Railway vehicle inspection robot teaching platform based on digital twinning

Abstract

The invention discloses a railway vehicle inspection robot teaching platform based on digital twinning, and relates to the technical field of intelligent transportation and digital twinning. Comprising a physical entity layer, a digital twinborn layer and a teaching resource layer, the physical entity layer comprises a movable chassis, a six-degree-of-freedom mechanical arm and a 10-gigabit internet access industrial area-array camera; the digital twin layer comprises a Blender three-dimensional modeling unit, a Unity virtual scene engine and a multi-protocol communication middleware; the teaching resource layer constructs a three-level experiment teaching system comprising a basic operation layer, an advanced practice layer and a comprehensive innovation layer based on a three-layer collaborative architecture. By constructing a three-layer collaborative closed loop of a physical entity, digital twinning and teaching resources, while the hardware cost and maintenance difficulty are remarkably reduced, the key technical features of the industrial-grade inspection robot are completely reproduced, and the contradiction between the teaching cost and the practical operation authenticity in traditional practical training is effectively solved.

Description

A teaching platform for rail vehicle inspection robots based on digital twins Technical Field

[0001] This invention relates to the fields of intelligent transportation and digital twin technology, and in particular to a teaching platform for a rail vehicle inspection robot based on digital twins. Background Technology

[0002] With the rapid expansion and accelerated intelligent transformation of my country's urban rail transit network, the rail vehicle operation and maintenance system is undergoing a historic leap from traditional manual inspection to a high-level automated and intelligent inspection mode. Since the establishment of the first demonstration application site for deploying intelligent inspection robots for urban rail vehicles in China in 2021, many core cities have successively carried out engineering trials of cutting-edge technologies such as trackside intelligent inspection robots, 360° image intelligent inspection systems, and even "flying train inspection" robots. These intelligent equipment generally integrate mobile chassis, multi-degree-of-freedom robotic arms, high-definition vision and 3D imaging modules, and rely on deep learning algorithms to achieve sub-millimeter-level image acquisition of key components under the car body such as bogies, brake discs, and pantographs, as well as real-time identification and graded alarms for typical defects such as cracks, loosening, and missing parts. This significantly improves maintenance efficiency and reliability, becoming a key enabling technology for building an intelligent operation and maintenance system for urban rail transit.

[0003] Against this backdrop, the competency structure of rail transit maintenance positions is undergoing profound changes. Personnel must not only master traditional mechanical disassembly and electrical fault diagnosis skills, but also possess comprehensive application capabilities in composite technologies such as Simultaneous Localization and Mapping (SLAM) navigation for inspection robots, robotic arm motion control, machine vision inspection, and embedded intelligent algorithm optimization. However, current urban rail vehicle maintenance training courses in most vocational schools and higher education institutions still highly focus on static equipment maintenance and basic circuit testing, lacking coverage of the core technology chain of intelligent inspection systems. This results in a significant gap between talent cultivation and actual industry needs. More critically, industrial-grade rail vehicle inspection robots, due to the integration of expensive components such as high-precision LiDAR, collaborative robotic arms, and 10-gigabit industrial cameras, are prohibitively expensive, far exceeding the equipment procurement budgets of ordinary educational institutions. This hinders the widespread adoption of hands-on training in real-world scenarios, severely restricting the large-scale cultivation of "AI + rail" composite technical and skilled personnel.

[0004] Existing teaching solutions often employ simplified simulation devices or purely software simulation platforms. While these reduce costs to some extent, they struggle to replicate the complex technical features of real inspection systems, such as multimodal sensor fusion, electromechanical collaborative control, and virtual-real data closed-loop systems. It's worth noting that such simplified solutions present a deep-seated technical contradiction in their teaching logic: over-reliance on physical entities leads to high costs and complex maintenance, failing to meet large-scale teaching needs; conversely, a complete shift to virtual simulation lacks dynamic interaction with real hardware, hindering students' intuitive understanding of engineering realities like robot kinematics, visual calibration errors, and environmental noise, resulting in a superficial understanding. Fundamentally, the problem lies in the failure of existing teaching methods to effectively bridge the gap between "physical realism" and "teaching accessibility"—that is, how to construct a cost-effective, fully functional training platform that supports multi-level teaching tasks while ensuring key technical parameters (such as positioning accuracy, image resolution, and control response latency) approach industrial-grade levels. This contradiction has become particularly prominent since the rise of digital twin technology: although digital twins can theoretically achieve virtual-real mapping, without a high-fidelity physical entity as a data source and feedback anchor, its virtual model will become a "castle in the air" that is detached from engineering constraints and cannot support advanced teaching activities such as hand-eye calibration, defect sample training, and fault injection debugging.

