Method and device for identifying state of steel rail joint of heavy haul railway and computer equipment

Abstract

The invention relates to a heavy haul railway steel rail joint state identification method and device and computer equipment. The method comprises the following steps: acquiring original acceleration data of an axle box of the heavy haul railway, and extracting section acceleration data corresponding to a steel rail joint of the heavy haul railway from initial acceleration data; inputting the section acceleration data into a pre-trained WDCNN network model to obtain a joint state of the steel rail joint; the pre-trained WDCNN network model is obtained by training a to-be-trained WDCNN network type by adopting sample acceleration data and a sample joint state level corresponding to the sample acceleration data; the sample joint state level is used as a sample tag of the sample acceleration data. By adopting the method, the identification accuracy of the steel rail joint state of the heavy haul railway can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of railway inspection and operation and maintenance, and particularly to a method, device, computer device, computer-readable storage medium, and computer program product for identifying the state of rail joints in heavy-haul railways. Background Art

[0002] The large axle load and high-density operation of heavy-haul railways cause the track foundation structure to frequently bear severe train dynamic loads, which in turn lead to various track diseases. Among them, the impact on the rail joints is the most prominent. Due to the discontinuity of the structure at the rail joints, the dynamic force between the wheel and the rail increases sharply, further accelerating the track damage near the joints, and hazards such as splint fractures, threaded hole cracks, and low joints appear. The joint state not only affects the running smoothness of the vehicle but also exacerbates the fatigue damage of the rails, threatening the safety of train operation. With the advancement of intelligent operation and maintenance technology for heavy-haul railways, developing efficient and economical track service state detection technology has become an important task for railway operation and maintenance workers.

[0003] Currently, track diseases at rail joints often occur along heavy-haul railway lines. Railway operation units usually use the method of manual on-track inspection to identify and process the joint state. However, this method is highly subjective, and the vehicle vibration response caused by rail joints is relatively complex. Sometimes it is difficult to accurately analyze the joint state of the rails, resulting in low accuracy in identifying the state of rail joints in heavy-haul railways. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a method, device, computer device, computer-readable storage medium, and computer program product for identifying the state of rail joints in heavy-haul railways, which can improve the accuracy of identifying the state of rail joints in heavy-haul railways.

[0005] In a first aspect, the present application provides a method for identifying the state of rail joints in heavy-haul railways, including:

[0006] Obtaining the original acceleration data of the axle boxes of heavy-haul railways;

[0007] Extracting the section acceleration data corresponding to the rail joints of the heavy-haul railways from the original acceleration data;

[0008] Inputting the section acceleration data into a pre-trained WDCNN network model to obtain the joint state of the rail joints; the pre-trained WDCNN network model is obtained by training a WDCNN network model to be trained using sample acceleration data and the sample joint state levels corresponding to the sample acceleration data; the sample joint state levels are used as the sample labels of the sample acceleration data.

[0009] In one embodiment, inputting the section acceleration data into the pre-trained WDCNN network model to obtain the joint state of the rail joint includes:

[0010] Input the section acceleration data into the pre-trained WDCNN network model. The pre-trained WDCNN network model calculates the wheel-rail vertical force, axle lateral force, and derailment coefficient corresponding to the rail joint according to the section acceleration data, and determines the joint state of the rail joint according to the wheel-rail vertical force, the axle lateral force, and the derailment coefficient.

[0011] In one embodiment, determining the joint state of the rail joint according to the wheel-rail vertical force, the axle lateral force, and the derailment coefficient includes:

[0012] Obtain three-level determination conditions; the three-level determination conditions include that the wheel-rail vertical force is higher than the first threshold, the axle lateral force is higher than the second threshold, and the derailment coefficient is higher than the third threshold;

[0013] Determine the number of satisfied conditions among the three-level determination conditions according to the wheel-rail vertical force, the axle lateral force, and the derailment coefficient;

[0014] Determine the joint state level of the rail joint according to the number of satisfied conditions to obtain the joint state of the rail joint.

[0015] In one embodiment, extracting the section acceleration data corresponding to the rail joint of the heavy-haul railway from the original acceleration data includes:

[0016] Perform detrending processing on the original acceleration data to obtain detrended data;

[0017] Perform low-pass filtering on the detrended data through a low-pass filter to extract the section acceleration data corresponding to the rail joint of the heavy-haul railway.

