Rail fault diagnosis method and system

Abstract

The application discloses a track fault diagnosis method and system, the method comprises the following steps: calculating the sample entropy, energy entropy, power spectrum and permutation entropy of the optimal solution, selecting the optimal entropy from the sample entropy, energy entropy, power spectrum and permutation entropy according to the preset optimal entropy weight selection strategy, and extracting typical features according to the optimal entropy; constructing an improved CNN-BiLSTM-SA neural network, and extracting sequence space features according to the improved CNN-BiLSTM-SA neural network; optimizing the hyperparameters of the improved CNN-BiLSTM-SA neural network according to the Cauchy-Euclidean clustering particle swarm optimization algorithm, and obtaining a track fault diagnosis model; fusing the optimal solution, typical features and sequence space features of the rail vibration signal to obtain fusion features, and inputting the fusion features into the track fault diagnosis model, so that the track fault diagnosis model outputs a track fault diagnosis result. The accuracy and efficiency of track fault category diagnosis can be improved.

Description

Technical Field

[0001] This invention belongs to the field of rail transit fault diagnosis technology, and in particular relates to a rail fault diagnosis method and system. Background Technology

[0002] In high-speed railway systems, the health of the tracks is crucial for ensuring the safety, comfort, and efficiency of train operation. With the increase in railway operating mileage, the tracks are increasingly affected by external environmental factors and wheel-rail friction, leading to increasingly prominent track defects. These defects include fastener abnormalities, track slab damage, rail corrugation, track irregularities, CA mortar problems, damage to the support layer or base plate, and structural defects caused by changes in the subgrade structure. Failure to promptly detect and repair these defects will severely impact the service life and operational efficiency of the tracks. Currently, track fault diagnosis has the following shortcomings:

[0003] Lack of real-time performance and accuracy: High-speed railways operate at high speeds, and existing fault diagnosis technologies cannot collect fault data in real time and obtain accurate diagnostic results. Summary of the Invention

[0004] This invention provides a method and system for diagnosing track faults, which solves the technical problem that existing fault diagnosis technologies cannot collect fault data in real time and obtain accurate diagnostic results.

[0005] In a first aspect, the present invention provides a method for diagnosing track faults, comprising:

[0006] Rail vibration signals are obtained based on a Beer-shaped adaptive filtering denoising strategy.

[0007] The rail vibration signal is decomposed using an improved scale-space guided variational mode decomposition method to obtain the optimal solution for the rail vibration signal. ;

[0008] Calculate the optimal solution The sample entropy, energy entropy, power spectrum and arrangement entropy are obtained, and the optimal entropy is selected from the sample entropy, energy entropy, power spectrum and arrangement entropy according to the preset optimal entropy weight selection strategy, and typical features are extracted according to the optimal entropy;

[0009] An improved CNN-BiLSTM-SA neural network is constructed, and sequence spatial features are extracted based on the improved CNN-BiLSTM-SA neural network;

[0010] The hyperparameters of the improved CNN-BiLSTM-SA neural network are optimized using the Cauchy-Euclidean clustering particle swarm optimization algorithm to obtain the track fault diagnosis model.

[0011] The optimal solution of the rail vibration signal The typical features and sequence space features are fused to obtain fused features, and the fused features are input into the track fault diagnosis model. The track fault diagnosis model outputs the track fault diagnosis result.

[0012] In a second aspect, the present invention provides a track fault diagnosis system, comprising:

[0013] The acquisition module is configured to acquire rail vibration signals based on a Beer morphology adaptive filtering and denoising strategy.

[0014] The decomposition module is configured to decompose the rail vibration signal according to the improved scale-space guided variational mode decomposition to obtain the optimal solution of the rail vibration signal. ;

[0015] Select the module and configure it to calculate the optimal solution. The sample entropy, energy entropy, power spectrum and arrangement entropy are obtained, and the optimal entropy is selected from the sample entropy, energy entropy, power spectrum and arrangement entropy according to the preset optimal entropy weight selection strategy, and typical features are extracted according to the optimal entropy;

[0016] The extraction module is configured to construct an improved CNN-BiLSTM-SA neural network and extract sequence spatial features based on the improved CNN-BiLSTM-SA neural network;

[0017] The optimization module is configured to optimize the hyperparameters of the improved CNN-BiLSTM-SA neural network according to the Cauchy-Euclidean clustering particle swarm optimization algorithm to obtain the track fault diagnosis model.

[0018] The output module is configured to output the optimal solution of the rail vibration signal. The typical features and sequence space features are fused to obtain fused features, and the fused features are input into the track fault diagnosis model. The track fault diagnosis model outputs the track fault diagnosis result.

[0019] Thirdly, 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, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the track fault diagnosis method according to any embodiment of the present invention.

[0020] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the track fault diagnosis method according to any embodiment of the present invention.