[0005] Therefore, how to construct a three-layer architecture teaching platform that deeply integrates physical entities, digital twins, and teaching resources to achieve systematic training in core skills such as SLAM mapping, robotic arm trajectory planning, multi-protocol data communication, and intelligent detection algorithm deployment has become a key challenge and an urgent technical problem for those skilled in the art. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a teaching platform for rail vehicle inspection robots based on digital twins. By constructing a three-layer collaborative closed loop of physical entities, digital twins, and teaching resources, it significantly reduces hardware costs and maintenance difficulties while fully replicating the key technical features of industrial-grade inspection robots. This effectively resolves the contradiction between teaching costs and the authenticity of practical training in traditional training. Based on this, the platform constructs a full-chain, scalable teaching system that spans basic operations, advanced practice, and comprehensive innovation. It also supports highly flexible virtual-real interaction and fault simulation, systematically cultivating students' abilities in fault diagnosis, system integration, and innovative application in complex engineering scenarios, thereby bridging the gap between talent cultivation and actual industry needs.

[0007] This invention provides a teaching platform for a rail vehicle inspection robot based on digital twins, comprising a physical entity layer, a digital twin layer, and a teaching resource layer. The physical entity layer includes a mobile chassis, a six-degree-of-freedom robotic arm, and a 10 Gigabit Ethernet industrial area array camera. The mobile chassis adopts a four-wheel differential drive structure, with a standard mounting flange on the top for fixing the robotic arm base. A LiDAR bracket is screwed to the front, and a single-line scanning LiDAR is fixed on the bracket. An inertial measurement unit is integrated inside the chassis and connected to an embedded navigation motherboard via an SPI bus. The navigation motherboard runs a Linux operating system and a ROS navigation stack, supporting dual-mode map construction with and without tags. Each joint of the six-degree-of-freedom robotic arm is equipped with a servo motor, which has a built-in high-resolution encoder and torque sensor. It communicates with the main controller via a CAN bus, and the end flange is connected to the industrial area array camera lens via a C-port thread. The industrial area array camera uses a global shutter CMOS sensor, supports hard trigger mode, and connects to the GigE... The Vision protocol transmits image data; the digital twin layer includes a Blender 3D modeling unit, a Unity virtual scene engine, and a multi-protocol communication middleware, used to construct high-fidelity rail vehicle component and robot body model, and to synchronize with the physical entity layer in terms of spatial pose, motion timing, and state parameters through MQTT or HTTP protocols; the teaching resource layer is based on a three-layer collaborative architecture, constructing a three-level experimental teaching system including a basic operation layer, an advanced practice layer, and a comprehensive innovation layer.

[0008] Preferably, the physical entity layer further includes: a mobile chassis generating a 2D grid map using an in-situ rotation plus U-shaped path planning strategy in the unlabeled map building mode; enhancing positioning robustness through pre-set QR codes in the labeled map mode; and supporting the setting of virtual walls in the ROS navigation system to limit activity boundaries after mapping is completed; a six-degree-of-freedom robotic arm supporting three operation modes: drag teaching, program control, and remote command; its motion control unit receiving PTP, LIN, or ARC motion commands and monitoring the angles, angular velocities, and end-effector center point pose of each joint in real time; and configured with collision detection, soft limit protection, and emergency stop de-enablement safety mechanisms; and an industrial area scan camera working with a ring LED light source component to achieve uniform illumination; image acquisition supporting both continuous shooting and single-frame triggering modes, with adjustable exposure time, gain parameters, and pixel format.

[0009] Preferably, the digital twin layer further includes: a Blender 3D modeling unit that performs high-fidelity modeling of the pantograph, bogie, brake disc, and inspection robot body and exports FBX format files; a Unity virtual scene engine that constructs a virtual environment based on the actual layout of the rail vehicle depot, enables a physics engine to achieve collision detection, and embeds a UI interactive interface to support virtual robot control, defect annotation, data visualization, and historical backtracking; and a multi-protocol communication middleware that configures differentiated data channels, with the pantograph model receiving attitude and state parameters via the MQTT protocol, and the inspection robot chassis and the six-degree-of-freedom robotic arm synchronizing position coordinates, speed information, joint angles, and end-effector pose data via the HTTP protocol.

