A portable railway track fault detection device and a fault detection method

Abstract

The present invention relates to the technical field of railway fault detection, and in particular to a portable railway fault detection device and a fault detection method. The device includes: a mounting plate, an ultrasonic guided wave transducer, a received signal component (including a first receiver and a second receiver), and a control unit; the ultrasonic guided wave transducer and the receiver are installed along the length direction of the mounting plate, and the receivers are symmetrically distributed on both sides of the transducer; the control unit is connected to each component, controls the transducer to send a signal, and determines the fault location and type by receiving the echo signal through the receiver. The present invention reduces the installation and maintenance costs, and comprehensively and accurately analyzes the fault information through echo signal acquisition, multi-feature extraction and screening, multi-dimensional data fusion and classifier diagnosis, providing a reliable basis for railway maintenance.

Description

Technical Field

[0001] The present invention relates to the technical field of railway track fault detection, and in particular to a portable railway track fault detection device and a fault detection method. Background Art

[0002] As the core load-bearing component of railway operation, railway tracks with diseases such as cracks, deformations, and fractures pose a serious threat to railway traffic safety. To ensure the stable and efficient operation of the railway transportation system, periodic non-destructive testing of railway tracks to promptly detect and handle potential fault hazards has become an important part of railway safety maintenance work.

[0003] Existing railway track detection technologies are mainly divided into two major systems: mobile detection and fixed detection. Mobile detection uses technical means such as ultrasonic detection, eddy current detection, and magnetic flux leakage detection to conduct regular inspections of railway tracks. These methods are easy to operate and have relatively high detection sensitivity, and can effectively detect defects on the surface and near the surface of railway tracks in specific scenarios. However, their detection distance is limited, and the detection operations can usually only be carried out during the idle periods of railway operation, resulting in relatively low overall detection efficiency. At the same time, due to the limitations of detection principles and equipment characteristics, there are inevitably detection blind spots during the detection process, and it is easy to miss detections.

[0004] Fixed detection is achieved by installing various sensors and monitoring devices on railway lines to realize real-time monitoring of the state of railway tracks. Common fixed detection technologies include stress detection, optical fiber detection, ultrasonic guided wave detection, and track circuit detection. Although stress detection can monitor the stress state of railway tracks, the detection distance is short, and the anti-interference ability is weak in a complex electromagnetic environment; optical fiber detection has the advantages of high precision and high sensitivity, but its installation and maintenance costs are high, and it is difficult to implement in railway areas with harsh environmental conditions and complex terrains; track circuit detection can detect changes in the electrical characteristics of railway tracks in real time and accurately, and has relatively high detection accuracy for some faults, but there are limitations in detecting partial fractures of railway tracks; ultrasonic guided wave detection has significant advantages such as long-distance propagation, full cross-section coverage, and fast detection speed. However, the installation position of the fixed ultrasonic guided wave excitation device is usually limited to the rail bottom or rail waist, so that the ultrasonic guided wave energy is mainly concentrated in the rail bottom and rail waist areas, and the detection effect on the rail head part is not good. Summary of the Invention

[0005] To solve the problems existing in the existing railway track detection technologies, such as low manual detection efficiency, poor accuracy, and difficulty in meeting the development needs of railways, the present invention provides a portable railway track fault detection device and a fault detection method. The portable railway track fault detection device is installed on the railway track and includes: a mounting plate, an ultrasonic guided wave transducer, a received signal component, and a control unit;

[0006] Among them, the ultrasonic guided wave transducer and the received signal assembly are both installed along the length direction of the mounting plate. The received signal assembly includes a first receiver and a second receiver, and the first receiver and the second receiver are symmetrically arranged on both sides of the ultrasonic guided wave transducer along the length direction of the mounting plate;

[0007] The control unit is connected to the ultrasonic guided wave transducer, the first receiver and the second receiver, controls the ultrasonic guided wave transducer to send ultrasonic guided wave signals, determines the fault location according to the time taken for the first receiver and the second receiver to receive the echo signals passing through the fault point, and judges the fault type according to the echo signals.

[0008] In an embodiment of the present invention, the minimum distance between the first receiver and the second receiver satisfies:

[0009] , , where V represents the transmission speed of the ultrasonic guided wave signal, represents the total duration of the ultrasonic guided wave signal generated by the ultrasonic guided wave transducer, T represents the period of the ultrasonic guided wave signal generated by the ultrasonic guided wave transducer once, and n represents the total number of times of generating the ultrasonic guided wave signal.