[0018] In one embodiment, before inputting the section acceleration data into the pre-trained WDCNN network model to obtain the joint state of the rail joint, the method further includes:

[0019] Input the sample acceleration data and the sample joint state level corresponding to the sample acceleration data into the WDCNN network model to be trained to obtain a predicted joint state level;

[0020] Calculate the difference between the predicted joint state level and the sample joint state level through a cross-entropy loss function, and train the WDCNN network model to be trained according to the difference to obtain the pre-trained WDCNN network model.

[0021] In one embodiment, training the to-be-trained WDCNN network model according to the difference to obtain the pre-trained WDCNN network model includes:

[0022] Calculating the gradient of the cross-entropy loss function according to the difference;

[0023] Calculating the learning rate according to the change of the gradient;

[0024] Updating the model parameters of the to-be-trained WDCNN network model according to the direction of the gradient and the learning rate to determine the pre-trained WDCNN network model.

[0025] In a second aspect, the present application further provides a device for identifying the state of a rail joint of a heavy-haul railway, including:

[0026] An acquisition module, configured to acquire the original acceleration data of the axle box of the heavy-haul railway;

[0027] An extraction module, configured to extract the sectional acceleration data corresponding to the rail joint of the heavy-haul railway from the original acceleration data;

[0028] A state identification module, configured to input the sectional acceleration data into the pre-trained WDCNN network model to obtain the joint state of the rail joint; the pre-trained WDCNN network model is obtained by training the to-be-trained WDCNN network model with sample acceleration data and the sample joint state level corresponding to the sample acceleration data; the sample joint state level is used as the sample label of the sample acceleration data.

[0029] In a third aspect, the present application further provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the above method are implemented.

[0030] In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0031] In a fifth aspect, the present application further provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0032] The above-mentioned method, device, computer equipment, computer-readable storage medium and computer program product for identifying the state of rail joints in heavy-haul railways obtain the original acceleration data of the axle boxes of heavy-haul railways; extract the sectional acceleration data corresponding to the rail joints of heavy-haul railways from the original acceleration data; input the sectional acceleration data into a pre-trained WDCNN network model to obtain the joint state of the rail joints. The pre-trained WDCNN network model is obtained by training the WDCNN network model to be trained with sample acceleration data and the corresponding sample joint state levels, and the sample joint state levels are used as the sample labels of the sample acceleration data. By obtaining the original acceleration data of the axle boxes and extracting the sectional acceleration data corresponding to the rail joints, the impact response characteristics at the rail joints can be effectively captured. Inputting the sectional acceleration data into the pre-trained WDCNN network model can more accurately identify the dynamic changes of the rail joint state through the deep learning model. At the same time, during the pre-training process of the model, the sample acceleration data and its corresponding joint state levels are used as labels, enabling the model to establish a mapping relationship from the acceleration signal to the joint state, improving the accuracy of identifying the state of rail joints in heavy-haul railways. Description of the Drawings

[0033] To more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments of the present application or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0034] Figure 1 It is an application environment diagram of a method for identifying the state of rail joints in heavy-haul railways in one embodiment;

[0035] Figure 2 It is a flowchart of a method for identifying the state of rail joints in heavy-haul railways in one embodiment;

[0036] Figure 3 It is an installation diagram of an axle box acceleration sensor in one embodiment;

[0037] Figure 4 It is a schematic diagram of sectional acceleration data corresponding to a rail joint in one embodiment;

[0038] Figure 5 It is a schematic diagram of the recognition result of the joint state by a WDCNN network model in one embodiment;

[0039] Figure 6 It is a structural block diagram of a device for identifying the state of rail joints in heavy-haul railways in one embodiment;

[0040] Figure 7 It is the internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0041] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0042] The method for identifying the state of a rail joint of a heavy-haul railway provided by an embodiment of the present application can be applied to, for example Figure 1 the application environment shown in the figure. Among them, the terminal 102 communicates with the server 104 through a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or placed in the cloud or other network servers. The terminal 102 acquires the original acceleration data of the axle box of the heavy-haul railway; the terminal 102 extracts the section acceleration data corresponding to the rail joint of the heavy-haul railway from the original acceleration data; the terminal 102 inputs the section acceleration data into the pre-trained WDCNN network model to obtain the joint state of the rail joint; the pre-trained WDCNN network model is obtained by training the to-be-trained WDCNN network model with sample acceleration data and the sample joint state level corresponding to the sample acceleration data; the sample joint state level is used as the sample label of the sample acceleration data. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, a smart glasses, etc. The server 104 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0043] In an exemplary embodiment, as Figure 2 shown in the figure, a method for identifying the state of a rail joint of a heavy-haul railway is provided. Taking the method applied to Figure 1 the terminal 102 in the figure as an example, it includes:

[0044] Step S202, acquiring the original acceleration data of the axle box of the heavy-haul railway.

