Intelligent track fault diagnosis method and system based on improved CWT-CNN

Abstract

The invention discloses an intelligent track fault diagnosis method and system based on an improved CWT-CNN, and the method comprises the steps: decomposing a steel rail vibration signal according to continuous wavelet transform, obtaining target steel rail vibration sub-signals containing fault features, and calculating the wavelet entropy of each target steel rail vibration sub-signal; reconstructing the wavelet entropy according to an entropy dissimilation reconstruction comparison strategy to obtain a target wavelet entropy; inputting the steel rail vibration signal into a preset compression excitation resonance network, outputting by the compression excitation resonance network to obtain feature information at the same latitude as the target wavelet entropy, and fusing the feature information with the target wavelet entropy to obtain a fused feature signal; and performing iterative training on the optimized convolutional neural network according to the fused feature signal and the fault tag corresponding to the fused feature signal to obtain a track fault diagnosis model. The track fault type identification accuracy can be improved, and the real-time diagnosis of the track fault is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of rail transit monitoring, and in particular relates to an intelligent rail fault diagnosis method and system based on improved CWT-CNN. Background Art

[0002] With the continuous development of railway transportation technology and the continuous growth of the number of railway trains, the importance of railway rail fault diagnosis technology has become increasingly prominent. Rail failure not only affects the safety of train operation, but may also cause serious accidents. There are various types of rail failures, including cracks, wear, and breakage of the rails, which may be caused by a variety of factors. First, high-intensity train operation will accelerate the wear of the rails and increase the risk of failure; second, severe weather conditions such as heavy rain, snow, and freezing may cause damage to the rails, thereby increasing the probability of failure; finally, insufficient maintenance and inspection may lead to long-term wear of the rails.

[0003] Convolutional neural network (CNN) is one of the most commonly used networks in intelligent diagnosis systems. It can automatically extract the intrinsic features of data. However, in the feature learning of CNN, more attention is paid to pure time domain or frequency domain data. The performance of CWT is highly dependent on wavelet basis functions, and its generalization ability is limited. CWT-CNN ignores multimodal data, and the framework lacks an effective multi-source information fusion mechanism, which restricts the comprehensiveness of diagnosis. Summary of the invention

[0004] The present invention provides an intelligent rail fault diagnosis method and system based on improved CWT-CNN, which are used to solve the technical problems of ignoring multimodal data, weak generalization ability, lack of information fusion mechanism, and restricting the comprehensiveness of diagnosis.

[0005] In a first aspect, the present invention provides an intelligent rail fault diagnosis method based on an improved CWT-CNN, comprising: Acquire rail vibration signals based on a preset multi-dimensional self-adjusting vibration signal acquisition strategy; Decomposing the rail vibration signal according to continuous wavelet transform to obtain target rail vibration sub-signals containing fault characteristics, and calculating the wavelet entropy of each target rail vibration sub-signal; Reconstruct the wavelet entropy according to the entropy alienation reconstruction comparison strategy to obtain the target wavelet entropy; The rail vibration signal is input into a preset compression excitation resonance network, the compression excitation resonance network outputs characteristic information of the same latitude as the target wavelet entropy, and the characteristic information is fused with the target wavelet entropy to obtain a fused characteristic signal; Iteratively training the optimized convolutional neural network according to the fused feature signal and the fault label corresponding to the fused feature signal to obtain a track fault diagnosis model; A real-time rail vibration signal is acquired, and the real-time rail vibration signal is input into the track fault diagnosis model, and the track fault diagnosis model outputs a fault diagnosis result.

[0006] In a second aspect, the present invention provides an intelligent track fault diagnosis system based on an improved CWT-CNN, comprising: An acquisition module configured to acquire rail vibration signals based on a preset multi-dimensional self-adjusting vibration signal acquisition strategy; A decomposition module is configured to decompose the rail vibration signal according to continuous wavelet transform to obtain target rail vibration sub-signals containing fault characteristics, and calculate the wavelet entropy of each target rail vibration sub-signal; A reconstruction module, configured to reconstruct the wavelet entropy according to an entropy dissimilation reconstruction comparison strategy to obtain a target wavelet entropy; A first output module is configured to input the rail vibration signal into a preset compression excitation resonance network, the compression excitation resonance network outputs characteristic information of the same latitude as the target wavelet entropy, and fuses the characteristic information with the target wavelet entropy to obtain a fused characteristic signal; A training module, configured to iteratively train the optimized convolutional neural network according to the fused feature signal and the fault label corresponding to the fused feature signal to obtain a track fault diagnosis model; The second output module is configured to obtain a real-time rail vibration signal and input the real-time rail vibration signal into the track fault diagnosis model, and the track fault diagnosis model outputs a fault diagnosis result.