[0010] In an embodiment of the present invention, determining the fault location according to the time taken for the first receiver and the second receiver to receive the echo signals passing through the fault point includes:

[0011] Comparing the time taken for the first receiver and the second receiver to receive the echo signals passing through the fault point, determining the direction where the rail fault point is located, and calculating the distance from the rail fault point to the ultrasonic guided wave transducer as follows:

[0012] , where, represents the total time for the ultrasonic guided wave to propagate from the transducer to any one receiver for the first time, represents the total time for the ultrasonic guided wave to propagate from the transducer to the rail fault point and then to the receiver close to the rail fault point; V represents the transmission speed of the ultrasonic guided wave signal.

[0013] In an embodiment of the present invention, a groove is provided on the mounting plate, the groove is clamped on the rail, and the ultrasonic guided wave transducer contacts the rail head or the rail web.

[0014] In an embodiment of the present invention, the resonance frequency range of the ultrasonic guided wave transducer is 35 kHz to 40 kHz.

[0015] Based on the above fault detection device, the present invention also provides a portable railway track fault detection method, which uses the portable railway track fault detection device for fault detection. The portable railway track fault detection method includes the following steps:

[0016] S1: Collect the ultrasonic guided wave echo signals generated due to railway track faults. Based on the ultrasonic guided wave echo signals, determine the fault orientation and obtain time-domain features, frequency-spectrum features, and audio features;

[0017] S2: Screen the time-domain features, the frequency-spectrum features, and the audio features to obtain an echo signal feature data set;

[0018] S3: Use a variety of statistical functions to perform fusion processing on the echo signal feature data set to obtain multi-dimensional feature data;

[0019] S4: Use the multi-dimensional feature data as the input of a fault classifier to obtain the fault type of the railway track.

[0020] In an embodiment of the present invention, in S2, the method for obtaining the echo signal feature data set is as follows:

[0021] S21: Collect an original echo signal data set with labels. Each data sample includes time-domain features, frequency-spectrum features, and audio features, and determine the sample class label, denoted as C classes;

[0022] S22: For each feature k , calculate the between-class variance and the within-class variance, and use the ratio of the between-class variance and the within-class variance as the importance score of this feature k ;

[0023] S23: Sort all features according to the importance score to obtain an echo signal feature data set.

[0024] In an embodiment of the present invention, the calculation method of the between-class variance is as follows:

[0025] Calculate the mean i of the samples of each class k on the feature and the global mean k of all samples on the feature ;

[0026] According to and , calculate the between-class variance k of the feature on the original echo signal data set :

[0027] where is the number of samples in class i, and n is the total number of samples in the original echo signal dataset.

[0028] In one embodiment of the present invention, the calculation method of the within-class variance is as follows:

[0029] Calculate for each class i the mean of the samples on the feature k and the value of the sample x on the k-th feature ; ;

[0030] According to and , calculate the within-class variance k of the feature on the original echo signal dataset: ,

[0031] wherein, is the sample set of class i , and n is the total number of samples in the original echo signal dataset.

[0032] In one embodiment of the present invention, the echo signal feature dataset includes the mean of the zero-crossing rate, fundamental frequency, spectral peak centroid, spectral centroid, spectral sharpness, spectral entropy, spectral skewness, spectral peak side ratio, spectral peak back ratio, the minimum value of the spectral peak back ratio, the variance of the spectral peak back ratio, and Mel cepstral coefficients.

[0033] The present invention also provides an electronic device, which includes a processor, a memory, and a bus system. The processor and the memory are connected through the bus system. The memory is used to store instructions, and the processor is used to execute the instructions stored in the memory to implement the portable rail fault detection method.

[0034] The present invention also provides a computer storage medium, which stores a computer software product. The computer software product includes several instructions for causing a computer device to execute the portable rail fault detection method.

[0035] The above technical solution of the present invention has the following advantages compared with the prior art:

[0036] The present invention combines the characteristics of simple operation and high sensitivity of mobile detection, and full cross-section coverage and long detection distance of fixed ultrasonic guided wave detection, making up for the disadvantages of limited detection distance and low efficiency of mobile detection, as well as the inability to accurately locate faults and high cost of fixed ultrasonic guided wave detection. The provided device can effectively detect rail head faults through the ultrasonic guided wave excitation device installed on the rail head, accurately locate faults using echo signals, introduce a deep learning network to process ultrasonic guided wave signal data in real time, not only with fast detection speed, but also can accurately identify faults such as micro-cracks, reduce human errors, achieve accurate discrimination of fault types, and greatly improve the accuracy and reliability of fault diagnosis. Description of the Drawings

[0037] In order to make the content of the present invention easier to be clearly understood, the following further describes the present invention in detail according to specific embodiments of the present invention in combination with the drawings, where

[0038] Figure 1 is a schematic structural diagram of a portable rail fault detection device provided in an embodiment of the present invention;

[0039] Figure 2 is a schematic connection diagram of the control unit and the signal transmitting and receiving device;