[0045] Among them, the axle box is a key component on the bogie of heavy-haul railways, used to install bearings and support the wheel set. It directly bears the wheel-rail interaction force and is a position where the wheel-rail dynamic response can be effectively obtained.

[0046] In practical applications, heavy-haul railways usually generate some vibrations due to various factors such as wheel-rail impact, vehicle body vibration, and track irregularity, and the acceleration sensor can collect acceleration data related to these vibrations.

[0047] In specific implementation, the acceleration sensor installed on the axle box can be used to collect the original acceleration data of the axle box. Among them, the original acceleration data can include vertical acceleration data and lateral acceleration data. In other words, the original acceleration data can be collected by the acceleration sensor installed on the axle box. Optionally, the acceleration sensor can be installed at the side position of the axle box, the vertical center line position of the axle box body, the lateral symmetry position, etc.

[0048] In one embodiment, obtaining the original acceleration data of the axle box of a heavy-haul railway may include: controlling the acceleration sensor set on the axle box of the heavy-haul railway to obtain the original acceleration data of the axle box of the heavy-haul railway according to the first acquisition accuracy, and obtaining the real-time longitude and latitude coordinates of the train through GPS positioning; calculating the distance between the real-time longitude and latitude coordinates and the position coordinates of the rail joints pre-calibrated in the electronic map of the heavy-haul railway, and determining the target rail joint closest to the train according to the Euclidean distance; when the distance between the train and the target rail joint is less than the preset threshold, controlling the acceleration sensor to obtain the original acceleration data of the axle box of the heavy-haul railway according to the second acquisition accuracy; the second acquisition accuracy is higher than the first acquisition accuracy. The technical solution of this embodiment matches through GPS and the electronic map to ensure that the acquisition window accurately covers the joint section, and only starts high-precision acquisition near the joint, reducing power consumption and storage requirements, and using a low sampling rate for data in non-joint sections to reduce the data transmission volume.

[0049] Step S204, extract the section acceleration data corresponding to the rail joints of the heavy-haul railway from the original acceleration data.

[0050] Among them, the rail joint can refer to the connection part between rails. Due to the structural discontinuity, local impact vibration is likely to occur.

[0051] Among them, the section acceleration data can refer to the acceleration signal within a specific time window directly related to the rail joint impact event extracted from the original data.

[0052] In practical applications, the impact characteristics of rail joints may include time-domain characteristics, frequency-domain characteristics, and energy characteristics. The joint impact is manifested as a short-term and high-amplitude acceleration pulse, and the duration can be 10 - 30 milliseconds; the main frequency of the joint impact is concentrated in the range of 500 - 3000 Hz, which is related to the wheel-rail resonance frequency; the short-term energy in the joint section is significantly higher than that in the normal track section. Therefore, according to the impact characteristics of the rail joint, the section acceleration data corresponding to the rail joint of the heavy-haul railway can be extracted from the original acceleration data.

[0053] In one embodiment, extracting the section acceleration data corresponding to the rail joint of the heavy-haul railway from the original acceleration data may include:

[0054] The original acceleration data is segmented into time windows of a fixed length to obtain multiple joint sections; when the peak acceleration in any joint section is higher than the acceleration threshold, the short-term energy is higher than the energy threshold, and the acceleration variance is higher than the variance threshold, that any joint section is determined as the section acceleration data corresponding to the rail joint. This technical solution can improve the accuracy of subsequent identification of the rail joint state by segmenting the time window and accurately identifying the section acceleration data corresponding to the rail joint through a series of judgment conditions.

[0055] Further optionally, determining that any joint section is the section acceleration data corresponding to the rail joint may be to use the point corresponding to the peak acceleration of that any joint section or the center point of that any joint section as the detection point, and expand a preset time period (such as 50 milliseconds) forward and backward with this detection point as the center to obtain the section acceleration data corresponding to the rail joint.