[0010] Preferably, in the teaching resource layer, the basic operation layer includes chassis navigation configuration, label-free map construction, detection point calibration, virtual wall setting, Blender modeling, Unity scene building, and MQTT protocol configuration; the advanced practice layer includes camera calibration, robotic arm trajectory planning, bogie defect detection, Blob analysis algorithm optimization, and virtual-real collaborative fault diagnosis; and the comprehensive innovation layer includes pantograph intelligent inspection process design, Unity visual interface development, and virtual-real closed-loop control under multi-protocol collaboration.

[0011] Preferably, the platform supports fault simulation teaching function. By modifying the pantograph regulating valve coefficient, the robotic arm joint angle offset, the chassis motor voltage parameters, the camera power supply status, or the solenoid valve logic status in the Unity scene, faults such as jamming, joint jamming, current overload, image interruption, or lifting failure are simulated respectively, and are simultaneously reflected in the physical entity layer for students to troubleshoot.

[0012] Preferably, the platform includes a navigation scheduling module; the navigation scheduling module is implemented based on the ROS navigation framework, supports dual-mode map construction with and without labels, and has multi-machine collaborative scheduling capabilities; target point calibration covers patrol points, charging points, and inspection points, and the distance between each calibration point and obstacles or virtual walls is not less than 20cm; path planning supports global route drawing and local obstacle avoidance dynamic adjustment, and the minimum passage width is set to 80cm; the multi-machine scheduling function is implemented through the ESP communication module, supporting two robots to coordinate collision avoidance in narrow areas based on preset avoidance points.

[0013] Preferably, the platform includes a robotic arm control module; the robotic arm control module integrates Robotmaster trajectory planning software, supports CAD model import and automatic generation of KRL control code; the teaching function provides single-axis jogging and multi-axis linkage modes for recording the spatial coordinates and attitude parameters of the teaching points, and allows adjustment of running speed and acceleration; motion command types include PTP, LIN and ARC, with configurable smooth transition time and spatial offset; safety mechanisms include collision detection, soft limit protection and emergency stop de-enablement, which require manual reset to resume operation after triggering; parameter configuration items include installation method, end effector load mass and tool coordinate system, supporting four-point and six-point methods for tool calibration.

[0014] Preferably, the platform includes a visual inspection module; the visual inspection module integrates algorithms for blob region extraction, template matching, and circular contour search, and can identify cracks with a minimum width of 0.1mm, loose bolts with an angle deviation of 5° or greater, and defects such as complete missing parts. The geometric dimension deviation detection error is controlled within ±0.2mm, and the detection results are stored in the form of image files with embedded metadata and support one-click export of standardized inspection reports.

[0015] Preferably, the platform's workflow includes four stages: map building, path planning, component inspection, and data feedback. In the map building stage, the mobile chassis activates a lidar scanner to scan the rail depot environment. In unlabeled mode, a two-dimensional grid map is generated using a scanning strategy combining in-situ rotation and U-shaped movement. In complex areas, it switches to labeled mode, using pre-set QR codes to improve positioning robustness. In the path planning stage, inspection points for the bogie, braking system, and pantograph are calibrated according to the inspection task. The navigation path points are spaced no more than 1 meter apart and at least 30 cm away from obstacles. Manual navigation is supported. The system employs two methods: dynamic rendering and automatic planning. During the component inspection phase, the robot travels along the planned path to the designated inspection point. The robotic arm adjusts its end-effector pose according to the preset trajectory, and the industrial camera acquires component images under the drive of a hard trigger signal. The vision control host calls Blob analysis and feature matching algorithms to identify typical defects such as cracks, looseness, and missing parts, and integrates 3D scanning data to complete the detection of geometric dimension deviations. During the data feedback phase, the inspection results are embedded in timestamps and exposure parameters and stored in a local database. At the same time, they are pushed to the Unity virtual scene via HTTP protocol for real-time annotation and visualization.

