Car coupler automatic unhooking and rehooking method based on artificial intelligence

Abstract

The invention relates to the technical field of railway related devices, in particular to an automatic coupler unhooking and rehooking method based on artificial intelligence, and the method comprises the following steps: a coupler unhooking and rehooking robot performs wireless magnetic attraction charging in a charging station, is in a standby state, performs self-inspection once every 2 hours in the charging standby state, and performs self-inspection once every 2 hours; a self-inspection result is transmitted to the centralized control center, the robot is started, the multi-mode sensing module scans the surrounding environment of the robot and detects obstacles in the surrounding environment of the robot and a travel path, and then the robot moves to an operation position. According to the scheme provided by the invention, the unhooking and rehooking robot can realize a composite detection effect in different weather and environment states by utilizing multiple detection means in a multi-mode detection mode, so that the accuracy of environment and coupler detection by the robot is improved, and meanwhile, the efficiency of the unhooking and rehooking process is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of railway related devices, in particular to a car coupler automatic uncoupling and recoupling method based on artificial intelligence. BACKGROUND

[0002] Uncoupling and recoupling are two basic operations in shunting operations: uncoupling and recoupling. The car coupler is a coupler at the two ends of a train car or a locomotive, mainly used to realize the connection between the locomotive and the vehicle or the vehicle and the vehicle, to transmit traction and impact force, and to maintain the distance between the vehicles. The car coupler has the functions of connection, traction and buffering. When two carriages need to be separated, the operator will lift the lifting rod of one car coupler to fully turn the tongue outward, then use the locomotive to pull the carriages apart, and the two car couplers can be separated. This process is called uncoupling. When two carriages need to be connected together, the operator first lifts the lifting rod of one car coupler to fully turn the tongue outward, then collides the two car couplers with each other to make the tongue turn into the hook head of the connected car coupler, and the two car couplers are connected together. This process is called recoupling or coupling. The current car coupler automatic uncoupling and recoupling method mainly relies on traditional mechanical control and simple sensor detection. In actual application, there are many deficiencies. The existing system has poor adaptability to complex environments (such as severe weather, night work, etc.). The detection accuracy of the sensor is greatly reduced under weather conditions such as rain, snow and fog. For example, in heavy snow weather, snow accumulates on the car coupler. At this time, visual recognition of the car coupler will have deviations, which reduces the stability of the subsequent uncoupling and recoupling process and affects the stability and efficiency of the automatic uncoupling and recoupling of the car coupler. SUMMARY

[0003] To overcome the problems in the related art, the present application provides a car coupler automatic uncoupling and recoupling method based on artificial intelligence.

[0004] To achieve the above-mentioned purpose, the first aspect of the present application provides a car coupler automatic uncoupling and recoupling method based on artificial intelligence. The specific method of the car coupler automatic uncoupling and recoupling includes the following steps: A1, the uncoupling and recoupling robot is wirelessly magnetically charged at the charging station and is in standby state. The charging standby state is self-checked every 2h, and the self-checking result is transmitted to the centralized control center; A2, the robot starts, the multi-modal perception module scans the environment around the robot, detects the obstacles in the environment around the robot and the route, and then the robot moves to the working position; A3, after the robot is at the working position, the multi-spectral camera starts to shoot the car coupler image, identifies the type of the car coupler, and comprehensively analyzes the car coupler. At the same time, the distance between the robot and the car coupler is measured by combining the multi-modal perception module, and the spatial coordinate position of the car coupler is obtained; A4. According to the identified coupling type and state, the robot arm moves to the starting position of uncoupling according to the preset uncoupling mode and coupling coordinate position, and adjusts the starting position, uncoupling path and arm thrust according to the real-time detected coupling state, then moves according to the uncoupling path to lift the coupling to the unlocked position, completing automatic uncoupling; A5. When the coupling needs to be recoupled, the robot moves to the recoupling operation position, scans and judges the recoupled coupling again, then moves to the starting position of recoupling according to the coupling type, and adjusts the starting position, uncoupling path and arm thrust according to the actual detected state of the recoupled coupling, then performs recoupling operation according to the coupling type; A6. After each operation of the uncoupling and recoupling robot, the data generated in this uncoupling and recoupling process is uploaded to the control center, the path planning parameters are optimized through deep learning, and the uncoupling and recoupling data is updated to the control chip of the robot.