[0056] Exemplarily, the window length of this time window can be determined according to the vehicle speed of the train and the joint length of the rail joint. The higher the vehicle speed, the smaller the window length; the longer the joint length, the longer the window length. Therefore, the window length can change dynamically with the vehicle speed.

[0057] Step S206: Input the section acceleration data into the pre-trained WDCNN network model to obtain the joint state of the rail joint.

[0058] Among them, the pre-trained WDCNN network model is obtained by training the WDCNN network model to be trained with sample acceleration data and the sample joint state level corresponding to the sample acceleration data; the sample joint state level is used as the sample label of the sample acceleration data.

[0059] Among them, the WDCNN network model can be a Wide and Deep Convolutional Neural Network, which adaptively extracts multi-scale features of signals through wide convolution kernels and deep structures. The pre-trained WDCNN network model can be a neural network model pre-trained on a large amount of sample data and can be directly used for inference of new data.

[0060] Optionally, the joint state can include the health level of the rail joint, such as categories like normal, slightly damaged, severely damaged, etc.

[0061] Among them, the sample acceleration data is the acceleration data used to train the WDCNN network model, including acceleration data under different joint states; the sample joint state level is used to label the joint state of the sample acceleration data. The joint state level corresponding to the sample acceleration data is used to supervise model training.

[0062] In specific implementation, the model structure of the WDCNN network model can include an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. Among them, the input layer is used to receive the section acceleration data; the convolutional layer uses a wide convolution kernel (such as 64×1) to extract local features of the signal; the pooling layer reduces the dimension through max pooling and retains significant features; the fully connected layer is used to map the features into the joint state probability distribution; the output layer is used to output the joint state category.

[0063] In specific implementation, the joint state level can be used as a label to construct a data set. The data set includes the sample acceleration data and its corresponding sample joint state level; the data set can be divided into a training set, a validation set, and a test set; build the WDCNN network model to be trained, train the network through the training set and the validation set, optimize the model, and obtain the pre-trained WDCNN network model; input the test set into the pre-trained WDCNN network model to realize the recognition of the rail joint state. Optionally, the ratio of the training set, the validation set, and the test set is 140:40:20.

[0064] In the above method for identifying the state of the rail joint of a heavy-haul railway, the original acceleration data of the axle box of the heavy-haul railway is obtained; the section acceleration data corresponding to the rail joint of the heavy-haul railway is extracted from the original acceleration data; the section acceleration data is input into a pre-trained WDCNN network model to obtain the joint state of the rail joint. Among them, the pre-trained WDCNN network model is obtained by training the WDCNN network model to be trained with sample acceleration data and the corresponding sample joint state level, and the sample joint state level is used as the sample label of the sample acceleration data. By obtaining the original acceleration data of the axle box and extracting the section acceleration data corresponding to the rail joint, the impact response characteristics at the rail joint can be effectively captured. Inputting the section acceleration data into the pre-trained WDCNN network model can more accurately identify the dynamic changes in the state of the rail joint through the deep learning model. At the same time, in the pre-training process of the model, the sample acceleration data and its corresponding joint state level are used as labels, enabling the model to establish a mapping relationship from the acceleration signal to the joint state, improving the accuracy of identifying the state of the rail joint of the heavy-haul railway.

[0065] In another embodiment, inputting the section acceleration data into the pre-trained WDCNN network model to obtain the joint state of the rail joint includes: inputting the section acceleration data into the pre-trained WDCNN network model, and the pre-trained WDCNN network model calculates the wheel-rail vertical force, wheel-axle lateral force, and derailment coefficient corresponding to the rail joint according to the section acceleration data, and determines the joint state of the rail joint according to the wheel-rail vertical force, wheel-axle lateral force, and derailment coefficient.

[0066] Among them, the wheel-rail vertical force can refer to the vertical acting force between the wheel and the rail, that is, the vertical pressure exerted by the wheel on the rail, reflecting the vertical dynamic load at the wheel-rail contact point. Optionally, the wheel-rail vertical force can be determined according to the vertical acceleration, the mass under the spring, and the general suspension dynamic force compensation amount.

