Train approaching early warning system and method

Abstract

The invention relates to a train approaching early warning system and method, and the system comprises an embedded rail vibration signal collection device which comprises a three-axis vibration sensor which can be adsorbed on a rail to collect the vibration data of the rail in the X axis, the Y axis and the Z axis; the neural network model is used for analyzing the vibration data, converting the neural network model trained at the server side into a format suitable for the embedded rail vibration signal acquisition equipment, and deploying the neural network model in the embedded rail vibration signal acquisition equipment; the neural network model comprises a dual-channel input convolution layer, a lower sampling layer, an upper sampling layer and an output layer; wherein the two-channel input convolution layer processes the high-frequency part and the low-frequency part in the vibration data, the features of the high-frequency part and the low-frequency part are fused, and a classification result is generated through the output layer after gradual compression of the lower sampling layer and reconstruction of the upper sampling layer. Rail vibration signals are processed on site in real time, and the accuracy and response speed of train approaching early warning are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of train approach monitoring equipment, and in particular to a train approach warning system and method. Background Art

[0002] The train approach warning system plays an important role in improving railway construction safety. The traditional technical solution has the following two shortcomings:

[0003] On the one hand, due to the insufficient computing power of traditional chips, it is difficult to complete complex spectrum transformations (such as Fourier transform, etc.) in real time, which makes it difficult to run complex neural network models on rail vibration signal acquisition equipment. To address this defect, the existing solution is to separate the rail vibration signal acquisition equipment from the computing unit, and then send the rail vibration signal to the remote computing unit for analysis and processing by transmitting data. This will cause a delay in rail vibration signal monitoring and is not suitable for train approach monitoring during construction.

[0004] On the other hand, there is a lack of specialized algorithms, that is, the existing rail vibration monitoring solutions lack algorithms that are specifically designed to monitor and analyze vibration signals generated by approaching trains. Therefore, when it is necessary to collect track vibration signals to analyze the approaching time of trains, it is necessary to develop new algorithms or customize existing algorithms to identify and process specific vibration signals when trains approach. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a train approach warning system for collecting rail vibration signals to analyze train approach warning information, the system comprising:

[0006] An embedded rail vibration signal acquisition device includes a three-axis vibration sensor, which can be attached to the rail to collect vibration data of the rail in three axes of XYZ; and

[0007] A neural network model for analyzing the vibration data converts the neural network model trained on the server side into a format suitable for the embedded rail vibration signal acquisition device and deploys it on the embedded rail vibration signal acquisition device; the input of the neural network model includes the vibration data; the neural network model includes a dual-channel input convolution layer, a downsampling layer, an upsampling layer and an output layer; wherein the dual-channel input convolution layer processes the high-frequency and low-frequency parts of the vibration data, and fuses the features of the high-frequency and low-frequency parts, which are gradually compressed by the downsampling layer and reconstructed by the upsampling layer, and a classification result is generated through the output layer.

[0008] Furthermore, the dual-channel input convolution layer extracts preliminary features of the vibration data through two convolution operations; wherein the convolution kernel size is set to (7, 3) and the stride is set to (3, 1).

[0009] Furthermore, the convolution kernel size in the downsampling layer is set to (3,3) and the stride is set to (2,1).

[0010] Furthermore, the upsampling layer is combined with batch normalization and a Sigmoid activation function to ensure that the accuracy of features is maintained during the restoration process and to generate a higher resolution output.

[0011] Furthermore, the output layer includes a convolution layer and a one-dimensional convolution layer, and generates a classification probability distribution through a Softmax function, and the category with the highest probability value will be selected as the final classification result.

[0012] Furthermore, the embedded rail vibration signal acquisition device also includes a housing having a housing, wherein the housing has an integrated circuit mainboard and a 4G module circuit mainboard built in; the integrated circuit mainboard includes a main control module, an Ethernet module, a WIFI module, an internal storage module and a power management module; the 4G module circuit mainboard includes a 4G module and an Internet of Things card module; wherein:

[0013] The main control module is used to receive and analyze the vibration data;

[0014] The 4G module is used to send the information of the embedded rail vibration signal acquisition device to the server side.

