Wheel-rail force measurement method and system based on multi-path dynamic information fusion

Abstract

The invention discloses a multi-path dynamic information fusion wheel-rail force measurement method and system, and relates to the technical field of rail transit vehicle monitoring, and the method comprises the steps: collecting multi-path dynamic data in the vehicle operation process through a sensor; carrying out normalization processing on the collected dynamic data; discretizing and visualizing the normalized data, and converting the normalized data into a two-dimensional matrix format suitable for the input of a convolutional neural network; performing feature extraction on the preprocessed data by using a multi-dimensional dynamic feature extraction network, and classifying and identifying the change of the wheel-rail force; analyzing the multi-path dynamic data through a multi-dimensional dynamic feature extraction network, and extracting key features related to the wheel-rail force; under the complex line working condition, the train wheel-rail force is monitored and recognized in real time through multi-path dynamic information fusion. According to the multi-path dynamic information fused wheel-rail force measurement method provided by the invention, through fusion of the multi-path dynamic information, wheel-rail force changes under different working conditions can be effectively processed, the adaptability of a measurement system is improved, and the maintenance and use cost is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of rail transit vehicle monitoring, in particular to a wheel-rail force measurement method and system based on multi-channel dynamic information fusion. BACKGROUND

[0002] With the acceleration of global urbanization, rail transit, as an efficient, safe and environmentally friendly mode of transportation, has developed rapidly. In particular, in high-speed rail and urban rail transit systems, the safety and stability of train operation have become the focus of the industry. Wheel-rail force, as an important mechanical parameter generated by the interaction between the wheel and the track during train operation, directly affects the running stability of the vehicle, the comfort of passengers and the safety of train operation. Therefore, accurate and real-time monitoring of wheel-rail force is crucial to ensuring the safe operation of trains and improving the maintenance efficiency of rail transit systems.

[0003] Currently, the measurement methods of wheel-rail force mainly include ground measurement method and on-board measurement method:

[0004] Ground measurement method: usually measures wheel-rail force by installing strain gauges or force rails on the track. This method has high precision and can measure wheel-rail force through the relationship between strain and force. However, ground measurement method has obvious limitations: it cannot achieve continuous monitoring over a large area due to the long-distance and high-speed operation of trains, and the strain gauges and other equipment are prone to aging, resulting in high maintenance costs, which makes it difficult to be widely applied in practice.

[0005] On-board measurement method: this method includes direct measurement method and indirect measurement method, and the more common technologies are force wheelset and bogie strain measurement. Force wheelset technology measures wheel-rail force by installing strain gauges on the wheel or bogie. Although this method has high measurement accuracy, it is expensive and requires frequent calibration, and due to the complexity of installation, it is not suitable for routine operation of conventional railways and urban rail transit. Indirect measurement method calculates wheel-rail force through the dynamic response of the vehicle, including through the acceleration, displacement, torque and other data of the vehicle to derive wheel-rail force. However, the identification accuracy of indirect measurement method is limited under complex line conditions, especially when facing small curve radii or high-speed vehicle operation, it is difficult to provide accurate wheel-rail force data.

[0006] In addition, the traditional measurement methods generally have the following problems:

[0007] Data acquisition is difficult: the installation of equipment is complex, especially the need for long-term maintenance and frequent calibration, resulting in high operating costs, making it difficult to be widely applied in daily operation.

[0008] Measurement accuracy is limited: Existing technologies cannot accurately identify subtle changes in wheel-rail forces under complex conditions such as track irregularities, curves, and high-speed operation.

[0009] High equipment cost: Whether it is ground measurement or vehicle-mounted measurement, the procurement, installation, and maintenance costs of traditional equipment are high, and real-time monitoring cannot be carried out on a large scale.

[0010] With the rapid development of artificial intelligence, especially deep learning technology, data processing and feature extraction based on neural networks have gradually become an important means to solve complex nonlinear problems. Traditional wheel-rail force measurement methods rely on physical models or simple signal processing algorithms, making it difficult to accurately identify changes in wheel-rail forces under complex conditions. Deep learning technology, especially convolutional neural networks (CNN), can automatically extract hidden features from large amounts of data and efficiently process complex, multi-variable dynamic data.

[0011] The successful application of CNN in image processing, speech recognition, and other fields shows its significant advantages in handling high-dimensional data and identifying complex patterns. By fusing and imaging multiple dynamic data (such as acceleration, vibration, and torque), convolutional neural networks can automatically learn the spatial relationships and features in the data, solving the problem of insufficient recognition accuracy of traditional methods under complex conditions. At the same time, deep learning models can improve the robustness of the system through continuous learning and adjustment of network structure, enabling real-time monitoring under various conditions.

