Intelligent station production method and system based on AI computer vision technology

Abstract

The invention belongs to the technical field of intelligent production, and particularly relates to an intelligent station production method and system based on the AI computer vision technology, and the method comprises the steps: obtaining order information, automatically extracting order data, and generating a structured order task; on the basis of the order tasks, order task distribution of the production line is carried out, and equipment operation parameters are correspondingly adjusted; physical station data are obtained, a digital twin virtual station simulation model is constructed in combination with the distributed order task and the standard operation process on the physical station, and the working state of each station is mapped in real time; performing product verification and operation detection based on the product state of the production line and the operation state of each physical station, and performing order task allocation optimization and employee operation track optimization in combination with the real-time operation state of each physical station; and on the basis of the verified product, electronic order signing is generated, and block chain evidence storage is carried out, so that full-link intelligence, automation and high efficiency of order processing, path planning, job execution and quality control are realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent production, and particularly relates to a production method and system of an intelligent station based on AI computer vision technology. BACKGROUND

[0002] In the field of intelligent manufacturing and intelligent warehousing, with the diversification of market demand and the increase of production complexity, the traditional operation mode faces the dual challenges of efficiency and quality. Specifically, in the traditional production mode, multi-variety line change relies on manual parameter adjustment, which takes as long as 30 minutes and is prone to errors due to human factors, affecting production efficiency; the manual verification method in the picking and assembly link has high missed detection rate and difficulty in tracing, leading to high error rate in picking and assembly, affecting product quality; at the same time, the employee turnover rate is high in the new era, and the training period of skilled workers is as long as 21 days, resulting in strong dependence on enterprise manpower and high training cost; and the traditional path planning method lacks dynamic optimization capability, resulting in long walking distance of personnel and high operation time cost; the traditional quality control relies on manual sampling inspection, and it is difficult to realize whole-process quality tracing and real-time monitoring. SUMMARY

[0003] The application provides a production method and system of an intelligent station based on AI computer vision technology, which realizes order task allocation by extracting order information, constructs a virtual station simulation model of digital twin by combining physical station data, supervises and controls the product state of products in each physical station and the operation state of each physical station, realizes product verification and operation detection, and then optimizes order task allocation and employee operation operation trajectory, completes electronic order signing, realizes the full-link intelligentization, automation and high efficiency of order processing, path planning, operation execution and quality control, reduces manual intervention, and improves operation efficiency, accuracy, quality control capability and man-machine cooperation efficiency.

[0004] A production method of an intelligent station based on AI computer vision technology, comprising: Obtaining order information, automatically extracting order data, and generating structured order tasks; Based on the order task, the order task of the production line is allocated, and the equipment operation parameters are adjusted correspondingly; Obtaining physical station data, combining the allocated order task and the standard operation process on the physical station, constructing a virtual station simulation model of digital twin, and mapping the operation state of each station in real time; Based on the product state of the production line and the operation state of each physical station, product verification and operation detection are performed, and order task allocation optimization and employee operation operation trajectory optimization are performed in combination with the real-time operation state of each physical station; Based on the verified product, an electronic order signing is generated, and a blockchain is stored.

[0005] By extracting order information for order task allocation, combining physical station data to build a virtual station simulation model of digital twin, supervising and controlling the product state of each physical station and the operation state of each physical station, realizing product verification and operation detection, and then optimizing order task allocation and employee operation trajectory optimization, completing electronic order receipt, realizing the full-link intelligentization, automation and high efficiency of order processing, path planning, operation execution and quality control, reducing manual intervention, and improving operation efficiency, accuracy, quality control ability and man-machine cooperation efficiency.

[0006] Further, the order information is acquired, order data is automatically extracted, and a structured order task is generated, including: An API interface is connected to an upper business system to directly acquire electronic order data; A scanning head is used to scan paper documents to identify and acquire paper order data; Based on the electronic order data and the paper order data, real-time fusion is performed to obtain order data; A BERT-NER natural language processing model is used to automatically extract order specification parameters and order delivery data in the order data, and to generate a structured order task; the order specification parameters include product, product quantity, product size, product material, and product industry requirements; the order delivery data includes delivery time limit, product packaging, and gift level.

[0007] Through automatic acquisition and processing of order information, the order processing efficiency is improved; by generating a structured order task, a clear and accurate data basis is provided for subsequent order task allocation.

[0008] Further, the order task is based on the order task to allocate the order task of the production line, and the equipment operation parameters are adjusted accordingly, including: Based on the order delivery data in the order task, the order delivery time limit is acquired; Based on the order delivery time limit, the order type is determined; the order type includes emergency order, regular order, and peak order; Based on the order type, the production line mode is selected, the order task allocation is completed, and the production line equipment operation parameters are adjusted accordingly; When it is determined that the current order is an emergency order, a manual mode is selected, the order task allocation is preferentially intervened by manual intervention, and the production line equipment operation parameters are adjusted accordingly; When it is determined that the current order is a regular order, an AI automatic mode is selected, the order task allocation is automatically performed, and the production line equipment operation parameters are adjusted accordingly; When it is determined that the current order is a peak order, the mixed-line mode is selected for multi-product co-line production, the product categories are automatically analyzed, and the production line equipment operation parameters corresponding to the product categories are switched.

