Switch rail crack detection method and device based on fused time-frequency domain features

Abstract

The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency domain features, and relates to the technical field of frog crack detection. The application provides a switch frog crack detection method and device based on fusion time-frequency

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of frog crack detection, in particular to a frog crack detection method and device based on fusion of time-frequency domain features. BACKGROUND

[0002] In recent years, the volume of railway transportation is also increasing. Under the cyclic action of wheel load and temperature load, the steel rail will be accelerated to break. This will cause the train to derail and cause great casualties. In the railway track structure, the turnout is an indispensable device. The railway turnout is composed of a frog, a nose rail, a sliding bed plate and a fastener. Compared with ordinary steel rails, the frog section of the turnout is very thin. Due to the special function of its structure, the contact trajectory of the vehicle through the turnout switch is discontinuous. This will result in a larger impact force between the wheel and the rail than on ordinary lines, thereby causing the frog to be more prone to cracking than ordinary section steel rails. However, more and more turnouts are laid in high-altitude uninhabited areas, making it difficult to detect their condition regularly. Cracks further cause the steel rail to break, which will directly affect the safety of train operation.

[0003] Traditional track crack detection is usually done manually and cannot detect small cracks or hidden damage. The real-time rail break detection of acoustic waves is poor in terms of detection sensitivity at mechanically insulated locations and is not suitable for turnout damage detection. Ultrasonic guided wave detection has high measurement costs and cannot achieve long-term real-time detection in turnouts due to problems such as the size of the measurement equipment and sensors. SUMMARY

[0004] The purpose of the present application is to provide a frog crack detection method and device based on fusion of time-frequency domain features to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present application is as follows:

[0005] In a first aspect, the present application provides a frog crack detection method based on fusion of time-frequency domain features, comprising:

[0006] Obtaining vibration signals of different knocking points of the frog, the vibration signals including force signals and acceleration signals;

[0007] Synchronously compressing wavelet transform on the vibration signals to obtain a synchronous compression wavelet transform result;

[0008] Calculating a frequency response function of the vibration signals;

[0009] Constructing a multi-modal vibration signal fusion model;

[0010] Training the multi-modal vibration signal fusion model based on the acceleration signals, the frequency response function and the synchronous compression wavelet transform result to obtain a crack damage identification model;

[0011] The crack damage recognition model is used for detecting actual switch frog cracks to obtain a damage result of the actual switch frog.

[0012] In a second aspect, the application further provides a switch frog crack detection device based on fused time-frequency domain features, comprising:

[0013] An acquisition module is configured to acquire vibration signals of different knocking points of a switch frog, wherein the vibration signals include force signals and acceleration signals.

[0014] A transformation module is configured to perform synchronous compressive wavelet transformation on the vibration signals to obtain synchronous compressive wavelet transformation results.

[0015] A calculation module is configured to calculate frequency response functions of the vibration signals.

[0016] A construction module is configured to construct a multi-modal vibration signal fusion model.

[0017] A training module is configured to train the multi-modal vibration signal fusion model based on the acceleration signals, the frequency response functions and the synchronous compressive wavelet transformation results to obtain a crack damage recognition model.

[0018] A prediction module is configured to detect actual switch frog cracks by using the crack damage recognition model to obtain a damage result of the actual switch frog.

[0019] The application has the following beneficial effects: the vibration signals are converted by calculating frequency response functions and synchronous compressive wavelet transformation, the features of multi-modal data are recognized by using a multi-modal vibration signal fusion model, the cracks can be recognized and judged only by using vibration signals, the method is applicable to complex switch structures and has good effects, the sensitivity and reliability in the detection of complex variable cross-section rails such as switches can be effectively avoided, the energy consumption of a measurement sensor is reduced by using low-frequency vibration signals, and the measurement cost and data transmission cost are significantly reduced, and long-time real-time detection can be realized in switches.

[0020] Other features and advantages of the application will be described in the following description, and some will become apparent from the description, or will be understood through implementation of the embodiments of the application. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some of the embodiments of the application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0022] Figure 1 This is a schematic diagram of the method for detecting cracks in the switch tip rail based on fused time-frequency domain features as described in this embodiment of the invention.

[0023] Figure 2 This is a schematic diagram of the crack damage identification model described in this embodiment of the invention;

[0024] Figure 3 This is a diagram showing the arrangement of some measuring points and striking points in an embodiment of the present invention.

