprinting to solve at least one of the above-mentioned technical problems.

[0005] In a first aspect, the present invention provides a rail damage monitoring method based on electronic fingerprinting, comprising the following steps: generating a first amplitude-frequency response curve of the rail based on an electrical signal acquired by a measuring end located at one end of the rail, wherein the electrical signal is acquired after a broadband sweeping electrical signal is emitted by an excitation end located at the other end of the rail; performing noise suppression and feature enhancement processing on the first amplitude-frequency response curve to obtain a second amplitude-frequency response curve, extracting a feature matrix from the second amplitude-frequency response curve, and using the feature matrix as the electronic fingerprint of the rail segment; wherein the noise suppression and feature enhancement are implemented using a Transformer-based linkage mechanism; calculating the feature difference degree between the electronic fingerprint and a preset electronic fingerprint, and calculating a damage index based on the feature difference degree; wherein the feature difference degree is calculated based on a combination of frequency difference degree and amplitude difference degree.

[0006] In a second aspect, the present invention provides a rail damage monitoring system based on electronic fingerprinting. The system includes an acquisition module, a preprocessing module, and a damage monitoring module connected in sequence. The acquisition module generates a first amplitude-frequency response curve of the rail based on an electrical signal acquired by a measuring end located at one end of the rail. The electrical signal is acquired after a broadband sweep signal is emitted by an excitation end located at the other end of the rail. The preprocessing module performs noise suppression and feature enhancement processing on the first amplitude-frequency response curve to obtain a second amplitude-frequency response curve, extracts a feature matrix from the second amplitude-frequency response curve, and uses this feature matrix as the electronic fingerprint of the rail segment. The noise suppression and feature enhancement are implemented using a Transformer-based linkage mechanism. The damage monitoring module calculates the feature difference degree between the electronic fingerprint and a preset electronic fingerprint, and calculates a damage index based on the feature difference degree. The feature difference degree is calculated based on a combination of frequency difference degree and amplitude difference degree.

[0007] In a third aspect, the present invention provides an electronic device comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program, when executed by the processor, implements the method as described in any of the preceding claims.

[0008] In a fourth aspect, the present invention provides a computer storage medium storing a computer program that can be executed by a processor to implement the method as described in any of the preceding claims.

[0009] In a fifth aspect, the present invention provides a computer program product comprising a computer program executable by a processor to implement the method as described in any of the preceding claims.

[0010] This invention deploys excitation and measurement ends at both ends of the rail, transmits broadband swept-frequency electrical signals and acquires amplitude-frequency response curves, and extracts electronic fingerprints by combining a Transformer-based noise suppression and feature enhancement linkage mechanism. Then, it calculates the feature difference to derive a damage index. This invention achieves full-band coverage detection of rail damage, effectively suppresses environmental noise, enhances weak damage features, improves detection sensitivity and accuracy, and can provide real-time early warning and quantify the degree of damage, providing efficient and reliable technical support for the operation and maintenance of rail transit. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic flowchart of the rail damage monitoring method based on electronic fingerprinting disclosed in an embodiment of the present invention.

[0013] Figure 2 This is a schematic diagram of the structure of the rail damage monitoring system based on electronic fingerprint disclosed in an embodiment of the present invention.

[0014] Figure 3 This is a schematic diagram of the structure of the electronic device disclosed in the embodiments of the present invention. Detailed Implementation

[0015] The following specific embodiments illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] Furthermore, the technical features involved in the different embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0017] like Figure 1 As shown, this embodiment of the invention discloses a rail damage monitoring method based on electronic fingerprint, including the following steps: S101, generating a first amplitude-frequency response curve of the rail based on the electrical signal collected by the measuring end located at one end of the rail, wherein the electrical signal is collected after the excitation end located at the other end of the rail transmits a broadband sweep frequency electrical signal.

[0018] This invention uses the principle of electrical signal detection to monitor whether there is damage to the rail. When the excitation signal propagates in the rail, the damage will cause the electrical signal to be reflected, scattered and attenuated. The electrical signal received by the measuring end contains the damage information. By comparing it with the electrical signal of a healthy rail, it is possible to determine whether there is damage.

