A railway freight yard whole-process intelligent inspection system and an inspection method thereof

Abstract

This invention discloses a full-process intelligent inspection system and method for railway freight yards, comprising a mobile inspection terminal, edge computing nodes, and a central server. The mobile inspection terminal includes a quadruped robot for performing inspection tasks within the railway freight yard. The edge computing nodes are deployed near the work area of ​​the railway freight yard to perform near-field real-time analysis of the inspection data collected by the mobile inspection terminal to identify the physical state characteristics of target objects, and then transmit the physical state characteristic data to the central server via a communication network. The central server receives the data and performs business logic judgments based on the physical state characteristic data. Based on the judgment results, it triggers hierarchical response operations, including the execution of alarm actions by the control alarm module, via the communication network. By replacing field freight workers with quadruped robots to complete the vast majority of inspection tasks, and combining 3D SLAM mapping and autonomous obstacle avoidance technology, close-range, refined inspection of the entire freight yard area and all train locations can be achieved.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection technology for railway freight yards, and in particular to an intelligent inspection system and method for the entire process of railway freight yards. Background Technology

[0002] Currently, safety inspections and loading / unloading operations verification at domestic railway freight yards mainly rely on on-site manual work by field freight workers. The core tasks include three inspections before and after freight car loading / unloading, comprehensive safety hazard investigation of the freight yard, verification of the status of operating machinery and equipment, control of on-site personnel's work behavior, and compliance checks of cargo stacking. These are important foundations for consolidating the safety baseline of railway freight operations and ensuring the standardized and orderly operation of on-site operations.

[0003] Currently, on-site operations at railway stations generally rely on traditional methods such as manual point-to-point verification, visual inspections, and paper-based record keeping, resulting in a relatively crude overall management approach. With the large-scale expansion of railway freight business, the normalization of daily operations, and the upgrading of transportation efficiency, the total volume of freight yard operations continues to rise, and the on-site operating environment and business scenarios are becoming increasingly complex.

[0004] While some freight yards have already deployed security monitoring facilities such as fixed video surveillance and fixed-point sensor data collection devices, and a few areas have piloted the use of wheeled inspection robots to complete basic inspection tasks, thus initially establishing an intelligent auxiliary inspection system, these semi-intelligent monitoring methods are limited to fixed-point data and video collection. They generally suffer from prominent shortcomings such as incomplete inspection coverage, blind spots in on-site inspections, insufficient mobile inspection capabilities of equipment, and weak flexibility in adapting to different scenarios. These shortcomings make it difficult to meet the modern management requirements of refined control, reduced manpower, and intelligent operation and maintenance of railway smart freight yards in the new era. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a fully intelligent inspection system and method for railway freight yards. This system relies on the highly adaptable mobility of quadruped robots, vision and SLAM fusion navigation technology, a dual-layer detection architecture combining image recognition strategies, and a decoupled intermediate instruction set design. It achieves fully automated, standardized, and intelligent operations for pre- and post-loading inspections, safety inspections of the freight yard area, vehicle status detection, and site safety verification. This replaces the traditional manual inspection mode, reduces labor intensity, eliminates blind spots, improves inspection accuracy and efficiency, and is suitable for complex and dynamic operational scenarios in railway freight yards.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] In one aspect, this application discloses a full-process intelligent inspection system for railway freight yards, including a communication network and mobile inspection terminals, edge computing nodes, and a central server interconnected through the communication network. The mobile inspection terminal includes a quadruped robot for performing inspection tasks within the railway freight yard, and the quadruped robot is equipped with a data acquisition module and an alarm module. The edge computing node is deployed near the operating area of ​​the railway freight yard to perform near-field real-time analysis of the inspection data collected by the mobile inspection terminal to identify the physical state characteristics of the target object, and sends the physical state characteristic data to the central server through the communication network. The central server receives the data and performs business logic judgment based on the physical state characteristic data, and triggers a hierarchical response operation, including controlling the alarm module to perform alarm actions, through the communication network according to the judgment result.

[0008] Secondly, this application discloses a method for intelligent inspection of the entire process of a railway freight yard, implemented based on the aforementioned intelligent inspection system for the entire process of a railway freight yard, including:

[0009] The central server acquires loading and unloading task instructions containing information such as track number, train type, number of carriages, and operation type. Based on the task type, the central server matches preset inspection item rules and dynamically generates an inspection task list using carriage and model parameters. Items in the inspection task list that can be automatically executed by equipment are assigned to the quadruped robot, while the remaining items are reserved for manual processing. Upon receiving the instructions, the quadruped robot uses its onboard visual algorithm to identify train end features and, combined with navigation and positioning data, locks the train end coordinates. Based on the train end coordinates, track alignment, number of carriages, and vehicle size parameters, a point offset algorithm automatically generates inspection point coordinates covering all inspection items on the entire train, forming an inspection path. The quadruped robot autonomously plans its walking route based on the inspection path. When an abnormal risk is detected, a tiered early warning execution task is triggered, including controlling the quadruped robot to issue audible and visual alarm signals and pushing alarm information to the dispatch terminal. After the inspection task is completed, the central server automatically summarizes the detection results, abnormal records, and image evidence, generates a standardized inspection report, and synchronously archives the task data.