[0025] Figure 4 This is a schematic diagram of the time-frequency vibration signal in an embodiment of the present invention;

[0026] Figure 5 This is the time-frequency diagram after synchronous compressed wavelet transform in an embodiment of the present invention;

[0027] Figure 6 This is a schematic diagram of the frequency response function in an embodiment of the present invention;

[0028] Figure 7 This is a time-domain comparison diagram of vibration signals under different damage states in an embodiment of the present invention;

[0029] Figure 8 This is a comparison chart of confusion matrices used to train models for classification in embodiments of the present invention;

[0030] Figure 9 This is a schematic diagram of the switch rail crack detection device based on fused time-frequency domain features as described in an embodiment of the present invention.

[0031] The diagram is labeled as follows: 800, Switch rail crack detection equipment based on fused time-frequency domain features; 801, Processor; 802, Memory; 803, Multimedia component; 804, I / O interface; 805, Communication component. Detailed Implementation

[0032] 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, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0033] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0034] Example 1:

[0035] This embodiment provides a method for detecting cracks in switch rails based on fused time-frequency domain features.

[0036] See Figure 1 The figure shows that the method includes steps S100, S200, S300, S400, S500 and S600.

[0037] Step S100: Obtain vibration signals from different impact points on the switch rail, wherein the vibration signals include force signals and acceleration signals;

[0038] In this embodiment, the acceleration signal is the vibration acceleration signal.

[0039] Step S100 includes:

[0040] Sensors are placed at different positions on the switch rail;

[0041] Multiple striking points are set up, and each striking point is struck multiple times by a hammer. The multiple striking points are located at different positions on the switch point rail.

[0042] Acceleration and force signals from multiple sensors are acquired when each striking point is struck.

[0043] In this embodiment, cracks with different characteristics are created on the complete switch rail, and the hammer is used to strike the switch rail on the sleeper and in the middle of the span. The hammer is used with random force. The uneven hammering method can avoid hitting the modal nodes of the turnout rail structure.

[0044] Step S200: Perform synchronous compressed wavelet transform on the vibration signal to obtain the synchronous compressed wavelet transform result;

[0045] In this embodiment, the vibration signal is subjected to wavelet transform, instantaneous frequency estimation, compression redistribution and signal reconstruction operations in sequence through synchronous compressed wavelet transform.

[0046] By redistributing the wavelet coefficients, the concentration of the time-frequency plot is enhanced. Specifically, after calculating the wavelet transform of the signal, the instantaneous frequency is estimated using the phase information of the wavelet coefficients. Then, the wavelet coefficients are redistributed to the corresponding frequency positions, thereby compressing the time-frequency plot and making it more concentrated.

[0047] Step S200 includes:

[0048] The vibration signal was subjected to continuous wavelet transform to obtain wavelet coefficients at different scales and times;

[0049] The formula for calculating the wavelet coefficients is as follows:

[0050]

[0051] In the formula, W x (·) represents wavelet coefficients, a represents the scaling parameter, and b represents the translation parameter. express The conjugate of complex numbers, Let x(t) represent the mother wavelet function, and let x(t) represent the signal at time t, i.e., the vibration signal.

[0052] Calculate the phase derivative of the wavelet coefficients, and obtain the instantaneous frequency of the vibration signal based on the phase derivative;

[0053] The expression for the instantaneous frequency is:

[0054]

[0055] In the formula, ω x (·) represents the instantaneous frequency, a represents the scale parameter, b represents the translation parameter, and arg(W x (a,b)) represents the phase of the wavelet coefficients, W x (·) represents wavelet coefficients. This indicates the partial derivative.

[0056] Based on the wavelet coefficients and the instantaneous frequency, the wavelet coefficients are compressed and redistributed to obtain a compressed time-frequency diagram;

[0057] In this embodiment, the wavelet coefficients are compressed and redistributed to obtain a compressed time-frequency diagram, which is the time-frequency representation after synchronous compressed wavelet transform. The compression and redistribution process can be regarded as an optimization problem, the objective of which is to minimize the objective function.

[0058] The time-frequency representation and objective function after synchronous compressed wavelet transform are as follows:

[0059]

[0060] In the formula, T x (·) represents the time-frequency representation after synchronous compressed wavelet transform, ω represents the angular frequency, a represents the scaling parameter, b represents the translation parameter, and W x (·) represents the wavelet coefficients, A represents the range of the scaling parameter, and ω x(·) represents the instantaneous frequency, δ(·) represents the Dirac delta function, J(a,b) represents the objective function with respect to a and b, and arg(W x (a,b)) represents the phase of the wavelet coefficients. The expression represents taking the partial derivative, and |·| represents taking the absolute value.

[0061] The compressed time-frequency graph is inversely transformed to obtain the synchronous compressed wavelet transform result.