[0016] Compared with the prior art, the present invention has the following advantages and beneficial effects: 1. By using high-fidelity dynamic mapping between the physical entity layer and the digital twin layer, while retaining the key technical features of industrial robots (such as high-precision positioning and real-time visual detection), the hardware investment and maintenance costs are greatly reduced, solving the problem that practical training is difficult to carry out due to expensive equipment in traditional training, and achieving the unity of physical authenticity and teaching accessibility.

[0017] 2. Relying on a three-tier collaborative architecture and structured teaching resources, a complete teaching loop is formed, from basic operation and advanced practice to comprehensive innovation. It covers core skills such as SLAM mapping, robotic arm control, machine vision, and multi-protocol communication, helping students gradually master the integrated design and engineering implementation capabilities of intelligent inspection systems, bridging the gap between talent cultivation and industry needs.

[0018] 3. Through the digital twin layer, various virtual faults (such as mechanical jamming and communication interruption) can be flexibly injected and synchronized to the physical entity in real time, enabling students to conduct fault diagnosis and troubleshooting training in near-real operation and maintenance scenarios, and strengthening their comprehensive analysis and problem-solving abilities regarding the working principles of complex systems and engineering practice problems. Attached Figure Description

[0019] Figure 1 is a schematic diagram of a teaching platform for a rail vehicle inspection robot based on digital twins.

[0020] Figure 2 is an architecture diagram of a teaching platform for rail vehicle inspection robots based on digital twins.

[0021] Figure 3 shows the Unity scene development interface of a teaching platform for rail vehicle inspection robots based on digital twins.

[0022] Figure 4 shows the practical operation interface of a teaching platform for rail vehicle inspection robots based on digital twins. Detailed Implementation

[0023] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0024] Example 1: To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0025] Please refer to Figures 1-4. This invention provides a teaching platform for a rail vehicle inspection robot based on digital twins. Its overall architecture consists of a physical entity layer, a digital twin layer, and a teaching resource layer. The three layers achieve high-fidelity dynamic mapping and a closed loop of teaching functions through a unified data communication mechanism and control logic. The following will describe in detail the specific implementation of this invention from the aspects of system composition, hardware integration, software architecture, data flow design, control logic, and typical workflow.

[0026] The physical entity layer uses the composite robot body as its core carrier, integrating a mobile chassis, a six-degree-of-freedom robotic arm, and a 10 Gigabit Ethernet industrial area array camera. The mobile chassis adopts a four-wheel differential drive structure, and a standard mounting flange is provided on the top of the chassis frame for fixing the multi-degree-of-freedom robotic arm base. A lidar bracket is screwed onto the front of the chassis, and a single-line scanning lidar with a wavelength of 905nm, a range of 25m, and a scanning angle of 230° is fixed on the bracket. The chassis integrates an inertial measurement unit, including a single-axis gyroscope and a three-axis accelerometer, which is connected to the embedded navigation system via an SPI bus. The motherboard communicates with the navigation motherboard, which runs a Linux operating system and a ROS navigation stack. It supports map construction in both tagless and tagged modes. In tagless mode, a U-shaped path planning strategy is used to complete the environment scan, while in tagged mode, pre-set QR codes are used to improve positioning robustness. The mobile chassis has a repeatability accuracy of ±0.1mm, a maximum load capacity of 300kg, and a moving speed range of 0.1 to 1m / s. The built-in lithium battery pack supports continuous operation for more than 6 hours and supports both manual charging and automatic recharging modes, which can meet the needs of long-term inspection operations of rail transit vehicles.

[0027] The six-degree-of-freedom robotic arm features a collaborative joint structure, with each joint equipped with a servo motor. Each motor incorporates a high-resolution encoder and torque sensor, enabling real-time data exchange with the main controller via a CAN bus. The robotic arm has a working radius of 1868 mm and a repeatability of ±0.1 mm. Its end effector flange conforms to the ISO 9409-1-50-4-M6 standard mounting interface. The robotic arm supports three operating modes: drag-and-teach, program control, and remote command. Its motion control unit receives PTP, LIN, or ARC motion commands from trajectory planning software and monitors the angles, angular velocities, and end effector center point pose of each joint in real time. The end effector flange connects to an industrial area array camera lens via a C-type thread, and a ring-shaped LED light source assembly provides uniform illumination.