[0015] Furthermore, the end face of the shell reveals a switch, an IoT card slot, a memory card slot, an Ethernet interface, and a USB Type-C standard interface for an external charger; a 4G antenna is provided on one side of the shell.

[0016] Furthermore, a strong magnet is provided on the contact surface between the three-axis vibration sensor and the rail, and the three-axis vibration sensor is connected to the integrated circuit mainboard by wires.

[0017] A train approach warning method is applied to the train approach warning system.

[0018] Furthermore, the method comprises the following steps:

[0019] S1. Collection of vibration data: using the embedded rail vibration signal collection device 100 to collect vibration data of the rail 200 in multiple dimensions;

[0020] S2. Preprocessing of vibration data: standardizing the vibration data, converting the time domain signal into frequency domain representation through short-time Fourier transform, and dividing the vibration data into high-frequency and low-frequency parts for separate processing;

[0021] S3. Conversion and deployment of the neural network model: using TFLiteConverter to convert the UNet model trained on the server into a .tflite format suitable for embedded devices, and deploy it to the embedded rail vibration signal acquisition device 100;

[0022] S4, edge computing node analysis: extract features from the vibration data and fuse the features of the high-frequency and low-frequency parts; after gradual compression by the downsampling layer and reconstruction by the upsampling layer, a classification result is finally generated through the FinalConv layer to determine whether there are potential safety hazards;

[0023] S5. Generation and transmission of warning signals: If the model predicts that a certain vibration pattern is associated with a safety hazard, that is, if the vibration pattern of an approaching train is identified, the system immediately generates a warning signal, which is instantly transmitted to nearby construction personnel via the local area network and uploaded to the cloud server via 4G and GPS modules.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The present invention uses an embedded rail vibration signal acquisition device and a neural network model on the device for analyzing vibration signals, which can process rail vibration signals in real time on site, convert time domain signals into frequency domain signals through short-time Fourier transform, and use the UNet model for analysis, effectively extracting time-varying features in the signals, which helps to identify potential train approach signals, improve the accuracy and response speed of train approach warnings, and ensure the safety of railway construction workers. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a schematic diagram of a state where a three-axis vibration sensor disclosed in an embodiment of the present invention is adsorbed on a rail;

[0027] Figure 2 It is a structural schematic diagram of an embedded rail vibration signal acquisition device disclosed in an embodiment of the present invention;

[0028] Figure 3 A schematic diagram of the composition of a 4G module circuit mainboard disclosed in an embodiment of the present invention;

[0029] Figure 4 A schematic diagram of the composition of an integrated circuit mainboard disclosed in an embodiment of the present invention;

[0030] Figure 5The present invention is a flowchart of a train approach warning algorithm disclosed in an embodiment of the present invention.

[0031] In the figure:

[0032] 100. Embedded rail vibration signal acquisition equipment;

[0033] 110. Three-axis vibration sensor;

[0034] 120, housing;

[0035] 121, integrated circuit main board; 121a, main control module; 121b, Ethernet module; 121c, WIFI module; 121d, internal storage module; 121e, power management module;

[0036] 122, 4G module circuit mainboard; 122a, 4G module; 122b, Internet of Things card module;

[0037] 123. IoT card slot; 124. Memory card slot; 125. Switch; 126. USB Type-C standard interface; 127. Ethernet interface; 128. 4G antenna;

[0038] 200. Railway tracks. DETAILED DESCRIPTION

[0039] In order to make the technical scheme and technical effect of the present invention clearer, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments.

[0040] The present invention aims to provide a train approach warning system for collecting vibration signals of rails 200 to analyze train approach warning information, thereby ensuring the safety of railway construction workers.