[0005] Preferably, if the self-checking in A1 is normal, the subsequent task can be executed, if there is an abnormality in the self-checking, the corresponding abnormal alarm information will be sent to the control platform, and the task will be continued after the abnormality is repaired; The A1 self-checking items include: The power supply voltage, current, power and remaining power of the charging; Self-checking of the robot arm, multi-modal perception module, communication link and environment state.

[0006] Preferably, the multi-modal perception module includes: Laser radar, infrared thermal imager, millimeter wave radar and mechanical arm end pressure sensor; When the robot in A2 moves to the operation position, it generates an environment three-dimensional map based on the laser radar, and combines the millimeter wave radar to monitor the moving path in real time, if an obstacle is detected, it will bypass to generate a new path, and if there is no obstacle, it will move to the coupling connection operation position of the car body according to the preset path; When it is a rainy and snowy low temperature weather, the A3 multi-spectral camera takes an image, at this time the infrared thermal imager is used to identify the coupling profile by infrared thermal imaging, and to compensate the precision loss caused by low temperature and snow cover of the multi-spectral camera; When the mechanical arm contacts the coupling handle, the end pressure sensor contacts the coupling at this time, and real-time feedback of the pressure value of the mechanical arm end is realized, avoiding excessive extrusion, and adjusting the real-time application thrust of the mechanical arm according to the real-time pressure value.

[0007] Preferably, the mechanical arm is a six-joint mechanical arm, mainly composed of a motor drive module, a hydraulic drive module and a pneumatic drive module; The motor drive module is specifically a servo motor, which is used for steering in the joint direction of the mechanical arm. The hydraulic drive module is specifically a hydraulic push rod, which is used for providing driving force for the hook at the end of the mechanical arm. The pneumatic drive module is specifically a pneumatic push rod, which is used for providing auxiliary adjustment driving force for the hook at the end of the mechanical arm.

[0008] Preferably, the specific process of identifying the type of car hook by using the multi-spectral camera is as follows: A3.1, the multi-spectral camera synchronously shoots the visible light and near-infrared images of the car hook, both with a resolution of 3840x2160, and then superimposes and fuses the two images to enhance the outline of the car hook in the image; A3.2, downsample the fused high-resolution image to 640x640 pixels, use bilinear interpolation to retain key details, and normalize the pixel values to meet the input requirements of the recognition model; A3.3, input the preprocessed image into the recognition model, which confirms the car hook bounding box coordinates, confidence, and car hook type probability through car hook feature extraction and car hook multi-scale feature fusion; A3.4, filter the multiple prediction boxes output by the model, eliminate redundant boxes with too high overlap with the highest confidence box, and retain detection results with a confidence of >0.8, select the result with the highest confidence from the remaining prediction boxes, and determine the type of car hook according to the class probability.

[0009] Preferably, the specific calculation formula of the multi-spectral image fusion algorithm is as follows: ; Wherein, is visible light, the weight is 0.7, is near-infrared light, the weight is 0.3, is the fused superimposed image.

[0010] Preferably, the specific steps of filtering the multiple prediction boxes output by the model are as follows: A3.4.1, sort all prediction boxes by confidence from high to low, select the highest confidence box, and calculate the ; A3.4.2, The calculation formula is as follows: ; Wherein, is the overlapping area of the prediction box and the real box, is the total area of the combined prediction box and real box; A3.4.3, eliminate the box whose exceeds the threshold value, and repeat the steps until there is no remaining box.

[0011] Preferably, the robot is installed with a wireless communication module, and the data transmission delay of the wireless communication module is less than 50 ms; the robot self-checking and operation data uploading are realized through the wireless communication module.

