Track fault identification method and system based on HSROA

Abstract

The invention discloses a track fault identification method and system based on HSROA, and the method comprises the steps: calculating a frequency band entropy value of at least one IMF component signal, and selecting a sensitive IMF component signal from the at least one IMF component signal according to the frequency band entropy value; filtering the sensitive IMF component signal for the second time according to a band-pass filter to obtain a target signal, and performing envelope power spectrum analysis on the target signal to obtain a fault characteristic frequency; fusing the fault feature frequency with a time domain feature, a frequency domain feature and a time-frequency domain feature in the track vibration signal to obtain a fusion feature vector, and constructing a fusion feature matrix according to the fusion feature vector; and inputting the fusion feature matrix into a preset deep belief network for iterative training to obtain a track fault recognition model, inputting the obtained real-time track vibration signal into the track fault recognition model, and outputting the track fault recognition model to obtain a track fault recognition result.

Description

Technical Field

[0001] The present invention belongs to the technical field of rail transit monitoring, and particularly relates to a method and system for identifying rail faults based on HSROA. Background Art

[0002] The development of high-speed railway systems has greatly improved traffic efficiency. However, its track structure faces multiple challenges during long-term operation. Due to factors such as mechanical stress generated by train operation, erosion of the natural environment, and material aging, various diseases often occur in the track, such as fatigue deterioration of the fastener system, cracks in the track slab, rail corrugation, and collapse of rail joints. These diseases not only threaten the safety and stability of train operation but may also cause serious traffic accidents, posing potential hazards to the lives and property of passengers.

[0003] Traditional rail fault detection methods, such as manual inspection and diagnostic methods based on vibration signals, can identify rail diseases to a certain extent, but have many limitations. Manual inspection relies on personnel experience and judgment, with low efficiency and being easily affected by subjective factors; while diagnostic methods based on vibration signals, although they can reflect changes in the track structure, are easily interfered by environmental noise and have limited ability to identify complex disease types. Therefore, it is particularly important to explore more efficient and accurate rail disease monitoring technologies. Summary of the Invention

[0004] The present invention provides a method and system for identifying rail faults based on HSROA to solve the technical problems of the limitations of traditional rail fault detection methods in terms of efficiency and accuracy.

[0005] In a first aspect, the present invention provides a method for identifying rail faults based on HSROA, including: Obtain rail vibration signals, and use an overband signal refining and optimizing algorithm as a pre-filter to perform the first filtering on the rail vibration signals to obtain pre-filtered narrowband signals; Perform EEMD decomposition on the narrowband signals to obtain at least one IMF component signal, calculate the band entropy values of the at least one IMF component signal, and select sensitive IMF component signals from the at least one IMF component signal according to the magnitudes of the band entropy values; Perform the second filtering on the sensitive IMF component signals according to a band-pass filter to obtain target signals, and perform envelope power spectrum analysis on the target signals to obtain fault characteristic frequencies; Fuse the fault characteristic frequencies with the time-domain characteristics, frequency-domain characteristics, and time-frequency domain characteristics in the rail vibration signals to obtain a fused feature vector, and construct a fused feature matrix according to the fused feature vector; Input the fused feature matrix into a preset deep belief network for iterative training to obtain an orbital fault identification model, and input the acquired real-time orbital vibration signal into the orbital fault identification model, and the orbital fault identification model outputs an orbital fault identification result.

[0006] In a second aspect, the present invention provides an orbital fault identification system based on HSROA, including: An acquisition module configured to acquire an orbital vibration signal and perform a first filtering on the orbital vibration signal by using an overband signal refinement and optimization algorithm as a pre-filter to obtain a pre-filtered narrowband signal; A calculation module configured to perform EEMD decomposition on the narrowband signal to obtain at least one IMF component signal, calculate the band entropy values of the at least one IMF component signal, and select sensitive IMF component signals from the at least one IMF component signal according to the magnitudes of the band entropy values; An analysis module configured to perform a second filtering on the sensitive IMF component signal by using a band-pass filter to obtain a target signal, and perform envelope power spectrum analysis on the target signal to obtain a fault characteristic frequency; A fusion module configured to fuse the fault characteristic frequency with the time-domain features, frequency-domain features, and time-frequency domain features in the orbital vibration signal to obtain a fused feature vector, and construct a fused feature matrix according to the fused feature vector; An output module configured to input the fused feature matrix into a preset deep belief network for iterative training to obtain an orbital fault identification model, and input the acquired real-time orbital vibration signal into the orbital fault identification model, and the orbital fault identification model outputs an orbital fault identification result.

