Ultrasonic guided wave based rail damage location method and apparatus
By preprocessing the ultrasonic guided wave signal and performing dual-stream time-frequency network analysis, combined with time-domain layered sampling and cross-correlation calculation, the problem of insufficient accuracy in long-distance rail damage location was solved, and higher accuracy damage location detection was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA RAILWAY SIYUAN SURVEY & DESIGN GRP CO LTD
- Filing Date
- 2025-07-16
- Publication Date
- 2026-07-14
Smart Images
Figure CN120870362B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of rail technology, and in particular to a method and apparatus for locating rail damage based on ultrasonic guided waves. Background Technology
[0002] During long-term use, steel rails are affected by various factors such as train loads, environmental corrosion, and fatigue wear, which may lead to internal defects such as cracks and cross-sectional damage. If these defects are not detected and addressed in time, they may cause serious safety accidents, such as train derailments and rail breakage, resulting in huge losses to railway transportation.
[0003] Related technologies include rail damage localization methods such as the pulse-echo method, which involves emitting ultrasonic guided waves into the rail and analyzing the echoes of the ultrasonic guided waves to locate rail damage.
[0004] However, ultrasonic guided wave signals in long-distance rails (such as rails longer than 1 km) exhibit dispersion effects. These dispersion effects cause wave packet broadening in the echoes of ultrasonic guided waves, which in turn leads to insufficient accuracy in determining the location of rail damage based on the analysis of ultrasonic echoes. Summary of the Invention
[0005] This disclosure provides a method and apparatus for locating rail damage based on ultrasonic guided waves, which can improve the accuracy of rail damage location. The technical solution includes at least the following:
[0006] In a first aspect, a method for locating rail damage based on ultrasonic guided waves is provided, comprising: sending an excitation signal from a first end of a first rail to a second end of the rail, and receiving an echo signal of the excitation signal at the first end, wherein the first end and the second end are two different endpoints of the first rail, and the first rail is a segment of the rail to be located for damage; preprocessing the echo signal, the preprocessing including Hilbert transform processing and noise reduction processing; processing the preprocessed echo signal using a dual-stream time-frequency network to obtain a first damage location; determining a first damage range in the preprocessed echo signal based on the first damage location; performing time-domain layered sampling on the first damage range to obtain multi-channel time-domain data; performing cross-correlation operation on the multi-channel time-domain data to obtain a cross-correlation sequence, and determining a second damage location based on the cross-correlation sequence, wherein the second damage location is the location where damage exists in the first rail.
[0007] Optionally, the step of performing cross-correlation operation on the multi-channel time-domain data to obtain cross-correlation sequences includes: performing cross-correlation operation on the multi-channel time-domain data and the reference data pairwise by channel to obtain multiple cross-correlation sequences. The reference data is obtained by performing time-domain layered sampling on the reference signal. The reference signal is the pre-processed echo signal of the first rail in a healthy state. The reference data and the multi-channel time-domain data have the same number of channels and sampling interval.
[0008] Optionally, in the multi-channel time-domain data, the first... The time-domain data of the first channel and the reference data of the second channel The cross-correlation sequence of the time-domain data of each channel is calculated using the following formula:
[0009]
[0010] in, Indicates the first channel of the multi-channel time-domain data The time-domain data of the first channel and the reference data of the second channel Cross-correlation sequences of time-domain data from each channel, and The values are the same. For time delay, The number of sampling points. Indicates the first channel of the multi-channel time-domain data The time domain data of the first channel One sampling point, Indicates the first in the reference data The time domain data of the first channel One sampling point.
[0011] Optionally, the step of performing time-domain layered sampling on the first damage range to obtain multi-channel time-domain data includes: dividing the first damage range according to a preset time window length and step size to obtain a time-domain data sequence, wherein the time-domain data sequence includes time-domain data of multiple channels, and the time window length of the time-domain data of each channel is the same; if the zero-crossing rate of the first time-domain data is greater than the zero-crossing rate threshold, or if there is an energy mutation point in the first time-domain data, reducing the time window length of the first time-domain data to obtain the multi-channel time-domain data, wherein the first time-domain data is the time-domain data of any one channel in the time-domain data sequence.
[0012] Optionally, the dual-stream time-frequency network includes a time-stream branch, a frequency-stream branch, a cross-modal attention module, a feature fusion module, and a convolutional layer; the time-stream branch is used to extract time features based on bidirectional LSTM, and the frequency-stream branch is used to extract spectral features based on continuous wavelet transform and ResNet-18; the cross-modal attention module is used to perform cross-modal attention operations on the time features and the spectral features to obtain spectral attention features; the feature fusion module is used to concatenate the time features and the spectral attention features to obtain fused features; and the convolutional layer is used to output the first damage location based on the fused features.
[0013] Optionally, the preprocessing of the echo signal includes: solving for the optimal parameters of the variational mode decomposition algorithm based on the particle swarm optimization algorithm, and performing noise reduction processing on the echo signal using the variational mode decomposition algorithm based on the optimal parameters.