[0009] By analyzing the order information to determine the order type and flexibly selecting the production line mode based on different order types, the device operation parameters are optimized, and the response speed and adaptability of the production line are improved. Different production line modes are selected to ensure that different order processing requirements can achieve efficient production.

[0010] Further, the physical station data is obtained, combined with the allocated order task and the standard operation process on the physical station, a virtual station simulation model of digital twin is constructed, and the operation state of each station is mapped in real time, including: Physical station modeling based on shelves, equipment, and operation areas; Real-time collection of sensor data based on sensor data on the physical station; Based on the allocated order task, the operation process on the physical station is modeled according to the standard operation process, and the OptiTrack motion capture system combined with the YOLOv8 pose recognition model is used to record the operation trajectory of the staff, construct the operation trajectory digital twin, and decompose to form an atomized action library; Real-time identification and positioning of products, staff, and equipment, and dynamic planning of the picking path combined with the allocated order task, constructing a virtual station simulation model of digital twin, and using a visual interactive terminal to map the operation state of each station in real time.

[0011] Through digital twin technology, the visual real-time mapping of the physical station and the virtual station simulation model is realized, which can provide intuitive and accurate data support for the monitoring and optimization of the production line, facilitating real-time control and optimization of operation behavior and operation process optimization.

[0012] Further, based on the allocated order task, the operation process on the physical station is modeled according to the standard operation process, and the OptiTrack motion capture system combined with the YOLOv8 pose recognition model is used to record the operation trajectory of the staff, construct the operation trajectory digital twin, and decompose to form an atomized action library, including: Based on the allocated order task, manual picking is performed according to the standard operation process, and the 3D vision system is used to record the operation trajectory of the staff of excellent staff; Based on the operation trajectory of the staff, the OptiTrack motion capture system combined with the YOLOv8 pose recognition model is used to construct the operation trajectory digital twin; Based on the operation trajectory digital twin, operation action atomization is performed, and each physical station standard time, key action and core quality inspection point are labeled to complete operation action timing modeling; the operation action atomization includes picking and placing, press fitting, screwing and plugging; Based on the operation action atomization disassembly process, operation action atomization combination is performed to construct an atomized action library.

[0013] Through the motion capture system and the posture recognition model, the operation trajectory of the operator is recorded and analyzed to construct the atomized action library, which provides data support for subsequent optimization of the operation trajectory of the operator, improves the operation efficiency and accuracy; at the same time, based on different operation action atomization combinations, cross-line process reuse can be realized through federated transfer learning, and cross-line rapid changeover can be realized.

[0014] Further, based on the product state of the production line and the operation state of each physical station, product verification is performed, including: Real-time product image information is captured by a camera, and a 3D vision system is used to count the products; Based on the product image information, product size measurement and product surface defect detection are performed; Based on the product image information, a deep learning model is used to identify the product category and extract product attribute information for automatic verification; the product attribute information includes product color, product specification and product packaging state.

[0015] Through the camera and the 3D vision system, real-time capture, counting, size measurement and surface defect detection of product image information are realized, which can improve the product picking quantity accuracy and the product integrity rate; at the same time, through the extraction of product attribute information, full-factor product verification is completed, which can improve the product verification accuracy and efficiency.

[0016] Further, based on the product state of the production line and the operation state of each physical station, operation detection is performed, including: Based on the 3D vision system, a space coordinate system is constructed to perform operation action frame-level analysis, and real-time skeletal key point recognition, product real-time positioning and three-way positioning verification of shelf location positioning are performed; Based on the three-way positioning verification results, the standard operation process in the process knowledge graph is compared in real time, and the operation deviation triggers an immediate correction strategy; The standard operation program execution index algorithm is configured to perform multi-dimensional quantitative evaluation of operation action standard degree, timing coincidence rate and tool compliance, and operation action quality data is recorded to generate a dynamic scoring report; Based on product part feature matching, product part assembly verification is performed, and an assembly pressure thermal map is generated through torque distribution analysis; Real-time monitoring and picking error detection are performed, so that when product picking errors occur, three-level early warning is triggered in turn, including shelf storage location light strip flashing, voice prompt and operation table vibration feedback, and dynamic adjustment of operation guidance is performed according to the early warning prompt to form a closed loop feedback.