[0021] This application discloses a track fault diagnosis method and system. It designs a vibration signal acquisition system based on Bilateral Morphological Adaptive Filtering Denoising (BMAFD); proposes an improved scale-space guided variational mode decomposition (MSVMD) to decompose the denoised rail vibration signal; and employs the Best Entropy Weight Selection (BEWM) strategy to select the most suitable entropy and calculate the decomposed... The optimal entropy is obtained; a CNN-BiLSTM-SA network model is proposed to extract sequence spatial features, capture their sequential dependencies, and introduce a self-attention mechanism to solve information overload and improve the attention to key information; a Cauchy-Euclidean distance Particle Swarm Optimization Algorithm (CEPSO) is proposed to optimize the hyperparameters of the CNN-BiLSTM-SA neural network, and an optimal entropy loss function is proposed. The fused features are input into the improved model for track fault diagnosis, which can improve the accuracy and efficiency of track fault category diagnosis and achieve accurate and efficient diagnosis of track fault categories. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart of a track fault diagnosis method provided in an embodiment of the present invention;

[0024] Figure 2 This is a structural block diagram of a track fault diagnosis system provided in an embodiment of the present invention;

[0025] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Please see Figure 1 The diagram shows a flowchart of a track fault diagnosis method according to this application.

[0028] like Figure 1 As shown, the track fault diagnosis method specifically includes the following steps:

[0029] Step S101: Obtain rail vibration signal based on Beer morphology adaptive filtering denoising strategy.

[0030] In this step, the vibration sensor is improved, and a Beer-shaped adaptive filtering denoising vibration signal acquisition system is designed to acquire rail vibration signals. Specifically, the Beer-shaped adaptive filtering denoising vibration signal acquisition system uses an 8-channel PCB triaxial digital accelerometer vibration sensor to acquire rail vibration signals. The signal preprocessing module performs signal amplification, Beer-shaped adaptive filtering denoising, bias adjustment, and single-ended to differential conversion. To reduce noise, eliminate scatter points and fill in gaps to improve the signal-to-noise ratio, the expression for the Beer-shaped adaptive filtering denoising strategy is as follows:

[0031] ,

[0032] ,

[0033] ,

[0034] In the formula, The signal obtained after Beer-shaped filtering and denoising. For opening and closing operations, weighting factors. This is the minimum value of the opening and closing operation. This is the original signal of rail vibration. For the weighting factor of the closing-opening operation, This represents the maximum value of the closing-opening operation. Let n be the sequence of the nth structuring element in the Beer morphological filter. For the (n+1)th structuring element sequence in the Beer morphological filter, For opening and closing operations , For opening and closing operations , for right Opening and closing operations, The original signal bandwidth, Pi Let the filter order be . Sampling frequency, For carrier frequency.

[0035] Step S102: Decompose the rail vibration signal according to the improved scale-space guided variational mode decomposition to obtain the optimal solution of the rail vibration signal. .

[0036] In this step, the acquired rail vibration signal is subjected to Fourier transform, and then convolved using a Gaussian kernel function to obtain the scale space set and spatial length, expressed as:

[0037] ,

[0038] ,

[0039] ,

[0040] In the formula, This is the rail vibration signal after Fourier transform. For signal length, This is the discrete-time signal of the rail vibration signal. For discrete time points, The imaginary unit, For frequency index, This is a spatial set of rail vibration signals. For each scale layer, the location of the local minimum value is... This represents the number of local minima in each scale layer. This represents the number of local minima at each scale in the previous recurrent convolution. The spatial length of the rail vibration signal. This represents the current number of iterations in the convolution loop.

[0041] The corresponding threshold is calculated using a coupled scale-space constrained rail vibration signal decomposition strategy, and the original scale-space curve is then transformed. The system is divided into two classes: Class 1, with the smallest intra-class variance, and Class 2, with the largest inter-class variance. Frequency bands between local minima represented by two scale-space curves exceeding a certain length (corresponding to a threshold T) are considered effective frequency bands. The frequency band boundaries corresponding to the threshold of this scale-space curve are then obtained. Furthermore, the number of resonant frequency bands, k, is calculated to guide variational mode decomposition. The expression is:

[0042] ,

[0043] ,

[0044] In the formula, This is a weighted sum of probabilities for the two types of variance. Let be the probability of the (t-1)th point in the i-th scale space curve. The probability of a first-class scale space curve. For tensor product operations, For the i-th heterogeneous scale space curve at point t, Let be the variance change at point (t-1) in the i-th scale space curve. Let be the mean change at point t-1 of the i-th scale space curve. Let be the probability change at point t in the i-th scale space curve. For inter-class variance, For class mean, Let be the probability of point t in the i-th scale space curve. Let be the probability that a first-type scale space curve lies at a point t in the set of scale space curves. The mean of the class at point t+1 in the first-class scale space curve. Let be the mean of the i-th scale space curve. Let t be the within-class mean of the first-class scale space curve. Let be the mean of the k-th IMF signal in the i-th scale space. For the k-th IMF signal in the i-th scale space, Let n be the number of categories, n=1, 2, The number of resonant frequency bands. To find the value of k that maximizes the objective function, Let be the probability of the i-th type of scale space curve. Let be the variance of type i. Let be the probability that the i-th type of scale space curve lies at point t in the set of scale space curves. The mean of the class;