[0007] In a third aspect, an electronic device is provided, which includes: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the steps of the orbital fault identification method based on HSROA according to any embodiment of the present invention.

[0008] In a fourth aspect, the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the program instructions are executed by a processor, the processor is enabled to execute the steps of the orbital fault identification method based on HSROA according to any embodiment of the present invention.

[0009] A method and system for rail fault identification based on HSROA in the present application effectively removes environmental noise and accurately screens out sensitive IMF components containing rail fault feature information by combining HSROA pre-filtering and EEMD decomposition. Envelope power spectrum analysis is used to extract fault characteristic frequencies, enhancing the recognition ability for complex disease types. By fusing time-domain, frequency-domain, and time-frequency-domain features, a comprehensive fusion feature matrix is constructed, significantly improving the feature expression ability and robustness. Finally, a fault identification model based on a deep belief network (DBN) realizes the intelligence and accuracy of rail fault identification, overcoming the limitations of traditional methods in terms of efficiency and accuracy, and providing a reliable technical guarantee for rail safety monitoring and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0011] Figure 1 is a flowchart of a method for rail fault identification based on HSROA provided by an embodiment of the present invention; Figure 2 is a structural block diagram of a system for rail fault identification based on HSROA provided by an embodiment of the present invention; Figure 3 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0012] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following clearly and completely describes the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0013] Please refer to Figure 1 , which shows a flowchart of a method for rail fault identification based on HSROA in the present application.

[0014] As Figure 1 shown, the method for rail fault identification based on HSROA (HyperBand Signal Refining Optimization Algorithm) specifically includes the following steps: Step S101: Obtain the track vibration signal, and use the over-band signal refining and optimization algorithm as a pre-filter to perform the first filtering on the track vibration signal to obtain a pre-filtered narrow-band signal.

[0015] In this step, the track vibration signal is decomposed into a first low-frequency signal and a first high-frequency signal through a low-pass filter and a high-pass filter by using multi-wavelet packet decomposition. The expression is: , , , , where, is the first low-frequency signal, are the low-pass filter coefficients, is the input signal sampling point index, is the output signal sampling point index, is the original track vibration signal, is the first high-frequency signal, are the high-pass filter coefficients, is the first low-frequency sub-signal, is the first high-frequency sub-signal, is the filter index, is the unit impulse function, is the low-frequency signal input to the next layer of decomposition, used to calculate the first low-frequency sub-signal ; The first low-frequency signal is further decomposed through a low-pass filter and a high-pass filter to obtain a first low-frequency sub-signal and a first high-frequency sub-signal, and the comprehensive score of each first low-frequency signal is calculated according to the non-linear combined energy and frequency characteristics of the first low-frequency sub-signal and the non-linear combined energy and frequency characteristics of the first high-frequency sub-signal. The expression is: , where, is the non-linear combined energy term, is the exponential decay term, is the frequency characteristic term, is the frequency characteristic exponent, is the comprehensive score of the i-th first low-frequency signal; The first low-frequency signals with comprehensive scores greater than the preset threshold are defined as target first low-frequency signals, and the target first low-frequency signals are signal-reconstructed through a preset reconstruction filter to obtain a pre-filtered narrow-band signal. Among them, the expression is: , where, is the narrowband signal after pre-filtering, is the target first low-frequency signal, is the optimized reconstruction low-pass filter coefficient, is the sampling point position of the first low-frequency sub-signal in the convolution operation, is the attenuation coefficient, is the comprehensive score of each first low-frequency signal, is the time window function, is the noise suppression term, is the optimized reconstruction high-pass filter coefficient, is the sampling point position of the first high-frequency sub-signal in the convolution operation.

[0016] It should be noted that the particle swarm is initialized and a group of particles are randomly generated. Each particle represents a group of reconstruction filter coefficients and , that is, , and each particle contains the corresponding position and velocity and the initial values are random values. Combining the two indicators of mean square error and signal-to-noise ratio, a comprehensive fitness function is created, and the expression is: , In the formula, is the comprehensive fitness function, is the signal-to-noise ratio, is the weight exponent of the signal-to-noise ratio, is the attenuation coefficient, is the mean square error, is the adjustment exponent of the mean square error, is the weight, and are learning factors; According to the calculation results of the fitness function, update the individual optimal position and the global optimal position. Then, according to the individual optimal position and the global optimal position, continuously update the velocity and position of the particles until the iteration reaches the preset number of times and then terminate. Obtain the final particle velocity and position. At the same time, obtain the optimized filter coefficient, and the expression is: , , , , In the formula, is the individual optimal position, is the global optimal position, is the position of particle i at time t, is the particle at position The comprehensive fitness value at is the comprehensive fitness value of the particle at the individual optimal position, is the velocity of the updated particle, is the weight, is the velocity of particle i at time t, and are the learning factors, and are random factors within [0, 1], is the position of the updated particle.

