Multi-sensor fusion rack rail railway foreign matter cleaning system and control method

Abstract

The invention relates to the technical field of rack rail railway operation and maintenance, in particular to a multi-sensor fusion rack rail railway foreign matter cleaning system which comprises a sensing module with a first radar, an image acquisition device and a second radar, and the first radar and the image acquisition device are installed at preset positions of the upper portion of a train head. The first radar and the image acquisition equipment are used for acquiring train front environment data; the second radar is installed at the bottom of the train head and used for collecting data of foreign matter between racks and tooth grooves. The decision-making module is electrically connected with the sensing module, and the decision-making module is used for processing data collected by the sensing module and outputting an instruction; and the execution module is electrically connected with the decision module, and is used for cleaning the rack rail according to an instruction of the decision module. According to the multi-sensor fusion rack rail railway foreign matter cleaning system, the operation safety, reliability and passing efficiency of a rack rail train in a complex mountain environment are improved.

Description

A multi-sensor fusion foreign object removal system and control method for rack rail railways Technical Field

[0001] This invention relates to the field of rack railway operation and maintenance technology, and in particular to a multi-sensor fusion rack railway foreign object removal system and control method. Background Technology

[0002] The traction and braking longitudinal force of ordinary rail vehicles mainly depends on the adhesion between the steel wheel and the rail. The maximum gradient that ordinary rail vehicles can climb is 40‰-60‰. When the gradient is too large, the adhesion will be insufficient due to the slippage of the wheel-rail contact surface.

[0003] A rack and pinion railway is a rail transit system specifically designed for extremely steep gradients. It provides powerful longitudinal traction to the train through the meshing of gears and racks fixed to the track. Relying on this mechanical meshing, rack and pinion railways can climb gradients up to 480‰, far exceeding that of ordinary railways. This makes it a crucial mode of transportation connecting numerous mountainous areas, with enormous application demand in mountainous and high-altitude regions. However, this also presents technical challenges to the operational support of rack and pinion railways operating in high-altitude, cold, and steep conditions, including the rack and pinion system, operating vehicles, and equipment. The key to the safe operation of mountain rack and pinion trains lies in effectively ensuring the interaction between the gears and racks. In complex operating environments such as mountains, high altitudes, and steep gradients, the rack and pinion system is constantly exposed to the elements and susceptible to environmental factors, such as: ballast, gravel, branches, and other hard objects becoming lodged in the rack grooves; and ice and snow accumulating in the grooves, potentially freezing into stubborn deposits. Foreign objects lodged in train gears can cause mechanical wear or even damage, seriously threatening the safety of train operation and the reliability of equipment.

[0004] Most foreign object removal devices on the market are designed for ordinary steel rails. For example, Chinese invention patent CN112391991A discloses a foreign object removal device for rail transit. The structure and principle of this device are not applicable to rack and pinion railways. Currently, there are few rack and pinion foreign object removal devices suitable for rack and pinion railways, and their functions are limited. Chinese invention patent CN110205973A discloses a rack and pinion cleaning device for rack and pinion railways. However, its gear-driven brush structure struggles to provide sufficient cutting force or melting ability when facing embedded snow, stubborn ice, and hard, stuck objects, resulting in incomplete removal and significantly limited cleaning efficiency. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing rack rail foreign object cleaning equipment, which has limited functionality and cleaning effect, and to provide a rack rail foreign object cleaning system and control method with multi-sensor fusion.

[0006] In a first aspect, the present invention provides a multi-sensor fusion foreign object removal system for rack and pinion railways, comprising: a sensing module, the sensing module including a first radar, an image acquisition device, and a second radar, the first radar and the image acquisition device being installed at a predetermined position on the upper part of the train head, the first radar and the image acquisition device being used to collect environmental data in front of the train; the second radar being installed at the bottom of the train head, the second radar being used to collect foreign object data between the rack teeth; a decision module, the decision module being electrically connected to the sensing module, the decision module being used to process the data collected by the sensing module and output instructions; and an execution module, the execution module being electrically connected to the decision module, the execution module being used to clean the rack and pinion according to the instructions of the decision module, the execution module being installed at the bottom of the train head, the execution module being located between the second radar and a first traction gear, and the execution module being provided with a cutting unit, a water spraying unit, and an air spraying unit sequentially from the direction near the second radar toward the rear of the train.

[0007] The multi-sensor fusion foreign object removal system for rack railways provided by this invention forms a three-stage composite cleaning mechanism consisting of a cutting unit, a water spraying unit, and an air spraying unit arranged sequentially from front to back. The cutting unit at the front end can mechanically cut and break long, hard foreign objects (such as branches and wood) to prevent them from getting stuck in the rack. For snow and ice, the water spraying unit can use high-temperature, high-pressure water jets to melt and wash them away. The air spraying unit at the back is used for final cleaning and drying. Whether it is blowing the debris after cutting / washing away from the track, using high-pressure airflow to remove fine gravel, or using hot airflow to further melt ice and snow, it can ensure that the rack is cleaned thoroughly and improve the removal efficiency.

[0008] The first radar and image acquisition equipment are positioned at predetermined locations on the upper part of the train's cab. The first radar is responsible for long-range detection of the distance, speed, and size of obstacles, providing early warnings. The image acquisition equipment can identify the specific type of obstacle (such as falling rocks, branches, snow, people, etc.), providing qualitative basis for decision-making. The second radar at the bottom of the cab is responsible for high-precision acquisition of foreign object data between the rack and pinion teeth, providing further evidence for determining the need for clearing. By integrating multiple sensing devices to form a system-wide intelligent detection method, the detection accuracy is improved.

[0009] The multi-sensor fusion foreign object removal system for rack railways provided by this invention improves the operational safety, reliability, and traffic efficiency of rack railway trains in complex mountainous environments.

[0010] Preferably, the first radar is a millimeter-wave radar, and the second radar is a lidar; the first radar operates in the frequency band of 76GHz~77GHz, has a detection accuracy of 0.25m, and a detection size of 15.6cm. 3The maximum detection range is 200m, and the obstacle resolution is ≤0.2m; the detection frequency of the second radar is 50Hz~100Hz, and the detection accuracy is ±1mm; the first radar and the image acquisition device are symmetrically installed at predetermined positions on the upper part of the train head; the second radar is 1.0±0.1m away from the automatic coupler head of the train head; the cutting unit is 1.5m±0.1m away from the automatic coupler head of the train head.