[0014] Secondly, an ultrasonic guided wave-based rail damage location device is also provided, comprising: an excitation module for sending an excitation signal from a first end of a first rail to a second end of the rail, and receiving an echo signal of the excitation signal at the first end, wherein the first end and the second end are two different endpoints of the first rail, and the first rail is a section of the rail to be damaged; a preprocessing module for preprocessing the echo signal, the preprocessing including Hilbert transform processing and noise reduction processing; a first damage location module for processing the preprocessed echo signal using a dual-stream time-frequency network to obtain a first damage location; a damage range determination module for determining a first damage range in the preprocessed echo signal based on the first damage location; a layered sampling module for performing time-domain layered sampling on the first damage range to obtain multi-channel time-domain data; and a second damage location module for performing cross-correlation operations on the multi-channel time-domain data to obtain a cross-correlation sequence, and determining a second damage location based on the cross-correlation sequence, wherein the second damage location is the location where damage exists in the first rail.
[0015] Optionally, the second damage localization module is further configured to perform cross-correlation operations on the multi-channel time-domain data and the reference data in pairs according to the channels to obtain multiple cross-correlation sequences. The reference data is obtained by performing time-domain layered sampling on the reference signal. The reference signal is the pre-processed echo signal of the first rail in a healthy state. The reference data and the multi-channel time-domain data have the same number of channels and sampling interval.
[0016] Optionally, in the second damage localization module, the first of the multi-channel time-domain data... The time-domain data of the first channel and the reference data of the second channel The cross-correlation sequence of the time-domain data of each channel is calculated using the following formula:
[0017]
[0018] in, Indicates the first channel of the multi-channel time-domain data The time-domain data of the first channel and the reference data of the second channel Cross-correlation sequences of time-domain data from each channel, and The values are the same. For time delay, The number of sampling points. Indicates the first channel of the multi-channel time-domain data The time domain data of the first channel One sampling point, Indicates the first in the reference data The time domain data of the first channel One sampling point.
[0019] Optionally, the hierarchical sampling module is further configured to divide the first damage range according to a preset time window length and step size to obtain a time-domain data sequence, wherein the time-domain data sequence includes time-domain data of multiple channels, and the time window length of the time-domain data of each channel is the same; if the zero-crossing rate of the first time-domain data is greater than the zero-crossing rate threshold, or if there is an energy mutation point in the first time-domain data, the time window length of the first time-domain data is reduced to obtain the multi-channel time-domain data, wherein the first time-domain data is the time-domain data of any one channel in the time-domain data sequence.
[0020] Optionally, in the first damage localization module, the dual-stream time-frequency network includes a time-stream branch, a frequency-stream branch, a cross-modal attention module, a feature fusion module, and a convolutional layer; the time-stream branch is used to extract time features based on bidirectional LSTM, and the frequency-stream branch is used to extract spectral features based on continuous wavelet transform and ResNet-18; the cross-modal attention module is used to perform cross-modal attention operations on the time features and the spectral features to obtain spectral attention features; the feature fusion module is used to concatenate the time features and the spectral attention features to obtain fused features; and the convolutional layer is used to output the first damage location based on the fused features.
[0021] Optionally, the preprocessing module is further configured to solve for the optimal parameters of the variational mode decomposition algorithm based on the particle swarm optimization algorithm, and to perform noise reduction processing on the echo signal using the variational mode decomposition algorithm based on the optimal parameters.
[0022] Thirdly, a computer device is also provided, comprising: a memory and a processor, wherein the memory stores at least one computer program, the at least one computer program being loaded and executed by the processor to perform the rail damage localization method based on ultrasonic guided waves described in the above embodiments.
[0023] Fourthly, a computer-readable storage medium is also provided, wherein at least one computer program is stored in the computer-readable storage medium, the at least one computer program being loaded and executed by a processor to perform the rail damage localization method based on ultrasonic guided waves described in the above embodiments.
[0024] Fifthly, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the method described in the first aspect.
[0025] The beneficial effects of the technical solutions provided in this disclosure include at least the following:
[0026] In this embodiment, after sending an excitation signal from the first end of the first rail to the second end, the echo signal of the excitation signal is received at the first end. The echo signal is preprocessed, and a dual-stream time-frequency network is used to process the preprocessed echo signal to obtain the first damage location. Due to the dispersion effect of ultrasonic guided wave signals in long-distance rails, the echo of the ultrasonic guided wave exhibits wave packet broadening, which leads to insufficient accuracy of the rail damage location (i.e., the first damage location) obtained from the analysis of the ultrasonic echo. Although the first damage location has insufficient accuracy, it still has a certain degree of reliability, and the accurate rail damage location can be considered to be near the first damage location. Based on the first damage location, the range of the rail damage location (i.e., the first damage range) can be roughly located from the echo signal. Then, the first damage range is subjected to time-domain layered sampling to obtain multi-channel time-domain data. Cross-correlation operation is then performed on the multi-channel time-domain data to obtain a cross-correlation sequence, and the second damage location is determined based on the cross-correlation sequence. The second damage location obtained by time-domain layering and cross-correlation calculation has high accuracy, thus achieving more accurate rail damage location. Attached Figure Description
[0027] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0028] Figure 1A flowchart illustrating a rail damage localization method based on ultrasonic guided waves provided in an exemplary embodiment of this disclosure is shown.