[0010] The beneficial effects of this invention are:

[0011] By replacing field freight workers with quadruped robots to complete the vast majority of inspection tasks, the number of personnel required for inspections can be reduced while improving efficiency and density. This solves the problems of high labor intensity, missed or false inspections, and high safety risks associated with manual inspections. Furthermore, quadruped robots have excellent adaptability to complex scenarios, eliminating blind spots. Leveraging the multi-degree-of-freedom motion characteristics of quadruped robots, they can adapt to complex terrains in freight yards, such as gravel roads, ramps, stairs, and track crossings. Combined with 3D SLAM mapping and autonomous obstacle avoidance technology, they can achieve close-range, detailed inspections of the entire freight yard and all train locations, addressing the pain points of poor terrain adaptability and numerous monitoring blind spots inherent in traditional equipment.

[0012] This application employs dynamic point generation technology, adapting to dynamic operations across all scenarios. It pioneers a train end-visual fusion recognition combined with a dynamic point generation algorithm, which can adapt to different train models, lengths, and track shapes (straight / curved) to accurately generate inspection points. This solves the technical challenge of traditional fixed points being unable to adapt to dynamic train parking, ensuring full inspection coverage without omissions.

[0013] A dual-layer detection architecture is adopted, improving both detection accuracy and compliance. Based on a separation architecture of intelligent image recognition and business strategy verification, it ensures the objectivity of image recognition while adhering to railway-specific operating regulations, significantly improving detection accuracy.

[0014] In addition, the hardware and software decoupling design, through an independent intermediate instruction set and robot execution subsystem, decouples the business system from the underlying equipment. Without modifying the upper-level business code, it can quickly adapt to different robot models, add inspection items, add new operating vehicle models and scenarios, and solve the problems of high coupling and high iteration costs in traditional systems.

[0015] This application adopts a closed-loop business process with a high degree of intelligence. It seamlessly connects to the railway freight management platform, realizing automatic reception of task instructions, automatic execution of inspections, automatic analysis of results, automatic risk warning, automatic data archiving, and automatic report feedback, forming a standardized business closed loop. It is fully compatible with the existing railway operation and maintenance management system. Furthermore, the equipment meets IP56-IP67 protection levels and can operate stably in complex outdoor environments such as -10℃ to 50℃, dust, light rain, and fog. It supports offline operation without network access, equipment self-recovery, and automatic switching between primary and backup systems, ensuring all-weather, highly reliable operation of inspection work. Attached Figure Description

[0016] Figure 1 A simplified schematic diagram of a railway freight yard end-to-end intelligent inspection system according to some embodiments of this application;

[0017] Figure 2 This is a schematic diagram illustrating the interaction methods according to some embodiments of this application;

[0018] Figure 3This is a phased schematic diagram of some intelligent inspection methods for the entire process of railway freight yards based on this application. Detailed Implementation

[0019] 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] An intelligent inspection system for the entire process of a railway freight yard, according to an embodiment of this application, includes a communication network and a mobile inspection terminal, an edge computing node, and a central server interconnected through the communication network. The mobile inspection terminal includes a quadruped robot for performing inspection tasks within the railway freight yard. The quadruped robot is equipped with a data acquisition module and an alarm module. The edge computing node is deployed near the operating area of ​​the railway freight yard to perform near-field real-time analysis of the inspection data collected by the mobile inspection terminal to identify the physical state characteristics of target objects, and sends the physical state characteristic data to the central server through the communication network. The central server receives the data and performs business logic judgment based on the physical state characteristic data. Based on the judgment result, it triggers a hierarchical response operation through the communication network, including controlling the alarm module to perform an alarm action.

[0021] Next, a detailed explanation of the intelligent inspection system for the entire railway freight yard process will be provided. (Reference) Figure 1 In this embodiment, the intelligent inspection system for the entire railway freight yard process includes a quadruped robot, an edge computing box, an automatic charging pile, and network equipment. Specifically, the mobile inspection terminal is a quadruped robot, the edge computing node is an edge computing box, and the central server includes a video streaming server, an AI server, and an application server deployed in the central computer room of the freight yard, as well as external systems such as the application server and the railway freight management platform within the railway bureau's dedicated network. The network equipment connects to the aforementioned servers, using Ethernet (wired and wireless) to connect cameras, the quadruped robot, drones, edge computing boxes, sensors, servers, and control terminals, enabling command transmission, data acquisition, and status data transmission. An open egress zone in the Ethernet connects to the railway bureau's dedicated network (i.e., the railway bureau's dedicated network) to transmit task instructions, receive inspection records and report data, and achieve secure exchange of wired data with the railway bureau's dedicated network.