[0045] Variational mode decomposition (VMD) is performed on the rail vibration signal. Using Hilbert transform, the one-sided spectrum of the corresponding analytic signal for each IMF signal is obtained, expressed as:

[0046] ,

[0047] In the formula, Let be the impulse function. The imaginary unit, For time, Pi The decomposed IMF signal;

[0048] Through complex exponential functions with their respective predicted center frequencies as parameters. Multiplying them together adjusts the frequencies of each IMF to baseband; the expression is:

[0049] ,

[0050] In the formula, The center frequency corresponding to each IMF signal after decomposition;

[0051] The bandwidth of each IMF signal is estimated using squared regularization and Gaussian smoothing of the offset signal, expressed as follows:

[0052] ,

[0053] In the formula, The bandwidth of the IMF signal. for, This is the set of all IMF components after decomposition. This is the set of center frequencies corresponding to each IMF component after decomposition. Let be the partial derivative with respect to t. This is the original rail vibration signal. The decomposed IMF signal, For IMF components, The original input signal;

[0054] Given a discrimination precision ,like If the iteration stops, the number of IMF signals k is obtained from the decomposition of the rail vibration signal constrained by the coupled scale space, and the strategy of selecting the center frequency through heterogeneous scale space is improved. , , and Variational mode decomposition yields the optimal solution and center frequency for each IMF signal. The expression for the center frequency selection strategy in heterogeneous scale space is:

[0055] ,

[0056] In the formula, For the k-th mode function in the (n+th)th iteration at frequency The value at that location, This is the frequency domain representation of the original signal f(t). For the modal function with modal index i, in the n+th iteration, the frequency... The value at that location, For the modal function with modal index i, in the nth iteration, the frequency... The value at that location, This is the time-domain representation of the signal after bandpass filtering. For the Lagrange operator corresponding to the constraint condition in the nth iteration, at frequency The value at that location, The coefficient for the penalty term. For frequency variables, The center frequency of the k-th mode function in the nth iteration;

[0057] ,

[0058] In the formula, The center frequency of the k-th mode is the indicator center frequency. In order to be in Within the closed area and Perform double integration. For the time variable used for integration, This represents the signal after bandpass filtering in the time-frequency domain. To single integral, This is the frequency domain representation of the signal after bandpass filtering. To The single integral;

[0059] ,

[0060] In the formula, In the (n+1)th iteration, at frequency Lagrange operators at the location, In the nth iteration, at frequency Lagrange operators at the location, This is the frequency domain representation of the signal after bandpass filtering. The Fourier transform of the original signal f(t) The number of IMF signals. For the k-th mode function in the (n+1)-th iteration at frequency The value at;

[0061] ,

[0062] In the formula, Let be the penalty term coefficient for the k-th mode. Sampling frequency, The signal frequency is the frequency after processing by the bandpass filter.

[0063] Step S103: Calculate the optimal solution The sample entropy, energy entropy, power spectrum, and arrangement entropy are calculated, and the optimal entropy is selected from the sample entropy, energy entropy, power spectrum, and arrangement entropy according to a preset optimal entropy weight selection strategy. Typical features are then extracted based on the optimal entropy.

[0064] In this step, the typical feature extraction target is the rail vibration signal. An initial evaluation index matrix is ​​constructed. i = 1, 2, ..., m, j = 1, 2, ..., n, that is, there are m evaluation samples in total, and each sample has n evaluation indicators;

[0065] Standardize the indicators to eliminate dimensional differences between them. The expression that the numerical value is proportional to the evaluation indicator is as follows:

[0066] ,

[0067] The expression that states the numerical value is inversely proportional to the evaluation index is:

[0068] ,

[0069] In the formula, For indicator feature weights, Let j be the value of the j-th feature of the i-th sample. Let be the minimum value of all features in the i-th sample. It represents the maximum value of all features in the i-th sample;

[0070] Calculate the weight matrix using the Neuerin coefficient-Heyer matrix method. The expression is:

[0071] ,

[0072] In the formula, The coefficient of neutron is 1. Let be the weighting value of the first evaluation index for the m-th evaluation sample under the weighting of n-1 Vinyl factor coefficients. Let be the weighting value of the first evaluation index for the m-th evaluation sample under 2D NUER coefficient weighting. Let n be the weight of the nth evaluation index for the mth evaluation sample, weighted by n-1 Vinyl coefficients. Let n be the weighting value of the nth evaluation index for the mth evaluation sample under the weighting of the second-dimensional neutron coefficient;

[0073] Calculate the information entropy of the j-th evaluation index. The weight of the j-th evaluation index The expression is:

[0074] ,

[0075] ,

[0076] In the formula, For the m-th evaluation sample, The indicator feature weight is the nth evaluation indicator of the mth evaluation sample.