[0029] Figure 2 This is a schematic diagram of the signal amplification result;
[0030] Figure 3 This is a schematic diagram showing the comparison between the detected signal and the reference signal;
[0031] Figure 4 A flowchart of a rail damage localization method based on ultrasonic guided waves provided in another exemplary embodiment of this disclosure is shown;
[0032] Figure 5 This is a schematic diagram of the structure of a two-stream time-frequency attention network;
[0033] Figure 6 A schematic diagram of the structure of a rail damage location device based on ultrasonic guided waves provided in an exemplary embodiment of this disclosure is shown.
[0034] Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this disclosure. Detailed Implementation
[0035] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this disclosure pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The terms “connected” or “linked” and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.
[0036] To make the objectives, technical solutions, and advantages of this disclosure clearer, the embodiments of this disclosure will be described in further detail below with reference to the accompanying drawings.
[0037] Figure 1 A flowchart illustrating an exemplary embodiment of this disclosure provides a rail damage localization method based on ultrasonic guided waves, which can be executed by a computer device. See also Figure 1 The method includes:
[0038] In step 101, after sending an excitation signal from the first end of the first rail to the second end of the rail, the echo signal of the excitation signal is received at the first end.
[0039] The first end and the second end are two different endpoints of the first rail, which is a section of the rail to be damaged and located.
[0040] In practical research, the rails to be damaged are often quite long. Because ultrasonic guided waves have distance limitations and attenuation, a single device cannot locate the damage across the entire rail. Therefore, the rail needs to be divided into multiple inspection areas, with each area using a separate set of devices for damage location detection, thus achieving comprehensive rail management. Here, the first rail represents one inspection area. All other inspection areas within the rail to be damaged can be processed using the same method as the first rail to achieve rail damage location.
[0041] In implementation, an ultrasonic transmitter and an ultrasonic receiver can be installed at the first end. The ultrasonic transmitter is used to send an excitation signal from the first end of the first rail to the second end of the rail, and the ultrasonic receiver is used to receive the echo signal of the excitation signal at the first end.
[0042] Ideally, the excitation signal would propagate along the rail, but in reality, due to the presence of welds and damage in the rail, the ultrasonic receiving device will inevitably be able to receive the echo signal of the excitation signal.
[0043] In this embodiment, Barker code is used as a pseudo-random sequence code, and BPSK (Binary Phase Shift Keying) technology is employed to encode and modulate the carrier signal to obtain the excitation signal, thereby achieving spread spectrum processing of the carrier signal. Correspondingly, the echo signal also needs to be despread before analysis.
[0044] The pseudo-random spreading code uses a 13-bit Barker code. Multiplying the signal code and the pseudo-random code yields a composite code, at which point the frequency is amplified to 13 times the original carrier signal frequency. Since the signal code is "1", the composite code is identical to the spreading code. The spread sequence is then used as the modulation code and modulated with the 200 kHz carrier signal. Modulation yields a carrier signal, i.e., an excitation signal, modulated by a spread spectrum sequence.
[0045] Before transmitting the excitation signal, it needs to be amplified. Optionally, a three-stage power amplifier architecture can be used to achieve signal amplification.
[0046] In a three-stage power amplifier architecture, the first stage uses an instrumentation amplifier (such as the AD620 instrumentation amplifier) as the preamplifier voltage amplifier unit, with an input impedance greater than [missing information]. Common-mode rejection ratio greater than By adjusting the gain resistor Achieving a 20x fixed voltage gain, boosting the signal to The second stage employs a push-pull emitter follower topology (such as a TIP41C / TIP42C complementary transistor pair) to eliminate crossover distortion and match the impedances of the preceding and following stages, reducing signal transmission loss. The third stage is a Class B power amplifier circuit built based on discrete components, stabilizing the output impedance through a negative feedback network. (Matching transducer impedance), final output peak voltage A high-voltage excitation signal with a peak current greater than 2A. This design achieves... Total gain, DAU 100 W instantaneous output power and 1kHz–100kHz ( The wide frequency response characteristics of the guided wave can increase the effective propagation distance of the guided wave in the rail by 40% and significantly improve the signal-to-noise ratio, thus solving the bottleneck of conversion from low voltage digital signal to high voltage power output.
[0047] Figure 2 This is a schematic diagram of the signal amplification result, such as... Figure 2 As shown, the excitation signal is amplified after signal enhancement. The amplitude and other parameters of the amplified signal are the same as those of the excitation signal before signal enhancement.
[0048] In step 102, the echo signal is preprocessed.
[0049] Preprocessing includes Hilbert transform and noise reduction.
[0050] First, the echo signal is subjected to Hilbert transform using formula (1), and then the upper envelope of the echo signal is extracted.
[0051] (1)
[0052] In formula (1), For echo signal, The echo signal after Hilbert transform. For integration variables, This is the normalization coefficient.
[0053] Will and Represented as a complex signal Afterwards, reply signal The amplitude is the upper envelope of the echo signal. The upper envelope of the echo signal is calculated using formula (2).
[0054] (2)
[0055] In formula (2), The upper envelope of the echo signal is given. The meanings of the other parameters in formula (2) are the same as those in formula (1), and will not be detailed here.
[0056] Optionally, the optimal parameters of the variational mode decomposition algorithm are solved based on the particle swarm optimization algorithm, and the variational mode decomposition algorithm is used to denoise the echo signal based on the optimal parameters.