[0040] Figure 3 is a schematic signal propagation diagram when the portable rail fault detection device provided by the present invention is in use, where (a) represents the signal propagation diagram when no fault occurs, and (b) represents the signal propagation diagram when a fault occurs;

[0041] Figure 4 is a schematic diagram of the formation and propagation process of ultrasonic guided waves in a flat plate;

[0042] Figure 5 is the acoustic wave attenuation curve in the rail under different frequency ranges;

[0043] Figure 6 is a schematic flow chart of a portable rail fault detection method provided in an embodiment of the present invention;

[0044] Figure 7 is an implementation flow chart of an echo signal feature extraction strategy provided in an embodiment of the present invention;

[0045] Figure 8 is a schematic diagram of the training and detection process of a rail fault detection model provided in an embodiment of the present invention;

[0046] Explanation of the reference numerals in the specification drawings:

[0047] 1. Mounting plate; 11. Groove; 2. Ultrasonic guided wave transducer; 3. Received signal assembly; 31. First receiver; 32. Second receiver; 4. Control unit; 5. Rail head; 6. Rail web. Detailed implementation manners

[0048] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, so that those skilled in the art can better understand the present invention and be able to implement it, but the embodiments cited are not intended to limit the present invention.

[0049] Embodiment 1:

[0050] See Figure 1 and Figure 2 As shown, the present invention also provides a portable rail fault detection device, which is installed on the rail and includes: a mounting plate 1, an ultrasonic guided wave transducer 2, a received signal assembly 3 and a control unit 4;

[0051] Wherein, the ultrasonic guided wave transducer 2 and the received signal assembly 3 are both installed along the length direction of the mounting plate 1, the received signal assembly 3 includes a first receiver 31 and a second receiver 32, and the first receiver 31 and the second receiver 32 are symmetrically arranged on both sides of the ultrasonic guided wave transducer 2 along the length direction of the mounting plate 1 respectively;

[0052] The control unit 4 is connected to the ultrasonic guided wave transducer 2, the first receiver 31 and the second receiver 32, controls the ultrasonic guided wave transducer 2 to send ultrasonic guided wave signals, determines the fault orientation according to the time taken by the first receiver 31 and the second receiver 32 to receive the echo signals passing through the fault point, and judges the fault type according to the echo signals.

[0053] As Figure 3 shown in (a) of, when there is no fault in the rail, after the ultrasonic guided wave transducer 2 generates an excitation, no echo signal of the ultrasonic guided wave can be generated, and both the first receiver 31 and the second receiver 32 can only receive a guided wave signal once and the time taken by the two receivers to receive the echo signals is approximately equal. As Figure 3 shown in (b) of, when there are faults such as cracks or fractures in the rail, echo signals will be generated. The first receiver 31 and the second receiver 32 can receive two guided wave signals. One is the echo signal directly emitted by the ultrasonic guided wave transducer 2, and the other is the echo signal received by the ultrasonic guided wave transducer 2 after passing through the fault point.

[0054] Obviously, for the present invention, it is only necessary to simply determine whether the receiver receives the echo signal to judge whether there is a fault in the railway track. Since the broken rail detector has two receivers, in order to make the signal reception times of the two receivers have a significant difference, taking the example of generating five cycles (i.e., n = 5) of ultrasonic guided wave signals, the minimum distance between the first receiver 31 and the second receiver 32 satisfies: , , where V represents the propagation speed of the ultrasonic guided wave signal, represents the total duration of the ultrasonic guided wave signal generated by the ultrasonic guided wave transducer, T represents the period of generating one ultrasonic guided wave signal by the ultrasonic guided wave transducer, and n represents the total number of times of generating the ultrasonic guided wave signal.

[0055] When a crack or fracture occurs on one side of the railway track, the fault location is determined according to the time taken for the first receiver 31 and the second receiver 32 to receive the echo signals passing through the fault point, including:

[0056] Judging and comparing the time taken for the first receiver 31 and the second receiver 32 to receive the echo signals passing through the fault point to determine the direction where the fault point of the railway track is located:

[0057] The time taken for the ultrasonic guided wave to propagate from the transducer to the fault point and then to the first receiver 31 is determined by peak detection and the time taken for the ultrasonic guided wave to propagate from the transducer to the fault point and then to the second receiver 32 . If , the fault point is close to the end of the first receiver 31; otherwise, the fault point is close to the end of the second receiver 32.