[0017] In step S102, perform EEMD decomposition on the narrowband signal to obtain at least one IMF component signal, calculate the band entropy values of the at least one IMF component signal, and select sensitive IMF component signals from the at least one IMF component signal according to the magnitudes of the band entropy values.

[0018] In this step, add Gaussian white noise to the pre-filtered narrowband signal and perform m times of EMD decomposition. Each decomposition obtains a series of initial intrinsic mode function (IMF) components. Calculate the mean values of the IMF components corresponding to the same mode in the m times of decomposition, and finally obtain a stable EEMD decomposition result, that is, a series of final IMF components. The expression is: , , where, is the signal after adding white noise for the mth time, is the amplitude coefficient of the white noise, is the Gaussian white noise sequence added for the mth time, is the kth IMF component signal obtained from the mth decomposition, is the residual term of the mth decomposition, k is the kth IMF component, and n is the total number of IMF component signals, is the kth IMF component signal, is the pre-filtered narrowband signal; Segment into segments, each segment with a length of L, to obtain , where is the value of the last sampling point of , is the value of the second sampling point of , is the value of the first sampling point of , and calculate the energy of each segment in , and calculate the proportion of the energy of each segment in the total energy , according to Calculate the band entropy value of the K-th IMF component signal, and the expression is: , In the formula, is the energy of each segment, is the proportion of the energy of each segment in the total energy, is the band entropy value, is the number of segments, is the value of the j-th sampling point of is the length of each segment; Normalize the IMF component signal whose band entropy value is greater than the entropy value threshold to obtain the sensitive IMF component signal, and the expression is: , In the formula, is the sensitive IMF component signal, is the parameter value that makes the objective function reach the maximum value, is the weight, is the band entropy value, is the total number of IMF components, is the indicator function, is the entropy value threshold.

[0019] Step S103, perform a second filtering on the sensitive IMF component signal according to the band-pass filter to obtain the target signal, and perform envelope power spectrum analysis on the target signal to obtain the fault characteristic frequency.

[0020] In this step, perform a fast Fourier transform on the sensitive IMF component signal to determine the passband frequency range , and the expression is: , In the formula, is the lower limit of the passband frequency, is the upper limit of the passband frequency, is the frequency range that makes the objective function reach the maximum value in the frequency interval [f1, f2], is the lower limit of the frequency range, is the upper limit of the frequency range, is the sensitive IMF component signal, is the Fourier transform of the sensitive IMF component signal; Design a band-pass filter and optimize the coefficients of the band-pass filter using the least mean square error algorithm to minimize the mean square error between the output of the band-pass filter and the output of the ideal band-pass filter, and apply the designed band-pass filter to the sensitive IMF component signal to obtain the target signal after the second filtering, and the expression is: , , , wherein, is the frequency response function, is the upper limit of the passband frequency, is the frequency, is the lower limit of the passband frequency, is the band-pass filter coefficient, is the parameter value that minimizes the objective function, is the target signal after the second filtering; Perform envelope demodulation on the target signal after the second filtering to obtain an envelope signal, and perform power spectrum analysis on the envelope signal to obtain the power spectral density , find the peak in the power spectrum, and the corresponding frequency is the fault characteristic frequency , the expression is: , , , wherein, is the signal 's envelope, is the target signal after the second filtering, is the width of the moving average window, is the time of the signal at, is the Hilbert transform, is the power spectral density of the envelope signal, is the Fourier transform of the envelope signal, is the variance of the envelope signal, is the number of sampling points of the signal.

[0021] Step S104, fuse the fault characteristic frequency with the time-domain characteristics, frequency-domain characteristics, and time-frequency domain characteristics in the track vibration signal to obtain a fused feature vector, and construct a fused feature matrix according to the fused feature vector.