[0022] For example, the system's business covers container handling lines, warehouse canopy handling lines, bulk loading lines, and park inspections. Functions include pre- and post-loading / unloading inspections, park inspections, loading / unloading process supervision, and loading / unloading equipment status checks. It is applicable to various vehicle types, including boxcars, open wagons, and flatcars. The accuracy rate for verification and analysis-related detections is 100% (excluding image recognition), the accuracy rate for protection signal and protection measure detections is no less than 95%, the accuracy rate for vehicle and vehicle status detection is no less than 92%, the accuracy rate for cargo status and key safety points in the freight yard is no less than 90%, and the accuracy rate for other detection items such as process supervision and equipment status checks is no less than 80%. Human-machine interaction terminals include handheld devices, PCs, large screens, and railway bureau PCs, accessed via a browser. Users can issue commands, view processes, check system status, and view inspection results and reports. Web interactive pages are built using HTML5 and related technologies, and data is transmitted via HTTP and WebSocket communication protocols. The terminal application includes all business functions such as task generation, device management, task scheduling, task instruction issuance, acquisition control, detection control, detection execution, inspection record generation, inspection report generation, and safety reminders. Specifically, it can provide service interfaces to the interactive terminal in the form of HTTP API, and use Java technology, microservice technology, distributed technology, MQ technology, etc. to build business service programs. It communicates with the interactive terminal through HTTP and WebSocket protocols, and transmits data with the image platform and execution subsystem through MQ (RocketMQ) and MQTT protocols.

[0023] The quadruped robot, serving as the terminal execution platform, is equipped with data acquisition and alarm modules, specifically including a high-definition pan-tilt acquisition device, an audible and visual alarm device, and a radar sensor. The quadruped robot is configured to receive inspection commands, perform autonomous navigation in complex terrain, dynamic recognition of train ends, full-point image data acquisition, local data caching, and preliminary edge computing inference. It can adapt to complex terrains such as freight yard ramps, stairs, track crossings, and gravel roads, and supports offline operation without network access and self-recovery from faults. Specifically, the quadruped robot has an execution subsystem that receives intermediate commands from the main business system, converts these commands into commands for the quadruped robot itself, manages and distributes command execution, and returns the collected results. It acts as an intermediary bridge between the main business system and the quadruped robot system. The execution subsystem service program is built using Java and Python technologies. It sends and replies to intermediate task commands with the main business system via the MQTT protocol, sends commands and receives results from the quadruped robot's navigation software via the MQTT protocol, and sends the collected results to the image algorithm platform of the edge computing box via the MQTT protocol.

[0024] The quadruped robot and related equipment must meet the requirements for operation in dusty environments and in light to moderate rain and foggy weather, with a protection rating of IP56~IP67; they must also meet the requirements for operation in direct sunlight and high temperatures, with an operating temperature range of -10°C to 50°C; and they must be usable at altitudes up to 1000 meters. The system can issue alarms when abnormal situations are detected, and the quadruped robot has a certain degree of self-recovery capability when trapped or tipped over; if a network interruption occurs during task execution, it can issue an alarm and temporarily switch to offline mode to complete the inspection task, automatically uploading the collected inspection data after the network is restored. The system supports primary / backup deployment, and the backup system can automatically switch over and take over when the primary system fails.

[0025] In this embodiment, the edge computing node is deployed at at least one of the container handling line, the warehouse canopy handling line, and the bulk loading line, and runs an image algorithm platform to perform target recognition on the image data collected by the data acquisition module, so as to output at least one physical state feature indicating the presence, absence, opening, damage, or displacement of the target object; the edge computing node also integrates a sound column, a loudspeaker, and an alarm for local sound and light alarm.

[0026] Specifically, there are three edge computing boxes, located on the container handling line, the warehouse canopy handling line, and the bulk loading line, respectively. Each edge computing box is equipped with a speaker, loudspeaker, and alarm. The edge computing boxes deploy an image algorithm platform to perform objective feature recognition of vehicle status, cargo stacking, personnel behavior, and safety protection measures, outputting basic status data such as presence / absence, open / closed, damage, and offset. For example, running an image algorithm platform based on frameworks such as YOLOv5 / v8 and PyTorch, it performs target recognition on the acquired images to derive objective statuses such as presence / absence, height, and open / closed.

[0027] The server runs the main business system, which performs policy calculations based on the recognition results returned by the edge computing boxes to determine whether there are violations or anomalies, and triggers tiered early warnings, on-site audio-visual alerts, background scheduling alarms, and task anomaly tracing. It supports real-time monitoring of equipment status, full-process control of tasks, and real-time recording of abnormal events to enable timely risk handling. Simultaneously, it aggregates all historical inspection data, task records, and risk data to complete big data statistical analysis, generate standardized inspection reports, and output operational optimization suggestions. The freight yard dispatch center is connected to the server and is equipped with business computers, a monitoring screen, and alert devices.