[0011] The multi-sensor fusion foreign object removal system for rack and pinion railways provided by this invention achieves complementary advantages in both long-range and short-range detection through the combined use of a first radar and a second radar.

[0012] Millimeter-wave radar has strong penetration capabilities through rain, snow, and fog, ensuring stable all-weather detection capabilities even in cold, foggy mountainous environments. With a maximum detection range of 200m, it provides ample advance warning time for trains. Optimal detection accuracy of 0.25m and obstacle resolution ≤0.2m are sufficient for long-distance hazard prediction and large obstacle detection.

[0013] The lidar has a detection accuracy of up to ±1mm and scans at a high frequency of 50Hz~100Hz, ensuring that the system can measure the height, shape and volume of foreign objects (such as ice, snow and gravel) in the rack grooves in real time and accurately, providing precise geometric basis for the decision-making module.

[0014] The first radar and image acquisition equipment are symmetrically mounted on the upper part of the front of the vehicle, which helps to provide a wide and balanced forward field of vision and reduce blind spots on both sides.

[0015] Through multiple experiments, it was determined that installing a high-precision lidar at a distance of 1.0 ± 0.1 m from the automatic coupler head of the train head is optimal. This allows for scanning the rack at the best distance and angle, ensuring the accuracy of foreign object data collection, while also allowing sufficient reaction time for the decision-making module and subsequent cleaning actions. The cutting unit is also optimally positioned at 1.5 m ± 0.1 m from the automatic coupler head of the train head, ensuring that the cleaning operation can begin rapidly when the train reaches the foreign object.

[0016] Preferably, the image acquisition device is an industrial camera; the sensor type of the image acquisition device is CCD, with a resolution of ≥250×250 pixels and a maximum detection distance of 100m.

[0017] The preferred image acquisition device is a CCD industrial camera, which helps ensure the accuracy and reliability of image acquisition.

[0018] The 100m detection range of the image acquisition device and the 200m detection range of the first radar form an effective relay and complement, significantly improving the system's accuracy in judging obstacle threats and avoiding misjudgments caused by relying solely on the first radar, thereby achieving more accurate early warning and decision-making.

[0019] Preferably, the cutting unit is a strip saw with a no-load rotation speed of 20m / s to 40m / s, a guide plate length of 40cm, and a voltage of 220V; the water spray unit has a water spray pressure adjustment range of 0.5MPa to 8MPa and a water temperature adjustment range of 50℃ to 90℃; the air jet unit has an air jet pressure adjustment range of 0.5MPa to 8MPa and a temperature adjustment range of 50℃ to 90℃.

[0020] The multi-sensor fusion foreign object removal system for rack railways provided by this invention, through parameterized configuration and adjustable temperature and pressure design of the three units of cutting, water spraying, and air spraying, can remove foreign objects such as hard objects (cutting), ice and snow (high temperature / high pressure water and air) and fine particles (high pressure air), ensuring that the rack remains clean in complex environments, greatly improving the effectiveness of foreign object removal and the operational reliability of rack railways.

[0021] In a second aspect, the present invention provides a control method for a multi-sensor fusion rack and pinion railway foreign object removal system, used to control the aforementioned multi-sensor fusion rack and pinion railway foreign object removal system, comprising the following steps: S1. The sensing module collects environmental data in front of the train: detecting the size of obstacles, the distance S0 between the obstacles and the rack and pinion train, and the speed v0 of the obstacles relative to the train via a first radar; identifying the type of obstacles via an image acquisition device; collecting foreign object data between the rack and pinion teeth via a second radar; transmitting the data collected by the sensing module to the decision module; S2. The decision module processes the data collected by the sensing module: when the first radar detects an obstacle... When the length of the obstacle is greater than 20cm, the third-level warning is activated, the train's power output is cut off, the train is brought to a stop, and an alarm message is sent. When the length of the obstacle detected by the first radar is ≤20cm, the obstacle information detected by the first radar is fused with the obstacle information identified by the image acquisition device: if only the first radar or only the image acquisition device detects the obstacle, the second-level warning is activated, the train slows down, and an alarm message is sent; if both the first radar and the image acquisition device detect the obstacle, and the image acquisition device identifies the obstacle as a tree branch, the second-level warning is activated, the train slows down, and the first command is sent to the execution module. If both the first radar and the image acquisition device detect an obstacle, and the image acquisition device identifies the obstacle as falling rocks, a second-level warning is activated, the train slows down, and a second command is sent to the execution module. If both the first radar and the image acquisition device detect an obstacle, and the image acquisition device identifies the obstacle as snow, and the second radar detects that the snow covers the rack teeth, a first-level warning is activated, the train slows down, and a third command is sent to the execution module. If neither the first radar nor the image acquisition device detects an obstacle, and the second radar detects foreign objects between the rack teeth, a first-level warning is activated, and the train slows down. The fourth instruction is sent to the execution module; S3. The execution module cleans the gear track according to the instruction from the decision module: When the execution module receives the first instruction: start the cutting unit and the jet unit; When the execution module receives the second instruction: start the jet unit, and the jet unit jet pressure is 7MPa~8MPa; When the execution module receives the third instruction: start the water spray unit, and the water spray unit spray temperature is 50℃~90℃; start the jet unit, and the jet unit jet temperature is 50℃~90℃; When the execution module receives the fourth instruction: start the jet unit, and the jet unit jet pressure is 0.5MPa~7MPa.

[0022] This invention provides a control method for a multi-sensor fusion-based foreign object removal system for rack railways. By controlling the multi-sensor fusion-based rack railway foreign object removal system, a three-level early warning control mechanism for foreign object removal is proposed. This mechanism enables graded responses to foreign objects of different threat levels, maximizing the operational safety of rack railway trains and improving the safety margin for mountainous operations. By fusing judgment using a first radar, image acquisition equipment, and a second radar, the system can accurately identify foreign object types (twigs, rocks, snow, etc.) and, combined with foreign object size and location information, effectively eliminates misjudgments from single sensors, avoiding resource waste and erroneous braking, and ensuring the accuracy of system detection. By mapping the accurate foreign object identification results to specific execution unit operations, the cleaning process can be customized and optimized.

