An unmanned electrolysis multifunctional crown based on AI and machine vision
The unmanned electrolysis multi-functional overhead crane, which integrates AI and machine vision, has solved the problem of manual operation in the aluminum electrolysis workshop, realized unmanned operation of the entire process, improved operation accuracy and safety, and reduced costs and energy consumption.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENYANG ALUMINUM & MAGNESIUM TECH CO LTD
- Filing Date
- 2026-02-03
- Publication Date
- 2026-05-29
AI Technical Summary
The existing crane operations in electrolytic aluminum workshops rely on manual operation, which has problems such as low precision, high safety risks, large efficiency fluctuations, weak functional coordination, and high labor costs. It lacks the real-time integration of AI deep decision-making and machine vision, and cannot achieve fully unmanned operation.
The unmanned electrolysis multifunctional overhead crane, based on AI and machine vision, integrates an intelligent decision-making system, a visual perception system, an execution control system, and a remote monitoring system to achieve unmanned control of the entire crane operation process, thereby improving operational accuracy, efficiency, and safety.
It has achieved fully automated operation of overhead cranes, improved operational accuracy and safety, reduced labor costs and energy consumption, increased production efficiency, and reduced the accident rate.
Smart Images
Figure CN122102001A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent technology of electrolytic aluminum production equipment, and in particular relates to an unmanned electrolytic multifunctional overhead crane based on AI and machine vision. It is suitable for the automated operation of large prebaked anode aluminum electrolytic cells, and realizes unmanned operation of all functions such as shell breaking, anode replacement, feeding, slag removal, aluminum discharge, and busbar lifting. Background Technology
[0002] The electrolytic multi-functional overhead crane is a core piece of equipment in the aluminum electrolysis workshop, undertaking key operations such as aluminum tapping, electrode changing, material unloading, and anode handling. Its operating efficiency and safety directly affect electrolytic production indicators. Currently, electrolytic overhead cranes mainly rely on manual operation, requiring operators to work continuously in high-temperature, dusty, and strong electromagnetic interference environments, which presents the following problems: 1. Low operational accuracy: Manual judgment of electrolyte shell surface, anode position, etc. by the naked eye, frequently leads to problems such as shell misalignment and anode installation misalignment, affecting electrolysis efficiency; 2. High safety risks: Operators are exposed to high temperatures, strong magnetic fields and harmful fumes for a long time, which poses a great risk to occupational health. In addition, manual operation is prone to safety accidents due to fatigue. 3. Large fluctuations in efficiency: The work rhythm depends on human experience, and the skill differences among different operators lead to poor production stability and difficulty in precise coordination with upstream and downstream processes; 4. Weak functional coordination: Switching between functions such as shell breaking, anode replacement, and aluminum tapping relies on manual operation, lacks intelligent scheduling, and has low overall work efficiency; 5. The training period for skilled overhead crane operators is long, and labor costs are rising year by year.
[0003] In existing technologies, some overhead cranes have achieved semi-automation for single functions (such as automatic shell breaking), but they lack the real-time fusion of AI deep decision-making and machine vision, making it impossible to achieve fully unmanned operation and still requiring human intervention in key stages. Therefore, developing an unmanned multi-functional overhead crane for electrolysis that integrates AI decision-making and machine vision perception has become a key requirement for improving the level of intelligence in electrolysis production. Summary of the Invention
[0004] To address the aforementioned technical issues, this invention provides an unmanned electrolysis multifunctional overhead crane based on AI and machine vision. By deeply integrating AI algorithms with machine vision, it achieves unmanned control of the entire crane operation process, improving operational accuracy, efficiency, and safety, while reducing labor costs and energy consumption.
[0005] The present invention adopts the following technical solution: an unmanned electrolysis multifunctional overhead crane based on AI and machine vision, characterized in that it includes an intelligent decision-making system, a visual perception system, an execution control system, and a remote monitoring system; The intelligent decision-making system is used for intelligent planning, decision-making and optimization of overhead crane operation tasks, and sends decision instructions; The visual perception system is used to collect images and depth information of the work scene to achieve target recognition and localization; The execution control system is used to receive decision instructions and drive the overhead crane to complete action execution and safety control. The remote monitoring system is used for real-time manual monitoring and emergency intervention. The intelligent decision-making system, visual perception system, execution control system, and remote monitoring system achieve data interaction through industrial Ethernet and wireless communication modules.
