Rail vehicle inspection method and system

Abstract

This invention relates to the field of rail vehicle technology, and provides a method and system for rail vehicle maintenance. The method includes generating and sending a first command to a drone; wherein the first command controls the drone to collect first image data of a rail vehicle on the maintenance track; the first command includes a first speed of the drone; receiving the first image data sent by the drone and performing identification processing on the first image data; if the first image data includes a target maintenance component, generating and sending a second command to the drone; wherein the second command controls the drone to collect second image data of the rail vehicle on the maintenance track; the second command includes a second speed of the drone, the second speed being less than the first speed; receiving the second image data sent by the drone to perform real-time maintenance on the target maintenance component based on the second image data, thereby improving the versatility and portability of rail vehicle maintenance.

Description

Technical Field

[0001] This invention relates to the field of rail vehicle technology, and provides a rail vehicle maintenance method and system. Background Technology

[0002] In scenarios such as rail transit depots, large freight yards, and logistics warehousing centers, regular and rapid condition checks on vehicles or large equipment parked on multiple tracks are a crucial and growing need. Traditional manual inspection methods suffer from inefficiency, limited coverage, significant environmental and time constraints, and potential safety risks.

[0003] Currently, common solutions for achieving automated inspection include deploying mobile inspection devices such as track robots and AGVs equipped with sensor devices, or building inspection systems based on fixed infrastructure. These systems typically rely on pre-defined, highly structured work paths. The systems require configuring and storing a complex set of path data for each track and even each parking space, lacking versatility and portability. Summary of the Invention

[0004] This invention provides a method and system for the maintenance of rail vehicles, which addresses the shortcomings of related technologies that require the separate configuration and storage of a complex set of path data for each track or even each parking space, thereby improving the versatility and portability of rail vehicle maintenance.

[0005] This invention provides a method for overhauling rail vehicles, comprising: Generate and send a first instruction to the drone; wherein the first instruction is used to control the drone to collect first image data of the track vehicle under maintenance; the first instruction includes the first speed of the drone; Receive the first image data sent by the drone and perform recognition processing on the first image data; If the first image data includes a target maintenance component, a second instruction is generated and sent to the drone; wherein the second instruction is used to control the drone to acquire second image data of the rail vehicle on the maintenance track; the second instruction includes a second speed of the drone, the second speed being less than the first speed; The system receives the second image data sent by the drone to perform real-time maintenance on the target repair component based on the second image data.

[0006] According to an embodiment of the present invention, the recognition processing of the first image data includes: The first image data is input into the maintenance identification model to obtain the first identification processing result output by the maintenance identification model; The recognition processing of the second image data includes: The second image data is input into the maintenance identification model to obtain the second identification processing result output by the maintenance identification model; Wherein, the first identification processing result is that the first image data includes the target maintenance component, or the first image data does not include the target maintenance component; The second identification processing result is that the second image data includes the target maintenance component, or the second image data does not include the target maintenance component. The maintenance identification model is trained based on sample image data and sample identification results.

[0007] According to an embodiment of the present invention, the real-time maintenance of the target maintenance component based on the second image data includes: Determine the type of the target maintenance component in the second image data, and invoke the corresponding anomaly detection rule according to the type of the target maintenance component; Based on the second image data, the target repair component is inspected in real time according to the anomaly detection rules.

[0008] According to an embodiment of the present invention, before receiving the first image data sent by the drone and performing recognition processing on the first image data, the method further includes: The corresponding target maintenance component is determined according to the maintenance procedures for the rail vehicles on the maintenance track.

[0009] According to an embodiment of the present invention, before receiving the first image data sent by the drone and performing recognition processing on the first image data, the method further includes: Obtain preliminary inspection results of the rail vehicles on the inspection track by at least one external inspection device; The target maintenance component of the rail vehicle on the maintenance track is identified from the preliminary inspection results.

[0010] According to one embodiment of the present invention, before generating and sending the first command to the drone, the method further includes: Obtain the point cloud of the maintenance track; Based on the point cloud, establish the maintenance trajectory between the vehicle maintenance start point and the vehicle maintenance end point of the maintenance track.

[0011] According to one embodiment of the present invention, before generating and sending the first command to the drone, the method further includes: The maintenance track selects a first route from at least two routes to the UAV nest to the vehicle maintenance start point of the maintenance track, and a second route from the vehicle maintenance end point of the maintenance track to the UAV nest. Both the first and second routes were obtained in advance by setting navigation points along the route based on historical flight data.