[0023] Preferably, in step S1, detecting the size of the obstacle, the distance S0 between the obstacle and the rack train, and the speed v0 of the obstacle relative to the train using the first radar includes the following steps: S11. The first radar uses frequency-modulated continuous wave sampling to convert the first radar data into a world coordinate system; the distance between the obstacle and the rack train is calculated. In the formula, S0 is the distance between the obstacle and the rack train; c is the speed of electromagnetic waves in the air; f0 is the frequency difference between the rising and falling edges of the transmitted signal; Δf is the frequency difference between the transmitted signal and the echo signal; T is the modulation period of the millimeter-wave radar; the speed of the obstacle relative to the train is calculated. In the formula, v0 is the speed of the obstacle relative to the train detected by the first radar; f - f is the frequency difference between the rising edges of the reflected signal and the transmitted signal from the obstacle; + This represents the frequency difference between the falling edge of the signal reflected from the obstacle and the transmitted signal.

[0024] The first radar uses frequency-modulated continuous wave sampling, enabling rapid and accurate measurement of the distance and relative speed of obstacles. This ensures a quick response to changes in the environment ahead, even at high speeds, meeting the train's need for advance warning. Converting radar data into a world coordinate system allows for precise alignment and fusion of obstacle information obtained by the radar with data from other sensors, such as image acquisition equipment.

[0025] Preferably, in S1, identifying obstacle categories using an image acquisition device includes the following steps: S12. The image acquisition device samples the data and converts the sampled photo data into a world coordinate system; S13. Image preprocessing is performed, including the following steps: S131. The original color image acquired by the image acquisition device is processed to grayscale, using the following formula:

[0026] In the formula, Gray(i,j) is the gray value of the pixel at point (i,j); R(i,j) is the red channel pixel value at point (i,j); G(i,j) is the green channel pixel value at point (i,j); B(i,j) is the blue channel pixel value at point (i,j); i is the x-coordinate of the pixel; j is the y-coordinate of the pixel; S132. Denoise the image processed in S131 using the following formula:

[0027] In the formula, g(i,j) is the two-dimensional median filter at point (i,j); f(i,j) is the original image value at point (i,j); Med() is the median calculation function; S14. Delineate the boundaries of the image processed in S13: S141. Extract the feature boundaries in the image using the edge detection method, and the calculation formula is as follows:

[0028] In the formula, y(i,j) is the output image at point (i,j); F(i,j) is the input image at point (i,j); S141. The midpoint between the rack and the tracks on both sides of the rack is selected as the boundary region, and the calculation formula is as follows:

[0029] In the formula, u r The position is the middle of the rack and the rail on its right; u l The position is the midpoint between the rack and the rail to its left; u rx The characteristic point of the right track line of the toothed rail; u lx is the feature point of the left track line of the toothed track; x is the feature point corresponding to the x-th track line's horizontal coordinate; S15. Crop the image processed in S14 and reduce it proportionally; S16. Perform sample training, using a convolutional neural network algorithm to train on at least 3000 sample datasets to identify obstacle categories.

[0030] By converting color images to grayscale through image acquisition equipment, the computational complexity of subsequent processing is reduced while preserving image brightness information, making it suitable for edge detection and feature extraction. Median filtering enhances the clarity and recognizability of image features, improving the accuracy of subsequent algorithms. Edge detection extracts the contour information of the track, performs boundary delineation, identifies key recognition areas, and eliminates irrelevant interference. By determining the boundary area of ​​the rack and its clearing region, subsequent recognition is performed only within the defined area, significantly reducing the interference of irrelevant information such as background, sky, and mountains on the recognition accuracy. The convolutional neural network algorithm has powerful feature learning capabilities, automatically extracting subtle features of foreign objects from complex mountain images, achieving qualitative classification of key clearing targets such as "fallen rocks, branches, and snow." A dataset of at least 3000 samples ensures the model's generalization ability, enabling it to maintain high recognition accuracy even when facing new, untrained scenes.

[0031] Preferably, S13. Image preprocessing further includes the following step: S133. Enhancing the image processed in S132, using the following calculation formula:

[0032] In the formula, p s (s) is the probability density function of s; p r (r) is the probability density function of r; r and s are random numbers between (0,1).

[0033] Mountainous environments exhibit dramatic changes in lighting, potentially resulting in strong shadows or localized overexposure. Histogram equalization enhances images, effectively reducing the impact of uneven lighting and visibility on the picture, thus improving the accuracy of obstacle category recognition.

[0034] Preferably, the obstacle category identified in S16 includes at least one of the following: people, animals, ice and snow, falling rocks, and tree branches.

[0035] Preferably, in S11, the first radar uses FMCW frequency-modulated continuous wave with a sampling frequency of 20Hz; in S12, the image acquisition device samples at a sampling frequency of 20Hz; in S15, the image processed in S14 is cropped and proportionally reduced to no more than 608×608 pixels.

[0036] By setting the sampling frequency of both the first radar and the image acquisition device to 20Hz, it was ensured that the information acquired by the first radar and the information acquired by the image acquisition device were time-aligned during data fusion in the decision module. Limiting the image size to 608×608 or smaller significantly reduced the computational load of the convolutional neural network and improved the system's response speed.