[0019] An excitation end and a measurement end are deployed at both ends of the rail to form a physical closed loop: the excitation end emits a wideband swept frequency electrical signal (covering the low-frequency to high-frequency band, such as 10Hz-10kHz) to excite the rail's vibration response at different frequencies; the measurement end collects electrical signals containing damage information and environmental noise, and generates the first amplitude-frequency response curve (the horizontal axis is frequency and the vertical axis is amplitude) through a fast Fourier transform (FFT).

[0020] It is understandable that rail damage (such as cracks and wear) will produce differentiated responses to excitation signals of different frequencies, and a single-frequency excitation is difficult to cover all damage characteristics. Therefore, this invention excites multimodal vibration of the rail with a wideband signal (e.g., 10Hz-10kHz), so that damage characteristics are presented in multiple frequency bands (e.g., microcracks resonate in the high-frequency band, while wear causes low-frequency energy attenuation), improving the comprehensiveness of damage detection and also significantly improving detection efficiency.

[0021] S102, noise suppression and feature enhancement are performed on the first frequency response curve to obtain the second frequency response curve. The feature matrix is ​​extracted from the second frequency response curve and used as the electronic fingerprint of the rail section. The noise suppression and feature enhancement are achieved using a Transformer-based linkage mechanism.

[0022] The original electrical signals collected contain multi-frequency mixed features: the amplitude-frequency response of healthy rails shows a stable resonance peak, while damage (such as cracks) will cause the resonance frequency to shift, the amplitude to decay, or the addition of stray frequency components; noise components include power frequency interference (50Hz), train vibration noise (low frequency band), electromagnetic pulse (high frequency band), etc., which are mixed with damage features. The above noises will mask weak damage features (such as the high frequency resonance amplitude change caused by microcracks), which will easily lead to errors in subsequent feature extraction. Therefore, it is necessary to perform noise suppression and feature enhancement processing on the first amplitude-frequency response curve to obtain a more accurate second amplitude-frequency response curve. Noise suppression can be achieved by adaptive notch filtering, wavelet threshold denoising, frequency domain energy normalization, etc., which are explained as follows: (1) Adaptive notch filtering: Power frequency interference is the most common periodic noise, and its energy is mainly concentrated around 50Hz. Notch filtering can accurately attenuate the noise in this frequency band and avoid affecting other frequency features. Therefore, an IIR filter (center frequency f0 = 50Hz) with a damping coefficient r = 0.95 was designed to suppress periodic noise by frequency domain notch filtering for power frequency interference (50Hz).

[0023] (2) Wavelet thresholding denoising: Wavelet transform can decompose a signal into different frequency sub-bands. Noise in the high-frequency sub-bands (usually Gaussian distributed) is thresholded while preserving low-frequency damage characteristics (e.g., energy attenuation due to wear). For example, a 5-level db4 wavelet decomposition of the signal can be performed to calculate noise estimates at each scale. Using empirical thresholds Suppress aperiodic random noise (e.g., electromagnetic pulses).

[0024] (3) Frequency domain energy normalization: The signal energy is calibrated by the environmental noise power spectrum compensation function G(f) to eliminate the influence of environmental noise intensity fluctuations at different times, so as to ensure that the signals collected in different time periods have comparable energy benchmarks and avoid distortion of the feature matrix due to changes in environmental noise power.

[0025] Early damage (e.g., microcracks) causes weak signal changes (e.g., amplitude attenuation <5%, frequency shift <10Hz), requiring signal processing techniques to amplify the difference between the feature and noise. Feature enhancement can employ synchronous averaging techniques, explained as follows: damage-related signals (e.g., fixed-frequency response caused by cracks) are periodic, while random noise is not. Multiple averaging allows the damage signal energy to accumulate, while the noise energy attenuates (noise power increases with frequency). (Reduction). Therefore, the denoised signal is subjected to synchronous averaging, for example, N≥32 times, with reference phase. Locking the phase of the excitation signal can further improve the signal-to-noise ratio by 5-8 dB, making the amplitude of weak damage features (such as the characteristic frequency components of early cracks) more significant, which is beneficial for differentiation and identification.