[0041] The train approach warning system mainly includes two parts: an embedded rail vibration signal acquisition device 100 and a neural network model for analyzing the vibration data, which are mainly introduced below:

[0042] 1. Embedded rail vibration signal acquisition equipment 100

[0043] refer to Figure 1 The embedded rail vibration signal acquisition device 100 mainly includes a triaxial vibration sensor 110 and a housing 120, and the triaxial vibration sensor 110 and the housing 120 are connected by wires. The triaxial vibration sensor 110 can be adsorbed on the side wall of the rail 200 to collect vibration data of the rail 200 in the XYZ three axes, and can capture subtle changes when a train approaches from three different directions, greatly improving the accuracy of recognition.

[0044] Preferably, a strong magnet is provided on the contact surface between the triaxial vibration sensor 110 and the rail 200. Figure 2 The housing 120 has a receiving chamber, in which an integrated circuit mainboard 121 and a 4G module circuit mainboard 122 are housed.

[0045] refer to Figure 3 The integrated circuit main board 121 includes a main control module 121a, an Ethernet module 121b, a WIFI module 121c, an internal storage module 121d and a power management module 121e. The three-axis vibration sensor 110 is connected to the integrated circuit main board 121 by wires; the main control module 121a is used to receive and analyze the vibration data of the rail 200; the internal storage module 121d is used to store the data sent by the three-axis vibration sensor 110; preferably, the power management module 121e is a rechargeable lithium battery pack.

[0046] refer to Figure 4 The 4G module circuit mainboard 122 includes a 4G module 122a and an IoT card module 122b. The 4G module 122a is used to send the information of the embedded rail vibration signal acquisition device 100 to the server.

[0047] The end surface of the shell 120 also reveals a switch 125, an Internet of Things card slot 123, a memory card slot 124, an Ethernet interface 127 and a USB Type-C standard interface 126 for an external charger; a 4G antenna 128 is set on one side of the shell 120.

[0048] 2. Neural Network Model

[0049] This model is a neural network model designed for the railway vibration signal warning system. It uses an improved version of the U-Net structure to adapt to complex vibration data analysis tasks. The UNet model trained on the server side is converted into a .tflite format suitable for the embedded rail vibration signal acquisition device 100 using TFLiteConverter, and deployed in the embedded rail vibration signal acquisition device 100. The input of the neural network model includes vibration data of the three axes of the rail 200XYZ, and is intended to capture and analyze the key features in these data to achieve accurate warning of train approach and abnormal conditions.

[0050] The advantage of the UNet model is that it has a lightweight network architecture, which makes it particularly suitable for processing resource-constrained environments such as embedded devices or mobile applications. At the same time, UNet can still achieve high-performance segmentation while keeping the model size and computational complexity small.

[0051] The neural network model includes a dual-channel input convolution layer, a downsampling layer, an upsampling layer and an output layer; the dual-channel input convolution layer processes the high-frequency and low-frequency parts of the vibration data, and fuses the features of the high-frequency and low-frequency parts, which are gradually compressed by the downsampling layer and reconstructed by the upsampling layer, and the classification results are generated through the output layer.

[0052] The core components of this neural network model are as follows:

[0053] 2.1. DoubleConv: a dual-channel input convolution layer that extracts preliminary features of the vibration signal through two convolution operations. Each convolution layer is followed by batch normalization, ReLU activation function, and Dropout layer to improve the learning ability of the model and reduce overfitting. The convolution kernel size is set to (7,3) and the stride is (3,1), which can effectively capture features in a larger range while maintaining the integrity of spatial information.

[0054] 2.2. Down: The downsampling layer is used to gradually reduce the spatial dimension of the feature map while increasing the depth of the feature. Downsampling is achieved through a small convolution kernel (3,3) and a stride (2,1), which helps the model extract more abstract features at different levels. This operation helps the model gradually focus on more important high-level information.