[0062] In this embodiment, the synchronous compressed wavelet transform result is obtained through inverse transform, which is the original vibration signal. The synchronous compressed wavelet transform result is a two-dimensional image.

[0063] The calculation formula for the synchronous compressed wavelet transform result is as follows:

[0064]

[0065] In the formula, x'(t) represents the result of synchronous compressed wavelet transform, t represents time, and T x (·) denotes the time-frequency representation after synchronous compressed wavelet transform, where ω represents the angular frequency, a represents the scaling parameter, and b represents the translation parameter. Represents the mother wavelet function. The normalization constant of the wavelet function is determined by the characteristics of the mother wavelet.

[0066] Step S300: Calculate the frequency response function of the vibration signal;

[0067] Step S300 includes:

[0068] Calculate the frequency response function curve of the vibration signal;

[0069] Perform Fourier transforms on the vibration signal and the frequency response function curve respectively to obtain the frequency domain representation of the vibration signal and the frequency domain representation of the frequency response function curve;

[0070] The frequency response function of the vibration signal is calculated using the frequency domain representation of the vibration signal and the frequency domain representation of the frequency response function curve.

[0071] The formula for calculating the frequency response function is:

[0072]

[0073] In the formula, H(·) represents the frequency response function, X(·) represents the frequency domain representation of the vibration signal, Y(·) represents the frequency domain representation of the frequency response function curve, and ω represents the angular frequency.

[0074] Frequency response functions play a crucial role in modal analysis, identifying a system's natural frequencies, damping ratios, and modal shapes. By analyzing the amplitude-frequency and phase-frequency characteristics of the frequency response function, we can gain a deeper understanding of the system's dynamic behavior, thereby enabling effective analysis and design.

[0075] The frequency response function is essentially the specific representation of a system's transfer function in the frequency domain, reflecting the system's response characteristics to input signals of different frequencies. To calculate the frequency response function from a time-domain signal, a common method is to transform the input and output signals from the time domain to the frequency domain using a Fourier transform, and then calculate it using the formula mentioned above. In experimental analysis, the frequency response function is usually obtained by applying a known excitation (such as an impact hammer or vibrator) and measuring the corresponding response signal.

[0076] Step S400: Construct a multimodal vibration signal fusion model;

[0077] Step S500: Based on the acceleration signal, the frequency response function, and the synchronous compressed wavelet transform result, train the multimodal vibration signal fusion model to obtain the crack damage identification model;

[0078] In this embodiment, before model training, each vibration signal is labeled with a label indicating the damage status of the switch rail.

[0079] Step S600: Detect the actual switch rail cracks using the crack damage identification model to obtain the damage results of the actual switch rail.

[0080] Step S600 includes:

[0081] The actual acceleration signal and the actual frequency response function are convolved using a convolution kernel to obtain the convolutional features.

[0082] The convolutional features are encoded through a weighted encoding layer to obtain encoded features. The weighted encoding layer includes a first branch, a second branch, and a feature vector dot product layer.

[0083] The feature encoding is transformed by a vector flattening layer to obtain the transformed features. The transformed features are then input into a long short-term memory network to obtain the first output features.

[0084] The actual synchronous compressed wavelet transform result is processed by a feature extraction layer to extract features and obtain the second output feature.

[0085] The first output feature and the second output feature are fused through a feature vector fusion layer to obtain the fused feature.

[0086] The fused features are sequentially input into the fully connected layer and the classifier to obtain the damage results of the actual switch rail.

[0087] The convolutional features are input into the weighted encoding layer for encoding to obtain encoded features, including:

[0088] The convolutional features are processed sequentially through the first branch, including batch normalization, first activation operation, max pooling operation, fully connected layer connection, and second activation operation, to obtain the first branch features. Both the first and second activation operations use the ReLU activation function.

[0089] The convolutional features are sequentially subjected to average pooling, fully connected layer connection, and mapping operations through the second branch to obtain the second branch features. The mapping operation uses the sigmoid function.

[0090] The first branch features and the second branch features are multiplied by a feature vector multiplication layer to obtain the encoded features.

[0091] In this embodiment, the trained multimodal vibration signal fusion model, i.e. crack damage identification model, combines a convolutional neural network, a long short-term memory network, and a weighted encoding layer designed in this invention, and uses multiple parallel branches to process multimodal inputs.