[0007] In a third aspect, an electronic device is provided, comprising: 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 perform the steps of the intelligent track fault diagnosis method based on improved CWT-CNN of any embodiment of the present invention.

[0008] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the program instructions are executed by a processor, the processor executes the steps of the intelligent track fault diagnosis method based on improved CWT-CNN of any embodiment of the present invention.

[0009] The intelligent rail fault diagnosis method and system based on improved CWT-CNN of the present application reconstructs wavelet entropy according to the entropy alienation reconstruction comparison strategy to obtain target wavelet entropy; inputs the rail vibration signal into a preset compression excitation resonance network, and the compression excitation resonance network outputs feature information with the same latitude as the target wavelet entropy, and fuses the feature information with the target wavelet entropy to obtain a fused feature signal; iteratively trains the optimized convolutional neural network according to the fused feature signal and the fault label corresponding to the fused feature signal to obtain a rail fault diagnosis model, which achieves high precision, strong adaptability, low latency and strong robustness in rail fault diagnosis, and can be widely used in preventive maintenance of rail transportation to improve the level of safe rail operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 A flowchart of an intelligent rail fault diagnosis method based on an improved CWT-CNN provided by an embodiment of the present invention; Figure 2 A structural block diagram of an intelligent rail fault diagnosis system based on an improved CWT-CNN provided by an embodiment of the present invention; Figure 3 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0013] See also Figure 1 , which shows a flow chart of an intelligent track fault diagnosis method based on improved CWT-CNN of the present application.

[0014] like Figure 1 As shown in FIG. 1 , the intelligent track fault diagnosis method based on the improved CWT-CNN specifically includes the following steps: Step S101, obtaining rail vibration signals based on a preset multi-dimensional self-adjusting vibration signal acquisition strategy.

[0015] In this step, the multi-dimensional adaptive vibration signal acquisition system uses IEPE vibration sensors to collect signals and obtain rail vibration signals; the vibration signal sensor performs signal preprocessing; in order to prevent the collected vibration signal from being confused in mode and causing signal distortion in the system, a multi-dimensional self-adjusting vibration signal acquisition strategy is proposed. The expression is: , , In the formula, is the signal adjustment degree, is the center frequency, is the sampling frequency, is the bandwidth of the collected rail vibration signal, is the reference frequency of the kth component dimension, is the dynamic weight of the kth signal component, is the weight adjustment parameter, is the filter order, is the number of signal sampling points, is the lower limit of angular frequency, is the upper limit of angular frequency, is the angular frequency, is the number of collected signal components, is the low-frequency component of the signal’s main frequency, It is the high frequency component of the signal.

[0016] Step S102, decomposing the rail vibration signal according to continuous wavelet transform to obtain target rail vibration sub-signals containing fault characteristics, and calculating the wavelet entropy of each target rail vibration sub-signal.

[0017] In this step, the signal analysis mother wavelet is selected. Morlet wavelet is one of the most commonly used mother wavelets, and its expression is: , , In the formula, is the Morlet mother wavelet, is an imaginary unit, is the center frequency, For time, is the complex conjugate form of the Morlet mother wavelet; The rail vibration signal is subjected to continuous wavelet transform (CWT) according to the mother wavelet, and the coefficient matrix of the continuous wavelet transform is calculated, and the expression is: , In the formula, is the coefficient matrix of continuous wavelet transform, is the rail vibration signal, is the scale factor, is the translation factor; The coefficient matrix of the continuous wavelet transform is smoothed according to the time window function to obtain the smoothed target coefficient matrix, which is expressed as: , In the formula, is the Gaussian smoothing window function in the time domain, is the target coefficient matrix after smoothing; The wavelet entropy of the target rail vibrator signal containing the fault characteristics is calculated according to the target coefficient matrix, and the expression is: , , , In the formula, is the wavelet entropy of the target rail vibration sub-signal containing the fault characteristics, is the normalized energy distribution, is the logarithmic correction value, is the regularization factor, is the total number of decomposition scales, For scale The energy on For scale The energy on For scale and time location The target coefficient matrix after smoothing.