[0017] By analyzing the operation action in the spatial coordinate system, triple positioning verification is realized; by comparing the process knowledge graph, operation deviation is corrected to improve operation accuracy and compliance; by real-time monitoring of the picking state, when product picking errors occur, three-level early warning is triggered in turn to form a closed loop feedback, thereby improving operation quality and efficiency.

[0018] Further, the order task allocation optimization is performed in combination with the real-time operation state of each physical station, including: Production data of each physical station is collected, and actual processing time of each physical station to complete actual production tasks is recorded; Based on the standard operation process and process requirements of the physical station, in combination with the production data of each physical station, the theoretical cycle time of each physical station is calculated by using the virtual station simulation model of digital twinning; In combination with the theoretical cycle time and actual processing time of each station, the time deviation is calculated; Based on the time deviation, a dynamic scheduling algorithm is used to reassign the order tasks, so that the production line efficiency meets the preset deviation.

[0019] By calculating the deviation between the processing time and the actual value and the theoretical value of each physical station, and using a dynamic scheduling algorithm to reassign the order tasks, the production efficiency can meet the preset deviation while achieving balanced production capacity of the entire line, thereby improving the overall efficiency and stability of production.

[0020] Further, the employee operation trajectory optimization is performed in combination with the real-time operation state of each physical station, including: Based on the shelves, equipment and operation area, in combination with the employee operation trajectory, a genetic algorithm is used to plan the initial employee operation trajectory; Based on a fixed time interval or a significant change in the warehouse environment, an employee operation trajectory re-planning mechanism is triggered, and according to the current order task situation, a wave merging strategy is used to merge order tasks, and a minimum completion time MCT scheduling strategy is used for load balancing to achieve order task allocation optimization, thereby obtaining an optimized employee operation trajectory.

[0021] By triggering the employee operation trajectory re-planning mechanism based on the initial employee operation trajectory according to the change in the warehouse environment or a fixed time interval, and using the wave merging strategy and the MCT scheduling strategy for load balancing, the employee operation trajectory is optimized, thereby improving the operation efficiency.

[0022] A system of an intelligent station production method based on AI computer vision technology, comprising: A perception layer structure deployed with a cluster of intelligent terminal devices; the intelligent terminal devices include a camera, a scanning head, a pressure sensor, a UWB positioning module, and a dynamic light guide module; the camera is used to capture product image information; the scanning head is used to scan paper order information; the pressure sensor is used to detect operating torque; the UWB positioning module is used to perform regional-level spatial positioning on products; the dynamic light guide module is installed on a shelf location and used to indicate product positions in real time; A network layer structure including edge digital twin nodes and industrial control networks; data of the perception layer structure is transmitted to the edge digital twin nodes through the industrial control networks; a plurality of the digital twin nodes are arranged in the perception layer structure and communicate with the intelligent terminal devices of the perception layer structure through the industrial control networks, receive and process perception layer structure data in real time, and perform model construction, simulation, and visual display; A platform layer structure including a digital twin module, an AI algorithm model library, and a blockchain storage module; the digital twin module is used to obtain perception layer structure data from the edge digital twin nodes to generate a virtual station simulation model of a digital twin, and perform simulation based on order tasks to realize order task allocation and employee operation trajectory planning; the AI algorithm model library is deployed in the cloud and stores different AI algorithm models; the blockchain storage module is used to realize operation data encryption and tamper-proofing traceability; An application layer structure that receives simulation results of the digital twin module and performs specific business implementation; the application layer structure includes an order module, a path planning module, a visual quality inspection module, a three-dimensional interactive guidance module, a video backtracking module, and an electronic signing module; the order module is used to extract and analyze order information and allocate order tasks to the digital twin module; the path planning module is used to plan employee operation trajectories; the visual quality inspection module is used to perform visual identification and verification on products; the three-dimensional interactive guidance module includes a touch screen and a tactile feedback operation table; the touch screen is used to integrate five-screen display functions of an electronic work order, a process drawing, a real-time clock, an operation animation, and real-time detection results; the tactile feedback operation table is used to perform operation table vibration early warning; the video backtracking module is used to backtrack and quickly locate operation behaviors; and the electronic signing module is used to realize paperless delivery and closed-loop delivery in the delivery link.

[0023] Through the system architecture design of the perception layer structure, the network layer structure, the platform layer structure, and the application layer structure, an end-edge-cloud collaborative intelligent station system is constructed, and data collection, transmission, processing, and application are realized, so that the modules work collaboratively to provide comprehensive and efficient technical support for intelligent station production, and the intelligent level and production efficiency of the production line are improved.