[0028] The industrial area scan camera with a 10 Gigabit Ethernet port uses a global shutter CMOS sensor with a resolution of 5120×5120 pixels and a frame rate of 68 frames per second. It supports hard trigger mode to synchronize with the movement of the robotic arm. Image data is transmitted to the vision control host via Gigabit Ethernet through the GigE Vision protocol.

[0029] The digital twin layer includes a Blender 3D modeling unit, a Unity virtual scene engine, and a multi-protocol communication middleware. The Blender 3D modeling unit performs high-fidelity 3D modeling of key components of the rail vehicle (including pantographs, bogies, and brake discs) and the inspection robot itself. It employs polygon subdivision surface technology to improve the geometric accuracy of the model and assigns physical material properties such as metal and rubber. Texture baking is used to simulate the optical properties of real surfaces. After modeling, the model is exported as an FBX file, retaining the original topology, UV layout, and material information. The Unity virtual scene engine imports the FBX model and constructs a virtual environment based on the actual maintenance trenches and trackside equipment layout of the rail vehicle depot. It configures a hybrid lighting system including parallel light and point light sources and enables a physics engine to perform collision detection and rigid body dynamics simulation. The scene includes an embedded UI interface that supports virtual robot motion control, defect location marking, inspection data visualization, and operation history backtracking. The multi-protocol communication middleware configures data links differently according to the model object type. The pantograph model receives attitude data and state parameters through the MQTT protocol, the inspection robot chassis synchronizes position coordinates and speed information through the HTTP protocol, and the six-degree-of-freedom robotic arm obtains the angles of each joint and the end pose through the HTTP protocol, thereby ensuring high-fidelity synchronization between the virtual model and the physical entity in terms of spatial pose, motion timing and state feedback.

[0030] The teaching resource layer, relying on the collaborative mapping mechanism of the physical entity layer and the digital twin layer, constructs a three-tiered experimental teaching system covering knowledge transfer, skills training, and literacy cultivation. Specifically, it includes a basic operation layer, an advanced practice layer, and a comprehensive innovation layer, each with clearly defined teaching objectives, experimental projects, key technical points, and quantitative assessment indicators. The basic operation layer focuses on equipment deployment and environment modeling, covering composite robot chassis navigation parameter configuration, label-free map construction, detection point calibration, virtual wall setup, Blender pantograph modeling, Unity scene construction, and MQTT protocol communication configuration. The advanced practice layer focuses on electromechanical collaboration and algorithm application, including machine vision camera calibration, robotic arm trajectory planning, bogie defect identification, Blob analysis algorithm tuning, and virtual-physical collaborative fault diagnosis. The comprehensive innovation layer emphasizes system integration and engineering implementation, requiring students to independently design intelligent pantograph inspection processes, develop a Unity platform visual interactive interface, and achieve virtual-physical closed-loop control under multi-protocol collaboration.

[0031] The platform's workflow comprises four stages: map building, path planning, component inspection, and data feedback. In the map building stage, the mobile chassis activates its LiDAR to scan the rail depot environment. In unlabeled mode, a scanning strategy combining in-situ rotation and U-shaped movement generates a 2D grid map. In complex areas, it switches to labeled mode, using pre-set QR codes to improve positioning robustness. After map building, virtual walls can be set in the ROS navigation system to define the robot's activity boundaries. In the path planning stage, inspection points for key components such as the bogie, braking system, and pantograph are calibrated in the environment according to the inspection task. The navigation path points are spaced no more than 1 meter apart and at least 30 cm away from obstacles, supporting both manual drawing and automatic planning. In the component inspection stage, the robot travels along the planned path to the designated inspection point. The robotic arm adjusts its end effector pose according to a preset trajectory, and an industrial camera acquires component images under hard-triggered signal drive. The vision control host calls Blob analysis and feature matching algorithms to identify typical defects such as cracks, looseness, and missing parts, and integrates 3D scan data to complete geometric dimension deviation detection. During the data feedback phase, all detection results (including defect type, location, size, and original image) are embedded with metadata such as timestamps and exposure parameters and stored in a local database. At the same time, they are pushed to the Unity virtual scene via HTTP protocol for real-time annotation and visualization. Teachers can access the detection report through the backend management system, and students can compare the standard data with the actual test results for error analysis.