[0067] Among them, the wheel-axle lateral force can refer to the lateral acting force between the wheel and the axle, that is, the lateral thrust or pull exerted by the wheel on the axle, reflecting the dynamic load during the lateral movement of the wheel. Optionally, the wheel-axle lateral force can be determined according to the lateral acceleration, the moment of inertia of the wheel, and the radius of the wheel.

[0068] Among them, the derailment coefficient can refer to the ratio of the wheel-rail vertical force to the wheel-axle lateral force, which is used to evaluate the derailment risk.

[0069] In specific implementation, the pre-trained WDCNN network model can extract feature vectors from the section acceleration data using a wide convolution kernel, and map the feature vectors to mechanical parameters (wheel-rail vertical force, wheel-axle lateral force, derailment coefficient) and the corresponding joint state probability distribution.

[0070] Exemplarily, if the vertical wheel-rail force ≤ 425 kN, the joint state can be determined as the normal state; if the vertical wheel-rail force is 425 - 480 kN, the joint state can be determined as slightly damaged; if the vertical wheel-rail force > 480 kN, the joint state can be determined as severely damaged. If the lateral wheel-axle force ≤ 98.4 kN, the joint state can be determined as the normal state; if the lateral wheel-axle force is 98.4 - 110 kN, the joint state can be determined as slightly damaged; if the lateral wheel-axle force > 110 kN, the joint state can be determined as severely damaged. If the derailment coefficient ≤ 0.8, the joint state can be determined as the normal state; if the derailment coefficient is 0.8 - 1.0, the joint state can be determined as slightly damaged; if the derailment coefficient > 1.0, the joint state can be determined as severely damaged.

[0071] The technical solution of this embodiment accurately calculates the mechanical parameters by calculating the mechanical parameters based on the WDCNN network and using the high-dimensional feature extraction ability of the deep learning model; based on the vertical wheel-rail force, lateral wheel-axle force and derailment coefficient, comprehensively determines the joint state, reduces the misjudgment risk of a single parameter through multi-parameter complementary verification, adapts to different working conditions and line conditions, and improves the stability of the determination.

[0072] In another embodiment, before inputting the section acceleration data into the pre-trained WDCNN network model, it further includes:

[0073] Performs normalization processing on the section acceleration data, and inputs the normalized section acceleration data into the pre-trained WDCNN network model.

[0074] ;

[0075] Wherein, is the normalized data, is the data to be normalized, and respectively represent the minimum and maximum values in the data.

[0076] In another embodiment, determining the joint state of the rail joint according to the vertical wheel-rail force, lateral wheel-axle force and derailment coefficient includes: obtaining three-level determination conditions; determining the number of satisfied conditions among the three-level determination conditions according to the vertical wheel-rail force, lateral wheel-axle force and derailment coefficient; determining the joint state level of the rail joint according to the number of satisfied conditions to obtain the joint state of the rail joint.

[0077] Among them, the three-level determination conditions are threshold conditions set respectively based on the vertical wheel-rail force, lateral wheel-axle force and derailment coefficient, and are used to evaluate the severity of the joint state. The number of satisfied conditions among the three-level determination conditions is used to comprehensively determine the joint state level.

[0078] Among them, the three-level determination conditions include that the wheel-rail vertical force is higher than the first threshold, the axle lateral force is higher than the second threshold, and the derailment coefficient is higher than the third threshold.

[0079] Exemplarily, the first threshold is 425 kN, the second threshold is 98.4 kN, and the third threshold is 0.8. Count the number of satisfied conditions (0, 1, 2, 3) among the three-level determination conditions to determine the joint state level. If one level determination condition is satisfied, the joint state level is determined as the first-level state; if two level determination conditions are satisfied, the joint state level is determined as the second-level state; if three level determination conditions are satisfied, the joint state level is determined as the third-level state; if none of the three level determination conditions are satisfied, the joint state level is determined as the normal state. Among them, the first-level state can be the state of slight joint damage, the second-level state can be the state of moderate joint damage, and the third-level state can be the state of severe joint damage.

[0080] The technical solution of this embodiment comprehensively determines the joint state based on the threshold conditions of the wheel-rail vertical force, the axle lateral force, and the derailment coefficient, and divides the joint state into four levels according to the number of satisfied conditions, providing more detailed joint state information, improving the accuracy of joint state recognition, and facilitating targeted maintenance.