[0024] The beneficial effects of the present application are: The present application extracts order information for order task allocation, constructs a virtual work station simulation model of digital twinning in combination with physical work station data, supervises and controls the product state of products at each physical work station and the operation state of each physical work station, realizes product checking and operation detection, and then performs order task allocation optimization and employee operation trajectory optimization, completes electronic order signing, realizes the full-link intelligentization, automation and high efficiency of order processing, path planning, operation execution and quality control, reduces manual intervention, and improves operation efficiency, accuracy, quality control capability and man-machine cooperation efficiency. Through data driving and flexible production design, the enterprise can break through the bottlenecks of low multi-variety changeover efficiency, high picking and assembly error rate, and strong dependence on manual operation, and significantly improve the production and warehousing operation efficiency and quality stability. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The flowchart of the present application is shown in the figure. Figure 2 The system structure diagram of the present application is shown in the figure. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0027] It should be noted that the various aspects of the embodiments described below are within the scope of the appended claims. It should be apparent that the aspects described herein can be embodied in a wide variety of forms and that any specific structure and / or function described herein is merely illustrative. Based on the disclosure provided, one skilled in the art should appreciate that an aspect described herein can be implemented independently of any other aspect and that two or more of these aspects can be combined in various ways. For example, an apparatus can be implemented and / or a method can be practiced using any number of the aspects set forth herein. In addition, such an apparatus can be implemented and / or such a method can be practiced using other structure and / or functionality in addition to or other than one or more of the aspects set forth herein.

[0028] In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. For one of ordinary skill in the art, the specific meaning of the above-mentioned terms in the present application can be understood in specific circumstances.

[0029] Example 1 Figure 1The production method of the intelligent station based on the AI computer vision technology is shown, order task allocation is performed by extracting order information, a virtual station simulation model of digital twinning is constructed in combination with physical station data, product states of products at various physical stations and operation states of various physical stations are supervised and controlled, product checking and operation detection are realized, order task allocation optimization and employee operation trajectory optimization are further performed, electronic order signing is completed, and full-link intelligentization, automation and high efficiency of order processing, path planning, operation execution and quality control are realized, manual intervention is reduced, and operation efficiency, accuracy, quality control capability and man-machine cooperation efficiency are improved. Specifically, the following steps are included: S1: obtaining order information, automatically extracting order data, and generating structured order tasks; S11: connecting an upper business system by using an API interface to directly obtain electronic order data; In this embodiment, the upper business system includes an ERP system and a warehouse management system (WMS).

[0030] S12: scanning a paper document by using a scanning head to identify and obtain paper order data; In this embodiment, the scanning head includes a 1D industrial scanning head and a 2D industrial scanning head, and the reading speed is 300 times per second. The paper document is subjected to OCR scanning and identification.

[0031] S13: based on the electronic order data and the paper order data, real-time fusion is performed to obtain order data; S14: using a BERT-NER natural language processing model to automatically extract order specification parameters and order delivery data in the order data, and generating structured order tasks; In this embodiment, the order specification parameters include products, product quantities, product sizes, product materials and product industry requirements; and the order delivery data includes delivery timeliness, product packaging and gift levels.

[0032] It should be noted that a hybrid algorithm based on a rule engine and deep learning can also be used to intelligently identify special processes such as anti-static packaging and serialization traceability.

[0033] S2: based on the order task, performing order task allocation of the production line and adjusting equipment operation parameters correspondingly; S21: based on the order delivery data in the order task, obtaining order delivery timeliness; S22: based on the order delivery timeliness, determining an order type; the order type includes an emergency order, a regular order and a peak order; S23: based on the order type, selecting a production line mode, completing order task allocation, and adjusting production line equipment operation parameters correspondingly; S231: When it is determined that the current order is an urgent order, the manual mode is selected, the order task assignment is prioritized for manual intervention, and the production line equipment operating parameters are adjusted accordingly; S232: When it is determined that the current order is a regular order, the AI automatic mode is selected, the order task assignment is automatically performed, and the production line equipment operating parameters are adjusted accordingly; S233: When it is determined that the current order is a peak order, the mixed-line mode is selected, multi-product co-line production is performed, product categories are automatically analyzed, and the production line equipment operating parameters corresponding to the product categories are switched; S3: Obtain physical station data, combine the assigned order tasks and the standard operation process on the physical station, and construct a virtual station simulation model of digital twin to real-time map the work station state; S31: Perform physical station modeling based on shelves, equipment, and operation areas; S32: Real-time collect sensor data based on sensor data on the physical station; S33: Based on the assigned order tasks, model the operation process on the physical station according to the standard operation process, and use the OptiTrack motion capture system combined with the YOLOv8 pose recognition model to record the employee work operation trajectory, construct an operation trajectory digital twin, and decompose to form an atomized action library; S331: Based on the assigned order tasks, perform manual picking according to the standard operation process, and use a 3D vision system to record the employee work operation trajectory of excellent employees; In this embodiment, the 3D vision system uses the YOLOv8 model to record the employee work operation trajectory.