[0012] According to an embodiment of the present invention, the real-time maintenance of the target maintenance component based on the second image data includes: Perform a field-of-view quality assessment on the video frames in the second image data; The target video frame is determined from the second image data based on the visual field quality assessment results, so as to perform real-time maintenance on the target maintenance component based on the target video frame.

[0013] The present invention also provides a rail vehicle maintenance system, comprising: The control module is configured to generate and send a first instruction to the drone; receive the first image data sent by the drone and perform recognition processing on the first image data; generate and send a second instruction to the drone when the first image data includes a target repair component; and receive the second image data sent by the drone to perform real-time repair on the target repair component based on the second image data. The drone is used to receive a first instruction sent by the control module to collect first image data of the track vehicle under maintenance and send the first image data to the control module; or it is used to receive a second instruction sent by the control module to collect second image data of the track vehicle under maintenance and send the second image data to the control module.

[0014] According to one embodiment of the present invention, the control module is further configured to: Determine the type of the target maintenance component in the second image data, and invoke the corresponding anomaly detection rule according to the type of the target maintenance component; Based on the second image data, the target maintenance component is detected in real time according to the anomaly detection rules.

[0015] The rail vehicle maintenance method and system provided by this invention generate and send a first instruction to a drone, receive first image data of rail vehicles on the maintenance track collected by the drone at a first speed, and perform identification processing on the first image data. When the first image data includes the target maintenance component, it can generate and send a second instruction to the drone, receive second image data of rail vehicles on the maintenance track collected by the drone at a second speed, and perform real-time maintenance. It can automatically control the speed at which the drone collects image data of rail vehicles on the maintenance track, without the need to set navigation and positioning points, and can facilitate the maintenance of rail vehicles on different maintenance tracks, improving the versatility and portability of rail vehicle maintenance. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a schematic flowchart of the rail vehicle maintenance method provided by the present invention.

[0018] Figure 2 This is a schematic framework diagram of the rail vehicle maintenance method provided by the present invention.

[0019] Figure 3 This is a schematic structural diagram of the rail vehicle maintenance system provided by the present invention. Detailed Implementation

[0020] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0021] like Figures 1 to 3 As shown, the present invention provides a method and system for the maintenance of rail vehicles.

[0022] Figure 1 This is a schematic flowchart of the rail vehicle maintenance method provided by the present invention, such as... Figure 1 As shown, the method includes the following steps.

[0023] Step 101: Generate and send a first command to the drone. The first command controls the drone to collect first image data of the track vehicle being inspected; the first command includes the drone's first speed. Step 102: Receive the first image data sent by the drone and perform recognition processing on the first image data.

[0024] Step 103: If the first image data includes the target maintenance component, generate and send a second command to the drone. The second command controls the drone to acquire second image data of the rail vehicle on the maintenance track; the second command includes a second speed of the drone, which is less than the first speed. Step 104: Receive the second image data sent by the UAV, so as to perform real-time maintenance on the target maintenance component based on the second image data.

[0025] The first instruction, also known as the inspection instruction, is the instruction that controls the UAV to fly to the next target component to collect the first image data of the track vehicle on the inspection track.

[0026] The second command, also known as the precision acquisition command, is a command that controls the UAV to fly when a target maintenance component appears within the detection field of view in order to acquire second image data of the rail vehicle on the maintenance track.

[0027] It should be noted that, in addition to the drone's flight speed, the first and second commands may also include the drone's flight direction, flight altitude, etc., but this embodiment does not limit this.

[0028] Among them, the maintenance track, also known as the track to be inspected or the track to be maintained, is a specific track section designated or planned for the inspection, testing or maintenance of rail vehicles.

[0029] Rail vehicles, also known as rail transit vehicles, can include trains, urban rail, and subways.

[0030] Target maintenance components refer to components or sets of components of a rail vehicle that require inspection, testing, or maintenance. For example, target maintenance components can be determined based on the rail vehicle's maintenance procedures or historical maintenance records.

[0031] The first speed, also known as cruising speed or normal flight speed, refers to the drone's flight speed when the target component to be repaired is not within the drone's detection field of view. The second speed, also known as repair speed or slow flight speed, refers to the drone's flight speed when the target component to be repaired is within the drone's detection field of view.