[0018] Step S103, reconstructing the wavelet entropy according to the entropy alienation reconstruction comparison strategy to obtain the target wavelet entropy.

[0019] In this step, the wavelet entropy is alienated to obtain alienated wavelet entropy, which is expressed as: , In the formula, is the wavelet entropy after alienation, is the number of signal components, is the alienation mode weight, For the alienation factor, is the wavelet entropy, is the frequency component, is the alienation time index, is the initial phase of the i-th component in the signal, is the alienation phase offset, is the adjustment coefficient, for The conjugated form of The dynamic characteristics of alienated wavelet entropy in frequency domain and time domain are decomposed according to m modes and n steps, and the modal vector of alienated wavelet entropy is obtained, which is expressed as: , In the formula, is the modal vector of the n-step alienated wavelet entropy of the m-th mode, is the wavelet entropy of the mth mode after alienation, is the right vector weight, is the left vector weight, For modal The left eigenvector of For the modal The relevant eigenvalues, for The complex conjugate form of For modal The right eigenvector of dimensional hyperbolic cosine function, is the frequency variable, is the time step, is the time interval; The alienated wavelet entropy modal vector is placed in the high-dimensional space in the increasing order of modal decomposition and step decomposition, and the wavelet entropy is reconstructed through the high-dimensional form matrix to obtain the reconstructed wavelet entropy, which is expressed as: , In the formula, To reconstruct wavelet entropy; Selected over-limit threshold accuracy , and judge whether the wavelet entropy and the reconstructed wavelet entropy meet the preset comparison condition, the expression of the comparison condition is: , In the formula, To compare weights; If the comparison condition is met, the reconstructed wavelet entropy is used to replace the wavelet entropy, otherwise it is not replaced.

[0020] Step S104, inputting the rail vibration signal into a preset compression excitation resonance network, the compression excitation resonance network outputs characteristic information of the same latitude as the target wavelet entropy, and fusing the characteristic information with the target wavelet entropy to obtain a fused characteristic signal.

[0021] In this step, the rail vibration signal is input into the preset compression excitation network. The compression block and excitation fast signal of the compression excitation network are used to output the noise suppression feature. Then, the noise suppression feature is dimensionally matched with the target wavelet entropy through identity mapping to make the two dimensions the same. The feature information is fused with the marked wavelet entropy to obtain a fused feature signal.

[0022] Specifically, the vibration signal is compressed and excited in the network, and the expression is: , , In the formula, For the Global average pooling of channels, is the high dimension of the vibration signal, is the wide dimension of the vibration signal, For the Channels at position The activation value of is the final channel weight, is the Sigmoid activation function, is the Relu activation function, is a dimension-raising matrix, is the dimension reduction matrix, is the bias weight; The vibration signal after compression and excitation is matched with the target wavelet entropy to obtain the fusion feature signal, which is expressed as: , In the formula, To fuse feature signals, is the channel dimension multiplication, Configure the matrix for the dimension; Step S105, iteratively training the optimized convolutional neural network according to the fused feature signal and the fault label corresponding to the fused feature signal to obtain a track fault diagnosis model.

[0023] In this step, a convolutional neural network is constructed, and appropriate parameters are selected to build the input layer, hidden layer (convolutional layer, aggregation layer, pooling layer) and output layer of the convolutional neural network. The relationship between the input and output sizes is as follows: , , in, is the output size, is the input size, is the size of the convolution kernel, is filling, is the stride; A transfer wavelet entropy dimensionality reduction loss function based on CWT continuous wavelet transform is proposed to improve the learning and training efficiency and classification accuracy of convolutional neural networks. The specific contents are as follows: , In the formula, is the transfer wavelet entropy dimensionality reduction loss function, is the migration weight factor, is the wavelet entropy dimension reduction factor, is the regularization factor, is the total number of decomposition scales, is the logarithmic correction value, For the The migration labels of samples, For the The true labels of samples, For the The normalized energy of the samples.