[0028] Specifically, quadruped robots, automatic charging piles, and charging rooms are deployed on container handling lines, warehouse canopy handling lines, and park inspection areas. Edge computing boxes, servers, and industrial Ethernet equipment are deployed on-site to build a local area network. Robot network bridging equipment connects the equipment's local area network with the freight yard's business local area network, and data interaction with the railway freight management platform is achieved through the railway's dedicated network safety exit. All outdoor equipment meets the requirements for adaptability to high temperature, dust, rain, and snow environments, and its protection level meets the standards. Charging piles are deployed in dedicated charging rooms to enable quadruped robots to automatically charge and standby.

[0029] The main business system, database, message queue, and file storage service are deployed on the on-site server; the image algorithm platform is deployed on the edge computing box to realize real-time inference at the edge; and the execution subsystem and customized navigation software are deployed on the quadruped robot body.

[0030] In this embodiment, the communication network uses a combination of wired and wireless industrial Ethernet to build a field local area network (LAN). This field LAN exchanges wired data with the railway bureau's freight management platform through the railway private network's security exit to transmit task instructions and receive inspection records and reports. Furthermore, at least one of the following security mechanisms is implemented: a device access security mechanism, which assigns authorization codes and capability lists to key devices accessing the system and verifies these codes and lists during device requests, instruction issuance, and result reception; a device access security mechanism, which deploys a security verification program on the quadruped robot or edge computing box to verify whether the external system is a designated legitimate system when the device accesses it; if not, it refuses to execute any instructions; a data security mechanism, which encrypts the transmission and storage of sensitive data and performs periodic hardware-level and software-level backups of important data; and an access security mechanism, which verifies the identity and permissions of all access requests, including personnel login status and permission verification, and interface authorization code and permission verification.

[0031] For example, the human-computer interaction terminal includes a PC, a handheld device, and a large-screen terminal. Remote task management, status viewing, and report retrieval are achieved by accessing the service through a browser. The system as a whole adopts a microservice architecture, supports Docker containerized deployment, and is compatible with mainstream operating systems such as Kylin and Windows Server.

[0032] The network equipment utilizes industrial Ethernet with both wired and wireless networking. Communication between devices is achieved through standard protocols such as TCP / IP, MQTT, and HTTP. The on-site local area network is independently isolated, exchanging data only with the railway's dedicated network via an authorized exit, ensuring data transmission security and meeting railway network security standards. For example, an authorized access mechanism is designed, requiring verification of identity and permissions for all access requests, including personnel and interface access. Personnel access verifies login status and permissions, while interface access verifies interface authorization codes and permissions. Device access security: All devices connected to the system are assigned authorization codes and capability lists. When a device requests system access, issues commands, or receives results, the authorization code is verified. All commands and results are verified to ensure they are within the capability list. Furthermore, secure device access: Quadruped robots, edge computing boxes, etc., deploy security verification programs through execution subsystems and algorithm platforms. When a device connects to any system, the security verification program verifies whether the connected system is a designated legitimate system. If not, the security verification program will refuse to execute any commands. Data security specifically includes: encrypting and storing sensitive data, such as user passwords, before transmission and after transmission. Important system data will be backed up regularly (daily, weekly, and monthly), including both hardware-level and software-level backups.

[0033] Based on the aforementioned intelligent inspection system for the entire railway freight yard process, the intelligent inspection method for the entire railway freight yard process includes the following steps:

[0034] Step 1: System initialization and environment adaptation.

[0035] After starting up, the quadruped robot uses the SLAM algorithm in its onboard navigation software to construct a 3D point cloud map of the cargo yard's operating area, marking the operating area, track position, and charging station location on the map. Simultaneously, the quadruped robot performs a self-check, detecting its own battery level, network connection, gimbal, sensors, and edge computing device status. Once the checks are successful, it enters standby mode. This step provides the operating environment for all subsequent inspection tasks.

[0036] Next, refer to Figure 3 Understand the following steps.

[0037] Step 2, the inspection task is automatically generated, that is Figure 3 The inspection task phase in the middle;

[0038] The system continuously monitors the railway freight management platform interface to receive task data, including loading and unloading operation instructions, track numbers, train models, number of carriages, and operation type information, and generates loading and unloading operation instructions. The central server dynamically generates an inspection task list based on the task type, matching preset inspection item rules, and carriage model parameters. Specifically, it continuously monitors the railway freight management platform interface to receive the aforementioned task data. If the platform does not issue real-time instructions, the server automatically generates park inspection tasks according to a pre-set schedule, or generates tasks by having operators manually enter task information. The server then automatically generates a complete inspection task list based on the task type, matching preset inspection item rules, and carriage model parameters. This list includes multiple items such as vehicle inspection, site verification, and safety protection checks.