[0032] The first image data was acquired by the drone at a first speed within the corresponding detection field of view, and the second image data was acquired by the drone at a second speed within the corresponding detection field of view.

[0033] It should be noted that before acquiring image data collected by the drone, the drone can be controlled at a default first speed to acquire first image data of the track vehicles being inspected while the drone is traveling at the first speed. Then, if the target component being inspected is identified in the first image data, the drone can be controlled to slow down to acquire second image data of the track vehicles being inspected. However, after the target component being inspected leaves the drone's detection field of view, the drone can be controlled to resume normal speed and continue acquiring first image data of the track vehicles being inspected at the first speed.

[0034] The number of target maintenance components is usually multiple. Therefore, each target maintenance component will repeatedly execute the aforementioned steps 101-104 as it enters and leaves the UAV's detection field of view, until the last target maintenance component leaves the UAV's detection field of view, thus completing the maintenance of the last target maintenance component.

[0035] There are many ways to determine whether a target maintenance component that has left the detection field of view of the UAV is the last target maintenance component. You can choose according to your actual needs. This embodiment does not limit this method.

[0036] The rail vehicle maintenance method provided in this invention generates and sends a first instruction to a drone, receives first image data of rail vehicles on the maintenance track collected by the drone at a first speed, and performs identification processing on the first image data. When the first image data includes the target maintenance component, it can generate and send a second instruction to the drone, receive second image data of rail vehicles on the maintenance track collected by the drone at a second speed, and perform real-time maintenance. It can automatically control the speed at which the drone collects image data of rail vehicles on the maintenance track, without the need to set navigation and positioning points, and can facilitate the maintenance of rail vehicles on different maintenance tracks, improving the versatility and portability of rail vehicle maintenance.

[0037] Furthermore, since the second speed is less than the first speed, clearer second image data, including the target maintenance component, can be acquired, thereby improving the accuracy of real-time maintenance of the target maintenance component based on the second image data.

[0038] Based on the above embodiments, the recognition processing of the first image data includes: The first image data is input into the maintenance identification model to obtain the first identification processing result output by the maintenance identification model; Wherein, the first identification processing result is that the first image data includes the target maintenance component, or the first image data does not include the target maintenance component; The maintenance identification model is trained based on sample image data and sample identification results.

[0039] A feasible training scheme for the initial recognition model may include: The pronunciation features of the sample are input into the initial recognition model to obtain the recognition result output by the initial recognition model. Then, based on the recognition result and the sample recognition result, the loss function value is calculated. Finally, based on the loss function value, the model parameters of the initial recognition model are updated. The above input process and calculation process are iteratively executed until the loss function converges or the preset number of iterations is reached, thus obtaining the inspection recognition model. The preset number of iterations can be set as needed and is not specifically limited here.

[0040] It is understandable that every variant of the target maintenance component is difficult to exhaust. Compared to manually programming rules for each variant of the target maintenance component and relying on those rules to identify the target maintenance component from image data, in this embodiment, the pre-trained maintenance recognition model can automatically understand and adapt to the deformation of the target maintenance component, thereby improving the accuracy of UAV maintenance in complex scenarios where the components of rail vehicles on maintenance tracks are distributed.

[0041] In some embodiments, the recognition processing of the second image data includes: The second image data is input into the maintenance identification model to obtain the second identification processing result output by the maintenance identification model; The second identification processing result is that the second image data includes the target maintenance component, or the second image data does not include the target maintenance component.

[0042] It is understandable that the working principle and technical effect of using the inspection and recognition module to identify and process the second image data are basically the same as those of using the inspection and recognition module to identify and process the first image data, and will not be elaborated here.

[0043] In some embodiments, the dataset used for model training includes positive and negative samples.

[0044] The positive samples include sample image data of the target maintenance component and sample recognition results of the target maintenance component; the sample image data is marked with the location and outline of the target maintenance component.

[0045] Negative samples include image data of samples without a target repair component, but with easily confused backgrounds or interfering objects. Easily confused backgrounds or interfering objects refer to objects whose similarity to the target repair component is greater than a set threshold.

[0046] The calculation of easily confused backgrounds or interferences refers to the method of similarity with the target repair component, and the specific value of the threshold can be set according to actual needs. This embodiment does not limit this.