[0012] Preferably, the method of deep learning in A6 is as follows: A6.1, the multi-spectral camera shoots the RGB and NIR images of the car hook, the laser radar obtains the three-dimensional point cloud of the car hook, and the mechanical arm end pressure sensor records the data of the hook contact pressure, which collects data under different weather, light conditions and car hook position deviation; A6.2, input multi-spectral images, laser point cloud and pressure sensor time series data into the deep learning model, initialize the vision and point cloud branch on the public data set, jointly train the multi-modal network, the learning rate is 3e-4, the batch size is 16, and the optimizer is used; first train the hook operation in simple scene and weather environment, and gradually increase the difficulty of hook operation in scene and weather environment; A6.3, the model simulates learning of the robot in different environment weather states, car hook 3D pose, environment parameters, mechanical arm joint angle, mechanical arm end moving speed, rotation angle, clamping force and hook pushing force; A6.4, then design a deep learning model reward mechanism: Success reward: when the hook / unhook operation is completed, R+50; Efficiency reward: R+10 for each second of operation time reduced; Recognition reward: R+20 for each 1% increase in accuracy of identifying the type of car hook; Efficiency penalty: R-30 for each second of operation time increased; Success penalty: R-100 for failure of the hook / unhook operation; Safety penalty: R-10 for contact force exceeding the threshold.

[0013] The second aspect of the technical scheme provided in the application provides an artificial intelligence-based car hook automatic unhooking and hooking system for executing an artificial intelligence-based car hook automatic unhooking and hooking method, the car hook automatic unhooking and hooking system comprising: a robot, a charging station and a centralized control center. The robot comprises a movable six-joint mechanical arm, a tracked autonomous mobile vehicle body, a mechanical arm and an end effector, a laser radar, an infrared thermal imager, a millimeter wave radar, a multi-spectral camera and a mechanical arm end pressure sensor, and the robot is in wireless communication connection with the centralized control center.

[0014] The technical scheme provided in the application can have the following beneficial effects: The scheme uses multiple detection means to achieve composite detection in different weather and environmental conditions, thereby improving the accuracy of the robot in detecting the environment and the coupler. Especially in extreme weather and complex environments, the robot can still accurately detect the coupler position and type, ensuring the stability and accuracy of the automatic uncoupling process. Through the recognition of the coupler type, the actual coupler position, the coupler state, and the presence of abnormal wear and deformation, path planning and control are used to effectively solve the problem of coupler deformation causing changes in the uncoupling path, resulting in small position deviations and angle errors. This greatly improves the reliability and safety of the coupler uncoupling connection. Through the artificial intelligence deep learning model, it has autonomous deep learning ability and can continuously optimize the operation strategy based on a large amount of historical operation data and real-time feedback information to improve the operation efficiency and quality. With the increase of usage time, the system performance will not decrease, but will continuously improve, thereby continuously optimizing the efficiency and accuracy of the automatic uncoupling of the coupler.

[0015] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and are not limiting to the present application. BRIEF DESCRIPTION OF DRAWINGS

[0016] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which like reference characters refer to like parts throughout the figures, and in which:

[0017] Fig. 1 is a flow chart of the automatic uncoupling method of the coupler according to the embodiment of the present application; Fig. 2 is a detailed step chart of the automatic uncoupling method of the coupler according to the embodiment of the present application. DETAILED DESCRIPTION

[0018] The preferred embodiments of the present application will be described in more detail by referring to the attached drawings. Although the preferred embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0019] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in this application and the appended claims, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the term "and / or," as used herein, refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0020] It should be noted that when an element is referred to as being "fixed" or "attached" to another element, it can be directly on the other element or indirectly on the other element, with one or more intervening elements. When an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or indirectly connected or coupled to the other element, with one or more intervening elements.

[0021] In the description of the application, it should be understood that the terms "thickness", "upper", "lower", "front", "back", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings based on the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the application.