[0077] Step S104: Construct an improved CNN-BiLSTM-SA neural network and extract sequence spatial features based on the improved CNN-BiLSTM-SA neural network.

[0078] In this step, sequence spatial features refer to the multi-level feature representation formed through different levels of feature extraction (local features of CNN, long-term dependencies of BiLSTM, and global relationships of SA). The object of extraction is the rail vibration signal. A Convolutional Neural Network (CNN) is constructed. CNN has powerful feature extraction capabilities. Its network composition includes an input layer, an output layer, activation layers, fully connected layers, pooling layers, and convolutional layers. The convolutional layers are used to extract features, and their operation expression is:

[0079] ,

[0080] In the formula, For input, This is the feature output after convolution by the i-th convolutional kernel in the first layer. and These represent the weights and biases of the i-th convolutional kernel in the first layer, respectively.

[0081] The activation layer incorporates a nonlinear ReLU activation function, enabling the model to learn nonlinear mappings across layers. Its operational expression is:

[0082] ,

[0083] In the formula, This is the input to the activation layer after processing by the i-th convolutional kernel in the first layer. The output after adding a non-linear ReLU activation function to the feature output after convolution with the i-th convolutional kernel of the first layer;

[0084] Max pooling is applied after the convolutional layers to downsample the data and extract the main features. These features are then integrated to form new features, avoiding overfitting. The computational expression is as follows:

[0085] ,

[0086] in, This refers to the new feature formed in the max pooling layer after processing by the i-th convolutional kernel in the first layer. The pooling kernel width, Represents step length, The characteristic length;

[0087] By embedding a Bidirectional Long Short-Term Memory (BiLSTM) network into the max-pooling layer and fully connected layer of a CNN, and using three gating structures to control the retention of information from previous units, preserve input information, and determine unit output, BiLSTM can fully acquire temporal data information and improve model dependency. Specifically:

[0088] ,

[0089] ,

[0090] ,

[0091] ,

[0092] ,

[0093] ,

[0094] ,

[0095] in, yes function, It is the hyperbolic tangent function. For input signal, This represents the weights of the input signal and the memory cell at the current time. The weights representing the hidden layer and the current memory unit value. Input signal and input gate weights, To hide the weights of the state to the input gate, The weights from the state of the memory cell to the input gate, The weights of the input signal to the forget gate; The bias of the memory unit at the current moment. For input gate bias, Forget Gate Offset Output gate bias, These are the network output biases; This is a temporary memory unit value. The value of the memory cell at the current moment; Indicates the value of the input gate. Indicates the value of the forget gate. and Output the gate and the output of the hidden layer at the current moment. This is the network output result;

[0096] BiLSTM consists of two LSTMs moving in opposite directions. The BiLSTM update process is as follows:

[0097] ,

[0098] ,

[0099] ,

[0100] in, and This represents a forward and backward update of the LSTM. and This is the updated result. and To output the forward and backward weights of the hidden layer, For output gate bias;

[0101] To improve the robustness and data feature characterization ability of the entire network, a fully connected layer is added before the output layer, specifically:

[0102] ,

[0103] in, for The activation function, the hidden layer weights and biases of the (m+1)th layer are respectively , ;

[0104] A self-attention module is introduced, which transforms the N sequences of a single sample into three vectors of length d through an adaptive linear mapping. These vectors are then concatenated into a matrix Q, representing the query matrix and the information that needs to be focused on; M, representing the key matrix and the information available for querying; and V, representing the value matrix and the information or features corresponding to the key. This addresses information overload and increases the focus on key information. The specific method is as follows:

[0105] ,

[0106] The Self-learning Variable Squeeze and Excitation Network (SVSEN) is used to enhance feature channels that are useful for the current task, specifically:

[0107] ,

[0108] ,

[0109] In the formula, For the compressed features, The total number of channels. Original features For element-wise multiplication, The total number of features To compress weights, Original features For compression ratio, For self-learning weights, For tensor product operations, To stimulate the calibrated features, For the characteristics after stimulation, Let be the learning rate for the i-th feature. Let be the incentive weight for the j-th channel.

[0110] Step S105: Optimize the hyperparameters of the improved CNN-BiLSTM-SA neural network according to the Cauchy-Euclidean clustering particle swarm optimization algorithm to obtain the track fault diagnosis model.