[0039] During the generation of inspection tasks, the system reads train information from the loading and unloading task management and dynamically expands the list of inspection items based on whether each inspection item applies to an object or a task: for items defined as applying to an object, an independent item record is generated for each carriage; for items defined as applying to a task, only one record is generated for the entire task. For example, if a train has 10 carriages and 10 inspection items, and 4 of these items apply to objects and 6 apply to tasks, the final total number of generated inspection items will be: Once a task is generated, it is managed, including viewing, editing, and tracking its status.

[0040] Step 3: Assign inspection tasks;

[0041] According to a preset human-machine collaboration strategy, the server assigns items in the inspection task list that can be automatically executed by the equipment to the quadruped robot, and reserves the remaining items as entry points for manual processing. Specifically, items that can be automatically executed by the equipment are assigned to the quadruped robot for automatic task execution; the remaining special items that the quadruped robot cannot cover are reserved as entry points for manual processing. Thus, the inspection task is divided into parallel branches of manual and automatic execution tasks.

[0042] Step 4, automatically perform task splitting and collaborative management, i.e. Figure 3The automatic task execution phase involves: retrieving the device capability list for matching based on the type of each point in the automatic task, and dividing the automatic task into one or more acquisition subtasks, detection subtasks, and execution subtasks according to the automatic splitting rules; prioritizing the execution of each subtask and configuring the execution coordination relationship between each subtask; wherein the acquisition subtask is assigned to the acquisition task phase for processing, the detection subtask is assigned to the detection task phase for processing, and the execution subtask is assigned to the execution task phase for processing.

[0043] Specifically, after receiving an assigned automated task, the quadruped robot manages the task and initiates task splitting. Based on the type of each point in the automated task—whether it's a detection or execution point—it retrieves the capability list configured in the device's product information. This capability list defines the types of points the device can execute. By matching points with the capability list and performing calculations based on automated splitting rules, a complete automated task is ultimately split into one or more acquisition subtasks, detection subtasks, and execution subtasks. Simultaneously, one or more device types are determined for each subtask according to the rules. After splitting, the system performs automated task collaborative management, which involves prioritizing the execution of each subtask and configuring their collaborative execution relationships.

[0044] Among them, the acquisition subtask is assigned to the acquisition task stage for processing, the detection subtask is assigned to the detection task stage for processing, and the execution subtask is assigned to the execution task stage for processing.

[0045] Step 5: Train end location and inspection point dynamic generation, refer to... Figure 3 The process involves two stages: the data collection phase and the quadruped robot phase. After receiving instructions, the quadruped robot uses its onboard vision algorithm to identify the features of the train's end and, combined with navigation and positioning data, locks the coordinates of the train's end. Based on the train's end coordinates, track alignment, number of carriages, and vehicle size parameters, it automatically generates inspection point coordinates covering all inspection items of the entire train using a point offset algorithm, thus forming an inspection path.

[0046] In detail, based on the foregoing, during the data collection phase, the received data collection tasks are first managed and scheduled.

[0047] To initiate field data collection, the received collection tasks are first managed and scheduled. To begin field data collection, the system performs a train end-finding operation: During the data collection phase, a finding instruction is generated and sent to the quadruped robot via an intermediate instruction set. Upon receiving the finding instruction, the quadruped robot manages and converts it into executable commands. The quadruped robot then begins inspection, using its onboard vision algorithm to identify the end features of boxcars, open wagons, and flatcars along the travel path in real time. Combined with SLAM navigation and positioning data, it accurately locates the train end coordinates, converts the coordinates, and forwards them back to the data collection phase. After obtaining the train end coordinates, the data collection phase generates the train data collection target point.

[0048] After obtaining the train end location results during the data acquisition phase, the system performs a train data acquisition target point generation operation. Based on the current track alignment, number of carriages, and vehicle size parameters, the system automatically generates inspection point coordinates that cover all inspection items of the entire train using a point offset algorithm, forming a complete inspection path.

[0049] Specifically, for straight tracks, the coordinates of each inspection point are calculated at equal intervals along the track direction, using the train end coordinates as a reference and combining the vehicle coupler offset and car length. For curved tracks, the system first measures the curvature of the on-site work line to determine key points and their angles. Then, based on the end coordinates, offset, and key point angles, coordinate transformation is performed to generate inspection points that fit the curved track. Through multiple rounds of on-site parameter optimization, the error of the generated points is ensured to be less than 5cm, meeting the data acquisition requirements.

[0050] After the points are generated, the system maps the target points collected by the train to the previously generated inspection points and centrally manages all the collection points.

[0051] Step 6, data collection task execution and multi-robot collaboration, continuing to combine... Figure 3 The data collection phase and the quadruped robot phase are understood in the context of the inspection path, whereby the quadruped robot autonomously plans its walking route based on the inspection path.