[0022] In the description of the application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connection" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium; it can be the internal communication of two elements or the interaction relationship between two elements. For those skilled in the art, the specific meaning of the above terms in this application can be understood according to the specific circumstances.

[0023] The technical solutions of the embodiments of the application are described in detail below with reference to the drawings.

[0024] Embodiments

[0025] Reference Figs. 1-2 The application provides an automatic uncoupling and recoupling method for a car coupler based on artificial intelligence. The specific method for automatic uncoupling and recoupling of the car coupler comprises the following steps: A1, the uncoupling and recoupling robot is wirelessly magnetically charged at the charging station and is in standby state. The charging standby state is self-checked every 2h, and the self-checking result is transmitted to the centralized control center; The robot stops at a dedicated charging station, which is equipped with a wireless magnetic charging device that uses electromagnetic induction principles and has automatic alignment and adsorption functions. It can quickly establish a charging connection when the robot stops. The wireless charging module has an IP67 protection level, an output power of 600W, a voltage of 48V, and a current range of 5A-50A, which is dynamically adjusted. The charging method uses voltage limiting and constant current to ensure safe and efficient charging of the battery. During the charging process, the charging display box on the robot's shell displays the charging progress percentage in real time, with a green progress bar for intuitive presentation. When the battery level is below 10%, the charging display box turns red, and the robot cannot perform any tasks, only maintaining basic charging and self-checking functions. When the battery level is between 50% and 100%, the green progress bar color gradually deepens as the battery level rises, and this interval is the optimal battery level range to ensure the robot performs tasks in the best state. In standby mode, the robot's built-in self-checking system starts every 2 hours on a fixed cycle. Self-checking covers key components such as remote control and telemetry modules, battery modules, drive modules, and communication equipment. The remote control and telemetry module detects whether the signal transceiver function is normal by sending test signals to the control center and receiving feedback to verify the stability of the communication link. The battery module monitors battery voltage, remaining capacity, charging and discharging current, internal temperature, and other parameters in real time to determine whether the battery has problems such as overcharging, overdischarging, and temperature abnormalities. The drive module detects the speed, torque, and operating state of the motor, as well as the hydraulic and pneumatic drive modules, to ensure that the motor and hydraulic and pneumatic drive modules can operate normally. The communication equipment checks wireless network connection strength, data transmission rate, and packet loss rate, and detects and calibrates the state of the multispectral camera, laser radar, infrared thermal imager, millimeter wave radar, and mechanical arm end pressure sensor. After self-checking is complete, the robot transmits detailed self-checking results to the control center via the 5G communication network. The monitoring system of the control center analyzes and processes the data, and if an anomaly is found, an alarm is immediately sent and maintenance personnel are prompted to repair.

[0026] A2, the robot starts, the multi-modal perception module scans the environment around the robot, detects obstacles in the environment around the robot and the route, and then the robot moves to the work position; Laser radar: scanning radius 30m, angle resolution 0.1°, generating a three-dimensional point cloud map; Millimeter wave radar: detection range 0.5-3m, dynamic tracking of moving obstacles; Infrared thermal imager: compensates for visual blind spots in rain, snow, and fog weather, and identifies the thermal characteristics of the coupler metal.

[0027] Fusion of multi-sensor data, prediction of obstacle position and motion trajectory of obstacles in Yunnan, real-time generation of optimal obstacle avoidance path; Repositioning error ≤±50mm, no need for two-dimensional code or reflector assistance.

[0028] Straight / retreat speed 0.5 m / s, turning radius ≥ 1.2 m; Ultrasonic radar (2 in front and back, detection angle 120°) detects obstacles within 1 m, triggers emergency stop and uploads alarm information.