[0024] To improve the convolutional neural network, the network activation function also needs to be added and modified: The hidden layer uses the Squareplus function as the activation function, and the formula is: , In the formula, is the input value; The output layer uses the SReLU function as the activation function, and the formula is: , In the formula, is the left threshold, is the right threshold, is the left slope, is the right slope; GoogLeNet is used for filter hyperparameter optimization. It introduces a small network, the Inception block, into the network. It consists of four parallel paths that do not interfere with each other. The weight of the branch with the best hyperparameter will continue to increase during the training process to help select the best hyperparameter. The particle swarm algorithm is introduced to optimize the convolutional neural network (CNN). The particle swarm is used as the hyperparameter of CNN, the CNN weights are randomly initialized, and the network training is performed to determine the speed and position of the particles updating themselves through individual extreme values ​​and group extreme values: , , in, It is a particle In the The speed of iterations, is the inertia weight, is the learning factor, is the learning factor, is a random number in the interval [0,1], A random number in the interval [0,1], It is a particle In the The individual optimal position of It is a particle In the The position of the iteration, It is a particle is the speed after iteration, It is a particle is the position after iteration; The fitness function of the particle swarm algorithm is determined as:

[0025] in, represents the number of samples, Model prediction value, is the reference value, is the prediction error weight, is the regularization adjustment weight, Indicates The weight matrix of the layer, Represents the total number of CNN layers, is the penalty factor, Indicates the training time of the current model. Indicates the maximum allowed training time; Based on the original particle swarm algorithm, a particle state constraint (RSC) strategy is proposed. That is, in the iteration process, a constraint threshold is proposed to constrain particles with abnormal position and speed updates, so that particles that have found the optimal hyperparameter value can update their own speed and position to ensure the training accuracy. The specific contents are as follows: , In the formula, is the position of the particle in k iterations, is the total number of particles in the particle swarm, is the particle state constraint weight, is the initial velocity of the particle, To achieve the local optimal component of the particle, To achieve the global optimal weight for particles, is the constraint threshold.

[0026] In addition, based on the particle state constraint strategy, a random value optimal adjustment (RVOA) strategy is proposed to randomly adjust the hyperparameters during the parameter transmission process, increase the diversity of hyperparameters, and prevent overfitting, which is conducive to speeding up the optimization speed, improving the analysis efficiency, and preventing the particle swarm from being over-optimized. The specific contents are as follows: , , In the formula, is the maximum random value, is the minimum random value, is the hyperbolic cotangent function, is the random proportion weight, is the time scale, To optimize the lower limit of the parameter, To optimize the upper limit of the parameters, is the fitness function, is the total number of particles in the particle swarm.

[0027] Set the normal random value and determine its location. If it falls on and In the elliptical domain composed of , the hyperparameter optimization is terminated, and the specific form is as follows:

[0028]

[0029] in, is a normal random value, yes The optimal adjustment factor under yes The optimal adjustment factor under .

[0030] Step S106, acquiring a real-time rail vibration signal, and inputting the real-time rail vibration signal into the track fault diagnosis model, and the track fault diagnosis model outputs a fault diagnosis result.

[0031] See also Figure 2 , which shows a structural block diagram of an intelligent track fault diagnosis system based on improved CWT-CNN in the present application.

[0032] like Figure 2 As shown, the intelligent track fault diagnosis system 200 includes an acquisition module 210 , a decomposition module 220 , a reconstruction module 230 , a first output module 240 , a training module 250 and a second output module 260 .

[0033] Among them, the acquisition module 210 is configured to acquire the rail vibration signal based on a preset multi-dimensional self-adjusting vibration signal acquisition strategy; the decomposition module 220 is configured to decompose the rail vibration signal according to continuous wavelet transform, obtain the target rail vibration sub-signal containing fault characteristics, and calculate the wavelet entropy of each target rail vibration sub-signal; the reconstruction module 230 is configured to reconstruct the wavelet entropy according to the entropy alienation reconstruction comparison strategy to obtain the target wavelet entropy; the first output module 240 is configured to input the rail vibration signal into a preset compression excitation resonance network, the compression excitation resonance network outputs characteristic information with the same latitude as the target wavelet entropy, and fuses the characteristic information with the target wavelet entropy to obtain a fused characteristic signal; the training module 250 is configured to iteratively train the optimized convolutional neural network according to the fused characteristic signal and the fault label corresponding to the fused characteristic signal to obtain a track fault diagnosis model; the second output module 260 is configured to obtain the real-time rail vibration signal, and input the real-time rail vibration signal into the track fault diagnosis model, and the track fault diagnosis model outputs the fault diagnosis result.