[0006] The intelligent decision-making system includes an AI decision server, a database server, and a task scheduling terminal; The AI decision server is used for dynamic modeling of work scenarios, priority ranking of multiple tasks, target trajectory planning, and generation of action instructions. The database server is used to store electrolytic cell parameter data, overhead crane equipment data, visual feature data, and historical operation data; The task scheduling terminal is used for task input, assignment, progress tracking, and parameter adjustment.
[0007] The database server contains a job optimization mathematical model with the objective function: minF = ω1f1 + ω2f2 + ω3(1 / f3); Wherein, minF is the minimum value of the overall objective function for optimizing unmanned crane operations, and ω1, ω2, ω3 are weight coefficients (ω1+ω2+ω3=1), which are dynamically adjusted according to the task type; f1 is the total time for a single task, f1 = t 移动 +t 操作 +t 校准 , t 移动 The time required for crane positioning is t 操作 t represents the time taken for the grab / drop action. 校准 Adjusting the time required for visual positioning; f2 represents the energy consumption of the operation, f 2= m is the load mass (kg), v is the operating speed (m / s), and s is the distance of the path segment (m). f3 represents the motion smoothness, f3 = 1 / a i The acceleration at each moment (m / s²) indicates that the larger the value, the smoother the motion. Each task is executed by only one overhead crane, and the task paths are uninterrupted. The mathematical model is as follows: ; Where, Path i Path j These are the sets of operation paths for the i-th and j-th overhead cranes, respectively. All tasks must be completed within the electrolytic cell process window: Start time t start ≥t window,start End time t end ≤t window,end ; Among them, t window,start t window,end This represents the endpoint of the time interval during which the electrolytic cell is allowed to operate.
[0008] The visual perception system includes an industrial vision sensor group and an image preprocessing module; The industrial vision sensor group includes a high-definition industrial camera deployed on the overhead crane beam, a laser vision sensor at the bottom of the trolley, and a binocular vision sensor on the workshop column to collect target images and three-dimensional depth information. The image preprocessing module is used for image noise reduction, distortion correction, feature enhancement, extraction of target edge features, and uploading to the intelligent decision-making system.
[0009] The industrial vision sensor group has an IP65 protection rating, is adaptable to environments from -20℃ to 80℃, and features a high-definition industrial camera with a 4K resolution and a frame rate of ≥30fps.
[0010] The visual perception system uses an improved Faster R-CNN algorithm to achieve target recognition, and the steps include: (1) Features are extracted from the preprocessed image using the ResNet50 network to generate a 1024-dimensional feature map; (2) The region proposal network generates candidate target boxes and filters candidate boxes with confidence IoU ≥ 0.75; (3) Perform bounding box regression on the candidate boxes, output the target category and two-dimensional coordinates, with a positioning error ≤ ±5mm; (4) The fusion depth data converts the two-dimensional coordinates into three-dimensional spatial coordinates with an accuracy of ≤ ±3mm.
[0011] The execution control system includes a crane drive unit, a positioning calibration module, and a safety braking device; The overhead crane drive unit adopts a servo motor drive system, which supports 1mm-level displacement adjustment; The positioning calibration module integrates visual data, laser ranging data and encoder feedback, and corrects position deviation through Kalman filtering, achieving a dynamic positioning accuracy of ≤±3mm. The safety braking device has a dual redundancy design, and can trigger emergency braking within 10ms in abnormal situations.