[0058] Meanwhile, the distance d from the fault point of the railway track to the ultrasonic guided wave transducer 2 is calculated as follows:

[0059] ,

[0060] where, represents the total time for the ultrasonic guided wave to propagate from the transducer to any one of the receivers for the first time, represents the total time for the ultrasonic guided wave to propagate from the transducer to the fault point of the railway track and then to the receiver close to the fault point of the railway track, . When n = 1, it indicates that the fault point is close to the end of the first receiver 31. At this time represents the total time for the ultrasonic guided wave to propagate from the transducer to the fault point and then to the first receiver 31; when n = 2, it indicates that the fault point is close to the end of the second receiver 32. At this time represents the total time for the ultrasonic guided wave to propagate from the transducer to the fault point and then to the second receiver 32; V represents the propagation speed of the ultrasonic guided wave signal.

[0061] Ultrasonic guided waves are essentially a type of mechanical wave, which are special ultrasonic waves generated after the interaction between ultrasonic waves and the boundaries of waveguide media. As Figure 4 shown, when ultrasonic waves propagate in elastic waveguide media such as cylindrical shells, rods, and layers, they will continuously reflect and refract with the medium boundaries, and at the same time, the conversion between transverse waves and longitudinal waves also occurs continuously, ultimately forming ultrasonic guided waves.

[0062] In this embodiment, an air compression impact device with adjustable frequency was used to conduct an attenuation experiment on the guided waves in the rail on an actual railway line. Three frequency ranges were set for this experiment, namely 0 kHz to 20 kHz, 20 kHz to 40 kHz, and 40 kHz to 60 kHz. The attenuation curve of ultrasonic guided waves in the rail was obtained through the experiment, as specifically Figure 5 shown.

[0063] It can be seen from Figure 5 that there are differences in the transmission distances of guided waves in different frequency ranges in the rail. Among them, when the frequency range is 20 kHz to 40 kHz, its attenuation curve is relatively flat. This means that under the condition of the same initial energy, the attenuation rate of ultrasonic guided waves in this frequency range is slower. In the actual application scenario of rail fault detection, the vibration intensity of the ultrasonic guided waves generated by the ultrasonic guided wave transducer is relatively weak. Based on this, it is more appropriate to select the ultrasonic guided wave signal in the frequency range of 20 kHz to 40 kHz.

[0064] In addition, there is a certain correlation with its size. The lower the frequency, the larger the size of the ultrasonic guided wave transducer 2. For example, the outer diameter of the ultrasonic guided wave transducer 2 with a resonance frequency of 25 kHz reaches 60 mm; while the outer diameter of the ultrasonic guided wave transducer 2 with a resonance frequency of 35 kHz is 48 mm.

[0065] During the process of exciting ultrasonic guided waves, in order to achieve efficient energy utilization, the energy should be concentrated on the rail head as much as possible. Since there are rail fasteners at the rail bottom, this will limit the propagation of vibration at the rail bottom, resulting in too fast vibration attenuation and then a sharp shortening of the detection distance, which is not conducive to the development of detection work. Therefore, in this embodiment, a groove 11 is provided on the mounting plate 1, the groove 11 is clamped on the rail, and the ultrasonic guided wave transducer 2 can usually only be installed at the positions of the rail head 5 and the rail web 6.

[0066] Considering this installation limitation and the requirement of energy propagation, the size of the ultrasonic guided wave transducer 2 should not be too large. After considering various factors, the resonance frequency range of the ultrasonic guided wave transducer 2 is finally determined to be 35 kHz to 40 kHz.

[0067] In actual detection, the amount of echo signal data obtained from the first receiver 31 or the second receiver 32 is huge. If these data are directly used for classification estimation of fault types, not only will there be great difficulties, but it is also difficult to ensure the accuracy rate. To solve this problem, the control unit 4 performs dimensionality reduction processing and feature extraction operations on the collected echo signal data according to the rail fault detection method described in the first embodiment. Through these processes, the data that can effectively characterize the fault features are screened out, and then these features are input into the classifier, so as to achieve accurate classification of various fault types such as rail cracks, fractures, and deformations.

[0068] Embodiment 2:

[0069] Since the amount of echo signal data generated by the actually collected rail faults is usually large, it is difficult to directly use these data for classification estimation of fault types, and the accuracy rate is also difficult to guarantee. In order to more effectively and accurately classify rail fault types, it is necessary to reduce the dimension of these original echo signal data, find a set of features that can best characterize these data, and then input these features into the classifier for classification estimation.

[0070] Therefore, as shown in Figure 6 and Figure 7 The present invention provides a portable rail fault detection method, which uses the portable rail fault detection device described in the first embodiment to detect faults. The method includes the following steps:

[0071] S1: Turn on the ultrasonic guided wave transducer 2 through the control unit 4, so that the ultrasonic guided wave transducer 2 sends ultrasonic guided wave signals along both ends of the rail, collect the ultrasonic guided wave echo signals generated due to rail faults through the receiving signal assembly 3, and calculate based on the ultrasonic guided wave echo signals to determine the fault orientation and obtain time domain features, frequency spectrum features, and audio features;

[0072] S2: Use the Fisher algorithm to screen the time domain features, the frequency spectrum features, and the audio features to obtain an echo signal feature data set;

[0073] S3: Use a variety of statistical functions to perform fusion processing on the echo signal feature data set to obtain multi-dimensional feature data;

[0074] S4: Use the multi-dimensional feature data as the input of the fault classifier to obtain the fault type of the rail.