[0026] In addition, the present invention also designs a linkage mechanism based on Transformer to achieve noise suppression and feature enhancement processing in a coordinated manner, thereby making the obtained second amplitude frequency response curve easier to analyze and obtain a more accurate feature difference degree, thus obtaining a more accurate damage index.

[0027] S103, calculate the feature difference degree between the electronic fingerprint and the preset electronic fingerprint, and calculate the damage index based on the feature difference degree; wherein, the feature difference degree is calculated based on the frequency difference degree and the amplitude difference degree.

[0028] The difference between the quantified current state (i.e., the aforementioned electronic fingerprint) and the preset electronic fingerprint corresponding to the healthy state is calculated, and a numerical assessment of the degree of damage is output. The feature difference is calculated based on a combination of frequency difference and amplitude difference, as detailed below: ;in, Let j be the j-th frequency feature in the electronic fingerprint. This refers to the j-th frequency feature in the preset electronic fingerprint; The frequency weights are derived from the Transformer's self-attention weights, for example, assigning higher weights to high-frequency bands that are sensitive to damage.

[0029] Amplitude difference ;in, For the j-th amplitude feature in this electronic fingerprint, This is the j-th amplitude feature in the preset electronic fingerprint; The frequency weights are learned through a self-attention mechanism, for example, assigning higher weights to frequencies that are sensitive to changes in the amplitude of the resonance peak.

[0030] The characteristic difference is obtained by combining the frequency difference and amplitude difference. ,in, ,For example .

[0031] After extracting the feature dissimilarity, for example, an adaptive threshold sigmoid function can be used to convert the feature dissimilarity into a damage index. The adaptive threshold sigmoid function is as follows: Where k=3.0 is used to control the slope of the curve and amplify the effect of the difference on the damage index.

[0032] This is a dynamic threshold, predicted by the Transformer based on environmental parameters (such as temperature and load). For example, threshold adaptation can be achieved by adding environmental features as input to the encoder. For instance, increased temperature causes thermal expansion and contraction of the rails, resulting in a slight shift in their natural frequency; a dynamic threshold... It can automatically calibrate such non-destructive differences.

[0033] The damage index I ∈ [0,1] indicates that the larger the value, the more severe the damage. An alert is triggered when the value exceeds a preset threshold (e.g., I > 0.5). For example, I < 0.3: healthy or minor damage, requiring regular monitoring.

[0034] 0.3≤I<0.7: Moderate damage, triggering an early warning, maintenance recommended.

[0035] I≥0.7: Severe damage, alarm immediately and emergency treatment.

[0036] Simultaneously combined and The distribution and location of damage types, for example: Dominant factor: Significant frequency shift, possibly due to rail cracks (stiffness changes).

[0037] Dominant indicator: Significant amplitude attenuation, possibly due to rail wear (material loss).

[0038] This invention deploys excitation and measurement ends at both ends of the rail, transmits broadband swept-frequency electrical signals and acquires amplitude-frequency response curves, and extracts electronic fingerprints by combining a Transformer-based noise suppression and feature enhancement linkage mechanism. Then, it calculates the feature difference to derive a damage index. This invention achieves full-band coverage detection of rail damage, effectively suppresses environmental noise, enhances weak damage features, improves detection sensitivity and accuracy, and can provide real-time early warning and quantify the degree of damage, providing efficient and reliable technical support for the operation and maintenance of rail transit.

[0039] Further, the step of performing noise suppression and feature enhancement processing on the first frequency response curve to obtain the second frequency response curve includes: sequentially performing adaptive notch filtering, wavelet threshold denoising, and frequency domain energy normalization processing on the first frequency response curve to achieve noise suppression, thereby obtaining the third frequency response curve; identifying the noise frequency bands in the noise suppression process by comparing the third frequency response curve with the first frequency response curve, and performing targeted enhancement on the damage features of adjacent frequency bands of each noise frequency band; wherein, the enhancement intensity of the targeted enhancement is determined by learning the association pattern between noise features and damage features through the self-attention mechanism of Transformer.