[0032] The platform further integrates fault simulation and troubleshooting teaching functions. In the digital twin layer, jamming faults can be simulated by modifying the regulating valve coefficient of the pantograph lifting mechanism within the Unity scene, joint jamming can be simulated by adjusting the joint angle offset of the robotic arm, an overload alarm can be triggered by setting an abnormal chassis motor drive voltage, image acquisition interruption can be simulated by cutting off the camera power supply link, or lifting failure can be simulated by manipulating the pantograph solenoid valve logic state. These virtual fault states are synchronously mapped to the physical entity layer. Students need to combine MQTT / HTTP protocol data stream parsing, on-site mechanical inspection, and electrical measurement to complete fault location and troubleshooting. The overall synchronization error between the virtual and physical states does not exceed 1 second.

[0033] The navigation scheduling module is implemented based on the ROS navigation framework, supporting both label-free and label-coded map construction and possessing multi-robot collaborative scheduling capabilities. Network communication supports 2.4GHz / 5GHz dual-band Wi-Fi access, and the navigation system's IP address can be quickly configured via FTP. Target point calibration covers departure points, charging points, and inspection points, with a minimum distance of 20cm between each calibration point and obstacles or virtual walls. Path planning supports global route drawing and dynamic adjustment for local obstacle avoidance, with a minimum passage width set at 80cm. The multi-robot scheduling function is implemented through the ESP communication module, supporting collision avoidance coordination between two robots in narrow areas based on preset avoidance points to prevent operational congestion.

[0034] The robotic arm control module integrates Robotmaster trajectory planning software, supports CAD model import, and automatically generates KRL control code. The teaching function offers single-axis jogging and multi-axis linkage modes, can record the spatial coordinates and attitude parameters of the teaching point, and allows adjustment of the running speed (0~100% standard speed) and acceleration. Motion command types include PTP (point-to-point), LIN (linear interpolation), and ARC (circular interpolation), with configurable smooth transition time and spatial offset. Safety mechanisms include collision detection, soft limit protection, and emergency stop de-enablement; manual reset is required to resume operation after triggering. Parameter configuration items include installation method (orthodox, inverted, or side-mounted), end-effector load mass, and tool coordinate system, supporting four-point and six-point methods for tool calibration.

[0035] The visual inspection module is built on the MVS client and SDK development kit. Image acquisition supports both continuous shooting and single-frame triggering modes, and the pixel format, exposure time, and gain parameters are adjustable. The maximum packet size for network communication is 8164 bytes. The defect detection algorithm integrates blob region extraction, template matching, and circular contour search functions. It can identify cracks with a minimum width of 0.1mm, loose bolts with an angle deviation of 5° or greater, and defects such as complete missing parts. Geometric accuracy analysis is achieved through 3D point cloud reconstruction and reverse modeling comparison. The detection data is stored in an image file format with embedded metadata, and a standardized inspection report containing defect details and original images can be exported with one click.

[0036] In the integrated innovation layer teaching practice, students are required to independently design an intelligent pantograph inspection process. This includes: creating a high-fidelity 3D model of the pantograph in Blender and importing it into Unity to build a virtual inspection scene; configuring the robotic arm's motion trajectory and camera acquisition parameters at the physical entity layer; achieving virtual-real state synchronization through multi-protocol communication middleware; and developing an interactive interface for defect annotation and data backtracking in Unity. This process requires students to comprehensively apply knowledge from multiple fields such as mechanical design, motion control, machine vision, communication protocols, and software development, demonstrating the systematic advantages of this invention in cultivating interdisciplinary intelligent rail transit professionals.

[0037] In the teaching example of fault simulation and troubleshooting, the teacher injects a pantograph jamming fault through the Unity backend, and the system automatically modifies the regulating valve coefficient and synchronizes it to the physical entity layer; the students listen to the pantograph status data stream through the MQTT protocol, find that the lifting action is abnormal, and confirm that the hydraulic circuit pressure is insufficient by combining on-site inspection, and finally locate the regulating valve fault.

[0038] The platform's multi-robot collaborative scheduling function has also been verified in the field. In a narrow passage 80cm wide, two robots performed different inspection tasks. When a collision risk was detected, the ESP communication module triggered a collision avoidance coordination mechanism, pausing one robot while the other passed through a pre-set avoidance point. Experimental results showed that the multi-robot scheduling success rate was 100%, with no blockages or collisions occurring.