[0057] When traditional variational mode decomposition (VMD) algorithms are applied to ultrasonic guided wave signal processing for rails, the key parameters rely on empirical settings, and fixed parameter combinations are prone to causing distortion of useful signals when operating conditions fluctuate.
[0058] In this embodiment of the disclosure, the optimal parameters of the variational mode decomposition algorithm are solved based on the particle swarm optimization algorithm. By introducing the particle swarm optimization algorithm, the multi-parameter optimization of the variational mode decomposition method is realized.
[0059] When using the particle swarm optimization algorithm to solve for the optimal parameters of the variational mode decomposition algorithm, the minimum envelope entropy is used as the fitness function, and the optimal values of the penalty factor and the number of decomposition layers in the VMD algorithm are solved iteratively through the particle swarm optimization algorithm.
[0060] Entropy is the probability of a certain type of information appearing, and envelope entropy reflects the sparsity of a signal. If the echo signal contains a large amount of defective signal components, the envelope entropy value is smaller; conversely, if the echo signal contains a large amount of noise components and the defective signal characteristics are not obvious, the envelope entropy value is larger. By minimizing the envelope entropy value as the fitness function, the optimal VMD parameters for denoising echo signals can be determined.
[0061] Optionally, the envelope entropy is expressed by formula (3).
[0062] (3)
[0063] In formula (3), Let N be the envelope entropy. By discretizing the envelope of the echo signal, the envelope can be divided into N segments. Let j be the j-th segment of the echo signal envelope. for The corresponding probability distribution sequence.
[0064] There are many implementation methods for the particle swarm optimization algorithm, which will not be detailed here.
[0065] For example, in the optimal parameters of the variational mode decomposition algorithm, the number of decomposition layers is... Take 6 as the penalty factor. Take 8162.
[0066] After obtaining the optimal parameters, these parameters can be substituted into the VMD algorithm, and the VMD algorithm can be used to denoise the echo signal, thereby achieving the denoising of the echo signal.
[0067] The preprocessed echo signal involved in subsequent steps is the upper envelope of the denoised echo signal, hereinafter referred to as the detection signal.
[0068] In the healthy state of the first rail, an excitation signal is transmitted and the echo signal of the excitation signal is received using the methods described in steps 101 to 102. After preprocessing the echo signal, a reference signal can be obtained. That is, the reference signal is the preprocessed echo signal of the first rail in the healthy state.
[0069] At locations where rail damage exists, there is a significant difference in waveform between the detection signal and the reference signal. Figure 3 This is a schematic diagram comparing the detected signal and the reference signal. For example... Figure 3 As shown, the part of the waveform that differs between the detected signal and the reference signal is the waveform corresponding to rail damage.
[0070] In step 103, a dual-stream time-frequency network is used to process the preprocessed echo signal to obtain the first damage location.
[0071] In step 104, the first damage range is determined in the preprocessed echo signal based on the first damage location.
[0072] Because of the dispersion effect of ultrasonic guided wave signals in long-distance rails (such as rails longer than 1km), the echo of ultrasonic guided waves has a wave packet broadening phenomenon, which in turn leads to insufficient accuracy of the rail damage location (i.e. the first damage location) obtained by analyzing the ultrasonic echo.
[0073] Although the first damage location has insufficient precision, it also possesses a degree of reliability. It can be assumed that the accurate location of the rail damage lies near this first damage location. Based on this first damage location, the approximate area of the rail damage (i.e., the first damage range) can be determined.
[0074] In step 105, time-domain layered sampling is performed on the first damage area to obtain multi-channel time-domain data.
[0075] Although the accuracy of rail damage location calculated after time-domain hierarchical sampling is high, the calculation is complex and highly uncertain, making it suitable only for calculations within a small range. Therefore, in this embodiment, the first damage range is determined through steps 101 to 104, and then a more refined calculation of the rail damage location is performed based on steps 105 to 106, resulting in a second damage location with higher accuracy.
[0076] In step 106, cross-correlation is performed on the multi-channel time-domain data to obtain a cross-correlation sequence, and the second damage location is determined based on the cross-correlation sequence.
[0077] The second damage location is the location where damage exists in the first rail.
[0078] In this embodiment, after sending an excitation signal from the first end of the first rail to the second end, the echo signal of the excitation signal is received at the first end. The echo signal is preprocessed, and a dual-stream time-frequency network is used to process the preprocessed echo signal to obtain the first damage location. Due to the dispersion effect of ultrasonic guided wave signals in long-distance rails, the echo of the ultrasonic guided wave exhibits wave packet broadening, which leads to insufficient accuracy of the rail damage location (i.e., the first damage location) obtained from the analysis of the ultrasonic echo. Although the first damage location has insufficient accuracy, it still has a certain degree of reliability, and the accurate rail damage location can be considered to be near the first damage location. Based on the first damage location, the range of the rail damage location (i.e., the first damage range) can be roughly located from the echo signal. Then, the first damage range is subjected to time-domain layered sampling to obtain multi-channel time-domain data. Cross-correlation operation is then performed on the multi-channel time-domain data to obtain a cross-correlation sequence, and the second damage location is determined based on the cross-correlation sequence. The second damage location obtained by time-domain layering and cross-correlation calculation has high accuracy, thus achieving more accurate rail damage location.