[0022] In this step, obtain the time-domain characteristics in the track vibration signal. The time-domain characteristics include the mean, standard deviation, mean square value, skewness, and kurtosis. The expression is: , wherein, is the time-domain characteristic, is the number of samples, is the i-th sample value, is the sample mean; Calculate the central frequency and the square frequency moment in the orbital vibration signal, integrate the central frequency and the square frequency moment to obtain a frequency-domain feature, and the expression is: , where, is the central frequency, is the square frequency moment, is the frequency at the k-th frequency point, is the power spectral density; Perform continuous wavelet transform on the orbital vibration signal to obtain at least one wavelet coefficient, calculate the energy and entropy of the at least one wavelet coefficient, integrate each energy and entropy to obtain a time-frequency domain feature, and the expression is: , where, is the time-frequency domain feature, is the energy, is the entropy, is the wavelet coefficient, is the translation factor, is the exponential decay function; Perform in-domain fusion on the time-domain feature, the frequency-domain feature, and the time-frequency domain feature to obtain a scalar value, perform weighted fusion on the scalar value and the fault characteristic frequency to obtain a fused feature vector, and integrate the fused feature vector to obtain a fused feature matrix, and the expression is: , , , where, is the result of time-domain feature fusion, is the mean feature function, is the standard deviation feature function, is the mean square value feature function, is the skewness feature function, , , , , are different weight coefficients in the time-domain feature fusion, is the kurtosis feature function, is the result of frequency-domain feature fusion, is the central frequency feature function, and are different weight coefficients in the frequency-domain feature fusion, is the square frequency feature function, is the result of time-frequency domain feature fusion, is the energy feature function, and are different weight coefficients in the time-frequency domain feature fusion, is the entropy feature function, is the final fused feature, is the weight coefficient of the time-domain feature, is the weight coefficient of the frequency-domain feature, is the weight coefficient of the time-frequency domain feature, is the weight coefficient of the fault feature, is the fault feature frequency, is the fused feature vector, is the q-th element in the fused feature vector.

[0023] Step S105: Input the fused feature matrix into a preset deep belief network for iterative training to obtain an orbit fault recognition model, and input the acquired real-time orbit vibration signal into the orbit fault recognition model, and the orbit fault recognition model outputs an orbit fault recognition result.

[0024] In this step, the deep belief network consists of an input layer, a hidden layer, and an output layer. Among them, the number of nodes in the input layer should be equal to the dimension of the feature vector, and the number of nodes in the output layer should be equal to the number of sample categories. If there are 4 types of faults, the number of output layers is 4, that is, a four-dimensional vector. Regarding the determination of the number of hidden layers, there is an empirical calculation formula: , wherein, is the number of input layers, is the number of hidden layers, is the number of output layers, q is an adjustment constant from 1 to 10, and z ranges from 8 to 17; The deep belief network contains a composite loss function based on information entropy and feature weights, and the expression is: , wherein, is the loss function, is the weight coefficient of the first term, is the true label of the i-th sample, is the predicted probability of the i-th sample, is the weight coefficient of the second term, is the number of samples, is the number of categories, is the weight of the i-th sample on the j-th feature, is the predicted probability of the i-th sample on the j-th feature, is the weight coefficient of the third term, is the weight of the j-th feature, is the target weight of the j-th feature.

[0025] The particle swarm optimization algorithm is used to optimize the deep belief network. Initialize the particle positions and velocities. Represent the weights and thresholds of the deep belief network as the position vectors of the particles. At the same time, determine the fitness function of the particle swarm algorithm to evaluate the fitness values of each particle. The higher the fitness value, the better the position of the particle. The expression of the fitness function is: , In the formula, is the fitness function, and are different adjustment parameters for controlling the sensitivity of the sigmoid function, is the classification accuracy, is the target threshold of the classification accuracy, is the cross-entropy error, is the target threshold of the cross-entropy error, is the regularization term weight, is the number of model parameters, is the number of samples, is the model complexity weight, is the regularization term; Introduce the intelligent core hybrid dynamic strategy (IHDS). Divide the particles into dominant particles and inferior particles, and set different inertia weights and learning factors for different types of particles. Dominant particles use larger inertia weights to enhance the global search ability, and inferior particles use smaller inertia weights to enhance the local search ability. The expression is: , , In the formula, is the inertia weight of the i-th particle, is the minimum value of the inertia weight, is the maximum value of the inertia weight, is the maximum number of iterations, is the current number of iterations, is the fitness value of the i-th particle, is the average fitness value of the population, is the maximum fitness value of the population, is the learning factor is the maximum value of is the learning factor is the minimum value of is the learning factor is the minimum value of is the learning factor is the maximum value of The learning factor of the i-th particle , is the learning factor of the i-th particle ; Intelligently update the velocity and position of each particle according to the inertia weight and learning factor adjusted by the intelligent core hybrid dynamic strategy (IHDS). The expression is as follows: , , wherein, is the velocity of the updated particle, is the velocity of particle i at time t, is the individual optimal position, and are random factors within [0, 1], is the position of particle i at time t, is the global optimal position, is the position of the updated particle; Introduce the intelligent core boundary termination strategy (ICBS) to judge whether the particle exceeds the boundary. If it exceeds, reflect its position back into the search space, set the minimum number of iterations. If the termination condition is judged successfully and the number of iterations is greater than the minimum number of iterations, the program runs to completion. The expression is as follows: , , , wherein, is the upper boundary of the search space, is the reflection coefficient, is the intensity coefficient of random perturbation, is a random number within the interval [0, 1], is the lower boundary of the search space, is the global optimal fitness value, is the fitness value threshold, is the standard deviation, is the standard deviation threshold, is the iteration stop number threshold, is the current iteration number, is the minimum iteration number.