[0052] Specifically, after the site preparation is completed, the data collection phase is responsible for managing the quadruped robot's data collection tasks; in other embodiments, if the task involves a drone, the drone data collection task management is also carried out simultaneously.

[0053] During the data acquisition phase, acquisition commands are sent to the quadruped robot via an intermediate command set. Upon receiving the commands, the quadruped robot performs command reception, queuing and scheduling, command conversion, and finally issues commands to drive the robot itself. Based on the received inspection path and locations, the quadruped robot autonomously plans the optimal walking route, adaptively completing actions such as crossing tracks, going up and down slopes, overcoming obstacles, and avoiding obstacles, accurately reaching each target location. Upon arrival, the quadruped robot uses its own gimbal to capture images and video data from multiple angles and directions, including the vehicle's exterior, the condition of the cargo compartment, cargo stacking, and site equipment.

[0054] During the data collection process, the quadruped robot reports the status of the collection task to the collection task stage in real time, and the collection task stage updates the collection task accordingly. If a network interruption occurs during the inspection, the quadruped robot automatically switches to offline mode, caches the collected data locally in the edge computing device, and automatically uploads it in batches after the network is restored.

[0055] Throughout the entire interaction process described above, the main business system on the server side and the quadruped robot communicate decoupled through a predefined set of intermediate instructions. Instructions issued by the main business system, such as "target navigation," "adjust gimbal angle," and "take a picture," are all intermediate instructions. The execution subsystem on the quadruped robot receives these instructions, translates them into proprietary control instructions for the robot itself, drives the robot's actions, and encapsulates the execution results and collected data in the intermediate instruction format before returning them. This design ensures that the upper-layer business logic is completely independent of the specific robot's proprietary protocol. When a robot model needs to be changed or added, only the intermediate instruction set needs to be adapted to the proprietary instructions within the execution subsystem of that model; no modification to the main business system is required.

[0056] After completing data collection, the quadruped robot receives and forwards the collected data, sending the final data to the data collection task phase. Simultaneously, the quadruped robot continuously collects its own status data to monitor the equipment's operational status. The data collection task phase manages the received data and performs a data collection result subscription operation, proactively pushing the data to the detection task phase.

[0057] Step 7, execution of the detection task based on the two-layer architecture, i.e. Figure 3 The detection task phase employs a two-layer architecture that integrates image algorithm recognition and strategy calculation for intelligent analysis. This includes: an image algorithm platform deployed on the edge computing node performs image recognition, calling an image algorithm model to perform fusion analysis on multiple frames of local images of the same inspection object taken by the quadruped robot at multiple points; and the business strategy module on the central server performs strategy calculation, combining railway freight operation regulations, calling a preset strategy calculation algorithm, and performing multi-dimensional fusion analysis and logical judgment on multiple image recognition results corresponding to a single point.

[0058] Specifically, after receiving the decomposed detection tasks, the detection task phase first performs detection task management, and then starts image algorithm start / stop management and strategy calculation start / stop management respectively, and prepares computing resources.

[0059] After receiving the subscription data from the acquisition task stage, the detection task stage uses a two-layer architecture that integrates image algorithm recognition and strategy calculation for intelligent analysis.

[0060] The first layer involves image recognition performed by an image algorithm platform deployed on an edge computing box. The system calls specific image algorithm models to fuse and analyze multiple frames of local images of the same inspection object taken by a quadruped robot at multiple points. This identifies objective conditions such as vehicle damage, cargo residue, lack of protective measures, excessive stacking, and violations of regulations, outputting objective status values ​​such as "present / absent," "high / low," and "open / closed." After image recognition is complete, the results are managed and, as needed, subscribed to by other modules.

[0061] The second layer involves the server-side business strategy module, which performs strategy calculations. This module, in conjunction with specific regulations for railway freight operations, calls a pre-defined strategy calculation algorithm to perform multi-dimensional fusion analysis and logical judgment on multiple image recognition results corresponding to a single item. Ultimately, it determines whether the item is in violation or abnormal, filters out abnormal risk items, and marks their risk level and the specific location of the abnormality on the carriage. For example, if an inspection item consists of multiple images, the strategy calculation algorithm will fuse the recognition results of these images and generate a final inspection result based on the business strategy. By separating the two layers, the stringent requirements for image acquisition integrity and image algorithm output accuracy are reduced. Furthermore, by directly linking the strategy calculation algorithm to operational regulations, the system significantly improves the adaptability of detection and analysis to different regulations. After completing the strategy calculation, the system manages the results. Finally, the detection task phase summarizes the detection task status and results and reports them to the automatic execution task phase.