[0092] An initial dataset was constructed by analyzing acceleration and force signals generated by impacts with different crack defects over time. Then, the individual time-series signals were enhanced using two different feature enhancement techniques: synchronous compressed wavelet transform, converting the impact-generated vibration signals into a time-spectrum based on a 2D image, and combining this with the force signal from the hammer vibration. This imbues the ordinary, single force signal with physical meaning, transforming it into a frequency response function, thus enhancing the interpretability of its physical properties. These two methods not only extend the single time domain to the frequency domain, but also combine the time and frequency domains, significantly increasing the interpretability of the training data itself and making it easier to identify even small crack defects.

[0093] In this embodiment, the specific structure of the crack damage identification model is as follows: Figure 2 As shown, it includes a convolutional kernel, a weighted encoding layer, a vector flattening layer, a long short-term memory network (LSTM), a feature extraction layer, a feature vector fusion layer, a fully connected layer, and a classifier. The convolutional kernel, weighted encoding layer, vector flattening layer, LSTM, feature vector fusion layer, fully connected layer, and classifier are connected sequentially. The feature vector fusion layer is also connected to the feature extraction layer.

[0094] The weighted encoding layer includes a first branch, a second branch, and a feature vector dot product layer. The first branch includes a batch normalization layer, a ReLU activation function, a max pooling layer, a fully connected layer, and a ReLU activation function connected in sequence. The second branch includes an average pooling layer, a fully connected layer, and a sigmoid function connected in sequence.

[0095] The feature extraction layer consists of a convolutional kernel, a batch normalization layer, a ReLU activation function, a max pooling layer, a convolutional kernel, a batch normalization layer, a ReLU activation function, a max pooling layer, and a vector flattening layer, connected in sequence.

[0096] In this embodiment, multiple convolutional layers are configured to extract complex key features from complex time-frequency images after synchronous compressed wavelet transform. Pooling layers are constructed to reduce the dimensionality of the extracted features while retaining key information. Based on the convolutional and pooling layers, fully connected layers are further introduced to integrate the extracted features, forming a more compact feature representation.

[0097] By utilizing the sigmoid function to map the output to a range of 0 to 1, and introducing a self-attention mechanism in the weighted encoding layer, this mechanism effectively enhances the model's ability to capture key information by assigning different weights to different time points in the input sequence, while reducing the influence of irrelevant or redundant information. This not only improves the model's global understanding of multimodal input data but also enhances the accuracy and robustness of crack detection, facilitating subsequent classification tasks.

[0098] Furthermore, to process time-series information, the model integrates a long short-term memory network to capture long-term dependencies and temporal dynamics in the time-series data. This step is crucial for the time-domain analysis of vibration signals, effectively identifying minute changes over time.

[0099] Example 2:

[0100] In this embodiment, the vibration characteristics of the combined rails in the turnout switch area under healthy and unhealthy conditions with cracks were investigated. A force hammer excitation test was carried out on a certain high-speed turnout switch area combined rail component with different crack damage conditions in the entire complete turnout switch point rail.

[0101] Multiple high-precision vibration acceleration sensors were installed on the rail surface. Based on the experimental data of this embodiment, a crack damage characteristic information database of vibration signals for turnout variable cross-section rails was established. Through vibration hammer tests, a large number of vibration signals with different damage degrees and locations on the turnout rails were obtained, and the mapping relationship between the damage characteristics and the vibration characteristics of the turnout rails was obtained.

[0102] In this embodiment, artificially created cracks of varying degrees and locations were applied to the switch rails of a complete turnout. Different locations (4.2m, 10m, and 15m from the tip) and different crack damage degrees (1cm, 2cm, 4cm, and 6cm, respectively) were considered. The rail cracks were located on the working edge side of the rail base. A complete hammer impact vibration test was conducted for each damage condition, and the vibration response characteristics at each location were obtained in detail.

[0103] This embodiment uses a high-performance data acquisition instrument, namely a PCB triaxial accelerometer, to acquire vibration signals. This sensor has a sensitivity of 50 mV / g, a range of ±100 g, and a resolution of 0.0002 grms, making it suitable for measuring vibration signals in the range of 1 Hz to 5 kHz.

[0104] A PCB triaxial accelerometer measures the temporal vibration signals of the switch rail in the longitudinal, transverse, and vertical directions, located at the tip rail web, heel rail web, and middle rail bottom, respectively. Including the force signal from the hammer, a total of 10 channels are used. That is, with one excitation using the hammer, the data acquisition instrument simultaneously collects temporal data from 10 channels, including the force signal, tip signal, heel signal, and middle signal in each of the three directions. The tip, heel, and middle signals are the acceleration signals within the vibration signal. Furthermore, to enrich the sources of acquired vibration signals and increase the sample size, each impact point is repeated 20 times. The locations of some impact points are shown in Table 1.