[0034] It should be understood that Figure 2 Modules and references documented in Figure 1 Therefore, the operations and features described above for the method and the corresponding technical effects are also applicable to Figure 2 The modules in it will not be described in detail here.

[0035] In some other embodiments, the embodiments of the present invention further provide a computer-readable storage medium having a computer program stored thereon, and when the program instructions are executed by a processor, the processor is caused to execute the intelligent rail fault diagnosis method based on the improved CWT-CNN in any of the above method embodiments; As an implementation mode, the computer-readable storage medium of the present invention stores computer-executable instructions, and the computer-executable instructions are configured as follows: Acquire rail vibration signals based on a preset multi-dimensional self-adjusting vibration signal acquisition strategy; Decomposing the rail vibration signal according to continuous wavelet transform to obtain target rail vibration sub-signals containing fault characteristics, and calculating the wavelet entropy of each target rail vibration sub-signal; Reconstruct the wavelet entropy according to the entropy alienation reconstruction comparison strategy to obtain the target wavelet entropy; The rail vibration signal is input into a preset compression excitation resonance network, the compression excitation resonance network outputs characteristic information of the same latitude as the target wavelet entropy, and the characteristic information is fused with the target wavelet entropy to obtain a fused characteristic signal; Iteratively training the optimized convolutional neural network according to the fused feature signal and the fault label corresponding to the fused feature signal to obtain a track fault diagnosis model; A real-time rail vibration signal is acquired, and the real-time rail vibration signal is input into the track fault diagnosis model, and the track fault diagnosis model outputs a fault diagnosis result.

[0036] The computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the intelligent track fault diagnosis system based on the improved CWT-CNN, etc. In addition, the computer-readable storage medium may include a high-speed random access memory, and may also include a memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the intelligent track fault diagnosis system based on the improved CWT-CNN via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0037] Figure 3 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as Figure 3 As shown, the device includes: a processor 310 and a memory 320. The electronic device may also 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 via a bus or other means. Figure 3 In the example, the bus connection is used. 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 the non-volatile software programs, instructions and modules stored in the memory 320, that is, the intelligent rail fault diagnosis method based on the improved CWT-CNN in the above-mentioned method embodiment is realized. The input device 330 can receive input digital or character information, and generate key signal input related to user settings and function control of the intelligent rail fault diagnosis system based on the improved CWT-CNN. The output device 340 may include display devices such as display screens.

[0038] The electronic device can execute the method provided by the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided by the embodiment of the present invention.

[0039] As an implementation mode, the electronic device is applied to an intelligent rail fault diagnosis system based on an improved CWT-CNN, and is used for a client, and includes: at least one processor; and a memory connected to the at least one processor in communication; 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: Acquire rail vibration signals based on a preset multi-dimensional self-adjusting vibration signal acquisition strategy; Decomposing the rail vibration signal according to continuous wavelet transform to obtain target rail vibration sub-signals containing fault characteristics, and calculating the wavelet entropy of each target rail vibration sub-signal; Reconstruct the wavelet entropy according to the entropy alienation reconstruction comparison strategy to obtain the target wavelet entropy; The rail vibration signal is input into a preset compression excitation resonance network, the compression excitation resonance network outputs characteristic information of the same latitude as the target wavelet entropy, and the characteristic information is fused with the target wavelet entropy to obtain a fused characteristic signal; Iteratively training the optimized convolutional neural network according to the fused feature signal and the fault label corresponding to the fused feature signal to obtain a track fault diagnosis model; A real-time rail vibration signal is acquired, and the real-time rail vibration signal is input into the track fault diagnosis model, and the track fault diagnosis model outputs a fault diagnosis result.