[0012] Based on traditional wheel-rail force measurement methods, the invention proposes a deep learning method based on multi-channel dynamic information fusion, which uses vehicle operation data collected by multiple sensors to fuse and analyze these data through convolutional neural networks, solving the problems of data acquisition difficulty, limited measurement accuracy, and high equipment cost in existing technologies. The invention has the following major innovative advantages:

[0013] Multi-channel dynamic information fusion: By fusing various dynamic data of the vehicle (including acceleration, vibration, and torque of the car body, bogie, and axle box), high-precision real-time monitoring of wheel-rail forces is achieved, especially under complex line conditions, significantly improving the recognition accuracy and robustness of the system.

[0014] Application of deep learning technology: Convolutional neural networks automatically extract and analyze the features of multi-dimensional data, providing efficient and accurate recognition results compared to traditional methods, which can better handle subtle changes in complex nonlinear systems.

[0015] Real-time and scalability: The present application can monitor in real time under various complex working conditions through multi-sensor and deep learning technology, and can further improve the robustness and accuracy of the system by continuously optimizing the network structure and algorithm, adapting to different vehicle types and line conditions.

[0016] To solve the problems of data collection difficulty, limited measurement accuracy, high equipment cost and other problems in the prior art, the present application proposes a wheel-rail force measurement method based on multi-channel dynamic information fusion. Through multi-channel sensor to collect a variety of dynamic data including vehicle vibration, acceleration and torque, and use convolutional neural network (CNN) in deep learning to fuse and identify these data, effectively improve the accuracy, real-time and reliability of wheel-rail force measurement, especially in complex line conditions. SUMMARY

[0017] In view of the above problems, the present application is proposed.

[0018] Therefore, the technical problem solved by the present application is to solve the problems of large data collection difficulty, high use cost, low recognition accuracy and inability to cope with complex line conditions of the existing wheel-rail force measurement technology. This method uses multi-channel dynamic information fusion and deep learning technology to monitor and identify train wheel-rail force in real time, especially in complex line conditions, which can significantly improve the measurement accuracy and system robustness.

[0019] To solve the above technical problems, the present application provides the following technical scheme: a wheel-rail force measurement method based on multi-channel dynamic information fusion, comprising:

[0020] Collecting multi-channel dynamic data during vehicle operation through sensors;

[0021] Normalizing the collected dynamic data;

[0022] Discretize and image the normalized data, and convert it into a two-dimensional matrix format suitable for convolutional neural network input;

[0023] Using a multi-dimensional dynamic feature extraction network to extract features from the preprocessed data and classify and identify changes in wheel-rail force; analyzing multi-channel dynamic data through a multi-dimensional dynamic feature extraction network to extract key features related to wheel-rail force;

[0024] Under complex line conditions, real-time monitoring and identification of train wheel-rail force is achieved through multi-channel dynamic information fusion.

[0025] As a preferred scheme of the wheel-rail force measurement method based on multi-channel dynamic information fusion, the multi-channel dynamic data comprises longitudinal vibration acceleration of a vehicle body, transverse vibration acceleration, vertical vibration acceleration, and torque data of a bogie and an axle box.

[0026] As a preferred scheme of the wheel-rail force measurement method based on multi-channel dynamic information fusion, the convolutional neural network comprises at least 8 input channels, and each channel corresponds to a dynamic data source.

[0027] As a preferred scheme of the wheel-rail force measurement method based on multi-channel dynamic information fusion, the wheel-rail force is identified in real time by using a multi-dimensional dynamic feature extraction network.

[0028] As a preferred scheme of the wheel-rail force measurement method based on multi-channel dynamic information fusion, the network architecture of the multi-dimensional dynamic feature extraction network comprises 3 convolutional layers, 2 fully connected layers and 1 output layer, each convolutional layer is followed by a ReLU activation function and a maximum pooling operation, and the network finally outputs a predicted wheel-rail force value, and an SGDM gradient descent optimization algorithm is used for network training.

[0029] As a preferred scheme of the wheel-rail force measurement method based on multi-channel dynamic information fusion, the real-time monitoring and identification of the train wheel-rail force comprises capturing the change of the wheel-rail force under complex working conditions by fusing multi-channel sensor data from an axle box, a vehicle body and a traction system.

[0030] In the case of a small curve radius, different networks are used for joint analysis of acceleration data and torque data to predict the change of the wheel-rail force.

[0031] Another object of the present application is to provide a wheel-rail force measurement system based on multi-channel dynamic information fusion, which can solve the problem of limited identification accuracy of a single information source in the prior art by constructing a wheel-rail force measurement system based on multi-channel dynamic information fusion.