[0105] Table 1

[0106]

[0107] As shown in Table 1, the impact points are not uniformly distributed. This serves two purposes: firstly, it avoids the impact points coinciding with vibration nodes; secondly, it ensures that the excitation points for the switch rail assembly include both the switch sleeper location and the mid-span location, allowing for more random excitation of the switch rail. A partial layout diagram of the measuring points and impact points is shown below. Figure 3 .

[0108] In this embodiment, the original hammer impact signals corresponding to different switch rail crack defects, as well as the time-frequency domain results after operations such as calculating the frequency response function and synchronous compressed wavelet transform, are as follows: Figures 4-6 As shown, Figure 4 Represents a vibration signal with time and frequency. Figure 5 This represents the time-frequency plot after synchronous compressed wavelet transform. Figure 6 This represents the frequency response function.

[0109] like Figures 4-6 As shown, the distribution of the original vibration signal did not differ significantly in each case. Although the attenuation trend after the peak showed some differences, no obvious pattern was revealed, making manual identification and analysis of these features impractical. Figure 7 The figure shown is a time-domain comparison of vibration signals under different damage conditions. Figure 7 The vibration signal time domain, specifically the vibration acceleration time domain, is shown under different damage conditions. Therefore, visually distinguishing and identifying key features from the raw time domain signal is difficult.

[0110] This invention integrates one-dimensional and two-dimensional features, as well as features from the time and frequency domains. It uses the fundamental time-domain acceleration signal and the frequency response function derived from the acceleration signal and the force signal of the hammer in the frequency domain as one source. Then, the time-frequency domain two-dimensional image after synchronous compressed wavelet transform is used as the other part of the multimodal source, and accurate feature extraction and recognition are performed through the established model.

[0111] In this embodiment, identification is performed using low-frequency vibration signals. 500 data points are extracted from each time-series vibration signal sample, which are then converted to 500*1 data points after frequency response function calculation. After synchronous compressed wavelet transform calculation, the result is 224*500 data points. To improve efficiency, each image is compressed to 32×32 by reducing the resolution ratio and color channels, and the processed images are saved as a dataset.

[0112] Before inputting the data into the model, the dataset was standardized and randomly divided into training, validation, and test sets. 70% of the data was selected for training, and the remaining 30% for testing. Conversely, 80% of the training data was used for training again, and the remaining 20% ​​for validation. The Adam algorithm with a learning rate of 0.001 and a batch size of 70 was used during model training. The number of iterations during training was set to 20.

[0113] In this embodiment, taking the classification results of different crack damage levels at 4.2m as an example, evaluation indicators were calculated, including accuracy, precision, recall, and F1 score. For tip signals, whether uniaxial or triaxial, all evaluation indicators were 100%. For heel signals, all triaxial evaluation indicators were 100%, and the accuracy, precision, recall, and F1 score of the uniaxial indicators all exceeded 99.6%. This shows that the classification effect of both uniaxial and triaxial sensors is very good. Furthermore, judging heel signals is more difficult for identifying microcracks than tip signals, but the vibration signals from triaxial sensors, when fused with the detection method of this invention, can significantly improve the various classification indicators.

[0114] Regarding the classification results of different crack damage locations, the model proposed in this invention shows extremely high accuracy in classifying different damage locations, whether it is the tip or the heel, or the data from single-axis or triaxial sensors, all indicators are 100%.

[0115] Meanwhile, under mixed conditions—that is, classification results for different combinations of injury severity and location—all evaluation metrics for tip signals, whether uniaxial or triaxial, reached 100%. For uniaxial signals, heel vibration signals exceeded 99.9% for all metrics across 2270 samples, and the accuracy of triaxial signals was also improved. This further demonstrates that tip signals perform slightly better than heel signals, and that fusing triaxial vibration signals significantly improves the classification and identification of vibration signals.

[0116] This embodiment organizes the vibration signal dataset into three datasets: a dataset containing only time-series vibration response, a dataset containing only synchronous compressed wavelet transform, and the multimodal dataset of this embodiment. The model proposed in this invention is trained using each of these three datasets, as follows: Figure 8 As shown, from left to right, the dataset consists of only time-series vibration response data, only synchronous compressed wavelet transform data, and the confusion matrix of the model classification results trained by the multimodal dataset in this embodiment. It can be seen that training the model of this invention using the multimodal dataset simultaneously yields a much better training effect than training the model using only single-modal data.