[0029] When receiving the start instruction from the control center, the robot enters the working state, and the multi-modal perception module starts quickly. The laser radar can perform 20 times of 360-degree omnidirectional scanning per second to construct a high-precision three-dimensional environment map, which can accurately identify the shape, position and distance of surrounding objects with an accuracy of millimeter level. The millimeter wave radar detects the speed and position of obstacles and other moving objects in real time, with a detection range of 0-100 meters, a speed detection accuracy of ±0.1 m / s, and a distance detection accuracy of ±1 cm. Two high-performance wide-angle and angle-adjustable ultrasonic radars are equipped in front and back. Once an obstacle is detected, the robot stops immediately and issues a sound and light alarm. The infrared thermal imager compensates for the visual blind area in extreme weather such as rain, snow and fog, identifies the thermal characteristics of the metal hook, and during the journey, when an obstacle is encountered, the multi-modal perception module identifies the obstacle and then re-plans the route to bypass the obstacle. At the same time, the information of the obstacle is uploaded to the control center for subsequent staff to handle. The feedback information of the real-time multi-modal perception module is used to adjust the speed and direction of the robot to ensure that the robot can efficiently and safely reach the work position.

[0030] A3、After the robot reaches the work position, start the multispectral camera to take pictures of the hook, identify the type of hook, and conduct a comprehensive analysis of the hook. At the same time, the multi-modal perception module measures the distance between the robot and the hook to obtain the spatial coordinate position of the hook. After the robot reaches the work position, the multispectral camera starts. This camera can capture image information in multiple spectral bands such as visible light, near-infrared, short-wave infrared, etc., which can effectively identify the material, wear degree, cracks and other subtle features on the surface of the hook. The hook photographed by the multispectral camera is processed by fusion and superposition, then compared with the existing hook type, and a comprehensive analysis of the hook's state is conducted, including the wear condition of the hook, the state of the locking mechanism, the connection gap, etc.

[0031] At the same time, the laser radar and millimeter wave radar in the multi-modal perception module accurately measure the distance between the robot and the hook, combine the robot's own positioning information, and through spatial coordinate conversion method, obtain the three-dimensional spatial coordinate position of the hook with an accuracy of ±1 cm.

[0032] A4、According to the identified coupling type and state, the robot arm moves to the starting position of the uncoupling according to the preset uncoupling mode and the coupling coordinate position, and adjusts the starting position, uncoupling path and arm thrust according to the real-time detected coupling state, then moves the coupling to the unlocking position according to the uncoupling path, and completes the automatic uncoupling; The artificial intelligence robot retrieves the corresponding uncoupling mode from the preset uncoupling scheme library according to the identified coupling type and state, and the uncoupling scheme library stores multiple uncoupling schemes for different types of couplings, each scheme contains detailed operation steps, arm motion parameters and force control parameters, etc. The robot arm is driven by high-precision servo motor and has six degrees of freedom, which can move flexibly in three-dimensional space.

[0033] The arm obtains the motion angle and displacement of each joint according to the preset uncoupling mode and coupling coordinate position, drives the arm to move to the starting position of uncoupling, and continuously monitors the coupling state in real time during the movement. If a slight change in the coupling position or an abnormal situation is detected, the arm will immediately pause the movement, the robot chip will recalculate and adjust the starting position, uncoupling path and arm thrust, the arm thrust is feedback and adjusted in real time through the pressure sensor, to ensure that the coupling connection resistance can be overcome during uncoupling without damaging the coupling. After adjustment, the arm moves according to the newly planned uncoupling path to lift the coupling to the unlocking position, triggers the unlocking mechanism of the coupling, and completes the automatic uncoupling operation.

[0034] A5、When the coupling needs to be recoupled, the robot moves to the recoupling operation position, scans and judges the recoupled coupling, then moves to the starting position of recoupling according to the coupling type, and adjusts the starting position, uncoupling path and arm thrust according to the actual detection state of the recoupled coupling, then performs recoupling operation according to the coupling type; After the uncoupling operation is completed, if re-coupling operation is needed, the robot moves to the re-coupling operation position (the gap between the empty train carriages) through track navigation, and after reaching the position, the multispectral camera and multi-modal perception module are started again for comprehensive scanning and detection of the re-coupling coupler, the coupler recognition and state analysis process in step A3 is repeated to obtain accurate information of the re-coupling coupler, according to the detected coupler type, the robot retrieves the corresponding re-coupling scheme in the re-coupling scheme library, the mechanical arm moves to the starting position of re-coupling according to the re-coupling scheme and the coupler coordinate position, and adjusts the starting position, re-coupling path and mechanical arm thrust according to the actual detection state of the re-coupling coupler, such as the inclination angle of the coupler and the flatness of the connecting end face, etc. After the adjustment is completed, the mechanical arm slowly moves according to the planned re-coupling path, accurately docks the coupler and completes the connection, and confirms the success of the re-coupling operation by detecting the locking state and connection gap of the coupler.