[0032] To solve the above technical problems, the application provides the following technical scheme: a multi-channel dynamic information fusion wheel-rail force measuring system, comprising a data acquisition module, a data processing module, a format conversion module, a feature extraction module and a wheel-rail force prediction module; the data acquisition module is used for acquiring multi-channel dynamic data of a vehicle during operation through a sensor; the data processing module is used for performing normalization processing on the acquired dynamic data; the format conversion module is used for discretizing and imaging the normalized data, and converting the data into a two-dimensional matrix format suitable for input of a convolutional neural network; the feature extraction module is used for performing feature extraction on the preprocessed data using a multi-dimensional dynamic feature extraction network, and classifying and identifying changes in wheel-rail force; the multi-dimensional dynamic feature extraction network is used for analyzing the multi-channel dynamic data, and extracting key features related to wheel-rail force; and the wheel-rail force prediction module is used for real-time monitoring and identifying wheel-rail force of a train under complex line conditions through multi-channel dynamic information fusion.

[0033] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the multi-channel dynamic information fusion wheel-rail force measuring method as described above when executing the computer program.

[0034] A computer readable storage medium stores a computer program, and the computer program implements the steps of the multi-channel dynamic information fusion wheel-rail force measuring method as described above when executed by a processor.

[0035] A computer program product comprises computer instructions, and a processor of a computer device reads the computer instructions, and the processor of the computer device implements the steps of the multi-channel dynamic information fusion wheel-rail force measuring method as described above when executing the computer instructions.

[0036] The application has the following beneficial effects: the multi-channel dynamic information fusion wheel-rail force measuring method provided by the application can effectively process wheel-rail force changes under different conditions through fusion of multi-channel dynamic information, improve adaptability of the measuring system, and reduce maintenance and use costs.

[0037] The application of multi-sensor fusion and convolutional neural network enables the system to still maintain high-precision real-time monitoring capability under complex conditions through acquisition, preprocessing and analysis of various dynamic data of a vehicle, and is suitable for urban rail transit, high-speed rail and other forms of rail vehicles. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0039] Figure 1 The overall flowchart of a multi-channel dynamic information fusion wheel-rail force measurement method provided by an embodiment of the present application.

[0040] Figure 2 The simulation model experiment setup flowchart of a multi-channel dynamic information fusion wheel-rail force measurement method provided by an embodiment of the present application. It shows the setup process of the simulation model, and explains how to simulate the real track and vehicle running state, so as to collect multi-channel dynamic data and conduct experiments. This flowchart provides a framework for system simulation and experimental verification, so as to evaluate the performance of the invention method under simulation conditions.

[0041] Figure 3 The arrangement schematic diagram of a multi-channel dynamic information acquisition device of a multi-channel dynamic information fusion wheel-rail force measurement method provided by an embodiment of the present application. It shows the position and manner of sensor arrangement, which is used to collect acceleration and vibration signals of the vehicle at different positions. Through multiple sensors to collect data in different directions, comprehensive fusion of dynamic information can be realized, which lays a foundation for subsequent data analysis and wheel-rail force identification.

[0042] Figure 4 The track line setting schematic diagram under composite conditions of a multi-channel dynamic information fusion wheel-rail force measurement method provided by an embodiment of the present application. It depicts the track line layout under composite conditions (such as small curve radius, track irregularity, etc.). This diagram helps to understand how to arrange sensors and collect data in complex environment, which provides an environmental reference for subsequent analysis of wheel-rail force.

[0043] Figure 5 The structure block diagram of a convolutional neural network of a multi-channel dynamic information fusion wheel-rail force measurement method provided by an embodiment of the present application. It shows the overall structure of the convolutional neural network used in the present application, including convolutional layer, pooling layer and fully connected layer, etc. This structure is used to extract key features in multi-channel dynamic data, and is used for real-time identification of changes in wheel-rail force.

[0044] Figure 6A multi-channel dynamic information fusion wheel-rail force measurement method provided for an embodiment of the present application provides a flowchart of the wheel-rail force identification process under different working conditions. The flowchart shows the whole process from data acquisition to final identification of the wheel-rail force under different working conditions. Through analysis of different working conditions, the performance and accuracy of the method of the present application under various track conditions can be evaluated.

[0045] Figure 7 The identification time-domain graph of the train vertical force when the vehicle speed is 15 m / s and the curve is 600 meters for a multi-channel dynamic information fusion wheel-rail force measurement method provided for an embodiment of the present application. This graph helps to understand how the system captures and identifies the dynamic changes of the wheel-rail force under a specific working condition, and verifies the accuracy and stability of the identification result.

[0046] Figure 8 The identification frequency spectrum graph of the train vertical force when the vehicle speed is 15 m / s and the curve is 600 meters for a multi-channel dynamic information fusion wheel-rail force measurement method provided for an embodiment of the present application. The frequency spectrum analysis of the vertical force under the same working condition is provided to show the identification accuracy of the system in the frequency domain. The frequency spectrum graph provides data support for further verifying the identification accuracy and the response capability of the system.