[0037] Compared with existing technologies, the beneficial effects of this invention are as follows: 1. This invention provides a multi-sensor fusion foreign object removal system for rack and pinion railways. A three-stage composite cleaning mechanism is formed by a cutting unit, a water spraying unit, and an air spraying unit arranged sequentially from front to back. The cutting unit at the forefront can mechanically cut and break long, hard foreign objects (such as branches and wood) to prevent them from getting stuck in the rack. For snow and ice, the water spraying unit can use high-temperature, high-pressure water jets to melt and flush them. The air spraying unit at the end is used for final cleaning and drying, whether it's blowing the cut / flushed debris off the track or using high-pressure airflow to remove fine particles. Small gravel or hot airflow can further melt ice and snow, ensuring the rack is thoroughly cleaned and improving removal efficiency. 2. This invention provides a multi-sensor fusion rack rail foreign object removal system. A first radar and image acquisition device are positioned at a predetermined location on the upper part of the train head. The first radar is responsible for long-range detection of the distance, speed, and size of obstacles, providing early warnings. The image acquisition device can identify the specific type of obstacle (such as falling rocks, branches, snow, people, etc.), providing qualitative basis for decision-making. A second radar at the bottom of the train head is responsible for high-precision acquisition of foreign object data between the rack teeth, providing further evidence for determining the need for cleaning. By integrating multiple sensing devices to form an intelligent detection method, the detection accuracy is improved.

[0038] 3. This invention provides a control method for a multi-sensor fusion-based foreign object removal system for rack railways. By controlling the multi-sensor fusion-based rack railway foreign object removal system, a three-level early warning control mechanism for rack railway foreign object removal is proposed. This enables graded response to foreign objects of different threat levels, maximizing the safety of rack train operation and improving the safety margin for mountain operations. By using a first radar, image acquisition equipment, and a second radar for fusion judgment, the system can accurately identify foreign object categories (twigs, rocks, snow, etc.) and, combined with foreign object size and location information, effectively eliminates misjudgments from single sensors, avoids resource waste and erroneous braking, and ensures the accuracy of system detection. By mapping the accurate foreign object identification results to specific execution unit operations, the cleaning process can be customized and optimized. Attached Figure Description

[0039] Figure 1 is a schematic diagram of the detection range of the millimeter-wave radar and industrial camera; Figure 2 is a schematic diagram of the arrangement of the first radar and image acquisition equipment; Figure 3 is a front view of the arrangement of the second radar and execution module; Figure 4 is a side view of the arrangement of the second radar and execution module; Figure 5 is a schematic diagram of the water spray unit; Figure 6 is a schematic diagram of the jet spray unit; Figure 7 is a cross-sectional view of the jet spray unit.

[0040] The markings in the diagram are: 1-First radar, 2-Second radar, 3-Image acquisition equipment, 4-Cut-off unit, 5-Water spray unit, 6-Air jet unit. Detailed Implementation

[0041] The present invention will now be described in further detail with reference to specific embodiments. However, this should not be construed as limiting the scope of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.

[0042] Unless otherwise specified, the use of terms such as "upper," "lower," "left," "right," "center," "inner," and "outer" to indicate orientation or positional relationships in the description of specific embodiments of the present invention is based on the orientation or positional relationships shown in the accompanying drawings, or the orientation or positional relationship in which the product / equipment / device is typically placed during use. These terms are merely for the purpose of facilitating the description of the present invention or simplifying the description in specific embodiments, enabling those skilled in the art to quickly understand the solution, and do not indicate or imply that a particular device / component / element must have a specific orientation, or be constructed and operated in a specific positional relationship. Therefore, they should not be construed as limitations on the present invention.

[0043] Furthermore, the use of terms such as "horizontal," "vertical," "suspended," and "parallel" does not imply that the corresponding device / component / element must be absolutely horizontal, vertical, suspended, or parallel, but rather that it can be slightly tilted or have a deviation. For example, "horizontal" merely means that its direction is more horizontal relative to "vertical," not that the structure must be completely horizontal, but that it can be slightly tilted. Alternatively, it can be simplified to mean that the corresponding device / component / element, when set in a "horizontal," "vertical," "suspended," or "parallel" direction, can have an error / deviation of ±10% relative to the corresponding direction, more preferably within ±8%, more preferably within ±6%, more preferably within ±5%, and more preferably within ±4%. As long as the corresponding device / component / element is within the error / deviation range, it can still achieve its function in the present invention.

[0044] Furthermore, the use of terms such as "first," "second," and "third" in terminology is merely for distinguishing descriptions of identical or similar components and should not be interpreted as emphasizing or implying the relative importance of a particular component.

[0045] Furthermore, in the description of the embodiments of the present invention, "several", "more than", and "a number of" represent at least two. The number can be any number, such as 2, 3, 4, 5, 6, 7, 8, or 9, and can even exceed nine.

[0046] Furthermore, in the description of the technical solution of this invention, unless otherwise explicitly specified / limited / restricted, the terms "set up," "install," "connect," "link," "provided with," "laid out," and "arranged" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to common connection methods in the art, such as welding, riveting, bolting, and threaded connections. Such connections can be mechanical, electrical, or communication connections; they can be direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components.

[0047] As shown in Figures 1 to 7, this embodiment provides a multi-sensor fusion foreign object removal system for toothed rail railways, including: a sensing module, which includes a first radar 1, an image acquisition device 3, and a second radar 2.

[0048] As shown in Figure 2, the first radar 1 and the image acquisition device 3 are symmetrically installed at predetermined positions on the front of the train. The first radar 1 and the image acquisition device 3 are used to collect environmental data in front of the train.

[0049] In this embodiment, the first radar 1 is a millimeter-wave radar with a frequency band of 76GHz~77GHz, a detection accuracy of 0.25m, and a detection size of 15.6cm. 3 The maximum detection distance is 200m, and the obstacle resolution is ≤0.2m.

[0050] The second radar 2 is installed at the bottom of the train head. The second radar 2 is used to collect data on foreign objects between the rack teeth. In this embodiment, the second radar 2 is a lidar. The detection frequency of the second radar 2 is 50Hz~100Hz and the detection accuracy is ±1mm, as shown in Figures 3 and 4. In this embodiment, the lidar can be installed on the bottom side of the train head.

[0051] Image acquisition device 3 is an industrial camera; the sensor type of image acquisition device 3 is CCD (Charge Coupled Device), with a resolution of ≥250×250 pixels and a maximum detection distance of 100m.

[0052] The second radar 2 is 1.0 ± 0.1 m away from the head of the automatic coupler (also known as the James coupler or automatic hook) of the train head; the cutting unit 4 is 1.5 m ± 0.1 m away from the head of the automatic coupler of the train head.