[0075] When using ultrasonic guided waves for rail fault detection, feature extraction of echo signals is a critical step. Key characteristic parameters of echo signals primarily include echo amplitude, phase variation, frequency attenuation, and waveform distortion. These parameters are crucial for comprehensively characterizing the extent and nature of track defects. In addition to the basic time and frequency domain characteristics described above, in-depth analysis of modal conversion and scattering phenomena during ultrasonic guided wave propagation is also essential. Ultrasonic guided waves of different modes convert into each other during propagation and, when encountering track defects, also produce scattering. Studying and analyzing these phenomena facilitates more accurate location of track defects and a deeper understanding of their essential characteristics. Given the diversity of track materials, the complexity of their geometries, and the variability of external environmental factors, a single feature extraction method often struggles to fully and accurately describe the complex characteristics of echo signals.

[0076] Therefore, in step S1, it is necessary to collect the ultrasonic guided wave echo signal generated by the rail fault, and perform correlation calculation based on the echo signal to determine the fault location, and at the same time obtain the time domain characteristics, spectrum characteristics and audio characteristics.

[0077] Furthermore, since dedicated feature sets typically have high feature dimensions, there are numerous possible combinations of their feature subsets. If classifier training is performed for each feature subset, the algorithm's computational overhead and computational complexity will increase significantly, making it difficult to implement in practical applications. To address this issue, in step S2, a feature filtering algorithm, such as the Fisher scoring algorithm, is used to filter the acquired time domain features, spectral features, and audio features to construct an echo signal feature dataset. The specific method is as follows:

[0078] S21: Collect a labeled raw echo signal dataset. Each data sample contains time domain features, spectrum features, and audio features. Determine the sample category label, which is recorded as C categories.

[0079] S22: For each feature k, calculate the between-class variance and the within-class variance, and use the ratio of the between-class variance to the within-class variance as the importance score of the feature k; wherein the between-class variance is calculated as follows:

[0080] Calculate each category i The sample has the following features: k The mean on And all samples in the feature k The global mean on ;

[0081] according to and , calculate the features k The inter-class variance on the original echo signal dataset : ,

[0082] in, is the number of samples of category i, and n is the total number of samples in the original echo signal dataset;

[0083] The intra-class variance is calculated as follows:

[0084] Calculate each category i The sample has the following features: k The mean on And the value of sample x on the kth feature ;

[0085] according to and , calculate the features k Intra-class variance on the original echo signal dataset : ,

[0086] in, For category i The sample set is n, and n is the total number of samples in the original echo signal data set;

[0087] S23: Sort all features according to the importance scores to obtain an echo signal feature dataset.

[0088] Furthermore, the echo signal feature dataset covers features such as zero crossing rate (ZCR), fundamental frequency (F0), spectral peak center of gravity, spectral centroid, spectral centroid, spectral sharpness, spectral entropy, spectral skewness, spectral peak-to-side ratio, mean spectral peak-to-background ratio, minimum spectral peak-to-background ratio, variance of spectral peak-to-background ratio, and Mel-frequency cepstral coefficients.

[0089] The zero-crossing rate (ZCR) represents the number of times the signal amplitude crosses zero within a frame. As the fault distance increases, the amplitude of the received transmitted signal decreases, and the environmental noise inside the rail gradually becomes dominant, resulting in an increase in the zero-crossing rate. Therefore, the zero-crossing rate (ZCR) can not only be used to characterize the time domain characteristics of the reflected signal, but also to some extent reflect the distance to the rail fault. Its calculation method is:

[0090] ,in, , N is the number of sampling points in each frame, is the amplitude of the qth sampling point.

[0091] The fundamental frequency F0 represents the lowest frequency component of the periodic vibration in the signal, which is the reciprocal of the time required for the signal to repeat a complete cycle and is defined as:

[0092] , where T is the periodic time interval of the signal. The fundamental frequency reflects the main vibration characteristics of the signal and is the frequency component with the most concentrated energy in the complex signal. In the rail vibration signal, the fundamental frequency corresponds to the natural vibration frequency of the rail after being impacted. When there are faults (such as cracks, looseness) in the rail, its vibration mode will change. When the fault is close to the detection point: the vibration energy is concentrated, the amplitude of the fundamental frequency component is significant and the frequency is stable. When the fault is far from the detection point: the attenuation of the signal causes the high-frequency components to weaken, the fundamental frequency may shift or split due to the superposition of reflected waves, and at the same time the noise interference increases.