[0047] Based on the above embodiments, the real-time maintenance of the target inspection component according to the second image data includes: Determine the type of the target maintenance component in the second image data, and invoke the corresponding anomaly detection rule according to the type of the target maintenance component; Based on the second image data, the target repair component is inspected in real time according to the anomaly detection rules.

[0048] The target maintenance component can be classified as a connection type, a surface wear type, or a complex assembly type. It should be noted that if the image data includes the target maintenance component, the type of the target maintenance component can be output simultaneously through the maintenance recognition model.

[0049] The specific content of the anomaly detection rules can be set according to the actual working conditions of the rail vehicle to be inspected, and this embodiment does not limit this.

[0050] For example, after identifying the target maintenance component and its type in the second image data, the template corresponding to the target maintenance component can be called from the template library to perform real-time detection by comparing the image of the target maintenance component and the template of the target maintenance component according to the anomaly detection rules.

[0051] It is understood that in this embodiment, by dynamically calling the corresponding anomaly detection rules for different types of target maintenance components, one-to-one accurate detection can be achieved, thereby enabling real-time detection to be performed under the most suitable rule for each target maintenance component, which greatly improves the accuracy of defect identification of the target maintenance component.

[0052] Furthermore, by modularly decoupling component identification and anomaly detection, when generating new anomaly detection rules for repaired components or adjusting existing anomaly detection rules for repaired components, it is possible to simply add or independently update the corresponding anomaly detection rules without retraining or adjusting the entire detection process.

[0053] Furthermore, because it eliminates the need for complex analysis of all image data, triggering complex analysis only when the target maintenance component is identified, it reduces computing power consumption, thereby reducing the risk of missed or false detections due to insufficient computing resources. This ensures the effectiveness of rail vehicle maintenance even with the limited onboard computing power of drones.

[0054] Based on the above embodiments, before receiving the first image data sent by the drone and performing recognition processing on the first image data, the method further includes: The corresponding target maintenance component is determined according to the maintenance procedures for the rail vehicles on the maintenance track.

[0055] Among them, the maintenance procedures can be the equipment maintenance and repair specifications for track vehicles on the maintenance track.

[0056] It should be noted that all maintenance components involved in the maintenance procedure can be identified as target maintenance components, or all maintenance components involved in the maintenance procedure can be divided according to historical experience data. For example, maintenance components that are prone to failure within a short interval can be identified as target maintenance components, while maintenance components that are prone to failure within a longer interval and the remaining maintenance components can be identified as target maintenance components.

[0057] Understandably, by identifying the corresponding target maintenance components based on the maintenance procedures for rail vehicles on the maintenance track, and training a maintenance identification model capable of recognizing the target maintenance components, drones can be automatically controlled to perform real-time maintenance on rail vehicles of the same model on different maintenance tracks in self-inspection mode, without the need to set separate navigation and positioning points for rail vehicles of the same model on each maintenance track, thus improving the versatility and portability of rail vehicle maintenance.

[0058] Based on the above embodiments, before receiving the first image data sent by the drone and performing recognition processing on the first image data, the method further includes: Obtain preliminary inspection results of the rail vehicles on the inspection track by at least one external inspection device; The target maintenance component of the rail vehicle on the maintenance track is identified from the preliminary inspection results.

[0059] Preliminary maintenance results can be obtained by inspecting rail vehicles on the maintenance track using a 360° trackside inspection system, by inspecting rail vehicles on the maintenance track using a ground robot, or by conducting joint inspections of rail vehicles on the maintenance track using a 360° trackside inspection system and a ground robot.

[0060] The results of each maintenance component in the preliminary maintenance results can be analyzed using rules to identify maintenance components with high uncertainty in the preliminary maintenance results and designate them as target maintenance components for real-time detection based on UAVs. Alternatively, the results of each maintenance component in the preliminary maintenance results can be analyzed using a pre-trained model to identify maintenance components with high uncertainty in the preliminary maintenance results and designate them as target maintenance components for real-time detection based on UAVs.

[0061] It should be noted that the entity executing the rail vehicle maintenance method provided in this embodiment can be a trackside integrated control center, or a trackside integrated maintenance platform.

[0062] The preliminary inspection results of the external inspection device can be sent to the trackside integrated control center to enable joint inspection between the external inspection device and the UAV.