[0053] The multi-sensor fusion foreign object removal system for rack railways provided in this embodiment achieves complementary advantages in both near and far detection through the combined use of the first radar 1 and the second radar 2.

[0054] Millimeter-wave radar has strong penetration capabilities through rain, snow, and fog, ensuring stable all-weather detection capabilities even in cold, foggy mountainous environments. With a maximum detection range of 200m, it provides ample advance warning time for trains. Optimal detection accuracy of 0.25m and obstacle resolution ≤0.2m are sufficient for long-distance hazard prediction and large obstacle detection.

[0055] The lidar has a detection accuracy of up to ±1mm and scans at a high frequency of 50Hz~100Hz, ensuring that the system can measure the height, shape and volume of foreign objects (such as ice, snow and gravel) in the rack grooves in real time and accurately, providing precise geometric basis for the decision-making module.

[0056] The first radar 1 and the image acquisition device 3 are symmetrically mounted on the upper part of the front of the vehicle, which helps to provide a wide and balanced forward field of vision and reduce blind spots on both sides.

[0057] Through multiple experiments, it was determined that installing a high-precision lidar at a distance of 1.0 ± 0.1 m from the automatic coupler head of the train head is optimal. This allows for scanning the rack at the best distance and angle, ensuring the accuracy of foreign object data collection, while also allowing sufficient reaction time for the decision-making module and subsequent cleaning actions. Furthermore, it was determined that the cutting unit 4 is positioned at a distance of 1.5 m ± 0.1 m from the automatic coupler head of the train head, ensuring that the cleaning operation can begin rapidly when the train reaches the foreign object.

[0058] The preferred image acquisition device 3 is a CCD industrial camera, which helps to ensure the accuracy and reliability of image acquisition.

[0059] The image acquisition device 3, with its 100m detection range, effectively complements the first radar 1, which has a 200m detection range. This significantly improves the system's accuracy in judging obstacle threats and avoids misjudgments caused by relying solely on the first radar 1, thereby enabling more precise early warning and decision-making.

[0060] The decision module is electrically connected to the sensing module. The decision module processes the data collected by the sensing module and outputs instructions. Specifically, the decision module can be a vehicle controller, PLC (Programmable Logic Controller), PAC (Programmable Automation Controller), or other commonly used devices in industrial production.

[0061] The execution module is electrically connected to the decision module. The execution module is used to clean the gear rail according to the instructions of the decision module. As shown in Figures 3 and 4, the execution module is installed at the bottom of the train head. The execution module is located between the second radar 2 and the first traction gear of the train. The execution module is provided with a cutting unit 4, a water spraying unit 5 and an air spraying unit 6 in sequence from the direction close to the second radar 2 towards the rear of the train (from front to back).

[0062] In this embodiment, the cutting unit 4 is a strip electric saw with a no-load rotation speed of 20m / s to 40m / s, a guide plate length of 40cm, and a voltage of 220V. As shown in Figure 5, the water spraying unit 5 can adopt a high-pressure micro-hole (e.g., hole diameter of 0.05mm to 1.0mm) water spraying array. The water spraying pressure adjustment range of the water spraying unit 5 is 0.5MPa to 8MPa, and the water temperature adjustment range is 50℃ to 90℃.

[0063] As shown in Figures 6 and 7, the jet unit 6 can adopt a high-pressure micro-orifice (e.g., orifice diameter of 0.1mm~2.0mm) jet array. The jet pressure adjustment range of the jet unit 6 is 0.5MPa~8MPa, and the temperature adjustment range is 50℃~90℃. Furthermore, the jet unit 6 can be set to have a strong mode, in which the jet pressure adjustment range is 7MPa~8MPa.

[0064] The multi-sensor fusion foreign object removal system for rack railways provided in this embodiment, through parameterized configuration and adjustable temperature and pressure design of the three units of cutting, water spraying, and air spraying, can remove foreign objects such as hard objects (cutting), ice and snow (high temperature / high pressure water and air), and fine particles (high pressure air), ensuring that the rack remains clean in complex environments, greatly improving the effectiveness of foreign object removal and the operational reliability of rack railways.

[0065] The multi-sensor fusion foreign object removal system for rack and pinion railways provided in this embodiment forms a three-stage composite cleaning mechanism consisting of a cutting unit 4, a water spraying unit 5, and an air spraying unit 6 arranged sequentially from front to back. The cutting unit 4, located at the front end, can mechanically cut and break long, hard foreign objects (such as branches and wood) to prevent them from getting stuck in the rack. For snow and ice, the water spraying unit 5 can use high-temperature, high-pressure water jets to melt and wash them away. The air spraying unit 6, located at the back, is used for final cleaning and drying. Whether it is blowing the debris after cutting / washing away from the track, using high-pressure airflow to remove fine gravel, or using hot airflow to further melt ice and snow, it can ensure that the rack is cleaned thoroughly and improve the removal efficiency.

[0066] The first radar 1 and image acquisition device 3 are positioned at predetermined locations on the upper part of the train's cab. The first radar 1 is responsible for long-range detection of the distance, speed, and size of obstacles, providing early warnings. The image acquisition device 3 can identify the specific type of obstacle (such as falling rocks, branches, snow, people, etc.), providing qualitative basis for decision-making. The second radar 2 at the bottom of the cab is responsible for high-precision acquisition of foreign object data between the rack and pinion teeth, providing further evidence for determining the need for clearing. By integrating multiple sensing devices to form a system-wide intelligent detection method, the detection accuracy is improved.

[0067] The multi-sensor fusion foreign object removal system for rack railways provided in this embodiment improves the operational safety, reliability, and traffic efficiency of rack railway trains in complex mountainous environments.