[0093] The centroid of the spectral peak can reflect the distance at which the receiver locates the rail fault by performing peak detection on the reflected signal. When no rail fault is detected, that is, when the signal line spectrum does not exist, the frequencies with large amplitudes are generated by the environmental noise inside the rail and are mainly concentrated in the low-frequency band; when a rail fault is detected, that is, when the signal line spectrum exists, the frequency amplitude of the environmental noise inside the rail is significantly lower than the amplitude of the line spectrum, and corresponding line spectra also appear in the high-frequency band. The method for obtaining the centroid of the spectral peak is as follows:

[0094] Obtain N maximum spectral peaks in the spectrum of the ultrasonic guided wave echo signal, and define the frequencies of the spectral peaks as , and the corresponding spectral amplitudes are , ; in this embodiment, N = 5;

[0095] Based on and , calculate the centroid of the spectral peak C: .

[0096] Furthermore, the spectral centroid represents the center of the signal frequency components and is one of the important physical parameters describing the timbre attribute. It is the frequency weighted and averaged by energy within a certain frequency range, and its calculation method is: , where, f is the frequency of the signal, E is the spectral energy corresponding to the frequency after performing the short-time Fourier transform. As the distance of the rail fault location increases, the amplitude of the low-frequency components of the reflected signal decreases rapidly, while the higher-frequency part is mainly noise components and the amplitude changes little, resulting in the spectral centroid moving towards the high frequency.

[0097] Furthermore, the spectral sharpness It represents the ratio of the medium and high frequency energy to the total energy in the signal, describes a feeling related to the frequency components of the sound, reflects the uncomfortable feeling of the sound, but is independent of the loudness of the sound. Its calculation method is as follows: , where is the energy at frequency f, is the sampling frequency. It can be seen from this that as the fault distance gradually becomes farther, the rapid attenuation of the low frequency components causes the sharpness of the signal spectrum to gradually increase.

[0098] Furthermore, the spectral entropy represents the orderliness of the signal spectrum, and its calculation method is as follows: , where x represents an event within a certain range of the spectral amplitude, is the probability of event x. The distance between the maximum and minimum values of the spectral amplitude is evenly divided into 100 segments, and each segment is regarded as an event. By calculating the probability of each event, the spectral entropy can be obtained. When the distance becomes farther, the spectral entropy increases, meaning that the orderliness of the spectrum decreases and the spectrum gradually becomes disordered. This is mainly because when the target distance is relatively far, the signal-to-noise ratio of the received signal is relatively low, and the signal is greatly affected by noise.

[0099] Furthermore, the spectral skewness can measure the skewness direction and degree of the spectrum, and is a numerical characteristic of the asymmetry degree of the spectrum distribution. Its calculation method is as follows:

[0100] , where and respectively represent the second-order central moment and the third-order central moment, X is the amplitude of the spectrum, and are the mean and variance of X respectively. Generally speaking, when the spectrum is right-skewed, the spectral skewness is greater than 0, and the larger the skewness value, the greater the right-skewed degree. The spectrum of the reflected signal of the rail fault can be regarded as a right-skewed distribution. When the distance becomes farther, the amplitude of the low frequency components rapidly attenuates, which is equivalent to a decrease in the right-skewed degree. Therefore, the spectral skewness value at a farther distance is less than that at a closer distance.

[0101] Furthermore, the spectral peak-side ratio represents the underlying characteristics describing a single spectral peak in the spectrum and is used to describe the details of a single spectral peak in the spectrum. Since the line spectrum corresponding to the reflected signal is high and narrow, comparing the peak value of a single spectral peak in the spectrum with the amplitude corresponding to the position obtained by raising a fixed frequency from the peak position of the spectrum, the ratio will be significantly higher than the ratio without a line spectrum. Therefore, this ratio is defined as the peak-side ratio, and the formula is as follows:

[0102] , where i represents the serial number of the maximum spectral peak in the spectrum of the ultrasonic guided wave echo signal obtained, represents the total number of maximum spectral peaks, represents the fixed frequency for enhancing the spectral peak position. When the line spectrum exists, the five largest spectral peaks extracted from the frequency spectrum are narrow peaks. Therefore, the peak side ratios of these 5 spectral peaks are all larger values and are more stable; when the line spectrum does not exist, the widths of the five largest spectral peaks vary and are basically wide peaks, so the peak side ratios are also smaller values and the values are quite different. To fully characterize the spectral characteristics of the reflected signal, the present invention uses the mean value of the 5 peak side ratios as the underlying characteristic for characterizing the spectral characteristics.