[0035] A6、After each operation of the uncoupling and re-coupling robot, the data generated in this uncoupling and re-coupling process is uploaded to the control center, the path planning parameters are optimized through reinforcement deep learning, and the uncoupling and re-coupling data is updated to the control chip of the robot.

[0036] After each uncoupling and re-coupling operation is completed, the robot uploads all data generated in this operation process, including environmental data collected by the multi-modal perception module, coupler recognition and state analysis data, mechanical arm motion parameters, force control data, and operation time, operation results and other information, to the control center through wireless 5G communication.

[0037] The big data analysis platform of the control center uses reinforcement deep learning algorithm to analyze and process the uploaded data, compares the actual operation data with the preset ideal data, evaluates various indicators in the operation process, finds out the existing problems and optimization space, improves the preset uncoupling and re-coupling scheme, and downloads the optimized path planning parameters and uncoupling and re-coupling data to the control chip of the robot through wireless communication, so that the robot can continuously learn and adapt to different operation scenes, and improve the efficiency and accuracy of operation.

[0038] The working principle of the automatic uncoupling and recoupling method for a car hook based on artificial intelligence provided by the application is as follows: the robot is wirelessly magnetically charged at the charging station, and is self-checked every 2 hours during charging to check the voltage, current, power, remaining power of the power supply, and the states of the mechanical arm, multi-modal sensing module and other key parts; after the vehicle enters the uncoupling and recoupling area, the robot receives a task instruction; if the robot is normal, the robot waits for the task; if the robot is abnormal, the robot alarms the centralized control platform and assigns another robot of the same type to uncouple and recouple, and the abnormal robot executes the task after being repaired; then the multi-modal sensing module of the robot starts to work, the laser radar generates an environment three-dimensional map, the millimeter wave radar monitors the path, and a new route is planned when an obstacle is encountered to ensure safe arrival at the work position; in rainy, snowy and low-temperature weather, the infrared thermal imager assists the multispectral camera to ensure that the car hook contour recognition is not affected; after the robot arrives at the work position, the multispectral camera shoots a car hook image, which is input into an identification model after fusion and preprocessing to determine the type and position of the car hook; the mechanical arm adjusts the action and thrust according to the car hook information and the feedback of the end pressure sensor to complete automatic uncoupling; during recoupling, the scanning, judging and adjusting process is repeated to realize accurate connection; after the work is completed, the robot uploads the work data to the centralized control center, and then returns to the charging station for standby charging.

[0039] As to the device in the above-mentioned embodiments, the specific manner in which each module performs an operation has been described in detail in the embodiments related to the method, and will not be described in detail here.

[0040] The solutions of the present application have been described in detail above with reference to the drawings. In the above-mentioned embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments. It should also be known by those skilled in the art that the actions and modules involved in the specification are not necessarily required by the present application. In addition, it can be understood that the steps in the method embodiments of the present application can be adjusted, combined and reduced in sequence according to actual needs, and the modules in the device embodiments of the present application can be combined, divided and reduced according to actual needs.

[0041] The above has described the embodiments of the present application, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes are obvious to those skilled in the art without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles, practical applications or improvements to the technology in the market of the embodiments, or to enable other ordinary skilled persons in the art to understand the embodiments disclosed herein.