[0039] In summary, this invention, through the construction of a three-layer collaborative architecture comprising a physical entity layer, a digital twin layer, and a structured teaching resource layer, fully replicates the core technical features of an industrial-grade rail vehicle inspection robot. The platform meets teaching application requirements in key indicators such as mobile navigation accuracy, robotic arm repeatability, visual acquisition synchronization, and defect identification capabilities. Furthermore, it effectively bridges the cognitive gaps in traditional practical training through a high-fidelity real-time mapping mechanism. The constructed three-level experimental teaching system covers the entire chain of capability development from basic operation to comprehensive innovation, providing a scalable, maintainable, and cost-effective systematic teaching platform for cultivating intelligent rail transit talent.

[0040] All content not described in detail in this specification is prior art known to those skilled in the art, and the model parameters of each electrical appliance are not specifically limited; conventional equipment can be used. Electrical control components not mentioned in this technical solution are not shown in the figures because they are prior art, and will not be described further here.

[0041] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A teaching platform for a rail vehicle inspection robot based on digital twins, characterized in that: The system comprises a physical entity layer, a digital twin layer, and a teaching resource layer. The physical entity layer includes a mobile chassis, a six-degree-of-freedom robotic arm, and a 10 Gigabit Ethernet industrial area array camera. The mobile chassis employs a four-wheel differential drive structure, with a standard mounting flange on top for securing the robotic arm base. A LiDAR bracket is screwed onto the front, and a single-line scanning LiDAR is mounted on the bracket. An inertial measurement unit is integrated within the chassis and connected to an embedded navigation motherboard via an SPI bus. The navigation motherboard runs a Linux operating system and a ROS navigation stack, supporting both tagless and tagged map construction. Each joint of the six-degree-of-freedom robotic arm is equipped with a servo motor, which incorporates a high-resolution encoder and torque sensor. The motor communicates with the main controller via a CAN bus, and the end flange is threaded to the industrial area array camera lens via a C-port thread. The industrial area array camera uses a global shutter CMOS sensor, supports hard triggering mode, and connects to the GigE... The Vision protocol transmits image data; the digital twin layer includes a Blender 3D modeling unit, a Unity virtual scene engine, and a multi-protocol communication middleware, used to construct high-fidelity rail vehicle component and robot body model, and to synchronize with the physical entity layer in terms of spatial pose, motion timing, and state parameters through MQTT or HTTP protocols; the teaching resource layer is based on a three-layer collaborative architecture, constructing a three-level experimental teaching system including a basic operation layer, an advanced practice layer, and a comprehensive innovation layer.

2. The teaching platform for rail vehicle inspection robots based on digital twins according to claim 1, characterized in that: The physical entity layer also includes: a mobile chassis that generates a 2D grid map using in-situ rotation and U-shaped path planning strategies in the unlabeled map building mode, and enhances positioning robustness through pre-set QR codes in the labeled map mode. After mapping is completed, it supports setting virtual walls in the ROS navigation system to limit the activity boundary; a six-degree-of-freedom robotic arm that supports three operation modes: drag teaching, program control, and remote command. Its motion control unit receives PTP, LIN, or ARC motion commands and monitors the angles, angular velocities, and end-effector center point pose of each joint in real time. It is also equipped with collision detection, soft limit protection, and emergency stop de-enablement safety mechanisms; an industrial area scan camera with a ring LED light source component achieves uniform illumination. Image acquisition supports both continuous shooting and single-frame triggering modes, and the exposure time, gain parameters, and pixel format can be adjusted.

3. The teaching platform for rail vehicle inspection robots based on digital twins according to claim 1, characterized in that: The digital twin layer also includes: a Blender 3D modeling unit that performs high-fidelity modeling of the pantograph, bogie, brake disc, and inspection robot body and exports FBX format files; a Unity virtual scene engine that constructs a virtual environment based on the actual layout of the rail vehicle depot, enables a physics engine to achieve collision detection, and embeds a UI interactive interface to support virtual robot control, defect annotation, data visualization, and historical backtracking; and a multi-protocol communication middleware that configures differentiated data channels, with the pantograph model receiving attitude and state parameters via the MQTT protocol, and the inspection robot chassis and the six-degree-of-freedom robotic arm synchronizing position coordinates, speed information, joint angles, and end-effector pose data via the HTTP protocol.