[0047] Figure 9 An experimental flowchart of a multi-channel dynamic information fusion wheel-rail force measurement method provided for an embodiment of the present application.

[0048] Figure 10 A general structure diagram of a multi-channel dynamic information fusion wheel-rail force measurement system provided for an embodiment of the present application. DETAILED DESCRIPTION

[0049] To make the above objectives, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application are described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the present application.

[0050] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application. However, the present application can be practiced according to other embodiments that do not require the specific details described herein. Therefore, the present application is not limited to the embodiments disclosed below, but rather, the scope of the present application is defined by the appended claims.

[0051] Embodiment 1

[0052] Reference Figures 1-9 For an embodiment of the present application, a multi-channel dynamic information fusion wheel-rail force measurement method is provided, comprising:

[0053] S1 Collecting multi-channel dynamic data during vehicle operation through multiple sensors, including acceleration signals and vibration signals of the vehicle body, bogie, and axle box;

[0054] S2 Normalizing the collected dynamic data to eliminate dimensional differences between sensors;

[0055] S3 Discretizing and imaging the processed data, converting it into a two-dimensional matrix format suitable for convolutional neural network input;

[0056] S4 Feature extraction using a multi-dimensional dynamic feature extraction network on the pre-processed data to classify and identify changes in wheel-rail force; Through the multi-dimensional dynamic feature extraction network, the multi-channel dynamic data (acceleration, vibration, etc.) is analyzed to extract key features related to wheel-rail force. These features are useful information hidden in complex multi-dimensional data, such as the trend of wheel-rail force changes under different conditions, peak values, etc.

[0057] S5 Real-time monitoring and identification of train wheel-rail force through multi-channel dynamic information fusion under complex line conditions, providing high-precision identification results.

[0058] It should be noted that the result of feature extraction: through the convolutional neural network, the multi-channel dynamic data (acceleration, vibration) is analyzed to extract key features related to wheel-rail force. These features are useful information hidden in complex multi-dimensional data, such as the trend of wheel-rail force changes under different conditions, peak values, etc.

[0059] Classification: The purpose of classification is to identify wheel-rail force under different conditions, which may be based on different force value changes, such as changes in vertical force and lateral force, so as to divide the data into different categories, reflecting the state of wheel-rail force under different environments. For example, the system may distinguish between wheel-rail force data under different conditions such as high speed, curve, track irregularity, etc.

[0060] Step S5 describes how to perform real-time monitoring and identification through the fusion of multi-channel dynamic information under complex line conditions. The key to this step is:

[0061] Fusion of multi-channel dynamic information:

[0062] The system combines information from different sensors (such as acceleration and torque sensors) to improve the ability to capture changes in wheel-rail force under complex conditions.

[0063] Real-time monitoring:

[0064] The system performs data collection and processing within a 0.1 second time window, ensuring real-time response to changes in train operating conditions.

[0065] Output of recognition results:

[0066] Through the trained convolutional neural network model, the system can output real-time wheel-rail force recognition results. These results can include specific values of vertical force and lateral force, and provide high-precision recognition results.

[0067] Multi-channel dynamic data includes vehicle body longitudinal vibration acceleration, lateral vibration acceleration, vertical vibration acceleration, and torque data of bogies and axle boxes.

[0068] The input layer of the convolutional neural network includes at least 8 input channels, each corresponding to a dynamic data source.

[0069] The deep learning network uses an optimization algorithm based on SGDM (momentum gradient descent) with an initial learning rate of 0.01 and automatically adjusts the learning rate after every 100 iterations.

[0070] The measurement method is suitable for different speeds, different curve radii, and complex working conditions of track irregularities.

[0071] The system performs data acquisition and processing every 0.1 seconds, ensuring real-time performance.

[0072] Multi-channel dynamic information acquisition device: Multiple sensors are arranged, including vibration sensors and acceleration sensors installed on vehicle axle boxes, vehicle bodies, and traction systems, for collecting dynamic data during vehicle operation. For sampling point arrangement, refer to GB / T 5599-2019, and based on correlation analysis, 18 acquisition signals as shown in Table 1 are arranged:

[0073] Table 1 Dynamic signal acquisition points

[0074] Serial number Name Symbol Unit 1 Vehicle body longitudinal vibration acceleration a cx ]]> m / s 2 ]] 2 Vehicle body lateral vibration acceleration a cy ]]> m / s 2 ]] 3 Vehicle body vertical vibration acceleration a cz ]]> m / s 2 ]] 4 Vehicle body roll angular velocity cx ]]> ​ rad / s 5 Vehicle body pitch angular velocity cy ]]> ​ rad / s 6 Vehicle body yaw angular velocity cz ]]> ​ rad / s 7 Bogie frame longitudinal vibration acceleration a tx ]]> m / s 2 ]]> 8 Bogie frame lateral vibration acceleration a ty ]]> m / s 2 ]] 9 Bogie frame vertical vibration acceleration a tz ]]> m / s 2 ]]> 10 Bogie frame roll angular velocity tx ]]> ​ rad / s 11 Bogie frame pitch angular velocity ty ]]> ​ rad / s 12 Bogie frame yaw angular velocity tz ]]> ​ rad / s 13 Axle box longitudinal vibration acceleration a bx ]]> m / s 2 ]] 14 Axle box lateral vibration acceleration a by ]]> m / s 2 ]]> 15 Axle box vertical vibration acceleration a bz ]]> m / s 2 ]]> 16 Primary spring vertical displacement [SA jz ]]> mm 17 Secondary spring lateral displacement [SA cy ]] mm 18 Secondary spring vertical displacement [SA cz ]]> mm

[0075] Data fusion and preprocessing: Normalize, discretize, and image the collected multi-channel dynamic data, and convert the data into a format suitable for deep learning network input.

[0076] Deep learning network design: Use convolutional neural network (CNN) to extract features and classify the preprocessed data, and establish a real-time wheel-rail force recognition model.

[0077] Complex condition processing: Under complex line conditions, by fusing multi-channel dynamic information, the robustness and accuracy of network recognition are improved.

[0078] Network optimization and model training: According to the characteristics of different working conditions and wheel-rail force changes, design corresponding network structure and training scheme, and verify the network recognition effect through experiments.

[0079] The method has the following beneficial effects:

[0080] Real-time and accurate measurement of wheel-rail force under complex line conditions can solve the problem of low recognition accuracy of existing technologies under complex working conditions.

[0081] Through the fusion processing of multiple dynamic information, the robustness and applicability of the recognition model are further improved.

[0082] Using deep learning technology, the maintenance cost and data acquisition difficulty of the system are effectively reduced.

[0083] The application adopts a multi-dimensional dynamic feature extraction network to extract features and classify and identify the preprocessed data, and establishes a real-time wheel-rail force recognition model. The network can include a convolutional neural network (CNN) as one of the specific implementation methods.

[0084] Existing wheel-rail force measurement methods, such as ground measurement and vehicle-mounted measurement, can provide high-precision data in some scenarios, but their application range is limited, especially in terms of measurement accuracy and real-time performance under complex line conditions (such as track curves, track irregularities, high-speed driving, etc.). To overcome these limitations, the application proposes a wheel-rail force measurement method based on multi-channel dynamic information fusion, which collects and processes multiple vehicle dynamic data in real time, uses a deep learning network for data fusion and analysis, and thus realizes high-precision, low-cost real-time wheel-rail force measurement.

[0085] The specific purposes of the application include providing a high-precision wheel-rail force measurement method suitable for complex line conditions.

[0086] By fusing multiple dynamic information, the problem of limited recognition accuracy of a single information source is solved.

[0087] Using deep learning technology, the real-time performance and robustness of wheel-rail force recognition are improved.

[0088] Reducing the equipment cost and maintenance difficulty of the wheel-rail force measurement system increases the applicability of the system.

[0089] Table 2 lists different working condition parameters during simulation, including distance, curve radius and vehicle speed, etc. This table is used to simulate different railway line conditions to test the recognition accuracy of the wheel-rail force measurement system under various complex working conditions. These parameters represent the real environment of vehicle operation, covering small-radius curves and high-speed driving scenarios, which facilitates verification of the applicability and stability of the system under various working conditions.

[0090] Table 2 Simulation Working Condition Parameters

[0091]

[0092]

[0093] Table 3 shows the specific architecture and specifications of the convolutional neural network used in the present application. The network structure includes input layer, convolutional layer, ReLU activation layer, fully connected layer, etc. Different network configurations help to extract multi-channel dynamic information features and predict wheel-rail forces in real time under different working conditions. The configuration of network parameters is optimized through experiments to ensure that the system can obtain high precision and real-time performance in practical applications.

[0094] Table 3 Specific specification table of real-time wheel-rail force identification network

[0095]

[0096] Table 4 provides the error and accuracy (MAE and R 2 The values in the table show the identification effect of the system on wheel-rail force under different working conditions, thereby verifying the effectiveness of the deep learning model in handling complex track conditions, especially in the scenarios of small radius and high speed driving, showing high precision and robustness.

[0097] Table 4 Identification effect table of vertical force and lateral force under different curve radii

[0098]

[0099] During vehicle operation, the present application installs multiple acceleration sensors and vibration sensors at key parts such as axle boxes, car bodies, and traction systems to collect dynamic data in real time during vehicle operation. These sensors are KISTLER 8763B series acceleration sensors, which have high sensitivity and a wide measurement range, ensuring accurate signals even under complex working conditions. These sensors are arranged at key nodes of the vehicle, and through an integrated signal acquisition unit, multi-channel sensor data is transmitted synchronously to the data processing module.