[0079] Figure 4 A flowchart illustrating a rail damage localization method based on ultrasonic guided waves, provided in another exemplary embodiment of this disclosure, is shown. This method can be executed by a computer device. See also Figure 4 The method includes:
[0080] In step 401, after sending an excitation signal from the first end of the first rail to the second end of the rail, the echo signal of the excitation signal is received at the first end.
[0081] The first end and the second end are two different endpoints of the first rail, which is a section of the rail to be damaged and located.
[0082] In step 402, the echo signal is preprocessed.
[0083] Preprocessing includes Hilbert transform and noise reduction.
[0084] The relevant content of steps 401 to 402 is the same as that of steps 101 to 102 mentioned above, and will not be described in detail here.
[0085] In step 403, a dual-stream time-frequency network is used to process the preprocessed echo signal to obtain the first damage location.
[0086] By analyzing the preprocessed echo signal using a Twin-Stream Temporal-Frequency Attention Network (TS-TFANet), the first damage location can be obtained, which is a preliminary damage location.
[0087] Figure 5 This is a schematic diagram of the structure of a two-stream time-frequency attention network. The following section combines... Figure 5 The structure of the dual-stream time-frequency attention network is explained.
[0088] Optionally, the dual-stream time-frequency network includes a temporal branch, a frequency branch, a feature fusion module, and convolutional layers. The outputs of the temporal branch and the frequency branch are respectively connected to the input of the cross-modal attention module, and the output of the cross-modal attention module, the feature fusion module, and the convolutional layers are connected sequentially.
[0089] The temporal branch is used to extract temporal features based on bidirectional LSTM (Bi-directional Long Short-Term Memory) network.
[0090] Optionally, the temporal branch includes dilated convolutional layers and bidirectional LSTM. The temporal branch can capture multi-scale temporal features and long-range dependencies of the detected signal.
[0091] The dilated convolutional layer consists of N cascaded 1D convolutional layers. For example, N is 4, and the dilation rates of these four cascaded convolutional layers are set to 1, 3, 9, and 27, respectively. The dilated convolutional layer can expand the receptive field without increasing the number of parameters, which helps bidirectional LSTM to more accurately capture long-range patterns when extracting temporal features.
[0092] After the detection signal is input to the time flow branch, it is dilated by a dilated convolutional layer, and then the dilated detection signal is input to a bidirectional LSTM to obtain the time features output by the time flow branch.
[0093] Frequency flow branch is used to extract spectral features based on continuous wavelet transform and ResNet-18.
[0094] For example, the wavelet basis functions used in the continuous wavelet transform are, for instance, Morlet wavelets. After the detection signal undergoes the continuous wavelet transform, it is converted into a two-dimensional time-frequency spectrum. This time-frequency spectrum is then input into ResNet-18 to obtain the spectral characteristics of the frequency stream branch output.
[0095] The cross-modal attention module is used to perform cross-modal attention operations on temporal and spectral features to obtain spectral attention features.
[0096] When performing cross-modal attention operations, the temporal features and spectral features are first projected onto the same dimension. Then, the temporal features output by the temporal stream branch are used as the query, and the spectral features output by the frequency stream branch are used as the key and value for cross-modal attention operations, thereby obtaining the spectral attention features.
[0097] By employing a cross-modal attention mechanism, the time-frequency regions related to impairment in the spectral features can be enhanced, thereby obtaining spectral attention features.
[0098] The feature fusion module is used to concatenate temporal and spectral features to obtain fused features. For example, this concatenation can be done along the channel dimension.
[0099] The convolutional layer is used to output the first damage location based on the fused features.
[0100] In implementation, the output of the convolutional layer includes a first probability distribution corresponding to the end face echo and a second probability distribution corresponding to the damage echo.
[0101] Before using this dual-stream time-frequency attention network, it is necessary to train the network specifically for railway damage identification tasks, so that the network can output the first probability distribution corresponding to the end face echo and the second probability distribution corresponding to the damage echo.
[0102] Here, both the first and second probability distributions refer to the entire detection signal. For each point on the detection signal, there exists a corresponding probability in both the first and second probability distributions. The maximum probability in the first probability distribution is the point corresponding to the end face in the detection signal; the maximum probability in the second probability distribution is the point corresponding to the damage in the detection signal.
[0103] Based on the maximum probability in the first probability distribution, the maximum probability in the second probability distribution, and the length of the first rail (the distance between the first end and the second end), the first damage location can be calculated using formula (4).
[0104] (4)
[0105] In formula (4), The distance between the first damaged location and the first end in the first rail. This can be indirectly considered as the location of the first injury. This represents the time point in the detected signal corresponding to the maximum probability value in the second probability distribution. This represents the time point in the detected signal corresponding to the maximum probability value in the first probability distribution. This is the length of the first rail.
[0106] In step 404, the first damage range is determined in the preprocessed echo signal based on the first damage location.
[0107] Because of the dispersion effect of ultrasonic guided wave signals in long-distance rails (such as rails longer than 1km), the echo of ultrasonic guided waves has a wave packet broadening phenomenon, which in turn leads to insufficient accuracy of the rail damage location (i.e. the first damage location) obtained by analyzing the ultrasonic echo.