[0040] First, the first frequency response curve obtained is subjected to adaptive notch filtering, wavelet threshold denoising, and frequency domain energy normalization in sequence to suppress the interference of power frequency interference (mainly 50Hz), random noise (such as electromagnetic pulse), and environmental noise intensity fluctuations.

[0041] Meanwhile, compared to traditional feature enhancement methods that only focus on the damage features themselves, this embodiment also uses the noise features extracted during noise suppression to guide subsequent feature enhancement. By directionally enhancing the damage features in adjacent frequency bands through noise, it is possible to capture weak damage signals that may be overlooked. Specifically, the process is as follows: First, the third frequency response curve after noise suppression is compared with the original first frequency response curve, and the frequency domain difference is calculated: ;if If the value is higher than the frequency domain threshold, then the frequency is determined. The frequency band in question is the noise band, and the set of noise bands is denoted as . .in, It is an identifier for the noise frequency band.

[0042] Then, for each noise band Construct noise feature vectors Includes the center frequency value of the frequency band, bandwidth, and difference curve. Information such as the average amplitude, i.e. All noise feature vectors constitute the noise feature matrix. .

[0043] Candidate frequency bands for damage characteristics were determined from damage-sensitive frequency bands marked in historical data (e.g., 3kHz for cracks, 500Hz for wear) and significant resonant peaks identified by frequency domain gradient analysis. For each candidate frequency band for damage characteristics... (in, (Identifiers for candidate frequency bands of damage features) are used to construct candidate vectors of damage features according to the encoding method of "frequency point location; health status amplitude; historical damage sensitivity". All damage feature candidate vectors constitute the damage feature candidate matrix. .

[0044] Next, the noise feature matrix As the query matrix Q and the key matrix K, the candidate matrix for damage features V is the value matrix. For the h-th head, the attention score is calculated. The attention weight matrix is ​​obtained by using the softmax function. , then calculate The results of H heads are concatenated and subjected to a linear transformation to obtain the multi-head self-attention output. .in, These are the projections of Q, K, and V onto the h-th head, respectively. The dimension of the key vector.

[0045] The correlation matrix between noise features and damage features is calculated using multi-head self-attention output. ,For example, , The weight of the h-th head can be learned through training. For the attention weight matrix of the h-th head, the element in the i-th row and j-th column is... This represents the correlation strength between the i-th noise frequency band and the j-th damage feature frequency band.

[0046] According to the correlation matrix Generate Enhanced Mask For example, through formula It is concluded that For activation functions, such as the Sigmoid function, This is the weight matrix. This is the bias vector.

[0047] The enhancement strength can be determined based on the correlation strength, for example, by using a piecewise function: Finally, based on the enhanced mask The corresponding element For the third frequency response curve The amplitude of this frequency band is adjusted. The specific calculation method is as follows: .in It is in the third frequency response curve The amplitude of the frequency band, It is the second amplitude frequency response curve after enhancement processing The amplitude of the frequency band is determined. The amplitudes of other frequency bands remain unchanged, ultimately yielding the complete second amplitude-frequency response curve. .

[0048] In this way, the enhancement mask is integrated into the entire feature enhancement calculation process, enabling it to play a role in determining the frequency band enhancement amplitude.

[0049] Furthermore, the targeted enhancement of the impairment features of adjacent frequency bands of each noise frequency band includes: determining the amplitude and bandwidth of each noise frequency band, calculating the frequency band distance between each adjacent frequency band and the corresponding noise frequency band, calculating the enhancement coefficient based on the amplitude, the bandwidth and the frequency band distance; calculating the target enhancement intensity using the enhancement coefficient and the enhancement intensity corresponding to the adjacent frequency band, and using the target enhancement intensity to target the enhancement features of the impairment features of adjacent frequency bands of each noise frequency band.