[0100] The collected multi-channel dynamic information is first normalized to eliminate the dimensional differences between different sensor data. Normalization uses the Z-score standardization formula to map all sensor data to the same numerical range. Next, the data is discretized, with a sampling frequency of 500 Hz and a sample time window of 0.1 seconds. In order to facilitate subsequent deep learning model training, the discretized data is converted into a two-dimensional matrix, forming an image format as the input of the convolutional neural network.

[0101] The application adopts a multi-dimensional dynamic feature extraction network for real-time wheel-rail force recognition. The network architecture includes 3 convolutional layers, 2 fully connected layers and 1 output layer, and each convolutional layer is followed by a ReLU activation function and a maximum pooling operation. The convolution kernel and the pooling layer parameters are shown in Table 1, and the final output of the network is the predicted wheel-rail force value. The network training uses the SGDM gradient descent optimization algorithm, and the initial learning rate is set to 0.01, which is dynamically adjusted after every 100 iterations. The training data is derived from the combination of SIMPACK simulation data and experimental acquisition data, ensuring the network's generalization ability under different working conditions.

[0102] The multi-channel dynamic information fusion method of the application is particularly suitable for complex line conditions such as high-speed turning, track irregularities, etc. By fusing multi-channel sensor data from the axle box, car body and traction system, the network can capture subtle changes in wheel-rail force under complex conditions. In the case of small curve radius, different networks can accurately predict the change of wheel-rail force through joint analysis of acceleration data and torque data. Experiments show that different network structure parameters have different performances in predicting wheel-rail force. Table 1 shows three different network parameters, and Table 5 gives the comparison of the identification performance of the three networks.

[0103] Table 5 Comparison of network identification performance

[0104]

[0105] In order to verify the effect of the application in practical application, a high-speed railway train is selected for wheel-rail force measurement experiment. The experiment selects running data under different curve radii, speed intervals and track irregularities and other complex line conditions, and collects signals including car body acceleration, bogie vibration and axle box acceleration through a multi-channel dynamic information acquisition device.

[0106] Referring to the flow of Figure 9 The train travels in a straight line at a speed of 20 m / s (72 km / h), and through multi-channel dynamic information fusion, the vibration acceleration signals of the train body and bogie are monitored in real time. The deep learning model is used to process the signals. The results show that the measurement method of the application under high-speed large-radius driving conditions shows a vertical force peak value of 2500N and a lateral force peak value of 1800N. After processing by the convolutional neural network (CNN), the average absolute error (MAE) between the predicted value and the true measured value is 120N, proving that the system has high recognition accuracy under high-speed working conditions.

[0107] The train enters a small curve track area with a radius of 400 meters, and due to the small curve radius, the wheel-rail force changes dramatically. Through the fusion of multi-channel dynamic information, the system can effectively capture the dynamic changes of acceleration and torque, and the wheel-rail force data recorded by the system shows that the maximum vertical force is 5600N and the maximum lateral force is 2400N. The average absolute error (MAE) of the predicted wheel-rail force value and the actual measurement result is 560N, and the determination coefficient R 2 is 0.988, showing extremely high recognition accuracy.

[0108] When the train passes through the track irregularity section, the wheel-rail force changes significantly. The system identifies the wheel-rail force fluctuation caused by track irregularity by fusing dynamic information from different parts, and through the optimization of the deep learning model, it can accurately predict the wheel-rail force change trend under these complex conditions. The experimental results show that the wheel-rail force fluctuates significantly, and the vertical force fluctuation amplitude is between 1000N and 5000N. The system can accurately capture these fluctuations and control the error within 5%, the recognition error of the system is small, and it has strong robustness, and the experimental results are shown in Table 6.

[0109] Table 6 Comparison of effects of three cases

[0110]

[0111] The present application proposes a new solution to the problems of traditional methods, such as high cost, low recognition accuracy in complex line conditions, etc. The method arranges multiple sensors on the vehicle body, bogie, axle box and other key parts to collect multi-channel dynamic data including acceleration, vibration and torque in real time, and combines convolutional neural network for feature extraction and classification identification. This method is especially suitable for real-time monitoring of wheel-rail force in complex line conditions such as high-speed railway and urban rail transit, which can significantly improve the recognition accuracy, reduce the maintenance cost of the system, and enhance the robustness and adaptability of the system. The experimental results show that the recognition accuracy of the present application in complex conditions such as small curve radius and track irregularity reaches 0.99, and the error is controlled within 5%, which has wide application prospect

[0112] The multi-channel dynamic information fusion network of the present application is mainly applied to real-time monitoring of wheel-rail force in rail transit system. This application scenario is especially suitable for train operation monitoring in complex line conditions such as high-speed railway, urban rail transit and small radius track. In these scenarios, the dramatic changes in wheel-rail force will directly affect the safety and stability of train operation, so real-time and accurate measurement of wheel-rail force is crucial to maintain and ensure the safety of rail transit system.