[0108] Although the first damage location has insufficient precision, it also possesses a degree of reliability. It can be assumed that the accurate location of the rail damage lies near this first damage location. Based on this first damage location, the approximate area of the rail damage (i.e., the first damage range) can be determined.
[0109] When determining the first damage range, the location M meters before and after the first damage location can be taken as the center as the location range corresponding to the first damage range. M is a preset error range size. For example, the value of M ranges from 5 to 20, such as 5, 40, 20, etc.
[0110] The two variables in the preprocessed echo signal are amplitude and time. However, the location range corresponding to the first damage range is actually a distance range. Therefore, the echo signal needs to be converted to determine the correspondence between the echo signal and the distance in order to obtain the first damage range from the preprocessed echo signal.
[0111] In practice, the distance value corresponding to each moment in the echo signal can be determined by multiplying the ultrasonic wave velocity by time, thereby determining the first damage range.
[0112] It should be noted that the ultrasonic wave velocity here refers to the speed at which ultrasonic waves propagate in the rail, for example, 2800 m / s.
[0113] In step 405, time-domain layered sampling is performed on the first damage area to obtain multi-channel time-domain data.
[0114] Optionally, the temporal-domain layered sampling of the first damage area can be achieved using the following two steps.
[0115] The first step is to divide the first damage range according to the preset time window length and step size to obtain the time domain data sequence.
[0116] The time-domain data sequence includes time-domain data from multiple channels, with each channel having the same time window length. Here, the time-domain data within each window represents the time-domain data of one channel.
[0117] For example, the time window length is The step size is half the length of the time window. This is equivalent to performing semi-overlapping uniform sampling in the first step.
[0118] The second step is to reduce the time window length of the first time domain data if the zero-crossing rate of the first time domain data is greater than the zero-crossing rate threshold, or if there is an energy mutation point in the first time domain data, to obtain multi-channel time domain data. The first time domain data is the time domain data of any channel in the time domain data sequence.
[0119] The second step allows for adaptive adjustment of the time window length for each channel, enabling precise segmentation of the damage-sensitive region.
[0120] In step 406, the multi-channel time-domain data and the reference data are cross-correlated pairwise by channel to obtain multiple cross-correlation sequences.
[0121] The reference data is obtained by performing time-domain layered sampling on the reference signal. The reference signal is the pre-processed echo signal of the first rail in a healthy state. The reference data and the multi-channel time-domain data have the same number of channels and sampling interval.
[0122] In this embodiment of the disclosure, the reference data is obtained by performing time-domain layered sampling on the reference signal corresponding to the first damage range. The time length of the reference signal corresponding to the first damage range is the same as the time length of the first damage range. Therefore, the number of channels and the sampling interval of the reference data obtained by performing time-domain layered sampling on the reference signal corresponding to the first damage range are the same as the number of channels and the sampling interval of the multi-channel time-domain data.
[0123] Optionally, in the multi-channel time domain data, the first The time domain data of the first channel and the reference data of the second channel The cross-correlation sequence of the time-domain data of each channel is calculated using formula (5):
[0124] (5)
[0125] In formula (5), Indicates the first channel in multi-channel time-domain data The time domain data of the first channel and the reference data of the second channel Cross-correlation sequences of time-domain data from each channel, and The values are the same. For time delay, The number of sampling points. Indicates the first channel in multi-channel time-domain data The time domain data of the first channel One sampling point, Indicating the first in the benchmark data The time domain data of the first channel One sampling point.
[0126] In step 407, the location of the second damage is determined based on multiple cross-correlation sequences.
[0127] The second damage location is the location where damage exists in the first rail.
[0128] Formula (5) can calculate multiple cross-correlation sequences, each containing multiple cross-correlation coefficients. For any given cross-correlation sequence, the trend of change in the cross-correlation coefficients within that sequence can be calculated. By comparing the trends of change in the cross-correlation coefficients in each cross-correlation sequence, the interval with the largest decrease in cross-correlation coefficients is taken as the interval corresponding to the second lesion location.
[0129] Here, the interval where the cross-correlation coefficient decreases the most is actually a time interval in the echo signal. Formula (4) can be used to convert this time interval into a distance interval, thereby obtaining the distance range between the second damage location and the first end of the first rail. This distance range between the second damage location and the first end of the first rail can accurately locate the second damage location.
[0130] The following are device embodiments of this application. For details not described in detail in the device embodiments, please refer to the above method embodiments.
[0131] Figure 6 A schematic diagram of a rail damage location device based on ultrasonic guided waves, provided in an exemplary embodiment of this disclosure, is shown. See also Figure 6 The rail damage location device 600 based on ultrasonic guided waves includes: an excitation module 601, a preprocessing module 602, a first damage location module 603, a damage range determination module 604, a layered sampling module 605, and a second damage location module 606.
[0132] The excitation module 601 is used to send an excitation signal from the first end of the first rail to the second end of the rail, and then receive the echo signal of the excitation signal at the first end. The first end and the second end are two different endpoints of the first rail, and the first rail is a section of the rail to be damaged and located.
[0133] The preprocessing module 602 is used to preprocess the echo signal, and the preprocessing includes Hilbert transform processing and noise reduction processing.