[0040] Through the description of the above implementation modes, those skilled in the art can clearly understand that each implementation mode can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on such an understanding, the above technical solution can essentially or in other words be embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for 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 embodiment.

[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent rail fault diagnosis method based on improved CWT-CNN, characterized in that: include: Acquire rail vibration signals based on a preset multi-dimensional self-adjusting vibration signal acquisition strategy; Decomposing the rail vibration signal according to continuous wavelet transform to obtain target rail vibration sub-signals containing fault characteristics, and calculating the wavelet entropy of each target rail vibration sub-signal; Reconstruct the wavelet entropy according to the entropy alienation reconstruction comparison strategy to obtain the target wavelet entropy; The rail vibration signal is input into a preset compression excitation resonance network, the compression excitation resonance network outputs characteristic information of the same latitude as the target wavelet entropy, and the characteristic information is fused with the target wavelet entropy to obtain a fused characteristic signal; Iteratively training the optimized convolutional neural network according to the fused feature signal and the fault label corresponding to the fused feature signal to obtain a track fault diagnosis model; A real-time rail vibration signal is acquired, and the real-time rail vibration signal is input into the track fault diagnosis model, and the track fault diagnosis model outputs a fault diagnosis result.

2. The intelligent rail fault diagnosis method based on improved CWT-CNN according to claim 1 is characterized in that: The expression of the multi-dimensional self-adjusting vibration signal acquisition strategy is: ， ， In the formula, is the signal adjustment degree, is the center frequency, is the sampling frequency, is the bandwidth of the collected rail vibration signal, is the reference frequency of the kth component dimension, is the dynamic weight of the kth signal component, is the weight adjustment parameter, is the filter order, is the number of signal sampling points, is the lower limit of angular frequency, is the upper limit of angular frequency, is the angular frequency, is the number of collected signal components, is the low-frequency component of the signal’s main frequency, It is the high frequency component of the signal.

3. The intelligent rail fault diagnosis method based on improved CWT-CNN according to claim 1 is characterized in that: Decomposing the rail vibration signal according to continuous wavelet transform to obtain target rail vibration sub-signals containing fault characteristics, and calculating the wavelet entropy of each target rail vibration sub-signal includes: Select the mother wavelet for signal analysis, the expression is: ， ， In the formula, is the Morlet mother wavelet, is an imaginary unit, is the center frequency, For time, is the complex conjugate form of the Morlet mother wavelet; The rail vibration signal is subjected to continuous wavelet transform according to the mother wavelet, and the coefficient matrix of the continuous wavelet transform is calculated, and the expression is: ， In the formula, is the coefficient matrix of continuous wavelet transform, is the rail vibration signal, is the scale factor, is the translation factor; The coefficient matrix of the continuous wavelet transform is smoothed according to the time window function to obtain the smoothed target coefficient matrix, which is expressed as: ， In the formula, is the Gaussian smoothing window function in the time domain, is the target coefficient matrix after smoothing; The wavelet entropy of the target rail vibrator signal containing the fault characteristics is calculated according to the target coefficient matrix, and the expression is: ， ， ， In the formula, is the wavelet entropy of the target rail vibration sub-signal containing the fault characteristics, is the normalized energy distribution, is the logarithmic correction value, is the regularization factor, is the total number of decomposition scales, For scale The energy on For scale The energy on For scale and time location The target coefficient matrix after smoothing.

4. The intelligent rail fault diagnosis method based on improved CWT-CNN according to claim 1 is characterized in that: The reconstructing the wavelet entropy according to the entropy alienation reconstruction comparison strategy to obtain the target wavelet entropy includes: The wavelet entropy is alienated to obtain alienated wavelet entropy, which is expressed as: ， In the formula, is the wavelet entropy after alienation, is the number of signal components, is the alienation mode weight, For the alienation factor, is the wavelet entropy, is the frequency component, is the alienation time index, is the initial phase of the i-th component in the signal, is the alienation phase offset, is the adjustment coefficient, for The conjugated form of The dynamic characteristics of alienated wavelet entropy in frequency domain and time domain are decomposed according to m modes and n steps, and the modal vector of alienated wavelet entropy is obtained, which is expressed as: ， In the formula, is the modal vector of the n-step alienated wavelet entropy of the m-th mode, is the wavelet entropy of the mth mode after alienation, is the right vector weight, is the left vector weight, For modal The left eigenvector of For the modal The relevant eigenvalues, for The complex conjugate form of For modal The right eigenvector of dimensional hyperbolic cosine function, is the frequency variable, is the time step, is the time interval; The alienated wavelet entropy modal vector is placed in the high-dimensional space in the increasing order of modal decomposition and step decomposition, and the wavelet entropy is reconstructed through the high-dimensional form matrix to obtain the reconstructed wavelet entropy, which is expressed as: , In the formula, To reconstruct wavelet entropy; Selected over-limit threshold accuracy , and judge whether the wavelet entropy and the reconstructed wavelet entropy meet the preset comparison condition, the expression of the comparison condition is: ， In the formula, To compare weights; If the comparison condition is met, the reconstructed wavelet entropy is used to replace the wavelet entropy, otherwise it is not replaced.