[0068] Example 2 This example provides a control method for a multi-sensor fusion rack railway foreign object removal system, used to control the multi-sensor fusion rack railway foreign object removal system provided in Example 1, including the following steps: S1. The sensing module collects environmental data in front of the train: the size of the obstacle, the distance S0 between the obstacle and the rack railway train, and the speed v0 of the obstacle relative to the train are detected by the first radar 1; specifically, it includes the following steps: S11. The first radar 1 uses FMCW frequency-modulated continuous wave for sampling, with a sampling frequency of 20Hz, and converts the data of the first radar 1 into the world coordinate system; the distance between the obstacle and the rack railway train is calculated. In the formula: S0 is the distance between the obstacle and the rack train, in meters; c is the speed of electromagnetic waves in air, in meters per second; f0 is the frequency difference between the rising and falling edges of the transmitted signal; Δf is the frequency difference between the transmitted signal and the echo signal; T is the modulation period of the millimeter-wave radar, in seconds; the speed of the obstacle relative to the train is calculated. In the formula: v0 is the speed of the obstacle relative to the train detected by the first radar 1, in m / s; f - f is the frequency difference between the rising edges of the reflected signal and the transmitted signal from the obstacle; + This represents the frequency difference between the falling edge of the signal reflected from the obstacle and the transmitted signal.

[0069] In this embodiment, the first radar 1 uses frequency-modulated continuous wave sampling, which can quickly and accurately measure the distance and relative speed of obstacles. This ensures a rapid response to changes in the environment ahead when the train is running at high speed, meeting the train's requirement for advance warning time. Converting the radar data into a world coordinate system allows the obstacle information obtained by the radar to be accurately aligned and fused with data from other sensors, such as the image acquisition device 3.

[0070] The obstacle category is identified by the image acquisition device 3, specifically including the following steps: S12. The image acquisition device 3 samples the data at a frequency of 20Hz and converts the sampled photo data into a world coordinate system. Setting the sampling frequency of both the first radar 1 and the image acquisition device 3 to 20Hz ensures that the information acquired by the first radar 1 and the information acquired by the image acquisition device 3 are time-aligned during data fusion in the decision module.

[0071] S13. Perform image preprocessing, including the following steps: S131. Perform grayscale processing on the original color image acquired by image acquisition device 3, using the following formula:

[0072] In the formula, Gray(i,j) is the gray value of the pixel at point (i,j); R(i,j) is the red channel pixel value at point (i,j); G(i,j) is the green channel pixel value at point (i,j); B(i,j) is the blue channel pixel value at point (i,j); i is the x-coordinate of the pixel; j is the y-coordinate of the pixel; S132. Denoise the image processed in S131 using the following formula:

[0073] In the formula, g(i,j) is the two-dimensional median filter at point (i,j); f(i,j) is the original image value at point (i,j); Med() is the median calculation function; S133. Enhance the image processed by S132, and the calculation formula is:

[0074] In the formula, p s (s) is the probability density function of s; p r (r) is the probability density function of r; r and s are random numbers between (0,1).

[0075] S14. Delineate the boundaries of the image processed in S13: S141. Extract the feature boundaries in the image using edge detection methods. The calculation formula is as follows:

[0076] In the formula, y(i,j) is the output image at point (i,j); F(i,j) is the input image at point (i,j); S141. The midpoint between the rack and the tracks on both sides of the rack is selected as the boundary region, and the calculation formula is as follows:

[0077] In the formula, u r The position is the middle of the rack and the rail on its right; u l The position is the midpoint between the rack and the rail to its left; u rx The characteristic point of the right track line of the toothed rail; u lx is the feature point of the left track line of the toothed track; x is the feature point corresponding to the x-th track line's horizontal coordinate; S15. Crop the image processed in S14 and proportionally reduce it to no more than 608×608 pixels. Limiting the image size to 608×608 or smaller significantly reduces the computational load of the convolutional neural network and improves the system's response speed.

[0078] S16. Conduct sample training by using a convolutional neural network algorithm to train on a dataset of at least 3000 samples to identify obstacle categories.

[0079] Specifically, the obstacle categories identified in S16 include at least one of the following: people, animals, ice and snow, falling rocks, and tree branches.

[0080] In this embodiment, the image acquired by the image acquisition device 3 is processed into a grayscale image, reducing the computational complexity of subsequent processing while preserving the image brightness information, making it suitable for edge detection and feature extraction. Median filtering is used for noise reduction, enhancing the clarity and recognizability of image features and improving the accuracy of subsequent algorithms. Edge detection extracts the contour information of the track, performs boundary delineation, locks in key recognition areas, and eliminates irrelevant interference. By determining the boundary area of ​​the rack and its clearing area, subsequent recognition is only performed within the defined area, significantly reducing the interference of irrelevant information such as background, sky, and mountains on the recognition accuracy. The convolutional neural network algorithm has powerful feature learning capabilities, automatically extracting subtle features of foreign objects from complex mountain images, achieving qualitative classification of key clearing targets such as "fallen rocks, branches, and snow." A dataset of at least 3000 samples ensures the model's generalization ability, enabling it to maintain high recognition accuracy even when facing new, untrained scenes. Mountain environments exhibit drastic lighting changes, potentially resulting in strong shadows or localized overexposure. Image enhancement using histogram equalization can effectively reduce the impact of uneven lighting and visibility on the image, thereby improving the accuracy of obstacle category recognition.

[0081] Data on foreign objects between the rack and pinion teeth is collected by the second radar 2; the data collected by the sensing module is transmitted to the decision module; S2. The decision module processes the data collected by the sensing module: when the length of the obstacle detected by the first radar 1 is greater than 20cm, the third-level warning is activated, the train power output is cut off, the train is brought to an emergency stop, and a graphic alarm message is sent to the human-machine interface in a timely manner. In order to prevent the prompt message from not being detected and processed in a timely manner due to the driver being busy or negligent, the on-board equipment also issues a voice alarm prompt. In this embodiment, the length of the obstacle specifically refers to the maximum projected size of the obstacle in the longitudinal direction of the track, that is, the direction of train operation.

[0082] When the length of the obstacle detected by the first radar 1 is ≤20cm, the obstacle information detected by the first radar 1 is fused with the obstacle information identified by the image acquisition device 3 for judgment: if only the first radar 1 or only the image acquisition device 3 detects the obstacle, the second-level warning is activated, the train slows down, and an alarm message is sent to the driver for analysis; if both the first radar 1 and the image acquisition device 3 detect the obstacle, and the image acquisition device 3 identifies the obstacle as a tree branch, the second-level warning is activated, the train slows down, the first instruction is sent to the execution module, and an alarm message is sent to the driver.