Claims

1. An artificial intelligence-based automatic coupler unhooking and rehooking method, characterized in that: The specific method of automatically unhooking and rehooking the coupler comprises the following steps: A1. The hook-removing and re-hooking robot is wirelessly magnetically charged at the charging station and is in standby mode. In the charging standby mode, it performs a self-test every 2 hours and transmits the self-test results to the centralized control center. A2. The robot starts up and the multimodal perception module scans the robot's surroundings, detecting obstacles in the robot's surroundings and path. The robot then moves to its operating position. A3. Once the robot is in the working position, it activates the multispectral camera to capture images of the coupler, identifies the coupler type, and performs a comprehensive analysis of the coupler. Simultaneously, it uses the multimodal perception module to measure the distance between the robot and the coupler to obtain the coupler's spatial coordinate position. A4. The uncoupling method is determined based on the identified coupler type and status. The robot's arm moves to the uncoupling starting position according to the preset uncoupling method and coupler coordinate position. Based on the real-time detected coupler status, the arm adjusts the starting position, uncoupling path, and arm thrust. The arm then moves along the uncoupling path to lift the coupler to the unlocked position, completing automatic uncoupling. A5. When the coupler needs to be re-hooked, the robot moves to the re-hooking operation position and scans and judges the re-hooking coupler again. Then, according to the coupler type, the robot arm moves to the re-hooking starting position and determines the starting position, unhooking path and robot arm thrust according to the actual detection status of the re-hooking coupler. Then, the re-hooking operation is carried out according to the coupler type. A6. After each operation, the hook-picking and re-hooking robot uploads the data generated during the process to the centralized control center. The path planning parameters are optimized through enhanced deep learning, and the hook-picking and re-hooking data is updated to the robot's control chip.

2. The method for automatically unhooking and rehooking a coupler based on artificial intelligence according to claim 1, characterized in that: If there is no abnormality in the self-test in A1, the subsequent tasks can be executed. If there is an abnormality in the self-test, an alarm message corresponding to the abnormality will be sent to the centralized control platform, and the task will continue to be executed after the abnormality is repaired; The A1 self-test items include: Charging power supply voltage, current, power and remaining capacity; Self-check of the robot arm, multimodal perception module, communication link and environmental status.

3. The method for automatically unhooking and rehooking a coupler based on artificial intelligence according to claim 1, characterized in that: The multimodal perception module includes: LiDAR, infrared thermal imager, millimeter-wave radar, and pressure sensor at the end of the robotic arm; When the robot A2 moves to the working position, it generates a three-dimensional map of the environment based on the laser radar and monitors the moving path in real time with the millimeter-wave radar. If an obstacle is detected, it will bypass it and generate a new path. If there is no obstacle, it will move to the car coupler connection working position according to the preset path; When the A3 multispectral camera captures images during rainy, snowy, and low-temperature weather, the infrared thermal imager uses infrared thermal imaging to identify the coupler outline, compensating for the accuracy loss caused by the low temperature and rain and snow obstruction of the multispectral camera. When the robotic arm contacts the coupler unhooking handle, the end pressure sensor contacts the coupler and provides real-time feedback of the pressure value at the end of the robotic arm to avoid excessive squeezing. At the same time, the unhooking thrust of the robotic arm is adjusted in real time according to the real-time pressure value.

4. The method for automatically unhooking and rehooking a coupler based on artificial intelligence according to claim 1, characterized in that: The robotic arm is a six-joint robotic arm, mainly composed of a motor drive module, a hydraulic drive module and a pneumatic drive module; The motor drive module is specifically a servo motor, which is used for steering the joint direction of the robotic arm; The hydraulic drive module is specifically a hydraulic push rod, which is used to provide driving force for unhooking at the end of the robotic arm; The pneumatic drive module is specifically a pneumatic push rod, which is used to provide auxiliary adjustment driving force for unhooking at the end of the robotic arm.

5. The method for automatically unhooking and rehooking a coupler based on artificial intelligence according to claim 1, characterized in that: The specific process of using the multispectral camera to identify the coupler type is as follows: A3.