[0012] The execution control system adopts an improved A * The algorithm plans the path, and the steps include: (1) Construct a grid map based on the digital twin model of the workshop, with an accuracy of 50mm×50mm; (2) Define the heuristic function h(n)=ɑ∙Euclidean(n,goal) +β∙risk(n), where ɑ and β are weights, Euclidean(n,goal) is the Euclidean distance between the current grid node (n) and the target grid node (goal) in the grid map of the overhead crane path planning, and risk(n) is the grid safety risk value; (3) Search for the optimal path through the priority queue and select the node with the smallest cost f(n) = g(n) + h(n) to expand; where f(n) is the total estimated cost of the path search, g(n) is the actual cost from the starting point to the current node, and h(n) is the heuristic function of the improved A* algorithm, which is only used to estimate the cost from the current node to the target node. (4) Use B-spline curves to smooth the path.
[0013] The remote monitoring system includes a monitoring client, an emergency operation terminal, and a data visualization platform; The monitoring client displays the operation screen, AI recognition results, and equipment operating parameters in real time. The emergency operation terminal supports emergency takeover of the overhead crane operation by management personnel, with a response delay of ≤200ms. The data visualization platform generates work reports and 3D situation maps.
[0014] The workflow of the AI- and machine vision-based unmanned electrolysis multi-functional overhead crane includes the following steps: Step 301: The task scheduling terminal generates a job task order and uploads it to the AI decision server; Step 302: The AI decision server determines the feasibility of the task and triggers the start of the visual perception system; Step 303: The visual perception system completes target recognition and 3D localization, and sends the data to the AI decision server; Step 304: The AI decision server plans the path and generates action instructions; Step 305: The control system drives the overhead crane to complete the operation, and the positioning calibration module corrects the deviation; Step 306: After the task is completed, the results are reported, the remote monitoring system updates the task status, and emergency intervention is triggered in case of anomalies.
[0015] The AI- and machine vision-based unmanned electrolysis multifunctional overhead crane of the present invention has at least the following beneficial effects: 1. Fully automated operation: For the first time, AI and machine vision are integrated to achieve unmanned driving of all core functions such as shell opening and anode replacement. Real-time visual monitoring and intelligent braking system eliminate blind spots of manual operation, replace manual operation and reduce safety risks; 2. Improved operational accuracy: Through AI vision positioning and precise control, the shell-making deviation is ≤5mm, the anode installation flatness is ≤3mm, the aluminum output error is ≤±20kg, and the anode verticality deviation during electrode replacement is ≤1°, reducing costs by more than 80% compared to manual operation; 3. Strong dynamic adaptability: The reinforcement learning decision model can adapt to changes in the electrolytic cell operating conditions in real time (such as electrolyte level and anode consumption), the vision sensor is adaptable to high temperature and dusty environments, and the algorithm supports multiple task types (aluminum tapping, electrode changing, and material feeding), making it suitable for the transformation of new and old electrolytic workshops. 4. High safety redundancy: Multi-sensor fusion and safety interlock design enable functions such as collision warning and overload protection, reducing the accident rate by 90%; 5. Improved production efficiency: AI intelligent scheduling reduces the overhead crane idle rate to below 5%, shortens the aluminum output time per cell by 15-20%, and improves the efficiency of pole changing operations by 25%; 6. Cost reduction: Achieve unmanned operation, reduce the number of overhead crane operators by more than 60%, and reduce energy consumption by 10-15% through path optimization. Attached Figure Description
[0016] Figure 1 This is a system architecture diagram of the unmanned electrolysis multifunctional overhead crane based on AI and machine vision of the present invention; Figure 2 This is a schematic diagram illustrating the implementation of the unmanned electrolysis multifunctional overhead crane based on AI and machine vision of the present invention; Figure 3 This is a schematic diagram of the workflow of the unmanned electrolysis multifunctional overhead crane based on AI and machine vision according to the present invention. Detailed Implementation
[0017] The present invention will be further explained below with reference to the accompanying drawings and embodiments.