[0081] In another embodiment, the section acceleration data corresponding to the rail joints of heavy-haul railways is extracted from the original acceleration data, including: performing detrending processing on the original acceleration data to obtain the detrended data; and performing low-pass filtering on the detrended data through a low-pass filter to extract the section acceleration data corresponding to the rail joints of heavy-haul railways.

[0082] In practical applications, due to current interference at the sensor joint, there is an acceleration trend term in the original acceleration data directly collected by the axle box acceleration sensor. Therefore, detrending processing is required.

[0083] Among them, the detrending processing can include removing the acceleration trend term (such as temperature drift, sensor zero drift, etc.) in the acceleration data. Specifically, it can be to remove the linear trend term (such as a slowly changing DC component) in the signal through a linear fitting method, or to remove the non-linear trend term (such as a slowly changing curve) in the signal through a high-order polynomial fitting method.

[0084] Then, perform low-pass filtering on the detrended data through a low-pass filter, which can retain the low-frequency signals (such as rail joint impact signals) in the data. The rail joint impact signals are mainly distributed in the low-frequency range. Optionally, the cut-off frequency of the low-pass filter can be 50 Hz, so as to suppress high-frequency noise and extract the low-frequency vibration signals related to rail joint impacts.

[0085] The technical solution of this embodiment can improve the signal quality through detrending and low-pass filtering, retain the signals related to joint impacts, facilitate subsequent feature extraction and model inference, and thus improve the recognition accuracy of joint states.

[0086] In another embodiment, before inputting the section acceleration data into the pre-trained WDCNN network model to obtain the joint state of the rail joint, it further includes: inputting the sample acceleration data and the corresponding sample joint state level of the sample acceleration data into the WDCNN network model to be trained to obtain the predicted joint state level; calculating the difference between the predicted joint state level and the sample joint state level through the cross-entropy loss function, and training the WDCNN network model to be trained according to the difference to obtain the pre-trained WDCNN network model.

[0087] Among them, the cross-entropy loss function is a loss function for classification tasks, which measures the difference between the predicted value and the true label.

[0088] The cross-entropy loss function can be expressed as:

[0089] ;

[0090] Among them, is the true label, is the predicted value.

[0091] The technical solution of this embodiment uses the sample acceleration data and the corresponding sample joint state level to train the WDCNN network model. Through supervised learning, the model can accurately learn the feature mapping relationship of the joint state; the cross-entropy loss function is used to measure the difference between the prediction result and the true label, and by minimizing the loss function, the prediction accuracy of the model is improved.

[0092] In another embodiment, training the WDCNN network model to be trained according to the difference to obtain the pre-trained WDCNN network model includes: calculating the gradient of the cross-entropy loss function according to the difference; calculating the learning rate according to the change of the gradient; updating the model parameters of the WDCNN network model to be trained according to the direction and learning rate of the gradient to determine the pre-trained WDCNN network model.

[0093] Among them, the gradient is the partial derivative of the loss function with respect to the model parameters, indicating the direction of parameter update. The learning rate is a hyperparameter that controls the step size of parameter update, determining the convergence speed and stability of model training. Adjusting the model parameters according to the direction and learning rate of the gradient gradually optimizes the model performance.

[0094] In a specific implementation, the partial derivative of the loss function with respect to the model parameters can be calculated through the backpropagation algorithm to obtain the gradient of the cross-entropy loss function. The learning rate is dynamically adjusted according to the change of the gradient. For example, the Adam optimizer is used to optimize and obtain the learning rate. According to the gradient direction and the learning rate, the model parameters are updated through the gradient descent algorithm. By iteratively updating the parameters, the cross-entropy loss function is gradually minimized, and the prediction accuracy of the model is improved.

[0095] Exemplarily, the formula for updating the model parameters through the gradient descent algorithm can be expressed as:

[0096] ;

[0097] where is the weight matrix, is the learning rate, is the gradient of the cross-entropy loss function with respect to the weight.

[0098] The technical solution of this embodiment indicates the direction of parameter update by calculating the gradient of the cross-entropy loss function, accelerates the convergence of the model; dynamically adjusts the learning rate according to the change of the gradient to ensure the training stability; updates the model parameters according to the gradient direction and the learning rate, gradually optimizes the model performance, and gradually improves the prediction accuracy of the model.