[0111] In this step, the particle swarm is initialized;

[0112] Based on the nonlinear dynamic inertia weight competition adjustment strategy using Euclidean distance, the Euclidean distances among all particles in the population are calculated, yielding the Euclidean distance matrix. The expression is:

[0113] ,

[0114] In the formula, For nonlinear dynamic inertia weights The Euclidean distance between the nth particle in group A and the nth particle in group B;

[0115] ,

[0116] In the formula, To perform a product operation on the number of iterations m. For tensor product operations, As a competing factor, The maximum number of iterations is set. Let d be the velocity of the i-th particle in the (m-1)th iteration. Let be the Euclidean distance between all particles in the population after the m-th iteration. Let Euclidean distance be the distance between all particles in the population after the (m-1)th iteration. For partial derivative operators, Let d be the velocity of the i-th particle in the m-th iteration. Let d be the velocity of the i-th particle in the (m+1)th iteration;

[0117] ,

[0118] In the formula, The inertia weights are adjusted after the (m-1)th iteration. Let be the nonlinear dynamic correlation variable of the Euclidean distance of the particles in the m-th iteration;

[0119] A stochastic trajectory correction strategy based on the Cauchy-Lorentz coefficient distribution dynamically corrects the trajectory of each particle relative to the globally optimal particle. And output when the required number of iterations or accuracy is reached. The expression is:

[0120] ,

[0121] ,

[0122] ,

[0123] In the formula, The result is a correction for the random trajectory of the Cauchy-Lorenz coefficient distribution. This represents the particle position in the m-th iteration before correction. Lorentz coefficient , Let be the probability density function of the Cauchy-Lorenz coefficient distribution. The global optimal particle position. Let be the scaling parameter for the m-th iteration. For the number of iterations, These are the initial scale parameters. is the time constant.

[0124] Step S106: Calculate the optimal solution of the rail vibration signal. The typical features and sequence space features are fused to obtain fused features, and the fused features are input into the track fault diagnosis model. The track fault diagnosis model outputs the track fault diagnosis result.

[0125] In this step, the optimal solution of the rail vibration signal is... The typical features and sequence space features are fused and normalized to calculate the optimal entropy loss function, which is then input into the improved network model for training and optimization, and used in subsequent networks for fault classification and diagnosis. Specifically:

[0126] ,

[0127] In the formula, The result of feature fusion, As the normalization factor, The weight coefficients of the i-th feature layer are... The total number of feature layers. For the fused features of the i-th feature layer, Let be the mean of the i-th feature layer. Let be the standard deviation of the i-th feature layer;

[0128] ,

[0129] In the formula, Let be the weighting factor for the predicted probability of belonging to class B in the a-th sample. Let be the weighting factor for the predicted probability that the a-th sample does not belong to category B. Let be the predicted probability that the a-th sample belongs to category B. for Expected value For regularization parameters, The weighting factor in regularization. This is the weighting factor in regularization.

[0130] In summary, the method of this application acquires rail vibration signals based on a Beer morphological adaptive filtering denoising strategy; and decomposes the rail vibration signals according to an improved scale-space guided variational mode decomposition to obtain the optimal solution of the rail vibration signals. ; Calculate the optimal solution The sample entropy, energy entropy, power spectrum, and arrangement entropy are calculated, and the optimal entropy is selected from these parameters according to a preset optimal entropy weight selection strategy. Typical features are extracted based on the optimal entropy. An improved CNN-BiLSTM-SA neural network is constructed, and sequence spatial features are extracted based on this network. The hyperparameters of the improved CNN-BiLSTM-SA neural network are optimized using the Cauchy-Euclidean clustering particle swarm optimization algorithm to obtain a track fault diagnosis model. The optimal solution of the rail vibration signal is then used to... The typical features and sequence space features are fused to obtain fused features, which are then input into the track fault diagnosis model. The track fault diagnosis model outputs the track fault diagnosis results, which can improve the accuracy and efficiency of track fault category diagnosis.

[0131] Please see Figure 2 The diagram shows a structural block diagram of a track fault diagnosis system according to this application.

[0132] like Figure 2As shown, the track fault diagnosis system 200 includes an acquisition module 210, a decomposition module 220, a selection module 230, an extraction module 240, an optimization module 250, and an output module 260.

[0133] The acquisition module 210 is configured to acquire rail vibration signals based on a Beer morphology adaptive filtering denoising strategy; the decomposition module 220 is configured to decompose the rail vibration signals according to an improved scale-space guided variational mode decomposition to obtain the optimal solution of the rail vibration signals. Select module 230 and configure it to calculate the optimal solution. The system calculates the sample entropy, energy entropy, power spectrum, and arrangement entropy, and selects the optimal entropy from these parameters according to a preset optimal entropy weight selection strategy. Typical features are extracted based on this optimal entropy. An extraction module 240 is configured to construct an improved CNN-BiLSTM-SA neural network and extract sequence spatial features based on this improved CNN-BiLSTM-SA neural network. An optimization module 250 is configured to optimize the hyperparameters of the improved CNN-BiLSTM-SA neural network using a Cauchy-Euclidean clustering particle swarm optimization algorithm to obtain a track fault diagnosis model. An output module 260 is configured to output the optimal solution of the rail vibration signal. The typical features and sequence space features are fused to obtain fused features, and the fused features are input into the track fault diagnosis model. The track fault diagnosis model outputs the track fault diagnosis result.