[0018] This invention relates to an unmanned electrolysis multifunctional overhead crane based on AI and machine vision, comprising an intelligent decision-making system, a visual perception system, an execution control system, and a remote monitoring system. The intelligent decision-making system includes an AI decision server, a database server, and a task scheduling terminal. The visual perception system includes an industrial vision sensor array and an image preprocessing module. The execution control system includes an overhead crane drive unit, a positioning calibration module, and a safety braking device. The remote monitoring system includes a monitoring client, an emergency operation terminal, and a data visualization platform. This invention achieves precise coordination between intelligent decision-making, environmental perception, action execution, and remote monitoring through the fusion of AI algorithms and machine vision technology. This enables fully unmanned operation of the entire process, including aluminum extraction, electrode changing, and material unloading, improving the efficiency and safety of electrolysis workshops while reducing labor and energy costs.
[0019] like Figure 1 As shown, the present invention is an unmanned electrolysis multifunctional overhead crane based on AI and machine vision, comprising: an intelligent decision-making system 1, a visual perception system 2, an execution control system 3, and a remote monitoring system 4.
[0020] The intelligent decision-making system 1 includes a task scheduling terminal 11, an AI decision server 12, and a database server 13. The task scheduling terminal 11 is used for task input, priority adjustment, and progress tracking. Managers can view the real-time status and historical data of the overhead crane through the terminal. The task scheduling terminal 11 uses an industrial-grade touchscreen and supports stable operation in high-temperature environments.
[0021] The AI decision server 12 is used for dynamic modeling of work scenarios, priority ranking of multiple tasks, target trajectory planning, and generation of action instructions. Equipped with a GPU acceleration module, it has the capability to process 500 frames per second of images, enabling parallel decision-making for multiple tasks and serving as the core computing power center of the system. It generates work paths based on visual data and production requirements, and issues instructions to the execution control system via industrial Ethernet.
[0022] Database server 13 adopts a time-series database architecture to store electrolytic cell parameters (updated ≥500 records daily), crane operation data (collected 10 records per second), visual feature templates (containing 1000+ target features), and task history records (supporting 3 years of data backtracking). It is used to store electrolytic cell parameter data, crane equipment data, visual feature data, and historical operation data.
[0023] The mathematical model for job optimization established in the database server 13 is minF=ω1f1+ω2f2+ω3(1 / f3), where minF is the minimum value of the overall objective function for optimizing the unmanned crane operation, and ω1, ω2, ω3 are weight coefficients (ω1+ω2+ω3=1), which are dynamically adjusted according to the task type. Specifically, ω1=0.4 for aluminum extraction tasks and ω3=0.5 for pole changing tasks.
[0024] f1 is the total time for a single task, f1 = t 移动 +t 操作 +t 校准 , t 移动 The time required for crane positioning is t 操作 t represents the time taken for the grab / drop action. 校准 Adjusting the time required for visual positioning; f2 represents the energy consumption of the operation, f 2= m is the load mass (kg), v is the operating speed (m / s), and s is the distance of the path segment (m). f3 represents the motion smoothness, f3 = 1 / a i The acceleration at each moment (m / s²) indicates that the larger the value, the smoother the motion. Each task is executed by only one overhead crane, and there is no interference in the task path: ; Where, Path i Path j These are the sets of operation paths for the i-th and j-th overhead cranes, respectively. All tasks must be completed within the electrolytic cell process window: Start time t start ≥t window,start End time t end ≤t window,end ; Among them, t window,start t window,end This represents the endpoint of the time interval during which the electrolytic cell is allowed to operate.
[0025] The objectives are to achieve "shortest operating time, lowest energy consumption, and smoothest operation." For example, in aluminum tapping tasks, the objective function weights are set to ω1=0.4, ω2=0.3, and ω3=0.3 to ensure rapid response to the aluminum tapping demand of the electrolytic cell while reducing the risk of aluminum sloshing. In the constraints, the safe distance between the overhead crane and the electrolytic cell wall is strictly controlled at ≥300mm to avoid collision accidents.
[0026] The visual perception system 2 includes an industrial vision sensor group 21 and an image preprocessing module 22. The industrial vision sensor group 21 has an IP65 protection rating, is adaptable to environments from -20℃ to 80℃, and features a high-definition industrial camera with a 4K resolution and a frame rate ≥30fps. Four high-definition industrial cameras on the overhead crane beam cover the electrolytic cell opening area (120° field of view) and the bag-lifting hook area (close-up shooting). A laser vision sensor at the bottom of the trolley has a range of 0-10m and an accuracy of ±1mm, used to measure the load height. A binocular vision sensor on the workshop columns enables global positioning calibration of the overhead crane.