5. The intelligent rail fault diagnosis method based on improved CWT-CNN according to claim 1 is characterized in that: The iterative training of the optimized convolutional neural network according to the fused feature signal and the fault label corresponding to the fused feature signal to obtain the track fault diagnosis model includes: Constructing a convolutional neural network, wherein the convolutional neural network comprises an input layer, a hidden layer and an output layer; The convolutional neural network is optimized according to an improved particle swarm algorithm, the particle swarm is used as a hyperparameter of the convolutional neural network, and the weight of the convolutional neural network is randomly initialized, wherein the improved particle swarm algorithm includes a particle state constraint strategy and a random value optimal adjustment strategy; The optimized convolutional neural network is iteratively trained according to the fused feature signal and the fault label corresponding to the fused feature signal to obtain a track fault diagnosis model.

6. The intelligent rail fault diagnosis method based on improved CWT-CNN according to claim 5 is characterized in that: The expression of the particle state constraint strategy is: ， In the formula, is the position of the particle in k iterations, is the total number of particles in the particle swarm, is the particle state constraint weight, is the initial velocity of the particle, To achieve the local optimal component of the particle, To achieve the global optimal weight for particles, is the constraint threshold.

7. The intelligent rail fault diagnosis method based on improved CWT-CNN according to claim 5 is characterized in that: The expression of the random value optimal adjustment strategy is: ， ， In the formula, is the maximum random value, is the minimum random value, is the hyperbolic cotangent function, is the random proportion weight, is the time scale, To optimize the lower limit of the parameter, To optimize the upper limit of the parameters, is the fitness function, is the total number of particles in the particle swarm.

8. The intelligent rail fault diagnosis method based on improved CWT-CNN according to claim 1 is characterized in that: The loss function of the track fault diagnosis model is the migration wavelet entropy dimensionality reduction loss function, which is expressed as: ， In the formula, is the transfer wavelet entropy dimensionality reduction loss function, is the migration weight factor, is the wavelet entropy dimension reduction factor, is the regularization factor, is the total number of decomposition scales, is the logarithmic correction value, For the The migration labels of samples, For the The true labels of samples, For the The normalized energy of the samples.

9. An intelligent track fault diagnosis system based on improved CWT-CNN, characterized in that: include: An acquisition module configured to acquire rail vibration signals based on a preset multi-dimensional self-adjusting vibration signal acquisition strategy; A decomposition module is configured to decompose the rail vibration signal according to continuous wavelet transform to obtain target rail vibration sub-signals containing fault characteristics, and calculate the wavelet entropy of each target rail vibration sub-signal; A reconstruction module, configured to reconstruct the wavelet entropy according to an entropy dissimilation reconstruction comparison strategy to obtain a target wavelet entropy; A first output module is configured to input the rail vibration signal into a preset compression excitation resonance network, the compression excitation resonance network outputs characteristic information of the same latitude as the target wavelet entropy, and fuses the characteristic information with the target wavelet entropy to obtain a fused characteristic signal; A training module, configured to iteratively train the optimized convolutional neural network according to the fused feature signal and the fault label corresponding to the fused feature signal to obtain a track fault diagnosis model; The second output module is configured to obtain a real-time rail vibration signal and input the real-time rail vibration signal into the track fault diagnosis model, and the track fault diagnosis model outputs a fault diagnosis result.

Images