[0134] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.

[0135] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the track fault diagnosis method in any of the above method embodiments.

[0136] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:

[0137] Rail vibration signals are obtained based on a Beer-shaped adaptive filtering denoising strategy.

[0138] The rail vibration signal is decomposed using an improved scale-space guided variational mode decomposition method to obtain the optimal solution for the rail vibration signal. ;

[0139] Calculate the optimal solution The sample entropy, energy entropy, power spectrum and arrangement entropy are obtained, and the optimal entropy is selected from the sample entropy, energy entropy, power spectrum and arrangement entropy according to the preset optimal entropy weight selection strategy, and typical features are extracted according to the optimal entropy;

[0140] An improved CNN-BiLSTM-SA neural network is constructed, and sequence spatial features are extracted based on the improved CNN-BiLSTM-SA neural network;

[0141] The hyperparameters of the improved CNN-BiLSTM-SA neural network are optimized using the Cauchy-Euclidean clustering particle swarm optimization algorithm to obtain the track fault diagnosis model.

[0142] The optimal solution of the rail vibration signal The typical features and sequence space features are fused to obtain fused features, and the fused features are input into the track fault diagnosis model. The track fault diagnosis model outputs the track fault diagnosis result.

[0143] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the track fault diagnosis system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely configured relative to a processor, which can be connected to the track fault diagnosis system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0144] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the 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, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the track fault diagnosis method described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the track fault diagnosis system. The output device 340 may include a display screen or other display device.

[0145] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0146] In one implementation, the above-described electronic device is applied to a track fault diagnosis system for a client, 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, the instructions being executed by the at least one processor to enable the at least one processor to:

[0147] Rail vibration signals are obtained based on a Beer-shaped adaptive filtering denoising strategy.

[0148] The rail vibration signal is decomposed using an improved scale-space guided variational mode decomposition method to obtain the optimal solution for the rail vibration signal. ;

[0149] Calculate the optimal solution The sample entropy, energy entropy, power spectrum and arrangement entropy are obtained, and the optimal entropy is selected from the sample entropy, energy entropy, power spectrum and arrangement entropy according to the preset optimal entropy weight selection strategy, and typical features are extracted according to the optimal entropy;

[0150] An improved CNN-BiLSTM-SA neural network is constructed, and sequence spatial features are extracted based on the improved CNN-BiLSTM-SA neural network;

[0151] The hyperparameters of the improved CNN-BiLSTM-SA neural network are optimized using the Cauchy-Euclidean clustering particle swarm optimization algorithm to obtain the track fault diagnosis model.

[0152] The optimal solution of the rail vibration signal The typical features and sequence space features are fused to obtain fused features, and the fused features are input into the track fault diagnosis model. The track fault diagnosis model outputs the track fault diagnosis result.

[0153] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence 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., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0154] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing track faults, characterized in that, include: Rail vibration signals are obtained based on a Beer-shaped adaptive filtering denoising strategy. The rail vibration signal is decomposed using an improved scale-space guided variational mode decomposition method to obtain the optimal solution for the rail vibration signal. ; Calculate the optimal solution The sample entropy, energy entropy, power spectrum and arrangement entropy are obtained, and the optimal entropy is selected from the sample entropy, energy entropy, power spectrum and arrangement entropy according to the preset optimal entropy weight selection strategy, and typical features are extracted according to the optimal entropy; An improved CNN-BiLSTM-SA neural network is constructed, and sequence spatial features are extracted based on the improved CNN-BiLSTM-SA neural network; The hyperparameters of the improved CNN-BiLSTM-SA neural network are optimized using the Cauchy-Euclidean clustering particle swarm optimization algorithm to obtain the track fault diagnosis model. The optimal solution of the rail vibration signal The typical features and sequence space features are fused to obtain fused features, and the fused features are input into the track fault diagnosis model. The track fault diagnosis model outputs the track fault diagnosis result.

2. The track fault diagnosis method according to claim 1, characterized in that, in, The expression for the Beer morphology adaptive filtering denoising strategy is: ， ， ， In the formula, The signal obtained after Beer-shaped filtering and denoising. For opening and closing operations, weighting factors. This is the minimum value of the opening and closing operation. This is the original signal of rail vibration. For the weighting factor of the opening and closing operations, This represents the maximum value of the closing-opening operation. Let n be the sequence of the nth structuring element in the Beer morphological filter. For the (n+1)th structuring element sequence in the Beer morphological filter, For opening and closing operations , For opening and closing operations , for right Opening and closing operations, The original signal bandwidth, Pi Let the filter order be . Sampling frequency, For carrier frequency.