[0062] Step 8, Risk Warning and On-site Handling, continue to refer to the Task Execution phase for further understanding. When an abnormal risk is detected, a tiered warning execution task is triggered, including controlling the quadruped robot to issue audible and visual alarm signals and pushing alarm information to the dispatch terminal.

[0063] After receiving the extracted tasks during the task execution phase, the execution results are managed. When the edge computing box detects an anomaly risk in step 7, it immediately triggers a risk warning execution task. Specifically, this includes controlling the quadruped robot on-site to issue an audible and visual warning signal, and simultaneously pushing alarm information to the scheduling terminal and the field handheld terminal. The information clearly displays the location of the anomaly, the type of anomaly, and the corresponding on-site image evidence to remind and assist the operators in timely rectification and handling. After the handling is completed, the task execution phase reports the status and results of the task execution to the automatic task execution phase.

[0064] Step 9: Automatic task status summary and feedback, for reference. Figure 3 The automatic task execution phase;

[0065] During the automatic task execution phase, it continuously receives data on the status and results of data acquisition tasks and equipment status from the data acquisition task phase, the status and results of detection tasks from the detection task phase, and the status and results of execution tasks from the execution task phase. By integrating the execution status of all these sub-tasks, the automatic task execution phase generates the overall status and results of the automatic task and reports it to the inspection task phase.

[0066] Step 10, Task Closure and Data Archiving, refer to Figure 3 During the inspection task phase; after the inspection task is completed, the central server automatically summarizes the detection results, anomaly records and image evidence, generates a standardized inspection report, and archives the task data synchronously.

[0067] During the inspection task phase, after receiving the automatic task status and result reports from the automated task execution phase, the system first performs reminder management, sending task completion or anomaly reminders to relevant personnel according to preset rules. Subsequently, the server automatically aggregates the inspection results, anomaly records, and image / video evidence for all inspection items, generating a standardized inspection report. The system synchronously transmits the task completion status, all inspection data, and report file of this inspection task back to the railway freight management platform on the server, completing the closed loop of the entire business process. Simultaneously, data backup and archiving are performed on the local server, and all historical inspection data, task records, and risk data are aggregated for statistical analysis, ultimately outputting optimization suggestions for on-site operations.

[0068] Step 11: Device reset and standby;

[0069] After the quadruped robot completes all assigned inspection tasks, it autonomously plans a return path, walks to the charging station and automatically docks for charging, resets the status of each component of the equipment, and re-enters standby mode to await the next task instruction.

[0070] It is worth noting that in the entire process described above, the manual task execution stage in step 3 involves operators receiving task information through handheld terminals or other devices and then performing the corresponding inspection items offline. This branch is carried out independently until its completion.

[0071] Through the above methods, this invention realizes intelligent closed-loop management of the entire process of railway freight yard inspection tasks, from generation, assignment, execution, detection, early warning to archiving, which significantly improves the coverage, accuracy and automation level of inspection operations.

[0072] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A full-process intelligent inspection system for railway freight yards, comprising a communication network, and mobile inspection terminals, edge computing nodes, and a central server interconnected through the communication network, characterized in that: The mobile inspection terminal includes a quadruped robot for performing inspection tasks within the railway freight yard. The quadruped robot is equipped with a data acquisition module and an alarm module. The edge computing node is deployed near the working area of ​​the railway freight yard to perform near-field real-time analysis on the inspection data collected by the mobile inspection terminal to identify the physical state characteristics of the target object, and send the physical state characteristic data to the central server through the communication network. The central server receives the data and performs business logic judgments based on the physical state characteristic data. Based on the judgment results, it triggers hierarchical response operations, including the control alarm module executing alarm actions, through the communication network.

2. The intelligent inspection system for the entire railway freight yard process according to claim 1, characterized in that, The edge computing node is deployed at least one of the container handling line, warehouse canopy handling line, and bulk loading handling line, and runs an image algorithm platform to perform target recognition on the image data collected by the data acquisition module, so as to output at least one physical state feature indicating the presence, absence, opening, damage, or displacement of the target object. The edge computing node also integrates a sound column, a loudspeaker, and an alarm for local sound and light alarms.

3. The intelligent inspection system for the entire railway freight yard process according to claim 1, characterized in that, The communication network is constructed using a combination of wired and wireless industrial Ethernet to form a field local area network. The field local area network exchanges wired data with the railway bureau's railway freight control platform through the safety exit of the railway private network to transmit task instructions and receive inspection records and reports. And set up at least one of the following security mechanisms: The device access security mechanism assigns authorization codes and capability lists to each device accessing the system, and verifies the authorization codes and capability lists when devices request, issue commands, and receive results. The device is connected to a security mechanism, and a security verification program is deployed on the quadruped robot or edge computing box. When the device connects to an external system, the security verification program verifies whether the external system is a designated legitimate system. If not, it refuses to execute any instructions. Data security mechanisms encrypt the transmission and storage of sensitive data, and perform regular hardware-level and software-level backups of important data. The access security mechanism verifies the identity and permissions of all access requests, including verifying the login status and permissions of personnel and the authorization code and permissions of interfaces.