1. A multispectral camera simultaneously captures visible light and near-infrared images of the coupler, both with a resolution of 3840 × 2160. The two images are then fused and superimposed to enhance the coupler outline. A3.

2. Downsample the fused high-resolution image to 640 × 640 pixels, use bilinear interpolation to preserve key details, and normalize the pixel values ​​to meet the input requirements of the recognition model. A3.

3. The preprocessed image is fed into the recognition model. The model extracts coupler features and fuses multi-scale coupler features to determine the coupler bounding box coordinates, confidence level, and coupler type probability. A3.

4. Filter the multiple prediction boxes output by the model, remove redundant boxes that overlap too much with the highest confidence box, retain the detection results with confidence > 0.8, select the result with the highest confidence from the remaining prediction boxes, and determine the coupler type based on its category probability.

6. The method for automatically unhooking and rehooking a coupler based on artificial intelligence according to claim 5, characterized in that: The specific calculation formula of the multispectral image fusion algorithm is as follows: ； in, is visible light, with a weight of 0.7, is near infrared light, with a weight of 0.3, To fuse the overlaid images.

7. The method for automatically unhooking and rehooking a coupler based on artificial intelligence according to claim 5, characterized in that: The specific steps for screening multiple prediction boxes output by the model are as follows: A3.4.

1. Sort all predicted boxes by confidence from high to low, select the box with the highest confidence, and calculate its confidence with the remaining boxes. ; A3.4.2 The calculation formula is as follows: ； in, is the overlapping area between the predicted box and the real box, It is the total area after the predicted box and the real box are merged; A3.4.

3. Elimination For boxes that exceed the threshold, repeat the steps until there are no remaining boxes.

8. The method for automatically unhooking and rehooking a coupler based on artificial intelligence according to claim 1, characterized in that: The robot is equipped with a wireless communication module, the data transmission delay of which is less than 50ms. The robot's self-test and operation data uploading are both realized through the wireless communication module.

9. The method for automatically unhooking and rehooking a coupler based on artificial intelligence according to claim 1, characterized in that: The deep learning method in A6 is as follows: A6.

1. A multispectral camera captures RGB and NIR images of the coupler, a lidar acquires a 3D point cloud of the coupler, and a pressure sensor at the end of the robotic arm records contact pressure data during coupler removal. This data is collected under different weather conditions, lighting conditions, and coupler position deviations. A6.

2. Input multispectral imagery, laser point cloud, and pressure sensor time series data into the deep learning model. Initialize the vision and point cloud branches on a public dataset and jointly train the multimodal network with a learning rate of 3e-4 and a batch size of 16. Use an optimizer. First, train the unhooking operation for simple scenes and weather conditions, then gradually increase the difficulty of the unhooking operation for scenes and weather conditions. A6.

3. Model simulation to learn the robot's 3D coupler pose, environmental parameters, arm joint angles, end-of-arm movement speed, rotation angle, gripping force, and hook removal thrust under different environmental and weather conditions. A6.

4. Then design the deep learning model reward mechanism: Success reward: R+50 when completing the unhooking / rehooking operation; Efficiency bonus: R+10 for every second of operation time reduced; Identification bonus: When the accuracy of identifying coupler type increases by 1%, R+20; Efficiency penalty: R-30 for every additional second of operation time; Success penalty: R-100 if the unhooking / rehooking operation fails; Safety penalty: When the contact force exceeds the threshold, R−10.

10. An artificial intelligence-based automatic coupler unhooking and rehooking system, characterized by: Used to execute the method for automatic coupler unhooking and rehooking based on artificial intelligence as claimed in claim 1, the automatic coupler unhooking and rehooking system comprises: a robot, a charging station and a centralized control center; The robot includes: a movable six-joint robotic arm, a tracked autonomous mobile body, a robotic arm and end effector, a laser radar, an infrared thermal imager, a millimeter-wave radar, a multispectral camera and a pressure sensor at the end of the robotic arm. The robot is wirelessly connected to the control center.

Images