[0050] The aforementioned embodiment derived the enhancement intensity for targeted enhancement of damage features in adjacent frequency bands through the Transformer's self-attention mechanism. Building upon this, this embodiment further considers the inherent characteristics (amplitude and bandwidth) of the noise frequency band itself and the distance relationship between adjacent frequency bands and the noise frequency band to determine a more suitable enhancement intensity, thereby highlighting damage features masked by noise and enabling more accurate rail damage monitoring. Specifically, after initial noise suppression processing, each identified noise frequency band is analyzed... Its amplitude was determined using methods such as spectrum analysis. The amplitude reflects the energy of the noise; at the same time, its bandwidth is determined. Bandwidth reflects the range of noise frequencies it covers. For example, by obtaining a frequency domain signal through Fourier transform, the maximum amplitude of the noise band can be measured in the frequency domain as the amplitude. Determine the start and end points of the noise frequency to calculate the bandwidth. .

[0051] For each noise frequency band Determine the adjacent frequency band range (a scaling factor multiplied by the bandwidth can be determined empirically or experimentally to represent the adjacent range). Within the adjacent frequency band range, for each candidate frequency band of the damage feature... Calculate its relationship with the noise frequency band. frequency band distance .

[0052] The calculation of the enhancement factor requires comprehensive consideration of amplitude, bandwidth, and frequency range. The following formula can be used: ;in, This is the enhancement factor. In the above formula, the noise amplitude... The larger the bandwidth, the stronger the interference with damage characteristics, and the more it needs to be enhanced; The larger the frequency, the wider the noise impact range, and the more dispersed the impact on individual frequency points; This reflects the influence of bandwidth distance; the closer the distance, the greater the enhancement factor. The enhancement factor, for example, falls within the range of [0, 10]. It is understandable that the amplitude, bandwidth, and bandwidth distance in the above formula can be normalized parameters.

[0053] The piecewise function described in the previous embodiments can be used to determine the initial enhancement intensity, which is applied to adjacent frequency bands at different frequency band distances. In this example, the enhancement coefficients corresponding to each adjacent frequency band are used for targeted adjustments, such as through multiplication operations, thereby achieving specific directional enhancement of the damage characteristics of each adjacent frequency band, which is beneficial for capturing weak damage signals that may be ignored.

[0054] Further, the step of extracting the feature matrix from the second frequency response curve includes: dividing the second frequency response curve into multiple frequency bands; for each frequency band, extracting feature parameters, including peak amplitude, center frequency, frequency band energy, and amplitude variance; selecting features sensitive to rail damage from the extracted feature parameters based on historical damage data; arranging the selected sensitive features in a preset order to construct a feature matrix; and normalizing the feature matrix.

[0055] This embodiment converts frequency domain signals into structured feature representations for subsequent identification, classification, and location of rail damage. Specifically: First, the second frequency response curve is divided into multiple non-overlapping frequency bands in the frequency domain. The width of each frequency band can be set according to signal characteristics and damage type (e.g., equal-width segmentation or logarithmic scaling). For each frequency band, the following feature is calculated: Peak Amplitude: The maximum amplitude within the frequency band, reflecting signal strength.

[0056] Center Frequency: The central location of energy distribution within a frequency band, reflecting the frequency shift of damage characteristics.

[0057] Band Energy: The total energy of a signal within a frequency band, reflecting the severity of the damage.

[0058] Amplitude variance: The degree of dispersion of amplitude within a frequency band, reflecting the stability of the signal.

[0059] Next, features strongly correlated with rail damage are selected from the extracted feature parameters based on historical damage data. This setup avoids interference from irrelevant features and enhances the sensitivity of the attention mechanism to early damage. The selected sensitive features are arranged in a preset order (e.g., by frequency range or feature type) to form a feature matrix. Each column of the feature matrix (i.e., each feature) is then subjected to min-max normalization to eliminate the influence of different feature units and numerical ranges.

[0060] It is understandable that the range of characteristic parameters extracted from the frequency band is determined based on domain knowledge. That is, based on knowledge of rail material mechanics, acoustic properties, etc., it is known in advance that there is a theoretical correlation between certain frequency ranges or characteristic parameters (such as energy of a specific frequency band, amplitude variation of a specific frequency) and damage types (such as cracks, wear).