[0027] The image preprocessing module 22 performs Gaussian noise reduction and contrast enhancement on the original image, extracts target edge features, and uploads the feature data (updated every 100ms) to the AI decision server via Ethernet.
[0028] Taking the pole-switching task as an example, the recognition process of the improved Faster R-CNN algorithm is as follows: First, the features of the preprocessed image are extracted through the ResNet50 network to generate a 1024-dimensional feature map; after generating candidate boxes, candidate boxes with a confidence of ≥0.75 are selected, and bounding box regression is performed on the candidate boxes to output the target category and two-dimensional coordinates with a positioning error of ≤±5mm; the two-dimensional coordinates are converted into three-dimensional spatial coordinates by fusing depth data with an accuracy of ≤±3mm, which meets the pole-switching installation accuracy requirements.
[0029] The execution control system 3 includes: a crane drive unit 31, a positioning calibration module 32, and a safety braking device 33. The servo driver of the crane drive unit 31 supports 1mm-level displacement control, controlling the ladle to precisely align with the aluminum outlet at a speed of 0.5m / min during aluminum unloading operations. The positioning calibration module 32 integrates visual data and encoder feedback, using Kalman filtering to correct path deviations in real time. For example, when the actual crane position deviates from the planned path by 3mm, the drive parameters are automatically adjusted. The safety braking device 33 immediately triggers braking when abnormal tension in the lifting cable (exceeding 120% of the rated value) is detected, with a response time ≤10ms.
[0030] Improvement A * In path planning, the algorithm automatically generates detour paths for obstacles such as temporarily stacked anode blocks in the workshop. For example, the planned path from the starting point to the target electrolytic cell is 12% shorter in total length than the traditional path and has no inflection points, reducing the overhead crane's operating energy consumption by 8%.
[0031] The steps include: (1) Construct a grid map based on the digital twin model of the workshop, with an accuracy of 50mm×50mm; (2) Define the heuristic function h(n) = ɑ∙Euclidean(n,goal) + β∙risk(n), where ɑ and β are weights, Euclidean(n,goal) is the Euclidean distance between the current grid node (n) and the target grid node (goal) in the grid map of the overhead crane path planning, and risk(n) is the grid safety risk value; (3) Search for the optimal path through the priority queue and select the node with the smallest cost f(n) = g(n) + h(n) to expand; where f(n) is the total estimated cost of the path search, g(n) is the actual cost from the starting point to the current node, and h(n) is the heuristic function of the improved A* algorithm, which is only used to estimate the "cost from the current node to the target node". (4) Use B-spline curves to smooth the path.
[0032] The remote monitoring system 4 includes: a monitoring client 41, an emergency operation terminal 42, and a data visualization platform 43. The 4K high-definition screen of the monitoring client 41 displays the crane operation footage and AI recognition results in real time, and managers can view key parameters such as the verticality of the anode blocks and the tilt angle of the lifting bag.
[0033] When AI recognition fails (e.g., dust obstruction causing confidence IoU < 0.6), the emergency operation terminal 42 supports one-click takeover, controlling the crane's movement via a handle, with a response latency of ≤ 200ms.
[0034] The data visualization platform generates daily work reports, showing an average accuracy of 98.5% for pole changing tasks and a 100% completion rate for aluminum extraction tasks.