[0099] For the convenience of those skilled in the art, Figure 3 an installation schematic diagram of an axle box acceleration sensor is exemplarily provided. The acceleration data of the axle box can be obtained through the acceleration sensor installed on the side of the axle box.

[0100] For the convenience of those skilled in the art, Figure 4 a schematic diagram of the acceleration data corresponding to a rail joint section is exemplarily provided. The original acceleration data is detrended, and the signal is filtered using a low-pass filter with a cut-off frequency of 50 Hz to extract the acceleration data corresponding to the rail joint section.

[0101] For the convenience of those skilled in the art, Figure 5 a schematic diagram of the recognition result of the WDCNN network model for the joint state is exemplarily provided. The recognition result is used to show the comparison between the prediction result and the actual label. The rows can represent the true label categories of the samples, and the columns can represent the label categories predicted by the model. The numbers in the cells can represent the number of samples of the corresponding categories. For example, "47" in the first row and the first column means that 47 samples with a true label of 0 are correctly predicted as 0.

[0102] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are displayed in sequence according to the indications of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0103] Based on the same inventive concept, an embodiment of the present application further provides a device for identifying the state of a rail joint of a heavy-haul railway for implementing the method for identifying the state of a rail joint of a heavy-haul railway described above. The solution provided by this device for solving the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the device for identifying the state of a rail joint of a heavy-haul railway provided below can refer to the limitations on the method for identifying the state of a rail joint of a heavy-haul railway in the above text, and will not be repeated here.

[0104] In an exemplary embodiment, as Figure 6 shown, a device for identifying the state of a rail joint of a heavy-haul railway is provided, including:

[0105] An acquisition module 610, configured to acquire the original acceleration data of the axle box of the heavy-haul railway.

[0106] An extraction module 620, configured to extract the section acceleration data corresponding to the rail joint of the heavy-haul railway from the original acceleration data.

[0107] A state recognition module 630, configured to input the section acceleration data into a pre-trained WDCNN network model to obtain the joint state of the rail joint; the pre-trained WDCNN network model is obtained by training the WDCNN network model to be trained with sample acceleration data and the sample joint state level corresponding to the sample acceleration data; the sample joint state level is used as the sample label of the sample acceleration data.

[0108] In one embodiment, the status recognition module 630 is specifically configured to input the section acceleration data into a pre-trained WDCNN network model, calculate the wheel-rail vertical force, wheel axle lateral force, and derailment coefficient corresponding to the rail joint through the pre-trained WDCNN network model according to the section acceleration data, and determine the joint status of the rail joint according to the wheel-rail vertical force, the wheel axle lateral force, and the derailment coefficient.

[0109] In one embodiment, the status recognition module 630 is specifically configured to obtain three-level determination conditions; the three-level determination conditions include that the wheel-rail vertical force is higher than a first threshold, the wheel axle lateral force is higher than a second threshold, and the derailment coefficient is higher than a third threshold; determine the number of satisfied conditions among the three-level determination conditions according to the wheel-rail vertical force, the wheel axle lateral force, and the derailment coefficient; and determine the joint status level of the rail joint according to the number of satisfied conditions to obtain the joint status of the rail joint.

[0110] In one embodiment, the extraction module 620 is specifically configured to perform detrending processing on the original acceleration data to obtain detrended data; and perform low-pass filtering on the detrended data through a low-pass filter to extract the section acceleration data corresponding to the rail joint of the heavy-haul railway.

[0111] In one embodiment, the rail joint status recognition device for a heavy-haul railway further includes a training module. The training module is specifically configured to input the sample acceleration data and the sample joint status level corresponding to the sample acceleration data into the WDCNN network model to be trained to obtain a predicted joint status level; calculate the difference between the predicted joint status level and the sample joint status level through a cross-entropy loss function, and train the WDCNN network model to be trained according to the difference to obtain the pre-trained WDCNN network model.

[0112] In one embodiment, the training module is specifically configured to calculate the gradient of the cross-entropy loss function according to the difference; calculate the learning rate according to the change of the gradient; and update the model parameters of the WDCNN network model to be trained according to the direction of the gradient and the learning rate to determine the pre-trained WDCNN network model.

[0113] Each module in the above rail joint status recognition device for a heavy-haul railway can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form so that the processor can call and execute the operations corresponding to the above respective modules.