[0061] Furthermore, domain knowledge can also be used to explain the causal relationship between characteristic parameters and damage mechanisms. For example, a shift in center frequency may reflect changes in structural stiffness, and changes in stiffness are directly related to crack propagation; an increase in amplitude variance may indicate enhanced non-stationarity of the signal, which is more pronounced during damage development.

[0062] Furthermore, based on historical damage data, features sensitive to rail damage are selected from the extracted feature parameters, including: calculating the first correlation coefficient between each feature parameter and the known damage type based on historical damage data, and selecting feature parameters whose absolute value of the correlation coefficient is greater than the first threshold; for the selected feature parameters, if the absolute value of the second correlation coefficient between any two feature parameters is greater than the second threshold, redundant feature parameters are retained, and feature parameters with clearer physical meaning or lower computational complexity are retained to obtain sensitive features sensitive to rail damage; wherein, the second threshold is greater than the first threshold.

[0063] This embodiment removes irrelevant and redundant features by calculating and comparing correlation coefficients, leaving only the key features that truly reflect rail damage, so as to construct a more effective feature matrix for subsequent damage monitoring. Specifically, the historical damage data consists of two parts: a feature parameter set: feature parameters extracted from historical monitoring signals (such as peak amplitude and center frequency of each frequency band); and damage labels: manually labeled damage types, i.e., damage labels (such as crack level, wear degree, etc., which need to be quantified into numerical data, such as a damage index of 0-10).

[0064] For each extracted feature parameter, a correlation analysis was performed with different known damage types (e.g., cracks, wear, spalling, etc.). The Pearson correlation coefficient method was used to calculate each feature parameter. With damage label First correlation coefficient: ;in, For the sample size, Features and damage labels The mean. The closer the absolute value is to 1, the stronger the characteristic parameter. With damage label The stronger the linear correlation.

[0065] The above calculations yield the first correlation coefficient between each feature parameter and various known damage types.

[0066] When the absolute value of the correlation coefficient between a certain feature parameter and any known damage type is greater than the first threshold, it indicates that the feature parameter has a strong correlation with the damage, and it is retained.

[0067] In addition, for the feature parameters initially selected above, a second correlation coefficient needs to be calculated between each pair of these feature parameters. For feature parameters whose absolute value of the second correlation coefficient is greater than the second threshold (e.g., the second threshold is 0.9 and the first threshold is 0.6), there is a strong correlation between them, and they are considered redundant features. Among these redundant features, feature parameters with clearer physical meaning or lower computational complexity are selected for retention. For example, if two feature parameters are highly correlated, but one of them is physically easier to explain the rail damage mechanism (e.g., closely related to the principles of material mechanics), that feature parameter is retained; or if a feature parameter is simpler to calculate and can improve the efficiency of subsequent processing, it is also selected for retention.

[0068] All the feature parameters obtained after the first two steps of screening are identified as sensitive features that are sensitive to rail damage. These features will be used to construct the feature matrix in the subsequent process, providing key information for rail damage monitoring.

[0069] Similar to the first correlation coefficient, the second correlation coefficient can also be calculated using the Pearson correlation coefficient method, which will not be elaborated further.

[0070] like Figure 2As shown, this embodiment of the invention also provides a rail damage monitoring system 100 based on electronic fingerprinting. The system includes an acquisition module 11, a preprocessing module 12, and a damage monitoring module 13 connected in sequence. The acquisition module 11 generates a first amplitude-frequency response curve of the rail based on an electrical signal acquired by a measuring end located at one end of the rail. The electrical signal is acquired after a broadband sweep signal is emitted by an excitation end located at the other end of the rail. The preprocessing module 12 performs noise suppression and feature enhancement processing on the first amplitude-frequency response curve to obtain a second amplitude-frequency response curve. A feature matrix is ​​extracted from the second amplitude-frequency response curve, and this feature matrix is ​​used as the electronic fingerprint of the rail segment. The noise suppression and feature enhancement are implemented using a Transformer-based linkage mechanism. The damage monitoring module 13 calculates the feature difference degree between the electronic fingerprint and a preset electronic fingerprint, and calculates a damage index based on the feature difference degree. The feature difference degree is calculated based on a combination of frequency difference degree and amplitude difference degree.