[0034] S332: Based on the employee work operation trajectory, use the OptiTrack motion capture system combined with the YOLOv8 pose recognition model to construct an operation trajectory digital twin; S333: Based on the operation trajectory digital twin, perform operation action atomization decomposition, and label the standard time of each physical station, key actions, and core quality inspection points to complete operation action timing modeling; In this embodiment, the operation action atomization decomposition includes picking and placing, press fitting, screwing, and plugging to form a plurality of basic operation actions.

[0035] S334: Based on the operation action atomization decomposition process, perform operation action atomization combination to construct an atomized action library; In this embodiment, a plurality of basic actions are combined to form a plurality of operation action atomization combinations to construct an atomized action library. In actual application, a process rule library can be constructed in combination with a knowledge graph, and then cross-line process reuse is realized through federated transfer learning to reduce changeover time.

[0036] S34: Real-time identification and positioning of products, employees, and equipment, dynamic planning of picking paths combined with assigned order tasks, construction of a virtual work station simulation model of digital twin, real-time mapping of each work station operation state using a visual interactive terminal; S4: Product verification and operation detection based on the product state of the production line and the operation state of each physical work station, order task allocation optimization and employee operation trajectory optimization combined with the real-time operation state of each physical work station; S41: Product verification based on the product state of the production line and the operation state of each physical work station, including: S411: Real-time capture of product image information using a camera and product counting using a 3D vision system; S412: Product size measurement and product surface defect detection based on product image information; S413: Product category identification using a deep learning model based on product image information, and extraction of product attribute information for automatic verification; In this embodiment, product attribute information includes product color, product specification, and product packaging state.

[0037] S42: Operation detection based on the product state of the production line and the operation state of each physical work station, including: S421: Construction of a spatial coordinate system based on a 3D vision system, operation action frame-level analysis, and real-time skeletal key point recognition, product real-time positioning, and three-dimensional positioning verification of shelf location; In this embodiment, a hybrid 3D vision scheme of binocular structured light and TOF fusion is used to construct a spatial coordinate system with a precision of 0.05 mm, and a time sequence action segmentation network (TS-ASNet) is used for continuous operation action frame-level analysis.

[0038] S422: Real-time comparison of the standard operation process in the process knowledge graph based on the three-dimensional positioning verification results, and immediate correction strategies triggered by operation deviation; In this embodiment, a multi-industry standard process rule library is configured in the process graph knowledge base.

[0039] S423: Configuration of a standard operation procedure execution index algorithm for multi-dimensional quantitative evaluation of operation action standard, time sequence compliance rate, and tool compliance, recording of operation action quality data, and generation of a dynamic scoring report; In the embodiment, the state detection of product component integrity detection, product component consistency detection, misloading and missing detection, key action detection, and label pasting specification detection is performed; based on a time series algorithm, real-time logic compliance detection of each physical station process execution is performed; and through Google key point recognition technology, operation standard evaluation of pressing depth, screwing angle and other operations is performed.

[0040] S424: Based on product part feature matching, product part assembly verification is performed, and an assembly pressure thermal map is generated through torque distribution analysis; In the embodiment, SURF (Speeded Up Robust Features) algorithm and deep learning algorithm are used for product part feature matching to verify the component model, and then an assembly pressure thermal map is generated through torque distribution analysis to facilitate optimization of operation behavior.

[0041] S425: Real-time monitoring and picking error detection are performed, so that when a product picking error occurs, a three-level early warning of shelf location light strip flashing, voice prompt, and operation table vibration feedback is triggered in turn, and operation guidance is dynamically adjusted according to the early warning prompt to form a closed-loop feedback; In the embodiment, when a product picking error occurs, an error response is performed within 1.5 seconds, the shelf location light strip flashes red to prompt the product picking and assembly error, and a voice prompt is played synchronously to guide error correction. After no response, the operation table vibrates to prompt the staff to correct the error.

[0042] S43: Order task allocation optimization is performed in combination with the real-time operation state of each physical station, including: S431: Production data of each physical station is collected, and actual processing time PT of each physical station completing actual production tasks is recorded; S432: Based on the standard operation process and process requirements of the physical station, in combination with the production data of each physical station, the theoretical cycle time CT of each physical station is calculated by using a virtual station simulation model of digital twinning; S433: In combination with the theoretical cycle time CT and actual processing time PT of each station, a discrete event simulation (DES) technology is used to calculate the time deviation in real time; S434: Based on the time deviation, a dynamic scheduling algorithm is used to perform order task reassignment, so that the production line efficiency meets the preset deviation; In the embodiment, the dynamic scheduling algorithm uses an improved ant colony algorithm to perform order task reassignment, so that the entire production line efficiency fluctuation is controlled within [-3%, 3%].

[0043] By calculating the deviation of the processing time and the actual value and the theoretical value of each physical station, using a dynamic scheduling algorithm to reassign order tasks, the production efficiency can meet the preset deviation while realizing the balance of the whole line capacity, thereby improving the overall efficiency and stability of production.