[0035] like Figure 3 As shown, the workflow of this embodiment of the invention is as follows: Step 301: The task scheduling terminal generates the "Electrolytic Cell No. 28 Electrolysis Electrode Replacement Task", specifies the old electrode location, new electrode model, and process window 9:00-9:30, and uploads it to the AI decision server; Step 302: The AI decision server confirms that crane No. 3 is idle and triggers the visual perception system to start; Step 303: The visual sensor acquires an image of the area of slot 28, the algorithm identifies the position of the old pole, the three-dimensional positioning error is 2mm, and the data is sent to the AI decision server; Step 304: The AI decision server plans the path (total length 32m) and generates action instructions: the crane moves to the top of the old pole → descends to grab → transfers to the waste area → moves to the new pole stacking area → grabs the new pole → transfers to slot 28 → precise installation; Step 305: The control system drives the overhead crane to perform actions according to instructions, and the positioning calibration module corrects the deviation in real time, ensuring that the verticality deviation is 0.8°. Step 306: After the task is completed, the results are reported, the remote monitoring system updates the task status, and emergency intervention is triggered in case of anomalies.
[0036] like Figure 2 As shown, in specific implementation, the intelligent decision-making system 1 is deployed in the workshop control center, the sensors of the visual perception system 2 are installed on the overhead crane body and workshop columns, the execution control system 3 is integrated into the overhead crane equipment, and the remote monitoring system 4 is located in the control room. During system operation, AI decision-making and machine vision are integrated in real time: AI is responsible for real-time positioning and decision-making, while machine vision assists in calibrating key parameters through a monitoring client; when extreme dust interference occurs, management personnel take over through an emergency operation terminal to ensure operational continuity.
[0037] The above description is only a preferred embodiment of the present invention and is not intended to limit the ideas of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A multi-functional unmanned electrolysis crane based on AI and machine vision, characterized in that, This includes intelligent decision-making systems, visual perception systems, execution control systems, and remote monitoring systems; The intelligent decision-making system is used for intelligent planning, decision-making and optimization of overhead crane operation tasks, and sends decision instructions; The visual perception system is used to collect images and depth information of the work scene to achieve target recognition and localization; The execution control system is used to receive decision instructions and drive the overhead crane to complete action execution and safety control. The remote monitoring system is used for real-time manual monitoring and emergency intervention. The intelligent decision-making system, visual perception system, execution control system, and remote monitoring system achieve data interaction through industrial Ethernet and wireless communication modules.
2. The unmanned electrolysis multi-functional overhead crane based on AI and machine vision according to claim 1, characterized in that, The intelligent decision-making system includes an AI decision server, a database server, and a task scheduling terminal; The AI decision server is used for dynamic modeling of work scenarios, priority ranking of multiple tasks, target trajectory planning, and generation of action instructions. The database server is used to store electrolytic cell parameter data, overhead crane equipment data, visual feature data, and historical operation data; The task scheduling terminal is used for task input, assignment, progress tracking, and parameter adjustment.
3. The unmanned electrolysis multi-functional overhead crane based on AI and machine vision according to claim 2, characterized in that, The database server contains a job optimization mathematical model with the objective function: minF = ω1f1 + ω2f2 + ω3(1 / f3); Wherein, minF is the minimum value of the overall objective function for optimizing unmanned crane operations, and ω1, ω2, ω3 are weight coefficients (ω1+ω2+ω3=1), which are dynamically adjusted according to the task type; f1 is the total time for a single task, f1 = t 移动 +t 操作 +t 校准 , t 移动 The time required for crane positioning is t 操作 t represents the time taken for the grab / drop action. 校准 Adjusting the time required for visual positioning; f2 represents the energy consumption of the operation, f 2= m is the load mass (kg), v is the operating speed (m / s), and s is the distance of the path segment (m). f3 represents the motion smoothness, f3 = 1 / a i The acceleration at each moment (m / s²) indicates that the larger the value, the smoother the motion. Each task is executed by only one overhead crane, and the task paths are uninterrupted. The mathematical model is as follows: ； Where, Path i Path j These are the sets of operation paths for the i-th and j-th overhead cranes, respectively. All tasks must be completed within the electrolytic cell process window: Start time t start ≥t window,start End time t end ≤t window,end ; Among them, t window,start t window,end This represents the endpoint of the time interval during which the electrolytic cell is allowed to operate.