[0113] Substantial adjustments made by the algorithm to suit specific application scenarios In order to adapt to the above specific application scenarios, the algorithm makes the following substantial adjustments:

[0114] The present application combines multi-channel kinetic information such as acceleration, vibration and other multi-dimensional data of multiple parts of the vehicle, which are collected by different sensors and reflect the running state of the train under complex line conditions. During the model training process, a combination of SIMPACK simulation data and experimental data is used to enhance the generalization ability of the model under different working conditions.

[0115] The network designed in the present application is specifically designed for processing and analyzing multi-channel kinetic data during the running of the train, and is particularly suitable for identifying changes in wheel-rail force under complex track conditions. This multi-sensor data fusion can accurately capture the dynamic changes of the train on curves, high-speed and irregular tracks.

[0116] In order to process multi-dimensional data, the network structure parameters are specially designed, such as the number of convolution layers, the size of the convolution kernel, the selection of the activation function, and the number of neurons in the fully connected layer, etc. are optimized through experiments to adapt to specific dynamic data input and changes under complex line conditions.

[0117] In order to improve the real-time performance and recognition accuracy of the algorithm, a convolutional neural network (CNN) is used as a specific implementation method, and multi-channel sensor data fusion processing is introduced.

[0118] The training process combines wheel-rail force data under various complex conditions, including track curves, high-speed driving and irregular conditions, to enhance the robustness of the model. At the same time, the algorithm optimizes the selection of the SGDM (momentum gradient descent) method, with an initial learning rate of 0.01, and dynamically adjusts it according to the working conditions.

[0119] The network architecture includes multiple convolutional layers, fully connected layers and output layers. The convolutional layers are used to extract spatial features from the kinetic data, and the fully connected layers integrate these features for classification and prediction. The model level is designed to improve processing speed and recognition accuracy.

[0120] Solves specific technical problems in new scenarios and brings beneficial effects The present application solves the problem of real-time monitoring of wheel-rail force in rail transit systems, especially the problem of high-precision, real-time monitoring under complex line conditions that traditional methods cannot achieve. The specific technical problems include:

[0121] The data fusion of multi-channel sensors effectively solves the problem that a single sensor cannot fully collect kinetic data.

[0122] By using deep learning algorithm for automatic feature extraction, the invention has significantly improved the recognition accuracy of wheel-rail force under complex working conditions. Experiments show that the system's R 2 value reaches 0.988 under the condition of curve radius as small as 400 meters.

[0123] The synchronous acquisition of multiple sensors and the efficient processing of convolutional neural network enable the system to complete data processing within a 0.1 second time window, ensuring the real-time performance of the system.

[0124] Through multi-dimensional data fusion and deep learning network, the system can efficiently identify the slight changes of wheel-rail force under complex working conditions, especially under high speed or track irregularity conditions, with an identification error within 5%.

[0125] The network can adapt to different line conditions and train models, reducing the maintenance difficulty and use cost of the equipment, and is suitable for a wide range of rail transit systems.

[0126] Through improved data processing and fusion technology, the system can provide real-time monitoring and feedback during vehicle operation, improving the safety and maintenance efficiency of rail transit.

[0127] Embodiment 3, refer to Figure 10 , provides a multi-channel dynamic information fusion wheel-rail force measurement system, comprising:

[0128] a data acquisition module 100, a data processing module 200, a format conversion module 300, a feature extraction module 400, and a wheel-rail force prediction module 500;

[0129] The data acquisition module 100 is used to acquire multi-channel dynamic data during vehicle operation through sensors;

[0130] The data processing module 200 is used to normalize the collected dynamic data;

[0131] The format conversion module 300 is used to discretize and image the normalized data, and convert it into a two-dimensional matrix format suitable for convolutional neural network input;

[0132] The feature extraction module 400 is used to extract features from the preprocessed data using a multi-dimensional dynamic feature extraction network, and classify and identify the changes of wheel-rail force; the multi-dimensional dynamic feature extraction network is used to analyze the multi-channel dynamic data and extract key features related to wheel-rail force;

[0133] The wheel-rail force prediction module 500 is used to monitor and identify the train wheel-rail force in real time under complex line working conditions through multi-channel dynamic information fusion.

[0134] Embodiment 4, one embodiment of the present application, which is different from the previous three embodiments is:

[0135] The functions described can be implemented in software, firmware, hardware, or any combination thereof. If implemented in software and as an independent application, the functions described can be stored in one or more of the storage devices described above. The functions described can also be implemented in platform-specific or other hardware execution environments on the device. The software implementation can be implemented in a targeted language such as, by way of example, assembly language, machine language, machine code, high-level languages such as C, C++, or HTTP, and / or others.