[0083] If both the first radar 1 and the image acquisition device 3 detect an obstacle, and the image acquisition device 3 identifies the obstacle as a falling rock, a second-level warning is activated, the train slows down, and a second instruction is sent to the execution module to send an alarm message to the driver.

[0084] If both the first radar 1 and the image acquisition device 3 detect an obstacle, and the image acquisition device 3 identifies the obstacle as snow, and the second radar 2 acquires information indicating that the snow has submerged the rack tooth grooves, a first-level warning is activated, the train slows down, and a third instruction is sent to the execution module to send an alarm message to the driver. In this embodiment, the second radar 2 can be a lidar. The principle of lidar in acquiring snow information is to utilize its high-precision distance measurement capability to acquire real-time three-dimensional point cloud data of the rack tooth groove area and compare it with a preset, clean rack geometric model to quickly determine whether the snow has submerged the rack tooth grooves.

[0085] If neither the first radar 1 nor the image acquisition device 3 detects an obstacle, and the second radar 2 detects foreign objects between the rack teeth, the first-level warning is activated, the train slows down, and the fourth instruction is sent to the execution module; S3. The execution module cleans the rack according to the instruction from the decision module: When the execution module receives the first instruction, it activates the cutting unit 4 and the jet unit 6. The cutting unit 4 can be a bar saw, which cuts off excess branches on both sides of the rack by extending the bar saw; the jet unit 6 can be a high-pressure micro-hole jet array, which uses jets to sweep away wood blocks and wood chips on the rack.

[0086] When the execution module receives the second instruction: start the jet unit 6, and the jet pressure of the jet unit 6 is 7MPa~8MPa, the jet will sweep away the fallen rocks on the rack.

[0087] When the execution module receives the third instruction: activate water spray unit 5, with the water spray temperature of water spray unit 5 being 50℃~90℃, to remove snow or ice from the rack. Activate air jet unit 6, with the air jet temperature of air jet unit 6 being 50℃~90℃, to remove water, residual snow, and residual ice from the rack.

[0088] When the execution module receives the fourth instruction: start the jet unit 6, and the jet pressure of the jet unit 6 is 0.5MPa~7MPa, to remove foreign objects between the teeth.

[0089] S4. Record the cleaning data (type of foreign object, cleaning energy consumption, equipment status) to the decision computer in real time.

[0090] This embodiment provides a control method for a multi-sensor fusion foreign object removal system on a rack railway. By controlling the multi-sensor fusion foreign object removal system, a three-level early warning control mechanism for rack railway foreign object removal is proposed. This mechanism enables graded responses to foreign objects of different threat levels, maximizing the operational safety of rack railway trains and improving the safety margin for mountain operations. By using a fusion judgment with a first radar 1, an image acquisition device 3, and a second radar 2, the system can accurately identify the type of foreign object (twigs, rocks, snow, etc.). Combined with the size and location information of the foreign object, it effectively eliminates misjudgments from single sensors, avoids resource waste and erroneous braking, and ensures the accuracy of system detection. By mapping the accurate foreign object identification results to specific execution unit operations, the cleaning process can be customized and optimized.

[0091] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-sensor fusion foreign object removal system for rack and pinion railways, characterized in that, include: The sensing module includes a first radar (1), an image acquisition device (3), and a second radar (2). The first radar (1) and the image acquisition device (3) are installed at a predetermined position on the upper part of the train head and are used to collect environmental data in front of the train. The second radar (2) is installed at the bottom of the train head and is used to collect data on foreign objects between the rack teeth. The decision module is electrically connected to the sensing module and is used to process the data collected by the sensing module and output instructions. The execution module is electrically connected to the decision module and is used to clean the rack according to the instructions of the decision module. The execution module is installed at the bottom of the train head and is located between the second radar (2) and the first traction gear. The execution module is provided with a cutting unit (4), a water spraying unit (5), and a jet spraying unit (6) in sequence from the direction close to the second radar (2) toward the rear of the train.

2. The multi-sensor fusion foreign object removal system for rack railways according to claim 1, characterized in that, The first radar (1) is a millimeter-wave radar, and the second radar (2) is a lidar; the first radar (1) has a frequency band of 76GHz~77GHz, a detection accuracy of 0.25m, and a detection size of 15.6cm. 3 The maximum detection distance is 200m, and the obstacle resolution is ≤0.2m; the detection frequency of the second radar (2) is 50Hz~100Hz, and the detection accuracy is ±1mm; the first radar (1) and the image acquisition device (3) are symmetrically installed at a predetermined position on the upper part of the train head; the second radar (2) is 1.0±0.1m away from the automatic coupler head of the train head; the cutting unit (4) is 1.5m±0.1m away from the automatic coupler head of the train head.

3. The multi-sensor fusion foreign object removal system for rack railways according to claim 2, characterized in that, The image acquisition device (3) is an industrial camera; the sensor type of the image acquisition device (3) is CCD, with a resolution of ≥250×250 pixels and a maximum detection distance of 100m.

4. The multi-sensor fusion foreign object removal system for rack railways according to claim 1, characterized in that, The cutting unit (4) is a strip electric saw with a no-load speed of 20m / s to 40m / s, a guide plate length of 40cm, and a voltage of 220V; the water spraying unit (5) has a water spraying pressure adjustment range of 0.5MPa to 8MPa and a water temperature adjustment range of 50℃ to 90℃; the air jet unit (6) has an air jet pressure adjustment range of 0.5MPa to 8MPa and a temperature adjustment range of 50℃ to 90℃.