[0071] Further, the preprocessing module 12 is specifically used to: sequentially perform adaptive notch filtering, wavelet threshold denoising, and frequency domain energy normalization on the first frequency response curve to achieve noise suppression, thereby obtaining the third frequency response curve; identify the noise frequency bands in the noise suppression process by comparing the third frequency response curve with the first frequency response curve, and perform targeted enhancement on the damage features of adjacent frequency bands of each noise frequency band; wherein, the enhancement intensity of the targeted enhancement is determined by learning the association pattern between noise features and damage features through the self-attention mechanism of Transformer.

[0072] Further, the preprocessing module 12 is specifically used for: determining the amplitude and bandwidth of each noise frequency band, calculating the frequency band distance between each adjacent frequency band and the corresponding noise frequency band, calculating the enhancement coefficient based on the amplitude, the bandwidth and the frequency band distance; calculating the target enhancement intensity using the enhancement coefficient and the enhancement intensity corresponding to the adjacent frequency band, and using the target enhancement intensity to perform targeted enhancement of the damage characteristics of the adjacent frequency bands of each noise frequency band.

[0073] Furthermore, the preprocessing module 12 is specifically used for: dividing the second amplitude response curve into multiple frequency band intervals; for each frequency band interval, extracting feature parameters, including peak amplitude, center frequency, frequency band energy, and amplitude variance; selecting features sensitive to rail damage from the extracted feature parameters based on historical damage data; arranging the selected sensitive features in a preset order to construct a feature matrix; and normalizing the feature matrix.

[0074] Further, the preprocessing module 12 is specifically used for: calculating the first correlation coefficient between each feature parameter and the known damage type based on historical damage data, and selecting feature parameters whose absolute value of the correlation coefficient is greater than the first threshold; for the selected feature parameters, if the absolute value of the second correlation coefficient between any two feature parameters is greater than the second threshold, the redundant feature parameters with clearer physical meaning or lower computational complexity are retained to obtain sensitive features that are sensitive to rail damage; wherein, the second threshold is greater than the first threshold.

[0075] like Figure 3 As shown, embodiments of the present invention also provide an electronic device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program, when executed by the processor, implements the method as described in any of the foregoing embodiments.

[0076] This invention also provides a computer storage medium storing a computer program that can be executed by a processor to implement the methods described in any of the foregoing claims.

[0077] This invention also provides a computer program product comprising a computer program that can be executed by a processor to implement the method as described in any of the foregoing embodiments.

[0078] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A rail damage monitoring method based on electronic fingerprinting, characterized in that: The method includes the following steps: First, a first amplitude-frequency response curve of the rail is generated based on an electrical signal acquired at a measuring end located at one end of the rail. This electrical signal is acquired after a broadband sweep signal is emitted from an excitation end located at the other end of the rail. Noise suppression and feature enhancement are performed on the first amplitude-frequency response curve to obtain a second amplitude-frequency response curve. A feature matrix is ​​extracted from the second amplitude-frequency response curve and used as the electronic fingerprint of the rail. Noise suppression and feature enhancement are implemented using a Transformer-based linkage mechanism. The feature difference between this electronic fingerprint and a preset electronic fingerprint is calculated, and a damage index is calculated based on the feature difference. The feature difference is calculated based on a combination of frequency difference and amplitude difference. The noise suppression and feature enhancement processing of the first amplitude-frequency response curve to obtain the second amplitude-frequency response curve includes: sequentially applying adaptive notch filtering to the first amplitude-frequency response curve. Noise suppression is achieved through filtering, wavelet thresholding denoising, and frequency domain energy normalization, resulting in a third amplitude frequency response curve. The noise frequency bands in the noise suppression process are identified by comparing the third amplitude frequency response curve with the first amplitude frequency response curve. The impairment features of adjacent frequency bands of each noise frequency band are then targeted for enhancement. The enhancement intensity of this targeted enhancement is determined by learning the association pattern between noise features and impairment features through the Transformer's self-attention mechanism. The targeted enhancement of the impairment features of adjacent frequency bands of each noise frequency band includes: determining the amplitude and bandwidth of each noise frequency band; calculating the band distance between each adjacent frequency band and its corresponding noise frequency band; calculating the enhancement coefficient based on the amplitude, bandwidth, and band distance; calculating the target enhancement intensity using the enhancement coefficient and the enhancement intensity corresponding to the adjacent frequency bands; and using the target enhancement intensity to target the impairment features of adjacent frequency bands of each noise frequency band.