[0117] like Figure 8 As shown, the amount of data on the diagonal from the top left to the bottom right of the confusion matrix can intuitively represent the classification effect. The horizontal and vertical axes represent the predicted and true labels for health status, respectively. The larger the value, the more accurate the classification. From the confusion matrix, the multimodal classification method shows the best performance, with a classification accuracy of 100% for all categories, demonstrating very strong classification ability. In contrast, the single-modal trained model exhibits lower classification accuracy in some categories, with overall test set classification accuracies of 85% and 93%, respectively. Significant classification errors occur in the 1cm and 2cm damage conditions. This is mainly because the crack size is small, and the difference in their physical properties is not significant. Identification using low-frequency vibration signals requires the model to have a more powerful feature recognition capability.

[0118] Therefore, the detection method proposed in this invention can identify micro-cracks without manual visual inspection. It can be identified and judged simply by inputting vibration signals into the training model. It is applicable to complex turnout structures and has good results. It can effectively avoid the situation of poor sensitivity and reliability in the detection of complex variable cross-section rails such as turnouts. Furthermore, by collecting low-to-medium frequency vibration signals, specifically the analysis frequency of 1600Hz, it has the advantage of reducing the energy consumption of the measurement sensor, and significantly reducing the measurement cost and data transmission cost. Therefore, it can achieve long-term real-time detection of turnouts.

[0119] Example 3:

[0120] This embodiment provides a switch rail crack detection device based on fused time-frequency domain features, the device comprising:

[0121] The acquisition module is used to acquire vibration signals from different impact points on the switch rail, the vibration signals including force signals and acceleration signals;

[0122] The transformation module is used to perform synchronous compressed wavelet transform on the vibration signal to obtain the synchronous compressed wavelet transform result.

[0123] A calculation module is used to calculate the frequency response function of the vibration signal;

[0124] Modules for building multimodal vibration signal fusion models;

[0125] The training module is used to train the multimodal vibration signal fusion model based on the acceleration signal, the frequency response function, and the synchronous compressed wavelet transform result, so as to obtain the crack damage identification model.

[0126] The prediction module is used to detect actual switch rail cracks using the crack damage identification model to obtain the damage results of the actual switch rail.

[0127] The acquisition module includes:

[0128] The first setting unit is used to arrange sensors at different positions on the switch rail;

[0129] The second setting unit is used to set multiple striking points, and each striking point is struck multiple times by a hammer. The multiple striking points are located at different positions on the switch point rail.

[0130] The first acquisition unit acquires the acceleration and force signals of the multiple sensors when each striking point is struck.

[0131] The transformation module includes:

[0132] The first transformation unit is used to perform continuous wavelet transform on the vibration signal to obtain wavelet coefficients at different scales and at different times;

[0133] The first calculation unit is used to calculate the phase derivative of the wavelet coefficients and obtain the instantaneous frequency of the vibration signal based on the phase derivative.

[0134] The redistribution unit is used to compress and redistribute the wavelet coefficients based on the wavelet coefficients and the instantaneous frequency to obtain a compressed time-frequency diagram.

[0135] The second transformation unit is used to perform an inverse transformation on the compressed time-frequency diagram to obtain a synchronous compressed wavelet transform result.

[0136] The prediction module includes:

[0137] A convolutional unit is used to convolve the actual acceleration signal and the actual frequency response function through a convolutional kernel to obtain convolutional features.

[0138] The encoding unit is used to encode the convolutional features through a weighted encoding layer to obtain encoded features. The weighted encoding layer includes a first branch, a second branch, and a feature vector dot product layer.

[0139] The third transformation unit is used to transform the feature encoding through the vector flattening layer to obtain the transformed features, and input the transformed features into the long short-term memory network to obtain the first output features.

[0140] The feature extraction unit is used to extract features from the actual synchronous compressed wavelet transform result through the feature extraction layer to obtain the second output feature;

[0141] The feature fusion unit is used to fuse the first output feature and the second output feature through the feature vector fusion layer to obtain the fused feature;

[0142] The prediction unit is used to input the fused features sequentially into the fully connected layer and the classifier to obtain the damage results of the actual switch rail.

[0143] The encoding unit includes:

[0144] The first encoding subunit is used to process the convolutional features sequentially through the first branch, performing batch normalization, first activation operation, max pooling operation, fully connected layer connection, and second activation operation to obtain the first branch features. The first and second activation operations both use the ReLU activation function.

[0145] The second encoding subunit is used to sequentially perform average pooling, fully connected layer connection and mapping operations on the convolutional features through the second branch to obtain the second branch features. The mapping operation uses the sigmoid function.