[0026] In summary, the method of this application combines HSROA pre-filtering with EEMD decomposition to effectively remove environmental noise and accurately screen out sensitive IMF components containing track fault feature information. Envelope power spectrum analysis is used to extract fault characteristic frequencies, enhancing the recognition ability for complex disease types. By fusing time-domain, frequency-domain, and time-frequency-domain features to construct a comprehensive fusion feature matrix, the expression ability and robustness of the features are significantly improved. Finally, the fault recognition model based on the deep belief network (DBN) realizes the intelligence and accuracy of track fault recognition, overcomes the limitations of traditional methods in terms of efficiency and accuracy, and provides a reliable technical guarantee for track safety monitoring and maintenance.

[0027] Please refer to Figure 2 , which shows a structural block diagram of a track fault recognition system based on HSROA of this application.

[0028] As Figure 2 shown, the track fault recognition system 200 includes an acquisition module 210, a calculation module 220, an analysis module 230, a fusion module 240, and an output module 250.

[0029] Among them, the acquisition module 210 is configured to acquire a track vibration signal and perform a first filtering on the track vibration signal using an overband signal refinement optimization algorithm as a pre-filter to obtain a pre-filtered narrowband signal; the calculation module 220 is configured to perform EEMD decomposition on the narrowband signal to obtain at least one IMF component signal, calculate the band entropy value of the at least one IMF component signal, and select a sensitive IMF component signal from the at least one IMF component signal according to the size of the band entropy value; the analysis module 230 is configured to perform a second filtering on the sensitive IMF component signal according to a band-pass filter to obtain a target signal, and perform envelope power spectrum analysis on the target signal to obtain a fault characteristic frequency; the fusion module 240 is configured to fuse the fault characteristic frequency with the time-domain feature, frequency-domain feature, and time-frequency-domain feature in the track vibration signal to obtain a fusion feature vector, and construct a fusion feature matrix according to the fusion feature vector; the output module 250 is configured to input the fusion feature matrix into a preset deep belief network for iterative training to obtain a track fault recognition model, and input the acquired real-time track vibration signal into the track fault recognition model, and the track fault recognition model outputs a track fault recognition result.

[0030] It should be understood that Figure 2 the various modules described in Figure 1 correspond to the respective steps in the method described in reference Figure 2 . Therefore, the operations, features, and corresponding technical effects described above for the method also apply to

[0031] In some other embodiments, the embodiments of the present invention further provide a computer-readable storage medium, on which a computer program is stored. When the program instructions are executed by a processor, the processor is enabled to execute the HSROA-based rail fault identification method in any of the above method embodiments; As an implementation manner, the computer-readable storage medium of the present invention stores computer-executable instructions, which are set as: Obtain a rail vibration signal, and use an overband signal refinement optimization algorithm as a pre-filter to perform a first filtering on the rail vibration signal to obtain a pre-filtered narrowband signal; Perform EEMD decomposition on the narrowband signal to obtain at least one IMF component signal, calculate the band entropy values of the at least one IMF component signal, and select a sensitive IMF component signal from the at least one IMF component signal according to the magnitude of the band entropy value; Perform a second filtering on the sensitive IMF component signal according to a band-pass filter to obtain a target signal, and perform envelope power spectrum analysis on the target signal to obtain a fault characteristic frequency; Fuse the fault characteristic frequency with the time-domain characteristics, frequency-domain characteristics, and time-frequency domain characteristics in the rail vibration signal to obtain a fused feature vector, and construct a fused feature matrix according to the fused feature vector; Input the fused feature matrix into a preset deep belief network for iterative training to obtain a rail fault identification model, and input the obtained real-time rail vibration signal into the rail fault identification model, and the rail fault identification model outputs a rail fault identification result.