[0063] Among them, external inspection devices have high maintenance efficiency, but they also have significant limitations. For example, the 360-degree trackside inspection system needs to take pictures when the vehicle passes through a specific location. Due to the limited shooting angle, the image clarity for some small parts is insufficient. Although ground robots can move and inspect, they are limited by the ground environment and cannot cover key areas such as the roof and couplers. This results in the low accuracy of external inspection devices in inspecting rail vehicles.

[0064] Understandably, compared to using drones to inspect all maintenance components of rail vehicles on the maintenance track, this embodiment first uses an external inspection device with higher maintenance efficiency to conduct preliminary inspections, identify target maintenance components with higher uncertainty, and then uses drones in re-inspection mode to conduct real-time inspections of the target maintenance components of rail vehicles on the maintenance track. This can reduce the number of components to be inspected by drones, improve the accuracy of maintenance, and ensure maintenance efficiency.

[0065] In some embodiments, the maintenance identification model may include multiple maintenance identification layers, each corresponding to a maintenance component. Thus, by setting the calling method of the maintenance identification model, the corresponding maintenance identification layer can be called to identify the target maintenance component in self-inspection mode or re-inspection mode from the first image data.

[0066] Based on the above embodiments, before generating and sending the first command to the drone, the method further includes: Obtain the point cloud of the maintenance track; Based on the point cloud, establish the maintenance trajectory between the vehicle maintenance start point and the vehicle maintenance end point of the maintenance track.

[0067] The vehicle maintenance start point is the initial spatial position where the drone begins real-time maintenance of rail vehicles parked on the maintenance track. The vehicle maintenance end point is the final spatial position where the drone begins real-time maintenance of rail vehicles parked on the maintenance track.

[0068] It should be noted that the environment of the maintenance track can be scanned manually using a drone to obtain a point cloud of the track. Then, a horizontal straight line or smooth curve path can be planned as a baseline between the vehicle maintenance start point and the vehicle maintenance end point, along a direction parallel to the track reference axis. Based on the point cloud near this baseline path, a maintenance trajectory can be obtained that ensures the minimum distance between the trajectory and the vehicle body, overhead contact line, support pillars, and other obstacles is always greater than a preset dynamic safety threshold.

[0069] Based on the above embodiments, before generating and sending the first command to the drone, the method further includes: The maintenance track selects a first route from at least two routes to the UAV nest to the vehicle maintenance start point of the maintenance track, and a second route from the vehicle maintenance end point of the maintenance track to the UAV nest. Both the first and second routes were obtained in advance by setting navigation points along the route based on historical flight data.

[0070] It should be noted that historical flight data can be obtained by first manually controlling the drone to fly from its nest to the vehicle maintenance start point on the maintenance track and back to the nest from the vehicle maintenance end point on the maintenance track. Based on this, navigation points can be set at fixed intervals or according to key vehicle components along the route to obtain the corresponding flight path. Each navigation point can include three-dimensional coordinates, heading, and hovering commands.

[0071] The maintenance trajectory, first route, and second route can be associated and stored with the maintenance track. In this way, after determining the maintenance track where the rail vehicle to be maintained is located, the corresponding maintenance trajectory, first route, and second route can be directly obtained to generate the corresponding track task. The track task is sent to the UAV, which then travels along the first route to the vehicle maintenance start point on the maintenance track. With the assistance of the maintenance recognition model, it automatically travels along the maintenance trajectory to the vehicle maintenance end point, and then travels along the second route to the UAV nest.

[0072] Understandably, compared to requiring UAVs to rely entirely on onboard sensor data for real-time global path planning and obstacle avoidance during flight, this embodiment pre-plans and stores the flight path between the UAV nest and the maintenance track, completing computationally intensive point cloud processing, global path search, and obstacle avoidance calculations offline. The maintenance identification model assists the UAV in automatically navigating along the maintenance trajectory within the maintenance track, reducing the absolute positioning accuracy requirements of the surrounding environment and thus avoiding the risk of collisions due to insufficient onboard computing power.

[0073] Furthermore, by pre-planning and storing the flight path between the drone nest and the maintenance track, the drone's navigation in the maintenance track can be automatically controlled by the target maintenance components of the maintenance vehicle. This enables the drone to quickly migrate between the same vehicle type on different tracks, thus achieving a highly efficient detection mode that covers multiple tracks, multiple vehicles, and fewer drones.