[0136] Logic and / or steps represented in flow diagrams or otherwise described herein, for example, can be embodied in computer-readable instructions, orders / writes of operations, and / or in any other implementation techniques for implementation, for example, for execution by a processing unit or other computing device. The computer-readable instructions can be executed by the processing unit using a computer program or other processor- executable software. The computer-readable instructions can be stored on a computer-readable medium, which can include any medium that can store or transfer programming code to be used by the processing unit or other computing device. The computer-readable medium can be tangible and / or non-transitory. A computer program product can include a computer-readable medium.

[0137] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electrical device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium can even be paper or other suitable medium upon which the program is printed, as the program can be electronically captured, for example, via an optical scanner, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

[0138] It should be understood that portions of the present application can be implemented in hardware, software, firmware, or combinations thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, implementation can be with any or a combination of the following technologies, which are all well known in the art: a discrete logic circuit having logic gates for implementing logic functions upon an application of data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0139] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and all should be covered in the scope of the claims of the present application.

Claims

1. A multi-path dynamic information fusion wheel-rail force measurement method, characterized in that, The method comprises the following steps: Collecting multi-channel dynamic data during vehicle operation through sensors; Normalizing the collected dynamic data; Discretizing and imaging the normalized data, and converting it into a two-dimensional matrix format suitable for input of a convolutional neural network; Extracting features from the preprocessed data using a multi-dimensional dynamic feature extraction network to classify and identify changes in wheel-rail force; Analyzing the multi-channel dynamic data through the multi-dimensional dynamic feature extraction network to extract key features related to wheel-rail force; Real-time monitoring and identification of train wheel-rail force under complex line conditions through multi-channel dynamic information fusion.

2. The multi-pass, kinematically-fused, wheel-rail force measurement method of claim 1, wherein: The multi-channel dynamic data includes vehicle body longitudinal vibration acceleration, lateral vibration acceleration, vertical vibration acceleration, and torque data of the bogie and axle box.

3. The multi-pass, kinematically-fused, wheel-rail force measurement method of claim 2, wherein: The convolutional neural network includes at least 8 input channels, each corresponding to a dynamic data source.

4. The multi-pass, kinematically-fused, wheel-rail force measurement method of claim 3, wherein: The method includes real-time wheel-rail force identification using a multi-dimensional dynamic feature extraction network.

5. The multi-pass, kinematically-fused, wheel-rail force measurement method of claim 4, wherein: The network architecture of the multi-dimensional dynamic feature extraction network includes 3 convolutional layers, 2 fully connected layers, and 1 output layer. Each convolutional layer is followed by a ReLU activation function and a max-pooling operation. The final output of the network is the predicted wheel-rail force value, and the network is trained using the SGDM gradient descent optimization algorithm.

6. The multi-pass, kinematically-fused, wheel-rail force measurement method of claim 5, wherein: The real-time monitoring and identification of train wheel-rail force includes capturing the changes in wheel-rail force under complex conditions by fusing multi-channel sensor data from the axle box, vehicle body, and traction system; In the case of small curve radius, different networks are used to analyze acceleration data and torque data to predict changes in wheel-rail force.

7. A system for measuring wheel-rail force using the multi-channel kinetic information fusion method according to any one of claims 1 to 6, characterized in that, The method comprises the following steps: A data acquisition module (100), a data processing module (200), a format conversion module (300), a feature extraction module (400), and a wheel-rail force prediction module (500); The data acquisition module (100) is used to collect multi-channel dynamic data during vehicle operation through sensors; The data processing module (200) is used to normalize the collected dynamic data; The format conversion module (300) is used to discretize and image the normalized data, and convert it into a two-dimensional matrix format suitable for input of a convolutional neural network; The feature extraction module (400) is used to extract features from the preprocessed data using a multi-dimensional dynamic feature extraction network to classify and identify changes in wheel-rail force; Through the multi-dimensional dynamic feature extraction network, the multi-channel dynamic data is analyzed, and key features related to wheel-rail force are extracted; The wheel-rail force prediction module (500) is used to real-time monitor and identify train wheel-rail force under complex line conditions through multi-channel dynamic information fusion.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to realize the steps of the multi-channel dynamic information fusion wheel-rail force measurement method in any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the multi-channel dynamic information fusion wheel-rail force measurement method in any one of claims 1 to 6.

10. A computer program product, characterised in that, The computer program product comprises computer instructions, and a processor of a computer device reads the computer instructions, and the processor of the computer device implements the steps of the multi-path dynamic information fusion wheel-rail force measurement method in any one of claims 1 to 6 when executed.

Images