[0032] The computer-readable storage medium may include a program storage area and a data storage area. Among them, the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the HSROA-based rail fault identification system, etc. In addition, the computer-readable storage medium may include a high-speed random access memory, and may further include a memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the computer-readable storage medium may optionally include a memory remotely provided with respect to the processor, and these remote memories may be connected to the HSROA-based rail fault identification system through a network. Examples of the above networks include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0033] Figure 3 is a schematic structural diagram of the electronic device provided by the embodiments of the present invention, as Figure 3As shown in the figure, the device includes a processor 310 and a memory 320. The electronic device may further include an input device 330 and an output device 340. The processor 310, the memory 320, the input device 330, and the output device 340 may be connected through a bus or other means. Figure 3 Taking the connection through the bus as an example. The memory 320 is the above-mentioned computer-readable storage medium. The processor 310 executes various functional applications and data processing of the server by running non-volatile software programs, instructions, and modules stored in the memory 320, that is, implementing the above-mentioned method embodiment of the HSROA-based track fault identification method. The input device 330 can receive input digital or character information, and generate key signal inputs related to the user settings and function controls of the HSROA-based track fault identification system. The output device 340 may include a display device such as a display screen.

[0034] The above electronic device can execute the method provided by the embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the method. For technical details not described in detail in this embodiment, reference can be made to the method provided by the embodiment of the present invention.

[0035] As an implementation manner, the above electronic device is applied to an HSROA-based track fault identification system and is used for a client, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Obtain a track vibration signal, and use an overband signal refinement optimization algorithm as a pre-filter to perform a first filtering on the track vibration signal to obtain a pre-filtered narrowband signal; Perform EEMD decomposition on the narrowband signal to obtain at least one IMF component signal, calculate the band entropy values of the at least one IMF component signal, and select sensitive IMF component signals from the at least one IMF component signal according to the magnitudes of the band entropy values; Perform a second filtering on the sensitive IMF component signal according to a band-pass filter to obtain a target signal, and perform envelope power spectrum analysis on the target signal to obtain a fault characteristic frequency; Fuse the fault characteristic frequency with the time-domain characteristics, frequency-domain characteristics, and time-frequency domain characteristics in the track vibration signal to obtain a fused feature vector, and construct a fused feature matrix according to the fused feature vector; Input the fused feature matrix into a preset deep belief network for iterative training to obtain a track fault identification model, and input the obtained real-time track vibration signal into the track fault identification model, and the track fault identification model outputs a track fault identification result.

[0036] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiments.

[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present invention.

Claims

1. An HSROA-based method for identifying track faults, characterized in that Including: Obtain the track vibration signal, and use the overband signal refinement and optimization algorithm as a pre-filter to perform the first filtering on the track vibration signal to obtain a pre-filtered narrow-band signal; Perform EEMD decomposition on the narrow-band signal to obtain at least one IMF component signal, calculate the band entropy value of the at least one IMF component signal, and select a sensitive IMF component signal from the at least one IMF component signal according to the magnitude of the band entropy value; Perform the second filtering on the sensitive IMF component signal according to a band-pass filter to obtain a target signal, and perform envelope power spectrum analysis on the target signal to obtain a fault characteristic frequency; Fuse the fault characteristic frequency with the time-domain characteristics, frequency-domain characteristics, and time-frequency domain characteristics in the track vibration signal to obtain a fused feature vector, and construct a fused feature matrix according to the fused feature vector; Input the fused feature matrix into a preset deep belief network for iterative training to obtain a track fault identification model, and input the obtained real-time track vibration signal into the track fault identification model, and the track fault identification model outputs a track fault identification result.

2. The method for identifying track faults based on HSROA according to claim 1, wherein, The step of using the overband signal refinement and optimization algorithm as a pre-filter to perform the first filtering on the track vibration signal to obtain a pre-filtered narrow-band signal includes: Use multi-wavelet packet decomposition to decompose the track vibration signal into a first low-frequency signal and a first high-frequency signal through a low-pass filter and a high-pass filter, and the expression is: ， ， ， ， Wherein, is the first low-frequency signal, are the low-pass filter coefficients, is the input signal sampling point index, is the output signal sampling point index, is the original track vibration signal, is the first high-frequency signal, are the high-pass filter coefficients, is the first low-frequency sub-signal, is the first high-frequency sub-signal, is the filter index, is the unit impulse function, is the low-frequency signal input to the next layer of decomposition for calculating the first low-frequency sub-signal ; Decompose the first low-frequency signal through a low-pass filter and a high-pass filter again to obtain a first low-frequency sub-signal and a first high-frequency sub-signal, and calculate the comprehensive score of each first low-frequency signal according to the non-linear combined energy and frequency characteristics of the first low-frequency sub-signal and the non-linear combined energy and frequency characteristics of the first high-frequency sub-signal, and the expression is: ， In the formula, is the non-linear combination energy term, is the exponential decay term, is the frequency feature term, is the frequency feature exponent, is the comprehensive score of the i-th first low-frequency signal; Define the first low-frequency signal with a comprehensive score greater than a preset threshold as the target first low-frequency signal, and perform signal reconstruction on the target first low-frequency signal through a preset reconstruction filter to obtain a pre-filtered narrow-band signal, where the expression is: ， Wherein, is the narrowband signal after pre-filtering, is the target first low-frequency signal, is the optimized reconstruction low-pass filter coefficient, is the sampling point position of the first low-frequency sub-signal in the convolution operation, is the attenuation coefficient, is the comprehensive score of each first low-frequency signal, is the time window function, is the noise suppression term, is the optimized reconstruction high-pass filter coefficient, is the sampling point position of the first high-frequency sub-signal in the convolution operation.