4. The teaching platform for rail vehicle inspection robots based on digital twins according to claim 1, characterized in that: The teaching resource layer includes the following layers: the basic operation layer includes chassis navigation configuration, tagless map construction, detection point calibration, virtual wall setting, Blender modeling, Unity scene building, and MQTT protocol configuration; the advanced practice layer includes camera calibration, robotic arm trajectory planning, bogie defect detection, Blob analysis algorithm optimization, and virtual-real collaborative fault diagnosis; and the comprehensive innovation layer includes pantograph intelligent inspection process design, Unity visual interface development, and virtual-real closed-loop control under multi-protocol collaboration.

5. The teaching platform for rail vehicle inspection robots based on digital twins according to claim 1, characterized in that: The platform supports fault simulation teaching functions. By modifying the pantograph regulating valve coefficient, robotic arm joint angle offset, chassis motor voltage parameters, camera power supply status, or solenoid valve logic status in the Unity scene, it can simulate faults such as jamming, joint jamming, current overload, image interruption, or lifting failure, and simultaneously reflect them to the physical entity layer for students to troubleshoot.

6. The teaching platform for rail vehicle inspection robots based on digital twins according to claim 1, characterized in that: The platform includes a navigation scheduling module; the navigation scheduling module is based on the ROS navigation framework, supports dual-mode map construction with and without labels, and has multi-robot collaborative scheduling capabilities; target point calibration covers patrol points, charging points, and inspection points, and the distance between each calibration point and obstacles or virtual walls is not less than 20cm; path planning supports global route drawing and local obstacle avoidance dynamic adjustment, and the minimum passage width is set to 80cm; the multi-robot scheduling function is implemented through the ESP communication module, which supports two robots to coordinate collision avoidance in narrow areas based on preset avoidance points.

7. The teaching platform for rail vehicle inspection robots based on digital twins according to claim 1, characterized in that: The platform includes a robotic arm control module; the robotic arm control module integrates Robotmaster trajectory planning software, supports CAD model import and automatic generation of KRL control code; the teaching function provides single-axis jog and multi-axis linkage modes, used to record the spatial coordinates and attitude parameters of the teaching point, and allows adjustment of running speed and acceleration; motion command types include PTP, LIN and ARC, and can be configured with smooth transition time and spatial offset; The safety mechanism includes collision detection, soft limit protection, and emergency stop de-enablement. After being triggered, manual reset is required to resume operation. The parameter configuration items include installation method, end load mass, and tool coordinate system. It supports four-point and six-point methods for tool calibration.

8. The teaching platform for rail vehicle inspection robots based on digital twins according to claim 1, characterized in that: The platform includes a visual inspection module; the visual inspection module integrates algorithms for blob region extraction, template matching and circular contour search, and can identify cracks with a minimum width of 0.1mm, loose bolts with an angle deviation of 5° or greater and complete missing defects. The geometric dimension deviation detection error is controlled within ±0.2mm. The detection results are stored in the form of image files with embedded metadata and support one-click export of standardized inspection reports.

9. The teaching platform for rail vehicle inspection robots based on digital twins according to any one of claims 1-8, characterized in that: The platform's workflow comprises four stages: map building, path planning, component inspection, and data feedback. In the map building stage, the mobile chassis activates a lidar scanner to scan the rail depot environment. In unlabeled mode, a two-dimensional grid map is generated using a scanning strategy combining in-situ rotation and U-shaped movement. In complex areas, it switches to labeled mode, using pre-set QR codes to improve positioning robustness. In the path planning stage, inspection points for the bogie, braking system, and pantograph are calibrated according to the inspection task. The navigation path has a point spacing of no more than 1 meter and a distance of more than 30 cm from obstacles, supporting both manual drawing and automatic planning. In the component inspection stage, the robot travels along the planned path to the designated inspection point. The robotic arm adjusts its end effector pose according to a preset trajectory, and the industrial camera acquires component images under hard-triggered signal drive. The vision control host calls Blob analysis and feature matching algorithms to identify typical defects such as cracks, loosening, and missing parts, and integrates 3D scan data to complete geometric dimension deviation detection. In the data feedback stage, the inspection results are embedded in timestamps and exposure parameters and stored in a local database. Simultaneously, they are pushed to the Unity virtual scene via HTTP protocol for real-time annotation and visualization.

Images