[0146] The feature vector dot product unit is used to perform a dot product operation on the first branch features and the second branch features through the feature vector dot product layer to obtain the encoded features.

[0147] It should be noted that the specific manner in which each module performs its operation in the apparatus described in the above embodiments has been described in detail in the embodiments of the method, and will not be elaborated here.

[0148] Example 4:

[0149] Corresponding to the above method embodiments, this embodiment also provides a switch rail crack detection device based on fused time-frequency domain features. The switch rail crack detection device based on fused time-frequency domain features described below and the switch rail crack detection method based on fused time-frequency domain features described above can be referred to in correspondence.

[0150] Figure 9 This is a block diagram illustrating a switch rail crack detection device 800 based on fused time-frequency domain features, according to an exemplary embodiment. Figure 9 As shown, the switch rail crack detection device 800 based on fused time-frequency domain features may include: a processor 801 and a memory 802. The switch rail crack detection device 800 may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0151] The processor 801 controls the overall operation of the switch rail crack detection device 800 based on fused time-frequency domain features to complete all or part of the steps in the switch rail crack detection method based on fused time-frequency domain features. The memory 802 stores various types of data to support the operation of the switch rail crack detection device 800 based on fused time-frequency domain features. This data may include, for example, instructions for any application or method operating on the switch rail crack detection device 800 based on fused time-frequency domain features, as well as application-related data such as contact data, sent and received messages, images, audio, video, etc. The memory 802 can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in the memory 802 or transmitted via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. I / O interface 804 provides an interface between processor 801 and other interface modules, such as keyboards, mice, and buttons. These buttons can be virtual or physical. Communication component 805 is used for wired or wireless communication between the switch rail crack detection device 800 based on fused time-frequency domain characteristics and other devices. Wireless communication includes, for example, Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof. Therefore, the corresponding communication component 805 may include a Wi-Fi module, a Bluetooth module, or an NFC module.

[0152] In an exemplary embodiment, the switch rail crack detection device 800 based on fused time-frequency domain features can be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the aforementioned switch rail crack detection method based on fused time-frequency domain features.

[0153] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0154] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A fusion time-frequency domain feature-based switch point rail crack detection method, characterized in that, The method comprises the following steps: acquiring vibration signals of different knocking points of a switch rail, the vibration signals comprising force signals and acceleration signals; performing synchronous compressive wavelet transform on the vibration signals to obtain synchronous compressive wavelet transform results; calculating a frequency response function of the vibration signals; constructing a multi-modal vibration signal fusion model; training the multi-modal vibration signal fusion model based on the acceleration signals, the frequency response function and the synchronous compressive wavelet transform results to obtain a crack damage identification model; detecting actual switch rail cracks through the crack damage identification model to obtain a damage result of the actual switch rail.

2. The fusion time-frequency domain feature-based switch rail crack detection method according to claim 1, characterized in that The method for acquiring vibration signals of different knocking points of a switch rail comprises the following steps: arranging sensors at different positions of the switch rail; setting multiple knocking points, knocking each knocking point multiple times with a force hammer, and the multiple knocking points being located at different positions of the switch rail; acquiring acceleration signals and force signals of the multiple sensors when each knocking point is knocked. 3.The fusion time-frequency domain feature based switch rail crack detection method according to claim 1, characterized in that The method for performing synchronous compressive wavelet transform on the vibration signals to obtain synchronous compressive wavelet transform results comprises the following steps: performing continuous wavelet transform on the vibration signals to obtain wavelet coefficients at different scales and different times; calculating a phase derivative of the wavelet coefficients, and obtaining an instantaneous frequency of the vibration signals based on the phase derivative; performing compressive reassignment on the wavelet coefficients based on the wavelet coefficients and the instantaneous frequency to obtain a compressed time-frequency graph; performing inverse transform on the compressed time-frequency graph to obtain synchronous compressive wavelet transform results.

4. The fusion time-frequency domain feature-based switch rail crack detection method according to claim 1, characterized in that The method for detecting actual switch rail cracks through the crack damage identification model to obtain a damage result of the actual switch rail comprises the following steps: convolving actual acceleration signals and actual frequency response functions through a convolution kernel to obtain convolution features; encoding the convolution features through a weight encoding layer to obtain encoded features, the weight encoding layer comprising a first branch, a second branch and a feature vector point multiplication layer; performing vector transformation on the feature encoding through a vector flattening layer to obtain transformed features, and inputting the transformed features into a long short-term memory network to obtain first output features; extracting features of actual synchronous compressive wavelet transform results through a feature extraction layer to obtain second output features; performing feature fusion on the first output features and the second output features through a feature vector fusion layer to obtain fused features; inputting the fused features into a fully connected layer and a classifier in sequence to obtain a damage result of the actual switch rail.