[0114] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in Figure 7 . The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, a mobile cellular network, near field communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a method for identifying the state of rail joints of a heavy-haul railway. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0115] Those skilled in the art can understand that Figure 7 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different component layout.

[0116] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0117] In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0118] In an embodiment, a computer program product is provided, including a computer program. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0119] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data that have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0120] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., and are not limited thereto. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., and are not limited thereto.

[0121] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this application.

[0122] The above-described embodiments merely represent several implementation manners of this application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of this application. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several modifications and improvements can still be made, and these all belong to the protection scope of this application. Therefore, the protection scope of this application shall be subject to the appended claims.

Claims

1. A method for identifying the state of rail joints in heavy-haul railways, characterized in that, The method includes: Obtaining the original acceleration data of the axle box of a heavy-haul railway; Extracting the sectional acceleration data corresponding to the rail joints of the heavy-haul railway from the original acceleration data; Inputting the sectional acceleration data into a pre-trained WDCNN network model to obtain the joint state of the rail joint; the pre-trained WDCNN network model is obtained by training a WDCNN network model to be trained using sample acceleration data and the sample joint state level corresponding to the sample acceleration data; the sample joint state level is used as the sample label of the sample acceleration data.

2. The method according to claim 1, wherein The step of inputting the sectional acceleration data into a pre-trained WDCNN network model to obtain the joint state of the rail joint includes: Inputting the sectional acceleration data into a pre-trained WDCNN network model, and calculating the wheel-rail vertical force, wheel-axle lateral force, and derailment coefficient corresponding to the rail joint by the pre-trained WDCNN network model according to the sectional acceleration data, and determining the joint state of the rail joint according to the wheel-rail vertical force, the wheel-axle lateral force, and the derailment coefficient.

3. The method according to claim 2, wherein The step of determining the joint state of the rail joint according to the wheel-rail vertical force, the wheel-axle lateral force, and the derailment coefficient includes: Obtaining three grade determination conditions; the three grade determination conditions include that the wheel-rail vertical force is higher than a first threshold, the wheel-axle lateral force is higher than a second threshold, and the derailment coefficient is higher than a third threshold; Determining the number of satisfied conditions among the three grade determination conditions according to the wheel-rail vertical force, the wheel-axle lateral force, and the derailment coefficient; Determining the joint state level of the rail joint according to the number of satisfied conditions to obtain the joint state of the rail joint.

4. The method according to claim 1, characterized in that, The step of extracting the sectional acceleration data corresponding to the rail joints of the heavy-haul railway from the original acceleration data includes: Performing detrending processing on the original acceleration data to obtain detrended data; Performing low-pass filtering on the detrended data through a low-pass filter to extract the sectional acceleration data corresponding to the rail joints of the heavy-haul railway.

5. The method according to claim 1, wherein Before the step of inputting the sectional acceleration data into a pre-trained WDCNN network model to obtain the joint state of the rail joint, the method further includes: Inputting the sample acceleration data and the sample joint state level corresponding to the sample acceleration data into the WDCNN network model to be trained to obtain a predicted joint state level; Calculating the difference between the predicted joint state level and the sample joint state level through a cross-entropy loss function, and training the WDCNN network model to be trained according to the difference to obtain the pre-trained WDCNN network model.

6. The method according to claim 5, wherein The step of training the WDCNN network model to be trained according to the difference to obtain the pre-trained WDCNN network model includes: Calculating the gradient of the cross-entropy loss function according to the difference; Calculating the learning rate according to the change of the gradient; Update the model parameters of the WDCNN network model to be trained according to the direction of the gradient and the learning rate, so as to determine the pre-trained WDCNN network model.

7. A device for identifying the state of a rail joint in a heavy-haul railway, characterized in that, The device includes: An acquisition module, configured to acquire the original acceleration data of the axle box of the heavy-haul railway; An extraction module, configured to extract the section acceleration data corresponding to the rail joints of the heavy-haul railway from the original acceleration data; A state recognition module, configured to input the section acceleration data into the pre-trained WDCNN network model to obtain the joint state of the rail joint; the pre-trained WDCNN network model is obtained by training the WDCNN network model to be trained with the sample acceleration data and the sample joint state level corresponding to the sample acceleration data; the sample joint state level is used as the sample label of the sample acceleration data.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of 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 computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 6.

Images