[0044] S44: In combination with the real-time operation state of each physical station, the employee operation trajectory is optimized, including: S441: Based on the shelves, equipment, and operation area, in combination with the employee operation trajectory, a genetic algorithm is used to plan the initial employee operation trajectory; In this embodiment, the genetic algorithm uses an improved NSGA-II genetic algorithm, which considers the fitness function of three dimensions of walking distance, task balance degree, and time window compliance rate, generates a set of initial feasible employee operation trajectory schemes through genetic operations, and selects the employee operation trajectory with the optimal fitness as the initial employee operation trajectory.

[0045] S442: Based on fixed time intervals or significant changes in the warehouse environment, the employee operation trajectory re-planning mechanism is triggered, and according to the current order task situation, the wave merging strategy is used to merge order tasks, and the minimum completion time MCT scheduling strategy is used for load balancing, to realize order task allocation optimization and obtain the optimized employee operation trajectory; In this embodiment, based on fixed time intervals or significant changes in the warehouse environment such as new order task insertion, shelf product location change, etc., the employee operation trajectory re-planning mechanism is triggered, and the current order task state, product location, employee location, etc. Information is re-evaluated and distinguished, so that the product picking time of each order is shortened, and the employee walking distance is reduced.

[0046] In this embodiment, the fixed time interval is set to 15 seconds; in the wave merging strategy, the maximum number of merged orders is 500 per wave.

[0047] S5: Based on the verified products, an electronic order receipt is generated and stored in a blockchain; In this embodiment, after the system automatically completes the full-factor verification of products, product quantity, product specification, product packaging state, etc., the electronic order receipt is generated after integrating the handwriting signature recognition and RFID tag verification, realizing the paperless closed loop of delivery welding, and using the alliance chain architecture technology of Hyperledger Fabric technology to store the hash value, reducing the probability of tampering with the stored data; At the same time, the video files collected by the camera are automatically indexed according to the order number, which is convenient for quick positioning of events.

[0048] Embodiment 2 Based on the same technical concept, the embodiment provides a production method system of an intelligent work station based on an AI computer vision technology, which comprises a system architecture design of a perception layer structure, a network layer structure, a platform layer structure and an application layer structure, an end-edge-cloud collaborative intelligent work station system is constructed, and data collection, transmission, processing and application are realized, so that the modules work collaboratively, comprehensive and efficient technical support is provided for the production of the intelligent work station, and the intelligent level and production efficiency of the production line are improved.

[0049] Specifically, the perception layer structure is deployed with a cluster of intelligent terminal devices. The intelligent terminal devices comprise: a camera, which adopts a 3D structured light camera, is used for capturing product image information, supports product size measurement and product surface defect detection; a scanning head, which is used for OCR scanning and identification of paper documents to obtain paper order data; a pressure sensor, which is used for operation torque detection; and a UWB positioning module, which is used for regional-level spatial positioning of products a UWB positioning module, which is used for regional-level spatial positioning of products; a dynamic light guide module, which comprises an RGB intelligent light strip; the RGB intelligent light strip is installed on a shelf location, and the target product position is indicated in real time through the color and flashing frequency of the light strip; Specifically, the network layer structure comprises edge digital twin nodes and industrial control networks; the data of the perception layer structure is transmitted to the edge digital twin nodes through the industrial control networks; a plurality of digital twin nodes are arranged in the perception layer structure, and communicate with the intelligent terminal devices of the perception layer structure through the industrial control networks, to receive and process the data of the perception layer structure in real time, perform model construction, simulation and visual display; Specifically, the platform layer structure comprises: a digital twin module, which is used for generating a virtual work station simulation model of a digital twin from the perception layer structure data obtained from each edge digital twin node, and simulating based on an order task to realize order task allocation and employee operation track planning; an AI algorithm model library, which is deployed in the cloud and stores different AI algorithm models; the AI algorithm models comprise a YOLOv8 model and a genetic algorithm model; a blockchain storage module, which is used for realizing operation data encryption and tamper-proofing traceability; Specifically, the application layer structure receives the simulation results of the digital twin module and realizes specific business; the application layer structure comprises: an order module, which is used for extracting and analyzing order information and allocating order tasks to the digital twin module; a path planning module, which is used for planning an employee operation track; a visual quality inspection module for visual identification and verification of the product; a three-dimensional interactive guidance module including a touch screen and a haptic feedback operation platform; the touch screen is used for integrating five-screen display functions of electronic work orders, process drawings, real-time clocks, operation animations, and real-time detection results, and the haptic feedback operation platform is configured with a 6-axis force pressure sensor and used for triggering operation platform vibration early warning in case of abnormal operation; In the embodiment, the three-dimensional interactive guidance module further includes an AR glasses terminal and an AR auxiliary operation module, which are used for synchronously displaying order tasks and operation guidance animations; the AR auxiliary operation module is used for real-time projection of assembly animations and gesture recognition verification of key actions.