[0103] Furthermore, since the line spectrum corresponding to the reflected signal is few and narrow, the line spectrum has little effect on the mean value of the entire spectral amplitude. However, compared with the mean value of the background amplitude of the entire frequency spectrum, the amplitude of the line spectrum is significantly increased, and the ratio of the maximum spectral amplitude to the spectral average amplitude when there is a line spectrum is significantly higher than the ratio when there is no line spectrum. Therefore, this ratio is defined as the underlying characteristic for characterizing the spectral characteristics, called the peak-back ratio, and its calculation method is as follows:

[0104] Obtain N maximum spectral peaks in the frequency spectrum of the ultrasonic guided wave echo signal, and the spectral amplitude of the spectral peak is [[ID=IO]] , ;

[0105] According to all , calculate its spectral average amplitude , according to the maximum spectral amplitude and the spectral average amplitude , calculate the spectral peak-back ratio R of each spectral peak: .

[0106] Mel-frequency cepstral coefficients (MFCCs) are used to describe the distribution of audio signal energy in different frequency ranges and are widely used in the field of speech recognition. Past research has shown that they are also effective in flaw detection. The present invention deeply analyzes the Mel-frequency cepstral coefficients. After calculating the first 16 coefficients, it is found that under the condition of the same fault distance, there are significant differences in the 1-14 Mel-frequency cepstral coefficients (MFCC 1-14) corresponding to different fault types. Therefore, these 1-14 Mel-frequency cepstral coefficients (MFCC1-14) can be used as key indicators to reflect the spectral characteristics of the reflected signal, so as to distinguish different fault types.

[0107] In summary, the present invention finally selects the zero crossing rate (ZCR), fundamental frequency (F0), spectral peak centroid, spectral centroid, spectral sharpness, spectral entropy, spectral skewness, spectral peak side ratio, spectral peak back ratio (mean, minimum value, variance), and 1 to 14 coefficients of Mel frequency cepstral coefficients (MFCC) (MFCC1-14), a total of 25 underlying features to characterize the spectral characteristics of the echo signal.

[0108] In addition, in step S3, the present invention uses the 10 statistical functions listed in Table 1 to conduct a comprehensive statistical description of the above-mentioned 25 underlying features analyzed, and actually returns 13 parameters. These 25 underlying features screened by the Fisher score algorithm, after being processed by 10 statistical functions respectively, finally generate 13×25 = 325-dimensional feature data that can fully characterize different rail faults through feature extraction and fusion. These feature data will be used as the input data for the deep learning algorithm for rail fault detection.

[0109] Table 1

[0110]

[0111] In step S4, a classifier is used to accurately classify the possible fault types of the rail, such as cracks, fractures, and deformations. The types of classifiers available for selection are diverse, including neural networks, support vector machines, random forests, naive Bayes classifiers, etc.

[0112] Taking the deep learning model as an example, it plays an important classification role in rail fault detection. When the deep learning model is used as a classifier, it has a set of rigorous training and detection processes. The specific process can refer to Figure 8 .

[0113] In the training stage, a large number of labeled sample data containing different fault types are used to continuously adjust the parameters of the model through algorithms such as backpropagation, so that the model learns the feature patterns corresponding to different fault types. In the detection stage, the 325-dimensional feature data obtained by feature extraction and fusion of the echo signal to be detected is input into the trained deep learning model. The model analyzes the data based on the learned knowledge and outputs the corresponding fault type judgment result, so as to achieve efficient and accurate identification of rail fault types.

[0114] Example 3:

[0115] The present invention also provides an electronic device, which includes a processor, a memory, and a bus system. The processor and the memory are connected through the bus system. The memory is used to store instructions, and the processor is used to execute the instructions stored in the memory to implement the portable rail fault detection method described in Embodiment 2.

[0116] Embodiment 4:

[0117] The present invention also provides a computer storage medium, which stores a computer software product. The computer software product includes a number of instructions for causing a computer device to execute the portable rail fault detection method described in Embodiment 2.

[0118] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0119] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the specified functions in Figure 1 one or more of the processes Figure 1 or multiple processes and / or blocks

[0120] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device that implements the specified functions in Figure 1 one or more of the processes Figure 1 or multiple processes and / or blocks

[0121] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are executed on the computer or other programmable apparatus to produce a computer-implemented process, thereby the instructions executed on the computer or other programmable apparatus provide steps for realizing the functions specified in one process or a plurality of processes and / or blocks Figure 1 one process or a plurality of processes and / or blocks Figure 1 steps of the functions specified in one block or a plurality of blocks.

[0122] Obviously, the above embodiments are only examples for clear illustration and are not limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all implementation manners here. And the obvious changes or modifications derived therefrom still fall within the protection scope of the present invention.