3. The track fault diagnosis method according to claim 1, characterized in that, The rail vibration signal is decomposed using an improved scale-space guided variational mode decomposition method to obtain the optimal solution for the rail vibration signal. include: The collected rail vibration signals are subjected to Fourier transform, and then convolved using a Gaussian kernel function to obtain the scale space set and spatial length, expressed as: ， ， ， In the formula, This is the rail vibration signal after Fourier transform. For signal length, This is the discrete-time signal of the rail vibration signal. For discrete time points, The imaginary unit, For frequency index, This is a spatial set of rail vibration signals. For each scale layer, the location of the local minimum value is... This represents the number of local minima in each scale layer. This represents the number of local minima at each scale in the previous recurrent convolution. The spatial length of the rail vibration signal. This represents the current number of iterations in the convolution loop. The corresponding threshold is calculated using a coupled scale-space constrained rail vibration signal decomposition strategy, and the original scale-space curve is then transformed. The classes are divided into two groups: the first class with the smallest intra-class variance and the second class with the largest inter-class variance, as expressed by: ， ， In the formula, This is a weighted sum of probabilities for the two types of variance. Let be the probability of the (t-1)th point in the i-th scale space curve. The probability of a first-class scale space curve. For tensor product operations, For the i-th heterogeneous scale space curve at point t, Let be the variance change at point (t-1) in the i-th scale space curve. Let be the mean change at point t-1 of the i-th scale space curve. Let be the probability change at point t in the i-th scale space curve. For inter-class variance, For class mean, Let be the probability of point t in the i-th scale space curve. Let be the probability that a first-type scale space curve lies at a point t in the set of scale space curves. The mean of the class at point t+1 in the first-class scale space curve. Let be the mean of the i-th scale space curve. Let t be the within-class mean of the first-class scale space curve. Let be the mean of the k-th IMF signal in the i-th scale space. For the k-th IMF signal in the i-th scale space, Let n be the number of categories, n=1, 2, The number of resonant frequency bands. To find the value of k that maximizes the objective function, Let be the probability of the i-th type of scale space curve. Let be the variance of type i. Let be the probability that the i-th type of scale space curve lies at point t in the set of scale space curves. The mean of the class; Variational mode decomposition (VMD) is performed on the rail vibration signal. Using Hilbert transform, the one-sided spectrum of the corresponding analytic signal for each IMF signal is obtained, expressed as: ， In the formula, Let be the impulse function. The imaginary unit, For time, Pi The decomposed IMF signal; Through complex exponential functions with their respective predicted center frequencies as parameters. Multiplying them together adjusts the frequencies of each IMF to baseband; the expression is: ， In the formula, The center frequency corresponding to each IMF signal after decomposition; The bandwidth of each IMF signal is estimated using squared regularization and Gaussian smoothing of the offset signal, expressed as follows: ， In the formula, The bandwidth of the IMF signal. for, This is the set of all IMF components after decomposition. It is the set of center frequencies corresponding to each IMF component after decomposition. Let be the partial derivative with respect to t. This is the original rail vibration signal. The decomposed IMF signal, For IMF components, The original input signal; Given a discrimination precision ,like If the iteration stops, the number of IMF signals k is obtained by decomposing the rail vibration signal according to the coupled scale space constraint, and the strategy of selecting the center frequency through heterogeneous scale space is improved. , , and Variational mode decomposition yields the optimal solution and center frequency for each IMF signal. The expression for the center frequency selection strategy in heterogeneous scale space is: ， In the formula, For the k-th mode function in the (n+th)th iteration at frequency The value at that location, This is the frequency domain representation of the original signal f(t). For the modal function with modal index i, in the n+th iteration, the frequency... The value at that location, For the modal function with modal index i, in the nth iteration, the frequency... The value at that location, This is the time-domain representation of the signal after bandpass filtering. For the Lagrange operator corresponding to the constraint condition in the nth iteration, at frequency The value at that location, The coefficient for the penalty term. For frequency variables, The center frequency of the k-th mode function in the nth iteration; ， In the formula, The center frequency of the k-th mode is the indicator center frequency. In order to be in Within the closed area and Perform double integration. For the time variable used for integration, This represents the signal after bandpass filtering in the time-frequency domain. To single integral, This is the frequency domain representation of the signal after bandpass filtering. To The single integral; ， In the formula, In the (n+1)th iteration, at frequency Lagrange operators at the location, In the nth iteration, at frequency Lagrange operators at the location, This is the frequency domain representation of the signal after bandpass filtering. The Fourier transform of the original signal f(t) The number of IMF signals. For the k-th mode function in the (n+1)-th iteration at frequency The value at; ， In the formula, Let be the penalty term coefficient for the k-th mode. Sampling frequency, The signal frequency is the frequency after processing by the bandpass filter.