[0074] Based on the above embodiments, the real-time maintenance of the target inspection component according to the second image data includes: Perform a field-of-view quality assessment on the video frames in the second image data; The target video frame is determined from the second image data based on the visual field quality assessment results, so as to perform real-time maintenance on the target maintenance component based on the target video frame.

[0075] Among them, the target video frame, also known as the abnormal point screenshot, is an image that can clearly, stably, and without obstruction capture the status of the target maintenance component.

[0076] There are many ways to evaluate the visual quality of video frames in the second image data, such as through evaluation rules, pre-trained evaluation models, etc., and this embodiment does not limit this.

[0077] It should be noted that when the visual field quality assessment result is classified as excellent, good, average, or poor, the video frame corresponding to a certain category of visual field quality assessment result can be determined as the target video frame; when the visual field quality assessment result is a score, the video frame with a score greater than a certain threshold can be determined as the target video frame, and so on.

[0078] For example, after obtaining the target video frame of the target detection component with the best field of view, the type of the target maintenance component in the target video frame can be identified and determined, so as to call the template corresponding to the target maintenance component from the template library, and compare the target video frame of the target maintenance component with the template of the target maintenance component for real-time detection according to the anomaly detection rules. The real-time detection results and the target video frame are sent to the trackside integrated control center, and the real-time detection status is displayed to the user through the display device.

[0079] Understandably, in real-time maintenance scenarios, the amount of video frame data collected by drones is enormous. In this embodiment, by screening target video frames that meet the quality requirements through field of view quality assessment, the processing burden of redundant data can be reduced, thereby ensuring the real-time response capability of maintenance under the condition of limited onboard computing power.

[0080] To illustrate the function of the rail vehicle maintenance method provided in this implementation, such as Figure 3 As shown, a specific example is provided below.

[0081] First, the point cloud of the maintenance track can be obtained; then, the maintenance trajectory between the vehicle maintenance start point and the vehicle maintenance end point of the maintenance track can be established based on the point cloud; finally, a first route from the UAV nest to the vehicle maintenance start point of the maintenance track and a second route from the vehicle maintenance end point of the maintenance track to the UAV nest can be selected from at least two routes based on the maintenance track. Then, a maintenance task is sent to the UAV, and a first instruction is generated and sent to the UAV. The first image data sent by the UAV is received, and the first image data is input into the maintenance identification model to obtain the first identification processing result output by the maintenance identification model. If the first identification processing result is that the first image data includes the target maintenance component, a second instruction is generated and sent to the UAV to automatically control the UAV and realize navigation control based on the maintenance identification model. The system determines the type of the target maintenance component in the second image data and invokes the corresponding anomaly detection rule based on the type of the target maintenance component; performs real-time maintenance on the target maintenance component based on the second image data and the anomaly detection rule; receives the second image data sent by the UAV and performs real-time maintenance on the target maintenance component based on the second image data; performs a field-of-view quality assessment on the video frames in the second image data; determines the target video frame from the second image data based on the field-of-view quality assessment result to determine the type of the target maintenance component in the target video frame and invokes the corresponding anomaly detection rule based on the type of the target maintenance component; performs real-time maintenance on the target maintenance component based on the second image data and the anomaly detection rule, and sends the real-time maintenance result to the trackside integrated control center.

[0082] The rail vehicle maintenance system provided by the present invention is described below. The rail vehicle maintenance system described below can be referred to in correspondence with the rail vehicle maintenance method described above.

[0083] Figure 3 This is a schematic diagram of the structure of the rail vehicle maintenance system provided by the present invention, as shown below. Figure 3 As shown, the system includes: The control module 310 is configured to generate and send a first instruction to the drone; receive the first image data sent by the drone and perform recognition processing on the first image data; generate and send a second instruction to the drone when the first image data includes a target repair component; and receive the second image data sent by the drone to perform real-time repair on the target repair component based on the second image data. The drone 320 is used to receive a first instruction sent by the control module to collect first image data of the track vehicle under maintenance and send the first image data to the control module; or it is used to receive a second instruction sent by the control module to collect second image data of the track vehicle under maintenance and send the second image data to the control module.