[0134] The first damage location module 603 is used to process the preprocessed echo signal using a dual-stream time-frequency network to obtain the first damage location.
[0135] The damage range determination module 604 is used to determine the first damage range in the preprocessed echo signal based on the first damage location.
[0136] The layered sampling module 605 is used to perform time-domain layered sampling on the first damage range to obtain multi-channel time-domain data.
[0137] The second damage location module 606 is used to perform cross-correlation calculation on the multi-channel time domain data to obtain a cross-correlation sequence, and determine the second damage location based on the cross-correlation sequence. The second damage location is the location where damage exists in the first rail.
[0138] Optionally, the second damage location module 606 is further configured to perform cross-correlation operations on the multi-channel time-domain data and the reference data in pairs according to the channels to obtain multiple cross-correlation sequences. The reference data is obtained by performing time-domain layered sampling on the reference signal. The reference signal is the pre-processed echo signal of the first rail in a healthy state. The reference data and the multi-channel time-domain data have the same number of channels and sampling interval.
[0139] Optionally, in the second damage localization module 606, the multi-channel time domain data of the first... The time-domain data of the first channel and the reference data of the second channel The cross-correlation sequence of the time-domain data of each channel is calculated using the following formula:
[0140]
[0141] in, Indicates the first channel of the multi-channel time-domain data The time-domain data of the first channel and the reference data of the second channel Cross-correlation sequences of time-domain data from each channel, and The values are the same. For time delay, The number of sampling points. Indicates the first channel of the multi-channel time-domain data The time domain data of the first channel One sampling point, Indicates the first in the reference data The time domain data of the first channel One sampling point.
[0142] Optionally, the hierarchical sampling module 605 is further configured to divide the first damage range according to a preset time window length and step size to obtain a time-domain data sequence, wherein the time-domain data sequence includes time-domain data of multiple channels, and the time window length of the time-domain data of each channel is the same; if the zero-crossing rate of the first time-domain data is greater than the zero-crossing rate threshold, or if there is an energy mutation point in the first time-domain data, the time window length of the first time-domain data is reduced to obtain the multi-channel time-domain data, wherein the first time-domain data is the time-domain data of any one channel in the time-domain data sequence.
[0143] Optionally, in the first damage localization module 603, the dual-stream time-frequency network includes a time-stream branch, a frequency-stream branch, a cross-modal attention module, a feature fusion module, and a convolutional layer; the time-stream branch is used to extract time features based on bidirectional LSTM, and the frequency-stream branch is used to extract spectral features based on continuous wavelet transform and ResNet-18; the cross-modal attention module is used to perform cross-modal attention operations on the time features and the spectral features to obtain spectral attention features; the feature fusion module is used to concatenate the time features and the spectral attention features to obtain fused features; and the convolutional layer is used to output the first damage location based on the fused features.
[0144] Optionally, the preprocessing module 602 is further configured to solve for the optimal parameters of the variational mode decomposition algorithm based on the particle swarm optimization algorithm, and to perform noise reduction processing on the echo signal using the variational mode decomposition algorithm based on the optimal parameters.
[0145] It should be noted that the above embodiments of the ultrasonic guided wave-based rail damage location device are only illustrated by the division of the functional modules described above. In practical applications, the functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the ultrasonic guided wave-based rail damage location device and the ultrasonic guided wave-based rail damage location method embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, which will not be repeated here.
[0146] The module division in this embodiment is illustrative and represents only one logical functional division. In actual implementation, other division methods are possible. Furthermore, the functional modules in the various embodiments of this disclosure can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.
[0147] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a terminal device (which may be a personal computer, mobile phone, or communication device, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0148] Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this disclosure. For example... Figure 7 As shown, the computer device 700 includes a processor 701 and a memory 702.
[0149] Processor 701 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 701 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 701 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 701 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, processor 701 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.
[0150] The memory 702 may include one or more computer-readable storage media, which may be non-transitory. The memory 702 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 702 is used to store at least one instruction, which is executed by the processor 701 to implement the ultrasonic guided wave-based rail damage localization method provided in the embodiments of this disclosure.
[0151] Those skilled in the art will understand that Figure 7 The structure shown does not constitute a limitation on the computer device 700, and may include more or fewer components than shown, or combine certain components, or use different component arrangements.
[0152] This disclosure also provides a non-transitory computer-readable storage medium, wherein when the instructions in the storage medium are executed by the processor of a computer device, the computer device is able to execute the rail damage localization method based on ultrasonic guided waves provided in this disclosure.
[0153] This disclosure also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the rail damage localization method based on ultrasonic guided waves provided in this disclosure.
[0154] The above description is merely an optional embodiment of this disclosure and is not intended to limit this disclosure. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the protection scope of this disclosure.