4. A method for intelligent inspection of the entire process of a railway freight yard, implemented based on the intelligent inspection system for the entire process of a railway freight yard as described in any one of claims 1 to 3, characterized in that, include: The central server obtains loading and unloading operation task instructions containing information such as track number, train type, number of carriages and operation type. The central server matches preset inspection item rules according to task type, dynamically generates an inspection task list in combination with carriage model parameters, and assigns the items in the inspection task list that are automatically executed by the equipment to the quadruped robot, while reserving the remaining items for manual processing. After receiving the instruction, the quadruped robot uses a vision algorithm to identify the features of the train end and combines it with navigation and positioning data to lock the coordinates of the train end. Based on the train end coordinates, track type, number of carriages and vehicle size parameters, the robot automatically generates inspection point coordinates covering all inspection items of the entire train through a point offset algorithm, forming an inspection path. The quadruped robot autonomously plans its walking route based on the inspection path; When an abnormal risk is detected, a tiered early warning execution task is triggered, including controlling the quadruped robot to issue an audible and visual alarm signal and pushing alarm information to the scheduling terminal. After the inspection task is completed, the central server automatically summarizes the detection results, abnormal records and image evidence, generates a standardized inspection report, and archives the task data synchronously.

5. The intelligent inspection method for the entire process of railway freight yards according to claim 4, characterized in that, Before the central server receives the loading and unloading task instructions, the system also includes system initialization and environment adaptation steps: After the quadruped robot is started, it uses the SLAM algorithm in the navigation software mounted on the quadruped robot to construct a three-dimensional point cloud map of the cargo yard operation area and marks the operation area, track position and charging pile position on the map. Meanwhile, the quadruped robot performs a self-test, checking its own battery level, network connection, gimbal, sensors, and edge computing device status. Once the test is successful, it enters a standby mode.

6. The intelligent inspection method for the entire process of railway freight yards according to claim 4, characterized in that, The process by which the central server dynamically generates the inspection task list includes: Train information is read from the loading and unloading task management system, and the list of inspection items is dynamically expanded according to whether each inspection item applies to an object or a task. For items defined as applying to an object, an independent item record is generated for each carriage. For items defined as applying to a task, only one record is generated for the entire task.

7. The intelligent inspection method for the entire process of railway freight yards according to claim 4, characterized in that, After the items in the inspection task list that are automatically executed by the equipment are assigned to the quadruped robot, the following further includes: The device capability list is retrieved and matched according to the type of each point in the automatic task, and the automatic task is divided into one or more acquisition subtasks, detection subtasks and execution subtasks in combination with the automatic splitting rules; the execution priority of each subtask is sorted and the execution coordination relationship between each subtask is configured. The acquisition subtask is assigned to the acquisition task stage for processing, the detection subtask is assigned to the detection task stage for processing, and the execution subtask is assigned to the execution task stage for processing.

8. The intelligent inspection method for the entire process of railway freight yards according to claim 7, characterized in that, In the detection task phase, a two-layer architecture integrating image algorithm recognition and policy computation is adopted for intelligent analysis, including: The image algorithm platform deployed on the edge computing node performs image recognition and calls the image algorithm model to perform fusion analysis on multiple frames of local images taken by the quadruped robot at multiple points in the same inspection object. The business strategy module on the central server performs strategy calculations, combines railway freight operation regulations, and calls preset strategy calculation algorithms to perform multi-dimensional fusion analysis and logical judgment on multiple image recognition results corresponding to a single point.

9. The intelligent inspection method for the entire process of railway freight yards according to claim 4, characterized in that, The automatic generation of inspection point coordinates covering all inspection items of the entire train through the point offset algorithm includes: For straight tracks, the map coordinates of each inspection point are calculated at equal intervals along the track direction, based on the coordinates of the train end and the vehicle coupler offset and the length of the carriage. For curved tracks, the curvature of the on-site work line is measured to determine key points and their angles. Based on the coordinates of the train end, the offset, and the angles of the key points, coordinate transformation is performed to generate inspection points that conform to the curved track.

10. The intelligent inspection method for the entire process of railway freight yards according to claim 4, characterized in that, The process of the quadruped robot autonomously planning its walking route according to the inspection path also includes: If a network interruption occurs during the inspection, the quadruped robot automatically switches to offline mode, caches the collected data locally, and automatically uploads it in batches after the network is restored. Furthermore, the main business system located on the server side communicates with the quadruped robot through a predefined set of intermediate instructions. The main business system issues intermediate instructions, and the execution subsystem on the quadruped robot receives them and translates them into proprietary control instructions for the robot body, driving the robot to perform actions. The execution results and collected data are then encapsulated in the intermediate instruction format and returned.

Images