3. The method for identifying track faults based on HSROA according to claim 2, wherein Where The filter coefficients of the reconstruction filter are optimized by an improved PSO algorithm, specifically including: Initialize the particle swarm and randomly generate a group of particles, where each particle represents a set of reconstruction filter coefficients and , that is , each particle contains the corresponding position and velocity with initial values being random values. Combining the two metrics of mean square error and signal-to-noise ratio, create a comprehensive fitness function, and the expression is as follows: ， In the formula, is the comprehensive fitness function, is the signal-to-noise ratio, is the weight exponent of the signal-to-noise ratio, is the attenuation coefficient, is the mean square error, is the adjustment exponent of the mean square error, is the weight, and are the learning factors; Update the individual optimal position and the global optimal position according to the calculation result of the fitness function, and then continuously update the velocity and position of the particle according to the individual optimal position and the global optimal position until the iteration reaches a preset number of times and then terminate to obtain the final particle velocity and position. At the same time, obtain the optimized filter coefficients, and the expression is: ， ， ， ， Wherein, is the individual optimal position, is the global optimal position, is the position of particle i at time t, is the comprehensive fitness value of the particle at position , is the comprehensive fitness value of the particle at the individual optimal position, is the velocity of the updated particle, is the weight, is the velocity of particle i at time t, and are the learning factors, and are random factors within [0, 1], is the position of the updated particle.

4. A method for identifying track faults based on HSROA according to claim 1, characterized in that The step of calculating the band entropy value of the at least one IMF component signal and selecting a sensitive IMF component signal from the at least one IMF component signal according to the magnitude of the band entropy value includes: Denote as , is the K-th IMF component signal, is the value at the first sampling point, is the value at the second sampling point, is the value at the z-th sampling point; Pair Segment it into segments, each with a length of L, to obtain , where is the value of the last sampling point of is the value of the second sampling point of is the value of the first sampling point of , and calculate the energy of each segment in , and calculate the proportion of the energy of each segment in the total energy . According to calculate the band entropy value of the Kth IMF component signal, and the expression is: ， In the formula, is the energy of each segment, is the proportion of the energy of each segment in the total energy, is the frequency band entropy value, is the number of segments, is the value of the j-th sampling point of is the length of each segment; Perform normalization processing on the IMF component signal with a band entropy value greater than the entropy threshold to obtain a sensitive IMF component signal, and the expression is: ， In the formula, is the sensitive IMF component signal, is the parameter value that maximizes the objective function, is the weight, is the band entropy value, is the total number of IMF components, is the indicator function, is the entropy threshold value.

5. A method for identifying track faults based on HSROA according to claim 1, characterized in that The step of performing the second filtering on the sensitive IMF component signal according to a band-pass filter to obtain a target signal, and performing envelope power spectrum analysis on the target signal to obtain a fault characteristic frequency includes: Perform a fast Fourier transform on the sensitive IMF component signal to determine the passband frequency range , and the expression is: ， Wherein, is the lower limit of the passband frequency, is the upper limit of the passband frequency, is the frequency range within the frequency interval [f1, f2] that maximizes the objective function, is the lower limit of the frequency range, is the upper limit of the frequency range, is the sensitive IMF component signal, is the Fourier transform of the sensitive IMF component signal; Design a band-pass filter and use the least mean square error algorithm to optimize the coefficients of the band-pass filter , so as to minimize the mean square error between the output of the band-pass filter and the output of the ideal band-pass filter. Apply the designed band-pass filter to the sensitive IMF component signal to obtain the target signal after the second filtering. The expression is as follows: ， ， ， In the formula, is the frequency response function, is the upper limit of the passband frequency, is the frequency, is the lower limit of the passband frequency, is the band-pass filter coefficient, is the parameter value that makes the objective function reach the minimum value, is the target signal after the second filtering; The envelope demodulation is performed on the target signal after the second filtering to obtain an envelope signal, and the power spectral analysis is performed on the envelope signal to obtain the power spectral density , and the peak is searched in the power spectrum, and the corresponding frequency is the fault characteristic frequency , and the expression is as follows: ， ， ， Wherein, is the envelope of the signal ; is the target signal after the second filtering; is the width of the moving average window; is the value of the signal at time ; is the Hilbert transform; is the power spectral density of the envelope signal; is the Fourier transform of the envelope signal; is the variance of the envelope signal; is the number of sampling points of the signal.