Claims
1. A method for locating rail damage based on ultrasonic guided waves, characterized in that, The method includes: After sending an excitation signal from the first end of the first rail to the second end of the rail, the echo signal of the excitation signal is received at the first end. The first end and the second end are two different endpoints of the first rail, and the first rail is a section of the rail to be damaged and located. The echo signal is preprocessed, including Hilbert transform processing and noise reduction processing; The preprocessed echo signal is processed using a dual-stream time-frequency network to obtain the first damage location; Based on the first damage location, the first damage range is determined in the preprocessed echo signal; The first damage area is subjected to time-domain layered sampling to obtain multi-channel time-domain data; Cross-correlation operation is performed on the multi-channel time-domain data to obtain a cross-correlation sequence, and a second damage location is determined based on the cross-correlation sequence. The second damage location is the location where damage exists in the first rail. The step of performing time-domain layered sampling on the first damage area to obtain multi-channel time-domain data includes: The first damage range is divided according to the preset time window length and step size to obtain a time domain data sequence. The time domain data sequence includes time domain data from multiple channels, and the time window length of the time domain data of each channel is the same. If the zero-crossing rate of the first time-domain data is greater than the zero-crossing rate threshold, or if there is an energy mutation point in the first time-domain data, the time window length of the first time-domain data is reduced to obtain the multi-channel time-domain data, wherein the first time-domain data is the time-domain data of any channel in the time-domain data sequence; The dual-stream time-frequency network includes a time-stream branch, a frequency-stream branch, a cross-modal attention module, a feature fusion module, and convolutional layers. The time-flow branch is used to extract time features based on bidirectional LSTM, and the frequency-flow branch is used to extract spectral features based on continuous wavelet transform and ResNet-18. The cross-modal attention module is used to perform cross-modal attention operations on the temporal features and the spectral features to obtain spectral attention features; The feature fusion module is used to concatenate the temporal features and the spectral attention features to obtain fused features; The convolutional layer is used to output the first damage location based on the fused features.
2. The method according to claim 1, characterized in that, The step of performing cross-correlation operation on the multi-channel time-domain data to obtain a cross-correlation sequence includes: The multi-channel time-domain data and the reference data are cross-correlated in pairs by channel to obtain multiple cross-correlation sequences. The reference data is obtained by time-domain layered sampling of the reference signal. The reference signal is the pre-processed echo signal of the first rail in a healthy state. The reference data and the multi-channel time-domain data have the same number of channels and sampling interval.
3. The method according to claim 2, characterized in that, The first of the multi-channel time-domain data The time-domain data of the first channel and the reference data of the second channel The cross-correlation sequence of the time-domain data of each channel is calculated using the following formula: in, Indicates the first channel of the multi-channel time-domain data The time-domain data of the first channel and the reference data of the second channel Cross-correlation sequences of time-domain data from each channel, and The values are the same. For time delay, The number of sampling points. Indicates the first channel of the multi-channel time-domain data The time domain data of the first channel One sampling point, Indicates the first in the reference data The time domain data of the first channel One sampling point.
4. The method according to any one of claims 1 to 3, characterized in that, The preprocessing of the echo signal includes: The optimal parameters of the variational mode decomposition algorithm are obtained by using the particle swarm optimization algorithm, and the echo signal is then denoised using the variational mode decomposition algorithm based on the optimal parameters.
5. A rail damage location device based on ultrasonic guided waves, characterized in that, The device includes: The excitation module is used to send an excitation signal from the first end of the first rail to the second end of the rail, and then receive the echo signal of the excitation signal at the first end. The first end and the second end are two different endpoints of the first rail, and the first rail is a section of the rail to be damaged and located. The preprocessing module is used to preprocess the echo signal, and the preprocessing includes Hilbert transform processing and noise reduction processing; The first damage localization module is used to process the preprocessed echo signal using a dual-stream time-frequency network to obtain the first damage location. The damage range determination module is used to determine the first damage range in the preprocessed echo signal based on the first damage location. The layered sampling module is used to perform time-domain layered sampling on the first damage range to obtain multi-channel time-domain data; The second damage location module is used to perform cross-correlation calculation on the multi-channel time domain data to obtain a cross-correlation sequence, and determine the second damage location based on the cross-correlation sequence. The second damage location is the location where damage exists in the first rail. The step of performing time-domain layered sampling on the first damage area to obtain multi-channel time-domain data includes: The first damage range is divided according to the preset time window length and step size to obtain a time domain data sequence. The time domain data sequence includes time domain data from multiple channels, and the time window length of the time domain data of each channel is the same. If the zero-crossing rate of the first time-domain data is greater than the zero-crossing rate threshold, or if there is an energy mutation point in the first time-domain data, the time window length of the first time-domain data is reduced to obtain the multi-channel time-domain data, wherein the first time-domain data is the time-domain data of any channel in the time-domain data sequence; The dual-stream time-frequency network includes a time-stream branch, a frequency-stream branch, a cross-modal attention module, a feature fusion module, and convolutional layers. The time-flow branch is used to extract time features based on bidirectional LSTM, and the frequency-flow branch is used to extract spectral features based on continuous wavelet transform and ResNet-18. The cross-modal attention module is used to perform cross-modal attention operations on the temporal features and the spectral features to obtain spectral attention features; The feature fusion module is used to concatenate the temporal features and the spectral attention features to obtain fused features; The convolutional layer is used to output the first damage location based on the fused features.
6. A computer device, characterized in that, The computer device includes a memory and a processor, wherein the memory stores at least one computer program, which is loaded and executed by the processor to implement the method according to any one of claims 1 to 4.
7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to implement the method of any one of claims 1 to 4.
8. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the method described in any one of claims 1 to 4.