[0055] 2.3. Up: The upsampling layer restores the spatial resolution of the feature map through transposed convolution and reduces the number of channels of the feature map. The upsampling layer combines batch normalization and Sigmoid activation function. Among them, batch normalization is a commonly used regularization technique that can keep the inputs of each layer of the network in the same distribution, thereby accelerating the convergence speed of the network and improving the stability of the model. Using batch normalization in the upsampling process can ensure the consistency of scales between different feature maps, reduce internal covariate shifts, and make the network more stable; the Sigmoid function is a common nonlinear activation function that compresses the input value into the range of (0,1) so that its output has the nature of probabilistic interpretation. This feature makes the Sigmoid function particularly suitable for binary classification problems, and can map features to a probability space to represent the possibility of a certain category. The advantages of combining batch normalization and Sigmoid activation function in the upsampling operation are: on the one hand, batch normalization reduces the differences between features through standardization and improves the generalization ability of the model; on the other hand, the nonlinear characteristics of the Sigmoid function enhance the expressiveness of the model, enabling it to better capture complex data patterns. This ensures that feature accuracy is maintained during the restoration process and produces higher resolution output.

[0056] 2.4. FinalConv: The output layer is responsible for generating the final output of the model. The output layer includes a convolution layer and a one-dimensional convolution layer. Finally, the classification probability distribution is generated through the Softmax function, and the category with the highest probability value will be selected as the final classification result. This process converts the features extracted by the model into predictions for specific categories, thereby achieving accurate classification and early warning of vibration signals.

[0057] refer to Figure 5 , the algorithm applied to the above train approach warning system includes the following steps:

[0058] S1. Collection of vibration data: using the embedded rail vibration signal collection device 100 to collect vibration data of the rail 200 in multiple dimensions;

[0059] S2. Preprocessing of vibration data: First, standardize the signal to make it have zero mean and unit variance. Then, convert the time domain signal into frequency domain representation through short-time Fourier transform (STFT), so that the frequency change over time can be observed more clearly. Then, separate the signal into high-frequency and low-frequency parts, and process them separately, which helps to capture the characteristics of different frequency bands of the signal and improve the sensitivity of the model to different types of signals.

[0060] S3. Conversion and deployment of neural network model: Use TFLiteConverter to convert the UNet model trained on the server into a .tflite format suitable for embedded devices, and deploy it to the embedded rail vibration signal acquisition device 100;

[0061] S4, edge computing node analysis: First, extract the features of the vibration data of the XYZ axes, then process the high-frequency and low-frequency parts through the DoubleConv layer, and fuse the features of the two parts; after gradual compression by the downsampling layer and reconstruction by the upsampling layer, finally generate the classification results through the FinalConv layer to determine whether there are potential safety hazards;

[0062] S5. Generation and dissemination of warning signals: Once the model completes the classification task and gives a prediction result, the warning mechanism can be implemented based on this result. If the model predicts that a certain vibration pattern is associated with a safety hazard, that is, it identifies the vibration pattern of an approaching train, then the system will immediately generate a warning signal. These signals are instantly transmitted to nearby construction workers through the local area network, reminding relevant personnel to take necessary measures, and uploaded to the cloud server through 4G and GPS modules.

[0063] In addition, the system also has information management and access functions, that is, users can access early warning data in real time and view historical records to assess the safety of the construction area and make corresponding adjustments.