[0084] Based on any of the above embodiments, the control module 310 is further configured to: Determine the type of the target maintenance component in the second image data, and invoke the corresponding anomaly detection rule according to the type of the target maintenance component; Based on the second image data, the target maintenance component is detected in real time according to the anomaly detection rules.

[0085] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0086] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method of inspecting a rail vehicle, characterized by, include: Generate and send a first instruction to the drone; wherein the first instruction is used to control the drone to collect first image data of the track vehicle under maintenance; the first instruction includes the first speed of the drone; Receive the first image data sent by the drone and perform recognition processing on the first image data; If the first image data includes a target maintenance component, a second instruction is generated and sent to the drone; wherein the second instruction is used to control the drone to acquire second image data of the rail vehicle on the maintenance track; the second instruction includes a second speed of the drone, the second speed being less than the first speed; The system receives the second image data sent by the drone to perform real-time maintenance on the target repair component based on the second image data.

2. The rail vehicle service method according to claim 1, characterized in that, The recognition processing of the first image data includes: The first image data is input into the maintenance identification model to obtain the first identification processing result output by the maintenance identification model; The recognition processing of the second image data includes: The second image data is input into the maintenance identification model to obtain the second identification processing result output by the maintenance identification model; Wherein, the first identification processing result is that the first image data includes the target maintenance component, or the first image data does not include the target maintenance component; The second identification processing result is that the second image data includes the target maintenance component, or the second image data does not include the target maintenance component. The maintenance identification model is trained based on sample image data and sample identification results.

3. The rail vehicle service method of claim 1, wherein, The real-time maintenance of the target repair component based on the second image data includes: Determine the type of the target maintenance component in the second image data, and invoke the corresponding anomaly detection rule according to the type of the target maintenance component; Based on the second image data, the target repair component is inspected in real time according to the anomaly detection rules.

4. The rail vehicle service method of claim 1, wherein, Before receiving the first image data sent by the drone and performing recognition processing on the first image data, the method further includes: The corresponding target maintenance component is determined according to the maintenance procedures for the rail vehicles on the maintenance track.

5. The rail vehicle service method according to claim 1 or 4, characterized in that, Before receiving the first image data sent by the drone and performing recognition processing on the first image data, the method further includes: Obtain preliminary inspection results of the rail vehicles on the inspection track by at least one external inspection device; The target maintenance component of the rail vehicle on the maintenance track is identified from the preliminary inspection results.

6. The rail vehicle service method according to any one of claims 1-5, characterized in that, Before generating and sending the first command to the drone, the method further includes: Obtain the point cloud of the maintenance track; Based on the point cloud, establish the maintenance trajectory between the vehicle maintenance start point and the vehicle maintenance end point of the maintenance track.

7. The rail vehicle service method according to any one of claims 1-5, characterized in that, Before generating and sending the first command to the drone, the method further includes: The maintenance track selects a first route from at least two routes to the UAV nest to the vehicle maintenance start point of the maintenance track, and a second route from the vehicle maintenance end point of the maintenance track to the UAV nest. Both the first and second routes were obtained in advance by setting navigation points along the route based on historical flight data.

8. The rail vehicle service method according to any one of claims 1-5, characterized in that, The real-time maintenance of the target repair component based on the second image data includes: Perform a field-of-view quality assessment on the video frames in the second image data; The target video frame is determined from the second image data based on the visual field quality assessment results, so as to perform real-time maintenance on the target maintenance component based on the target video frame.

9. A rail vehicle maintenance system, characterized in that, include: The control module is used to generate and send a first command to the drone; receive the first image data sent by the drone; and perform recognition processing on the first image data. If the first image data includes a target repair component, a second instruction is generated and sent to the UAV; the second image data sent by the UAV is received to perform real-time repair on the target repair component based on the second image data; The drone is used to receive a first instruction sent by the control module to collect first image data of the track vehicle under maintenance and send the first image data to the control module; or it is used to receive a second instruction sent by the control module to collect second image data of the track vehicle under maintenance and send the second image data to the control module.

10. The rail vehicle maintenance system according to claim 9, characterized in that, The control module is also used for: Determine the type of the target maintenance component in the second image data, and invoke the corresponding anomaly detection rule according to the type of the target maintenance component; Based on the second image data, the target maintenance component is detected in real time according to the anomaly detection rules.

Images