6. The method for identifying track faults based on HSROA according to claim 1, wherein, Fusing the fault characteristic frequency with time-domain features, frequency-domain features, and time-frequency domain features in the rail vibration signal to obtain a fused feature vector, and constructing a fused feature matrix based on the fused feature vector includes: Obtaining the time-domain features in the rail vibration signal, where the time-domain features include mean, standard deviation, mean square value, skewness, and kurtosis, and the expressions are: ， In the formula, is the time-domain feature, is the number of samples, is the value of the i-th sample, is the sample mean; Calculating the central frequency and square frequency moments in the rail vibration signal, integrating the central frequency and the square frequency moments to obtain frequency-domain features, and the expressions are: ， In the formula, is the center frequency, is the square frequency moment, is the frequency of the k-th frequency point, is the power spectral density of; Performing continuous wavelet transform on the rail vibration signal to obtain at least one wavelet coefficient, calculating the energy and entropy of the at least one wavelet coefficient, and integrating each energy and entropy to obtain time-frequency domain features, and the expressions are: ， In the formula, is the time-frequency domain feature, is the energy, is the entropy, is the wavelet coefficient, is the translation factor, is the exponential decay function; Performing in-domain fusion on the time-domain features, the frequency-domain features, and the time-frequency domain features to obtain scalar values, performing weighted fusion on the scalar values and the fault characteristic frequency to obtain a fused feature vector, and integrating the fused feature vector to obtain a fused feature matrix, and the expressions are: ， ， ， In the formula, is the result of time-domain feature fusion, is the mean feature function, is the standard deviation feature function, is the mean square value feature function, is the skewness feature function, , , , , are different weight coefficients in time-domain feature fusion, is the kurtosis feature function, is the result of frequency-domain feature fusion, is the center frequency feature function, and are different weight coefficients in frequency-domain feature fusion, is the square frequency feature function, is the result of time-frequency domain feature fusion, is the energy feature function, and are different weight coefficients in time-frequency domain feature fusion, is the entropy feature function, is the final fusion feature, is the weight coefficient of time-domain feature, is the weight coefficient of frequency-domain feature, is the weight coefficient of time-frequency domain feature, is the weight coefficient of fault feature, is the fault feature frequency, is the fusion feature vector, is the q-th element in the fusion feature vector.

7. A method for identifying track faults based on HSROA according to claim 1, characterized in that, Wherein, The deep belief network includes a composite loss function based on information entropy and feature weights, and the expressions are: ， In the formula, is the loss function, is the weight coefficient of the first term, is the true label of the i-th sample, is the predicted probability of the i-th sample, is the weight coefficient of the second term, is the number of samples, is the number of classes, is the weight of the i-th sample on the j-th feature, is the predicted probability of the i-th sample on the j-th feature, is the weight coefficient of the third term, is the weight of the j-th feature, is the target weight of the j-th feature.

8. An HSROA-based track fault identification system, characterized in that, Including: An acquisition module configured to acquire a rail vibration signal and perform first filtering on the rail vibration signal using an overband signal refinement optimization algorithm as a pre-filter to obtain a pre-filtered narrowband signal; A calculation module configured to perform EEMD decomposition on the narrowband signal to obtain at least one IMF component signal, calculate the band entropy value of the at least one IMF component signal, and select a sensitive IMF component signal from the at least one IMF component signal according to the magnitude of the band entropy value; An analysis module configured to perform second filtering on the sensitive IMF component signal according to a band-pass filter to obtain a target signal, and perform envelope power spectrum analysis on the target signal to obtain a fault characteristic frequency; A fusion module configured to fuse the fault characteristic frequency with time-domain features, frequency-domain features, and time-frequency domain features in the rail vibration signal to obtain a fused feature vector, and construct a fused feature matrix based on the fused feature vector; An output module configured to input the fused feature matrix into a preset deep belief network for iterative training to obtain a rail fault identification model, and input the acquired real-time rail vibration signal into the rail fault identification model, and the rail fault identification model outputs a rail fault identification result.

9. An electronic device, characterized in that, Including: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the method according to any one of claims 1 to 7.

Images