4. The unmanned electrolysis multifunctional overhead crane based on AI and machine vision according to claim 1, characterized in that, The visual perception system includes an industrial vision sensor group and an image preprocessing module; The industrial vision sensor group includes a high-definition industrial camera deployed on the overhead crane beam, a laser vision sensor at the bottom of the trolley, and a binocular vision sensor on the workshop column to collect target images and three-dimensional depth information. The image preprocessing module is used for image noise reduction, distortion correction, feature enhancement, extraction of target edge features, and uploading to the intelligent decision-making system.
5. The unmanned electrolysis multi-functional overhead crane based on AI and machine vision according to claim 4, characterized in that, The industrial vision sensor group has an IP65 protection rating, is adaptable to environments from -20℃ to 80℃, and features a high-definition industrial camera with a 4K resolution and a frame rate of ≥30fps.
6. The unmanned electrolysis multifunctional overhead crane based on AI and machine vision according to claim 4, characterized in that, The visual perception system uses an improved Faster R-CNN algorithm to achieve target recognition, and the steps include: (1) Features are extracted from the preprocessed image using the ResNet50 network to generate a 1024-dimensional feature map; (2) The region proposal network generates candidate target boxes and filters candidate boxes with confidence IoU ≥ 0.75; (3) Perform bounding box regression on the candidate boxes, output the target category and two-dimensional coordinates, with a positioning error ≤ ±5mm; (4) The fusion depth data converts the two-dimensional coordinates into three-dimensional spatial coordinates with an accuracy of ≤ ±3mm.
7. The unmanned electrolysis multifunctional overhead crane based on AI and machine vision according to claim 1, characterized in that, The execution control system includes a crane drive unit, a positioning calibration module, and a safety braking device; The overhead crane drive unit adopts a servo motor drive system, which supports 1mm-level displacement adjustment; The positioning calibration module integrates visual data, laser ranging data and encoder feedback, and corrects position deviation through Kalman filtering, achieving a dynamic positioning accuracy of ≤±3mm. The safety braking device has a dual redundancy design, and can trigger emergency braking within 10ms in abnormal situations.
8. The unmanned electrolysis multi-functional overhead crane based on AI and machine vision according to claim 7, characterized in that, The execution control system adopts an improved A * The algorithm plans the path, and the steps include: (1) Construct a grid map based on the digital twin model of the workshop, with an accuracy of 50mm×50mm; (2) Define the heuristic function h(n)=ɑ∙Euclidean(n,goal) +β∙risk(n), where ɑ and β are weights, Euclidean(n,goal) is the Euclidean distance between the current grid node (n) and the target grid node (goal) in the grid map of the overhead crane path planning, and risk(n) is the grid safety risk value; (3) Search for the optimal path through the priority queue and select the node with the smallest cost f(n) = g(n) + h(n) to expand; where f(n) is the total estimated cost of the path search, g(n) is the actual cost from the starting point to the current node, and h(n) is the heuristic function of the improved A* algorithm, which is only used to estimate the cost from the current node to the target node. (4) Use B-spline curves to smooth the path.
9. The unmanned electrolysis multi-functional overhead crane based on AI and machine vision according to claim 1, characterized in that, The remote monitoring system includes a monitoring client, an emergency operation terminal, and a data visualization platform; The monitoring client displays the operation screen, AI recognition results, and equipment operating parameters in real time. The emergency operation terminal supports emergency takeover of the overhead crane operation by management personnel, with a response delay of ≤200ms. The data visualization platform generates work reports and 3D situation maps.
10. The unmanned electrolysis multifunctional overhead crane based on AI and machine vision according to claim 1, characterized in that, The workflow includes the following steps: Step 301: The task scheduling terminal generates a job task order and uploads it to the AI decision server; Step 302: The AI decision server determines the feasibility of the task and triggers the start of the visual perception system; Step 303: The visual perception system completes target recognition and 3D localization, and sends the data to the AI decision server; Step 304: The AI decision server plans the path and generates action instructions; Step 305: The control system drives the overhead crane to complete the operation, and the positioning calibration module corrects the deviation; Step 306: After the task is completed, the results are reported, the remote monitoring system updates the task status, and emergency intervention is triggered in case of anomalies.