[0064] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A train approach warning system for collecting rail (200) vibration signals to analyze train approach warning information, characterized in that: The system comprises: An embedded rail vibration signal acquisition device (100) comprises a three-axis vibration sensor (110), wherein the three-axis vibration sensor (110) can be adsorbed on a rail (200) to acquire vibration data of the rail (200) in three axes (X, Y, and Z); and A neural network model for analyzing the vibration data converts the neural network model trained on the server side into a format suitable for the embedded rail vibration signal acquisition device (100), and deploys it on the embedded rail vibration signal acquisition device (100); the input of the neural network model includes the vibration data; the neural network model includes a dual-channel input convolution layer, a downsampling layer, an upsampling layer and an output layer; wherein the dual-channel input convolution layer processes the high-frequency and low-frequency parts of the vibration data, and fuses the features of the high-frequency and low-frequency parts, which are gradually compressed by the downsampling layer and reconstructed by the upsampling layer, and a classification result is generated through the output layer.

2. The train approach warning system according to claim 1, characterized in that: The dual-channel input convolution layer extracts preliminary features of the vibration data through two convolution operations; wherein the convolution kernel size is set to (7, 3) and the stride is set to (3, 1).

3. The train approach warning system according to claim 2, characterized in that: The convolution kernel size in the downsampling layer is set to (3,3) and the stride is set to (2,1).

4. The train approach warning system according to claim 3, characterized in that: The upsampling layer combines batch normalization and sigmoid activation function to ensure that the accuracy of features is maintained during the restoration process and generate higher resolution output.

5. The train approach warning system according to claim 4, characterized in that: The output layer includes a convolution layer and a one-dimensional convolution layer, and generates a classification probability distribution through a Softmax function, and the category with the highest probability value will be selected as the final classification result.

6. The train approach warning system according to claim 1, characterized in that: The embedded rail vibration signal acquisition device (100) further comprises a housing (120) having a receiving chamber, wherein the housing (120) has an integrated circuit mainboard (121) and a 4G module circuit mainboard (122) built therein; the integrated circuit mainboard (121) comprises a main control module (121a), an Ethernet module (121b), a WIFI module (121c), an internal storage module (121d) and a power management module (121e); the 4G module circuit mainboard (122) comprises a 4G module (122a) and an Internet of Things card module (122b); wherein: The main control module (121a) is used to receive and analyze the vibration data; The 4G module (122a) is used to send information of the embedded rail vibration signal acquisition device (100) to the server side.

7. The train approach warning system according to claim 6, characterized in that: The end surface of the shell (120) is exposed with a switch (125), an Internet of Things card slot (123), a memory card slot (124), an Ethernet interface (127), and a USB Type-C standard interface (126) for an external charger; a 4G antenna (128) is arranged on one side of the shell (120).

8. The train approach warning system according to claim 6, characterized in that: A strong magnet is provided on the contact surface between the three-axis vibration sensor (110) and the rail (200), and the three-axis vibration sensor (110) is connected to the integrated circuit mainboard (121) by wires.

9. A train approach warning method, characterized in that: The method is applied to the train approach warning system described in any one of claims 1-8.

10. The train approach warning method according to claim 9, characterized in that: The method comprises the following steps: S1. Collection of vibration data: using an embedded rail vibration signal collection device (100) to collect vibration data of the rail (200) in multiple dimensions; S2. Preprocessing of vibration data: standardizing the vibration data, converting the time domain signal into frequency domain representation through short-time Fourier transform, and dividing the vibration data into high-frequency and low-frequency parts for separate processing; S3. Conversion and deployment of the neural network model: using TFLiteConverter to convert the UNet model trained on the server into a .tflite format suitable for embedded devices, and deploy it to the embedded rail vibration signal acquisition device 100; S4, edge computing node analysis: extract features from the vibration data and fuse the features of the high-frequency and low-frequency parts; after gradual compression by the downsampling layer and reconstruction by the upsampling layer, a classification result is finally generated through the FinalConv layer to determine whether there are potential safety hazards; S5. Generation and transmission of warning signals: If the model predicts that a certain vibration pattern is associated with a safety hazard, that is, if the vibration pattern of an approaching train is identified, the system immediately generates a warning signal, which is instantly transmitted to nearby construction personnel via the local area network and uploaded to the cloud server via 4G and GPS modules.

Images