[0050] a video backtracking module for backtracking and quick positioning of operation behaviors; an electronic signing module for realizing paperless delivery and closing loop in the delivery link.

[0051] It should be noted that the modules of the application layer structure can be expanded and added in function according to actual needs. Meanwhile, based on different application environments, 200Lus, 500Lus, and 800Lus different environment lighting modes are set, and automatic switching is performed through a photosensitive sensor to realize intelligent hierarchical regulation and control; meanwhile, 5Lus precision compensation is realized through DLP projection technology to facilitate support for automatic light compensation in shadow areas.

[0052] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily mean the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0053] Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A production method of an intelligent station based on AI computer vision technology, characterized in that, The method comprises the following steps: Obtain order information, automatically extract order data, and generate structured order tasks; Based on the order task, the order task of the production line is allocated, and the equipment operation parameters are adjusted accordingly; Obtain physical station data, combine the allocated order task and the standard operation process on the physical station, and construct a virtual station simulation model of digital twin to real-time map the operation state of each station; Based on the product state of the production line and the operation state of each physical station, product verification and operation detection are performed, and order task allocation optimization and employee operation trajectory optimization are performed in combination with the real-time operation state of each physical station; Based on the verified product, an electronic order receipt is generated, and a blockchain is stored.

2. The intelligent station production method based on AI computer vision technology according to claim 1, characterized in that, The method comprises the following steps: An API interface is used to connect to the upper business system to directly obtain electronic order data; A scanning head is used to scan paper documents to identify and obtain paper order data; Based on electronic order data and paper order data, real-time fusion is performed to obtain order data; A BERT-NER natural language processing model is used to automatically extract order specification parameters and order delivery data from the order data, and to generate structured order tasks; the order specification parameters include product, product quantity, product size, product material, and product industry requirements; the order delivery data includes delivery time limit, product packaging, and gift level.

3. The intelligent station production method based on AI computer vision technology according to claim 1, characterized in that, The method comprises the following steps: Based on the order delivery data in the order task, the order delivery time limit is obtained; Based on the order delivery time limit, the order type is determined; the order type includes emergency order, regular order, and peak order; Based on the order type, the production line mode is selected, the order task allocation is completed, and the production line equipment operation parameters are adjusted accordingly; When it is determined that the current order is an emergency order, the manual mode is selected, the order task allocation is prioritized for manual intervention, and the production line equipment operation parameters are adjusted accordingly; When it is determined that the current order is a regular order, the AI automatic mode is selected, the order task allocation is automatically performed, and the production line equipment operation parameters are adjusted accordingly; When it is determined that the current order is a peak order, the mixed line mode is selected, multi-product co-line production is performed, product categories are automatically analyzed, and the production line equipment operation parameters corresponding to the product categories are switched.

4. The intelligent station production method based on AI computer vision technology according to claim 1, characterized in that, The method comprises the following steps: Based on the shelves, equipment, and operation area, the physical station is modeled; Based on the sensor data on the physical station, real-time collection of each sensor data is performed; Based on the allocated order task, the operation process on the physical station is modeled according to the standard operation process, and the OptiTrack motion capture system is used in combination with the YOLOv8 posture recognition model to record the employee operation trajectory, construct an operation trajectory digital twin, and decompose to form an atomized action library; Real-time identification and positioning of products, employees, and equipment, combined with assigned order tasks, dynamically plan the picking path, build a virtual work station simulation model of digital twin, and use visual interactive terminals to map the real-time work station operation status.

5. The intelligent station production method based on AI computer vision technology according to claim 4, characterized in that, Based on the assigned order tasks, the operation process on the physical work station is modeled according to the standard operation process, and the OptiTrack motion capture system is used in combination with the YOLOv8 pose recognition model to record the employee operation trajectory, build an operation trajectory digital twin, and decompose it into an atomic action library, including: Based on the assigned order tasks, manual picking is performed according to the standard operation process, and a 3D vision system is used to record the operation trajectory of excellent employees; Based on the operation trajectory of employees, an OptiTrack motion capture system is used in combination with a YOLOv8 pose recognition model to build an operation trajectory digital twin; Based on the operation trajectory digital twin, operation action atomic decomposition is performed, and the standard time, key actions, and core quality inspection points of each physical work station are labeled to complete operation action timing modeling; the operation action atomic decomposition includes picking and placing, pressing, screwing, and plugging; Based on the operation action atomic decomposition process, operation action atomic combination is performed to build an atomic action library.