5. A control method for a multi-sensor fusion foreign object removal system for rack and pinion railways, characterized in that, The system for controlling a multi-sensor fusion rack rail foreign object removal system according to any one of claims 1-4 includes the following steps: S1. The sensing module collects environmental data in front of the train: the size of the obstacle, the distance S0 between the obstacle and the rack rail train, and the speed v0 of the obstacle relative to the train are detected by the first radar (1); the obstacle type is identified by the image acquisition device (3); foreign object data between the rack teeth is collected by the second radar (2); and the data collected by the sensing module is transmitted to the decision module; S2. The decision module processes the data collected by the sensing module: when the length of the obstacle detected by the first radar (1) is >20cm, the third-level warning is activated and the train power output is cut off. The train brakes and stops, and sends an alarm message. When the length of the obstacle detected by the first radar (1) is ≤20cm, the obstacle information detected by the first radar (1) is fused with the obstacle information identified by the image acquisition device (3) for judgment: if only the first radar (1) or only the image acquisition device (3) detects the obstacle, the second-level warning is activated, the train slows down, and an alarm message is sent; if both the first radar (1) and the image acquisition device (3) detect the obstacle, and the image acquisition device (3) identifies the obstacle as a tree branch, the second-level warning is activated, the train slows down, and the first instruction is sent to the execution module; if both the first radar (1) and the image acquisition device (3) detect the obstacle, the train slows down, and the first instruction is sent to the execution module. If an obstacle is detected and the image acquisition device (3) identifies the obstacle as falling rocks, a second-level warning is activated, the train slows down, and a second instruction is sent to the execution module. If both the first radar (1) and the image acquisition device (3) detect an obstacle, and the image acquisition device (3) identifies the obstacle as snow, and the second radar (2) collects information that the snow has covered the rack tooth groove, a first-level warning is activated, the train slows down, and a third instruction is sent to the execution module. If neither the first radar (1) nor the image acquisition device (3) detects an obstacle, and the second radar (2) collects information that there is a foreign object between the rack tooth groove, a first-level warning is activated, the train slows down, and a fourth instruction is sent to the execution module. Block; S3. The execution module cleans the gear track according to the instructions of the decision module: When the execution module receives the first instruction: start the cutting unit (4) and the jet unit (6); When the execution module receives the second instruction: start the jet unit (6), and the jet pressure of the jet unit (6) is 7MPa~8MPa; When the execution module receives the third instruction: start the water spray unit (5), and the water spray temperature of the water spray unit (5) is 50℃~90℃; start the jet unit (6), and the jet temperature of the jet unit (6) is 50℃~90℃; When the execution module receives the fourth instruction: start the jet unit (6), and the jet pressure of the jet unit (6) is 0.5MPa~7MPa.

6. The control method for a multi-sensor fusion foreign object removal system for rack and pinion railways according to claim 5, characterized in that, In S1, the size of the obstacle, the distance S0 between the obstacle and the rack train, and the speed v0 of the obstacle relative to the train are detected by the first radar (1), including the following steps: S11. The first radar (1) uses frequency-modulated continuous wave for sampling, and the data of the first radar (1) is converted into the world coordinate system; the distance between the obstacle and the rack train is calculated. In the formula, S0 is the distance between the obstacle and the rack train; c is the speed of electromagnetic waves in the air; f0 is the frequency difference between the rising and falling edges of the transmitted signal; Δf is the frequency difference between the transmitted signal and the echo signal; T is the modulation period of the millimeter-wave radar; the speed of the obstacle relative to the train is calculated. In the formula, v0 is the speed of the obstacle relative to the train detected by the first radar (1); f - f is the frequency difference between the rising edges of the reflected signal and the transmitted signal from the obstacle; + This represents the frequency difference between the falling edge of the signal reflected from the obstacle and the transmitted signal.

7. The control method for a multi-sensor fusion foreign object removal system for rack and pinion railways according to claim 5, characterized in that, In S1, the obstacle category is identified by the image acquisition device (3), including the following steps: S12. The image acquisition device (3) samples the image and converts the sampled photo data into the world coordinate system; S13. Image preprocessing is performed, including the following steps: S131. The original color image acquired by the image acquisition device (3) is processed for grayscale, and the calculation formula is: In the formula, Gray(i,j) is the gray value of the pixel at point (i,j); R(i,j) is the red channel pixel value at point (i,j); G(i,j) is the green channel pixel value at point (i,j); B(i,j) is the blue channel pixel value at point (i,j); i is the x-coordinate of the pixel; j is the y-coordinate of the pixel; S132. Denoise the image processed in S131 using the following formula: In the formula, g(i,j) is the two-dimensional median filter at point (i,j); f(i,j) is the original image value at point (i,j); Med() is the median calculation function; S14. Delineate the boundaries of the image processed in S13: S141. Extract the feature boundaries in the image using the edge detection method, and the calculation formula is as follows: In the formula, y(i,j) is the output image at point (i,j); F(i,j) is the input image at point (i,j); S141. The midpoint between the rack and the tracks on both sides of the rack is selected as the boundary region, and the calculation formula is as follows: In the formula, u r The position is the middle of the rack and the rail on its right; u l The position is the midpoint between the rack and the rail to its left; u rx The characteristic point of the right track line of the toothed rail; u lx is the feature point of the left track line of the toothed track; x is the feature point corresponding to the x-th track line's horizontal coordinate; S15. Crop the image processed in S14 and reduce it proportionally; S16. Perform sample training, using a convolutional neural network algorithm to train on at least 3000 sample datasets to identify obstacle categories.

8. The control method for a multi-sensor fusion foreign object removal system for rack and pinion railways according to claim 7, characterized in that, S13. Image preprocessing includes the following steps: S133. Enhancing the image processed in S132, using the following formula: In the formula, p s (s) is the probability density function of s; p r (r) is the probability density function of r; r and s are random numbers between (0,1).

9. The control method for a multi-sensor fusion foreign object removal system for rack railways according to claim 7, characterized in that, The obstacle categories identified in S16 include at least one of the following: people, animals, ice and snow, falling rocks, and tree branches.

10. The control method for a multi-sensor fusion foreign object removal system for rack railways according to claim 7, characterized in that, In S11, the first radar (1) uses FMCW frequency-modulated continuous wave with a sampling frequency of 20Hz; in S12, the image acquisition device (3) samples at a sampling frequency of 20Hz; in S15, the image processed in S14 is cropped and proportionally reduced to no more than 608×608 pixels.

Images