Claims

1. A portable railway fault detection device, characterized in that, The described portable rail fault detection device is installed on the rail and includes: a mounting plate, an ultrasonic guided wave transducer, a received signal assembly, and a control unit; Among them, the ultrasonic guided wave transducer and the received signal assembly are both installed along the length direction of the mounting plate. The received signal assembly includes a first receiver and a second receiver. The first receiver and the second receiver are symmetrically arranged on both sides of the ultrasonic guided wave transducer along the length direction of the mounting plate respectively. A groove is provided on the mounting plate, and the groove is clamped on the rail, and the ultrasonic guided wave transducer contacts the rail head or the rail web; The control unit is connected to the ultrasonic guided wave transducer, the first receiver, and the second receiver, controls the ultrasonic guided wave transducer to send ultrasonic guided wave signals, determines the fault orientation according to the time taken for the first receiver and the second receiver to receive the echo signals passing through the fault point, and determines the fault type according to the echo signals; Among them, the fault orientation is determined through the following steps: Judge and compare the time taken for the first receiver and the second receiver to receive the echo signals passing through the fault point, and determine the direction where the rail fault point is located: Determine the time t1 taken for the ultrasonic guided wave to propagate from the transducer to the fault point and then to the first receiver and the time t2 taken for the ultrasonic guided wave to propagate from the transducer to the fault point and then to the second receiver by means of peak detection. If t2>t1, the fault point is close to one end of the first receiver; otherwise, the fault point is close to one end of the second receiver; Calculate the distance d from the rail fault point to the ultrasonic guided wave transducer as follows: d = (t0 + t n ) · V / 2, where t0 represents the total time for the ultrasonic guided wave to first propagate from the transducer to any receiver, and t n represents the total time for the ultrasonic guided wave to propagate from the transducer to the rail fault point and then to the receiver close to the rail fault point; V represents the propagation speed of the ultrasonic guided wave signal; Determine the fault type according to the echo signals through the following steps: S1: Based on the echo signals, obtain time domain features, frequency spectrum features, and audio features; S2: Screen the time domain features, the frequency spectrum features, and the audio features to obtain an echo signal feature data set, including the mean values of the zero crossing rate, fundamental frequency, spectral peak centroid, spectral centroid, spectral sharpness, spectral entropy, spectral skewness, spectral peak side ratio, spectral peak back ratio, the minimum value of the spectral peak back ratio, the variance of the spectral peak back ratio, and the Mel cepstral coefficients; S3: Use a variety of statistical functions to perform fusion processing on the echo signal feature data set to obtain multi-dimensional feature data; S4: Use the multi-dimensional feature data as the input of the fault classifier to obtain the fault type of the rail.

2. The portable rail fault detection device according to claim 1, wherein The minimum distance D between the first receiver and the second receiver min satisfies: D min = T0·V, T0 = n·T, Among them, V represents the transmission speed of the ultrasonic guided wave signal, T0 represents the total time for the ultrasonic guided wave transducer to generate the ultrasonic guided wave signal, T represents the period of the ultrasonic guided wave transducer to generate an ultrasonic guided wave signal once, and n represents the total number of times of generating the ultrasonic guided wave signal.

3. The portable rail fault detection device according to claim 1, characterized in that, The resonance frequency range of the ultrasonic guided wave transducer is 35 kHz to 40 kHz.

4. The portable rail fault detection device according to claim 3, characterized in that, In S2, the method for obtaining the echo signal feature data set is as follows: S21: Collect the original echo signal data set with labels. Each data sample contains time domain features, frequency spectrum features, and audio features, and determine the sample category label, denoted as C categories; S22: For each feature k, calculate the between-class variance and the within-class variance, and use the ratio of the between-class variance to the within-class variance as the importance score of the feature k; S23: Sort all features according to the importance score to obtain an echo signal feature data set.

5. The portable rail fault detection device according to claim 4, wherein The calculation method of the between-class variance is as follows: Calculate the mean of samples in each category \(i\) on feature \(k\) and the global mean \(m\) of all samples on feature \(k\) (k) ; According to and m (k) , calculate the between-class variance of feature k on the original echo signal dataset where n i is the number of samples of class i, and n is the total number of samples in the original echo signal dataset.

6. The portable rail fault detection device according to claim 4, characterized in that, The calculation method of the within-class variance is as follows: Calculate the mean of the samples of each category \(i\) on feature \(k\) and the value \(x\) of sample \(X\) on the \(k\)-th feature (k) ; According to and x (k) , calculate the within-class variance of feature k on the original echo signal dataset where ω i is the sample set of class i, and n is the total number of samples in the original echo signal data set.

Images