2. The rail damage monitoring method based on electronic fingerprinting according to claim 1, characterized in that: The feature matrix is ​​extracted from the second frequency response curve, including: dividing the second frequency response curve into multiple frequency bands; for each frequency band, extracting feature parameters, including peak amplitude, center frequency, frequency band energy, and amplitude variance; selecting features sensitive to rail damage from the extracted feature parameters based on historical damage data; arranging the selected sensitive features in a preset order to construct a feature matrix; and normalizing the feature matrix.

3. The rail damage monitoring method based on electronic fingerprinting according to claim 2, characterized in that: Based on historical damage data, features sensitive to rail damage are selected from the extracted feature parameters. This includes: calculating the first correlation coefficient between each feature parameter and the known damage type based on historical damage data, and selecting feature parameters whose absolute value of the correlation coefficient is greater than the first threshold; for the selected feature parameters, if the absolute value of the second correlation coefficient between any two feature parameters is greater than the second threshold, redundant feature parameters are retained, and feature parameters with clearer physical meaning or lower computational complexity are retained to obtain sensitive features sensitive to rail damage; wherein, the second threshold is greater than the first threshold.

4. A rail damage monitoring system based on electronic fingerprinting, characterized in that, The system includes a data acquisition module, a preprocessing module, and a damage monitoring module, which are electrically connected in sequence. The data acquisition module generates a first amplitude-frequency response curve for the rail based on an electrical signal acquired at a measuring end located at one end of the rail. This electrical signal is acquired after a broadband sweep signal is emitted from an excitation end located at the other end of the rail. The preprocessing module performs noise suppression and feature enhancement on the first amplitude-frequency response curve to obtain a second amplitude-frequency response curve. A feature matrix is ​​extracted from the second amplitude-frequency response curve and used as the electronic fingerprint of the rail. Noise suppression and feature enhancement are implemented using a Transformer-based linkage mechanism. The damage monitoring module calculates the feature difference between this electronic fingerprint and a preset electronic fingerprint, and calculates a damage index based on the feature difference. The feature difference is calculated based on a combination of frequency difference and amplitude difference. The preprocessing module is specifically used to: sequentially perform adaptive notch filtering, wavelet threshold denoising, and frequency domain energy normalization on the first frequency response curve to achieve noise suppression, thereby obtaining the third frequency response curve; identify the noise frequency bands in the noise suppression process by comparing the third frequency response curve with the first frequency response curve, and perform targeted enhancement on the damage features of adjacent frequency bands of each noise frequency band; wherein, the enhancement intensity of the targeted enhancement is determined by learning the association pattern between noise features and damage features through the self-attention mechanism of Transformer; The preprocessing module is specifically used for: determining the amplitude and bandwidth of each noise frequency band, calculating the frequency band distance between each adjacent frequency band and the corresponding noise frequency band, calculating the enhancement coefficient based on the amplitude, the bandwidth and the frequency band distance; calculating the target enhancement intensity using the enhancement coefficient and the enhancement intensity corresponding to the adjacent frequency band, and using the target enhancement intensity to perform targeted enhancement of the damage characteristics of the adjacent frequency bands of each noise frequency band.

5. An electronic device, characterized in that: The electronic device includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program, when executed by the processor, implements the method as described in any one of claims 1-3.

6. A computer storage medium, characterized in that: The computer storage medium stores a computer program that can be executed by a processor to implement the method as described in any one of claims 1-3.

7. A computer program product, characterized in that: The computer program product includes a computer program that can be executed by a processor to implement the method as described in any one of claims 1-3.

Images