6. The intelligent station production method based on AI computer vision technology according to claim 1, characterized in that, Based on the product state and the operation state of each physical work station on the production line, product verification is performed, including: Using a camera to capture product image information in real time, and using a 3D vision system to count products; Based on product image information, product size measurement and product surface defect detection are performed; Based on product image information, a deep learning model is used for product category recognition and product attribute information extraction for automatic verification; the product attribute information includes product color, product specification, and product packaging state.

7. The intelligent station production method based on AI computer vision technology according to claim 1, characterized in that, Based on the product state and the operation state of each physical work station on the production line, operation detection is performed, including: Based on the 3D vision system, a spatial coordinate system is built for operation action frame-level analysis, and real-time skeletal key point recognition, product real-time positioning, and three-dimensional positioning verification of shelf location are performed; Based on the three-dimensional positioning verification results, the standard operation process in the process knowledge graph is compared in real time, and the operation deviation triggers an immediate correction strategy; Configure the standard operation program execution index algorithm to perform multi-dimensional quantitative evaluation of operation action standard, timing compliance, and tool compliance, record operation action quality data, and generate a dynamic scoring report; Based on product part feature matching, product part assembly verification is performed, and an assembly pressure thermal map is generated through torque distribution analysis; Real-time monitoring and picking error detection are performed so that when a product is picked incorrectly, a three-stage early warning of shelf location light strip flashing, voice prompt, and operation table vibration feedback is triggered in sequence, and the operation guidance is dynamically adjusted according to the warning prompt to form a closed-loop feedback.

8. The intelligent station production method based on AI computer vision technology according to claim 1, characterized in that, The combination of real-time operation status of each physical work station optimizes order task allocation, including: Collect production data from each physical work station and record the actual processing time of each physical work station to complete the actual production task; Based on the physical station standard operation process and process requirements, combined with the production data of each physical station, the theoretical cycle time of each physical station is calculated by using the virtual station simulation model of digital twinning; Combined with the theoretical cycle time and actual processing time of each station, the time deviation is calculated; Based on the time deviation, a dynamic scheduling algorithm is used for order task reassignment, so that the production line efficiency meets the preset deviation.

9. The intelligent station-based production method of claim 1, wherein The combination of real-time operation state of each physical station is used to optimize the employee operation trajectory, including: Based on the shelves, equipment and operation area, combined with the employee operation trajectory, a genetic algorithm is used for initial employee operation trajectory planning; Based on fixed time interval or significant change in warehouse environment, trigger employee operation trajectory re-planning mechanism, and according to current order task situation, use wave merging strategy for order task merging, and based on minimum completion time MCT scheduling strategy for load balancing, realize order task allocation optimization, and get optimized employee operation trajectory.

10. A system for implementing the production method of the intelligent station based on the AI computer vision technology according to claim 1, characterized in that, It includes: A perception layer structure is deployed with a cluster of intelligent terminal devices; the intelligent terminal devices include a camera, a scanning head, a pressure sensor, a UWB positioning module, and a dynamic light guide module; the camera is used to capture product image information; the scanning head is used to scan paper order information; the pressure sensor is used to detect the operation torque; the UWB positioning module is used for regional level space positioning of products; the dynamic light guide module is installed on the shelf location, which is used to indicate the product position in real time; A network layer structure includes edge digital twinning nodes and industrial control networks; the perception layer structure data is transmitted to the edge digital twinning nodes through the industrial control network; multiple digital twinning nodes are arranged in the perception layer structure and communicate with the intelligent terminal devices of the perception layer structure through the industrial control network, and real-time receive and process the perception layer structure data, and construct, simulate and visualize the model; A platform layer structure includes a digital twinning module, an AI algorithm model library, and a blockchain storage module; The digital twinning module is used to generate a virtual station simulation model of digital twinning from the perception layer structure data of each edge digital twinning node, and simulate based on order tasks to realize order task allocation and employee operation trajectory planning; the AI algorithm model library is deployed in the cloud and stores different AI algorithm models; the blockchain storage module is used to realize operation data encryption and tamper-proofing; An application layer structure receives the simulation results of the digital twinning module and realizes specific business; the application layer structure includes an order module, a path planning module, a visual quality inspection module, a three-dimensional interactive guidance module, a video backtracking module, and an electronic signature module; the order module is used to extract and analyze order information and assign order tasks to the digital twinning module; the path planning module is used to plan the employee operation trajectory; the visual quality inspection module is used for visual identification and verification of products; The three-dimensional interaction guiding module comprises a touch screen and a tactile feedback operation table, the touch screen is used for integrating five-screen display functions of electronic work orders, process drawings, real-time clocks, operation animations and real-time detection results, and the tactile feedback operation table is used for operation table vibration early warning reminding. The video backtracking module is used for backtracking and rapid positioning of operation behaviors, and the electronic signing module is used for realizing paperless delivery closed loop in the delivery link.

Images