4. The track fault diagnosis method according to claim 1, characterized in that, The selection of the optimal entropy from sample entropy, energy entropy, power spectrum, and arrangement entropy according to the preset optimal entropy weight selection strategy includes: Constructing the initial evaluation index matrix i = 1, 2, ..., m, j = 1, 2, ..., n, that is, there are m evaluation samples in total, and each sample has n evaluation indicators; Standardize the indicators to eliminate dimensional differences between them. The expression that the numerical value is proportional to the evaluation indicator is as follows: ， The expression that states the numerical value is inversely proportional to the evaluation index is: ， In the formula, For indicator feature weights, Let j be the value of the j-th feature of the i-th sample. Let be the minimum value of all features in the i-th sample. It represents the maximum value of all features in the i-th sample; Calculate the weight matrix using the Neuerin coefficient-Heyer matrix method. The expression is: ， In the formula, The coefficient of neutron is 1. Let be the weighting value of the first evaluation index for the m-th evaluation sample under the weighting of n-1 Vinyl factor coefficients. Let be the weighting value of the first evaluation index for the m-th evaluation sample under 2D NUER coefficient weighting. Let n be the weight of the nth evaluation index for the mth evaluation sample, weighted by n-1 Vinyl coefficients. Let n be the weighting value of the nth evaluation index for the mth evaluation sample under the weighting of the second-dimensional neutron coefficient; Calculate the information entropy of the j-th evaluation index. The weight of the j-th evaluation index The expression is: ， ， In the formula, For the m-th evaluation sample, The indicator feature weight is the nth evaluation indicator of the mth evaluation sample.

5. The track fault diagnosis method according to claim 1, characterized in that, The improved CNN-BiLSTM-SA neural network includes a self-learning compression and activation network, expressed as follows: ， ， In the formula, For the compressed features, The total number of channels. Original features For element-wise multiplication, The total number of features To compress weights, Original features For compression ratio, For self-learning weights, For tensor product operations, To stimulate the calibrated features, For the characteristics after stimulation, Let be the learning rate for the i-th feature. Let be the incentive weight for the j-th channel.

6. The track fault diagnosis method according to claim 1, characterized in that, The optimization of the hyperparameters of the improved CNN-BiLSTM-SA neural network using the Cauchy-Euclidean clustering particle swarm optimization algorithm includes: Initialize the particle swarm; Based on the nonlinear dynamic inertia weight competition adjustment strategy using Euclidean distance, the Euclidean distances among all particles in the population are calculated, yielding the Euclidean distance matrix. The expression is: ， In the formula, For nonlinear dynamic inertia weights The Euclidean distance between the nth particle in group A and the nth particle in group B; ， In the formula, To perform a product operation on the number of iterations m. For tensor product operations, As a competing factor, The maximum number of iterations is set. Let d be the velocity of the i-th particle in the (m-1)th iteration. Let be the Euclidean distance between all particles in the population after the m-th iteration. Let be the Euclidean distance between all particles in the population after the (m-1)th iteration. For partial derivative operators, Let d be the velocity of the i-th particle in the m-th iteration. Let d be the velocity of the i-th particle in the (m+1)th iteration; ， In the formula, The inertia weights are adjusted after the (m-1)th iteration. Let be the nonlinear dynamic correlation variable of the Euclidean distance of the particles in the m-th iteration; A stochastic trajectory correction strategy based on the Cauchy-Lorentz coefficient distribution dynamically corrects the trajectory of each particle relative to the globally optimal particle. And output when the required number of iterations or accuracy is reached. The expression is: ， ， ， In the formula, The result is a correction for the random trajectory of the Cauchy-Lorenz coefficient distribution. This represents the particle position in the m-th iteration before correction. Lorentz coefficient , Let be the probability density function of the Cauchy-Lorenz coefficient distribution. The global optimal particle position. Let be the scaling parameter for the m-th iteration. For the number of iterations, These are the initial scale parameters. is the time constant.

7. A track fault diagnosis system, characterized in that, include: The acquisition module is configured to acquire rail vibration signals based on a Beer morphology adaptive filtering and denoising strategy. The decomposition module is configured to decompose the rail vibration signal according to the improved scale-space guided variational mode decomposition to obtain the optimal solution of the rail vibration signal. ; Select the module and configure it to calculate the optimal solution. The sample entropy, energy entropy, power spectrum and arrangement entropy are obtained, and the optimal entropy is selected from the sample entropy, energy entropy, power spectrum and arrangement entropy according to the preset optimal entropy weight selection strategy, and typical features are extracted according to the optimal entropy; The extraction module is configured to construct an improved CNN-BiLSTM-SA neural network and extract sequence spatial features based on the improved CNN-BiLSTM-SA neural network; The optimization module is configured to optimize the hyperparameters of the improved CNN-BiLSTM-SA neural network according to the Cauchy-Euclidean clustering particle swarm optimization algorithm to obtain the track fault diagnosis model. The output module is configured to output the optimal solution of the rail vibration signal. The typical features and sequence space features are fused to obtain fused features, and the fused features are input into the track fault diagnosis model. The track fault diagnosis model outputs the track fault diagnosis result.

8. An electronic device, characterized in that, include: 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 to enable the at least one processor to perform the method according to any one of claims 1 to 6.

9. 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 described in any one of claims 1 to 6.

Images