5. The fusion time-frequency domain feature based switch rail crack detection method according to claim 4, characterized in that The method for encoding the convolution features in the weight encoding layer to obtain encoded features comprises the following steps: inputting the convolution features into the first branch in sequence to perform batch normalization processing, a first activation operation, a maximum pooling operation, fully connected layer connection and a second activation operation to obtain first branch features, and the first activation operation and the second activation operation both adopt a relu activation function; inputting the convolution features into the second branch in sequence to perform an average pooling operation, fully connected layer connection and a mapping operation to obtain second branch features, and the mapping operation adopts a sigmoid function; performing point multiplication on the first branch features and the second branch features through a feature vector point multiplication layer to obtain encoded features.

6. A fusion time-frequency domain feature-based switch point rail crack detection device, characterized in that, The method comprises the following steps: The acquisition module is used to acquire vibration signals of different knocking points of a switch rail, wherein the vibration signals include force signals and acceleration signals; The transformation module is used to perform synchronous compressive wavelet transformation on the vibration signals to obtain a synchronous compressive wavelet transformation result; The calculation module is used to calculate a frequency response function of the vibration signals; The construction module is used to construct a multi-modal vibration signal fusion model; The training module is used to train the multi-modal vibration signal fusion model based on the acceleration signals, the frequency response function and the synchronous compressive wavelet transformation result to obtain a crack damage identification model; The prediction module is used to detect an actual switch rail crack through the crack damage identification model to obtain a damage result of the actual switch rail.

7. The fusion time-frequency domain feature based switch rail crack detection device according to claim 6, characterized in that, The acquisition module comprises: The first setting unit is used to arrange sensors at different positions of the switch rail; The second setting unit is used to set a plurality of knocking points, and a force hammer is used to knock each knocking point multiple times, wherein the plurality of knocking points are located at different positions of the switch rail; The first acquisition unit is used to acquire acceleration signals and force signals of a plurality of sensors when each knocking point is knocked.

8. The fusion time-frequency domain feature based switch rail crack detection device according to claim 6, characterized in that, The transformation module comprises: The first transformation unit is used to perform continuous wavelet transformation on the vibration signals to obtain wavelet coefficients at different scales and different times; The first calculation unit is used to calculate a phase derivative of the wavelet coefficients, and an instantaneous frequency of the vibration signals is obtained based on the phase derivative; The redistribution unit is used to perform compressive redistribution on the wavelet coefficients based on the wavelet coefficients and the instantaneous frequency to obtain a compressed time-frequency graph; The second transformation unit is used to perform inverse transformation on the compressed time-frequency graph to obtain a synchronous compressive wavelet transformation result.

9. The fusion time-frequency domain feature based switch rail crack detection device according to claim 6, characterized in that, The prediction module comprises: The convolution unit is used to perform convolution on actual acceleration signals and actual frequency response functions through a convolution kernel to obtain convolution features; The encoding unit is used to encode the convolution features through a weight encoding layer to obtain encoded features, wherein the weight encoding layer comprises a first branch, a second branch and a feature vector point multiplication layer; The third transformation unit is used to perform vector transformation on feature encoding through a vector flattening layer to obtain transformed features, and the transformed features are input into a long short-term memory network to obtain first output features; The feature extraction unit is used to perform feature extraction on actual synchronous compressive wavelet transformation results through a feature extraction layer to obtain second output features; The feature fusion unit is used to perform feature fusion on the first output features and the second output features through a feature vector fusion layer to obtain fusion features; The prediction unit is used to sequentially input the fusion features into a fully connected layer and a classifier to obtain a damage result of an actual switch rail.

10. The fusion time-frequency domain feature based switch rail crack detection device according to claim 9, characterized in that, The encoding unit comprises: The first encoding subunit is used to sequentially perform batch normalization processing, a first activation operation, a maximum pooling operation, a fully connected layer connection and a second activation operation on the convolution features through the first branch to obtain first branch features, wherein the first activation operation and the second activation operation both adopt a relu activation function; The second encoding subunit is configured to sequentially perform an average pooling operation, a fully connected layer connection and a mapping operation on the convolutional features through a second branch to obtain second branch features, and the mapping operation adopts a sigmoid function. The feature vector point multiplication unit is configured to perform a point multiplication operation on the first branch features and the second branch features through a feature vector point multiplication layer to obtain the encoding features.

Images