Auxiliary material dispensing plate pre-performance method and system based on digital twinning

Abstract

The present application relates to the field of auxiliary material tray, and in particular to an auxiliary material tray pre-rehearsal method and system based on digital twinning, which collects real-time data of a physical environment by deploying a sensor network, and constructs a synchronous high-fidelity digital twin; generates multiple tray schemes based on an improved genetic algorithm; pre-rehearses the schemes in parallel in the digital twin and evaluates them through a multi-dimensional constraint rule base; selects an optimal scheme using a multi-objective optimization algorithm; compiles the optimal scheme into device executable instructions; and compares the physical and virtual states in real time during execution to realize closed-loop verification and dynamic correction. The present application realizes virtual pre-rehearsal and optimization of tray schemes by constructing a digital twin, can discover and avoid physical conflicts and efficiency bottlenecks in advance, and effectively solves the problems of relying on artificial experience, being unable to predict conflicts, information fragmentation, and simulation and control disconnection in the prior art.

Description

Technical Field

[0001] This invention relates to the field of auxiliary material preparation, specifically to an auxiliary material preparation method and system based on digital twins. Background Technology

[0002] In discrete manufacturing, material palletizing currently relies primarily on the personal experience of schedulers for manual planning. This approach has significant technical limitations when faced with application scenarios involving a wide variety of materials, varying sizes and shapes, and complex stacking constraints. Due to the lack of a scientific and systematic planning method, it is difficult for manual methods to quickly develop optimal palletizing schemes that achieve the best balance in terms of space utilization, center of gravity stability, and operational convenience. More seriously, existing technologies lack the ability to conduct comprehensive virtual verification before actual implementation, failing to anticipate potential physical conflicts such as unstable material stacking, interference between equipment, and obstructed transport paths. These problems often only surface during actual operation, leading to work interruptions, rework, and even safety accidents. While some manufacturing companies have attempted to introduce simulation technology to assist planning, existing simulation systems are mostly offline static simulations, severely disconnected from the real-time changing physical environment. They cannot acquire real-time production line status data, nor can they directly translate simulation results into executable control commands for equipment, creating a technological gap between virtual planning and physical execution. This current technological situation results in palletizing operations consistently facing prominent problems such as low efficiency, insufficient reliability, and high safety risks, severely hindering the overall improvement of intelligent manufacturing efficiency. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for pre-simulating auxiliary material batching based on digital twins, so as to solve the problems of unscientific planning, unpredictable physical conflicts, information fragmentation between systems, and disconnect between simulation and real-time control caused by reliance on human experience in the existing auxiliary material batching process.

[0004] To achieve the above objectives, the following technical solution is adopted.

[0005] The method for pre-simulating ingredient batching based on digital twins includes the following steps:

[0006] S1. Real-time data of auxiliary material pallets, automated guided vehicles and shelves are collected through a sensor network deployed in the physical environment. A high-fidelity digital twin synchronized with the physical environment is constructed based on a lightweight incremental update algorithm to obtain a real-time virtual scene.

[0007] S2. Receive the palletizing task from the production management system using the current state of the real-time virtual scene, and generate multiple initial palletizing schemes encoded with material picking order, placement position and transportation path based on the improved genetic algorithm.

[0008] S3. In the digital twin, the multiple initial disk configuration schemes are pre-performed in parallel. A comprehensive evaluation is performed by integrating a multi-dimensional constraint rule base of geometric constraints, physical constraints and business constraints to obtain comprehensive evaluation data for each scheme.

[0009] S4. Based on the comprehensive evaluation data, a multi-objective optimization algorithm is used to select the optimal disk allocation scheme from the multiple initial disk allocation schemes;

[0010] S5. Input the optimal disk configuration scheme into the instruction compiler, which will automatically parse and compile it into a sequence of low-level control instructions that can be directly executed by the execution device;

[0011] S6. During the execution of the underlying control command sequence, the feedback data from the physical sensors is compared with the expected state of the digital twin in real time. Once a deviation exceeding the tolerance is detected, a dynamic replanning program is triggered to generate an adjustment scheme, thereby achieving closed-loop verification and correction.

[0012] Optionally, step S1 specifically includes:

[0013] By collecting real-time position and attitude data of auxiliary material pallets, automated guided vehicles and shelves through ultra-wideband positioning tags and inertial measurement unit sensors deployed in the physical environment, a multi-sensor fusion algorithm is used to fuse the absolute position data provided by the ultra-wideband positioning tags and the relative motion data provided by the inertial measurement unit sensors. By correcting the measurement error of individual sensors through data complementarity, the precise three-dimensional coordinates and spatial orientation of each tracked entity in virtual space are calculated to obtain high-precision real-time pose information.

[0014] Meanwhile, by collecting the identification data of objects through radio frequency identification devices deployed at key nodes in the warehouse, the identification data is spatiotemporally correlated and mapped with real-time pose information to establish an accurate correspondence between the virtual model and the physical entity.

[0015] Based on real-time pose information and correspondence, a lightweight 3D engine incremental update algorithm is used to continuously monitor the state changes of physical data. By setting a state change recognition mechanism, the difference between physical data and the stored state of the virtual world is detected. Only the target virtual objects whose state has changed are updated in a directional pose, while the virtual objects whose state has not changed remain unchanged. A high-fidelity digital twin that is highly consistent with the real environment is constructed, resulting in a virtual scene that can accurately reflect the real-time state of the physical environment.

[0016] Optionally, step S2 specifically includes:

[0017] Utilizing the current state of the real-time virtual scene provided by the digital twin, receive palletizing tasks from the manufacturing execution system or warehouse management system, and obtain material type, quantity information, and current available resource status data from the task list;

[0018] Multiple initial palletizing schemes are generated based on an improved genetic algorithm. The improved genetic algorithm encodes the picking order of materials, their placement and orientation on the pallet, and the transportation path of the automated guided vehicle into a chromosome data structure. It adopts a diversity initialization strategy to generate an initial population and introduces an adaptive genetic operator to dynamically adjust the crossover and mutation probabilities according to the evolutionary state of the population in order to balance global search and local search capabilities.

[0019] The improved genetic algorithm is used to iteratively optimize the initial population. In each iteration, excellent individuals are selected based on the fitness evaluation results, and a new generation of population is generated through adaptive genetic operators. Finally, multiple initial mating schemes with differences are output.

[0020] Optionally, step S3 includes:

[0021] In the digital twin, multiple initial disk configuration schemes are simultaneously simulated using a parallel computing architecture, and each scheme is comprehensively evaluated through a multi-dimensional constraint rule base;

[0022] Among them, geometric constraints simulate the interference between materials, between materials and vehicles, and between automated guided vehicles and the environment in a virtual environment by integrating continuous collision detection algorithms, and detect and record collision events in real time during the execution of the scheme.

[0023] Physical constraints, by defining the weight attributes of materials and setting a center of gravity offset threshold, calculate the dynamic center of gravity position of the pallet in real time in the virtual environment, and determine whether the center of gravity projection exceeds the stable area of ​​the pallet, thus marking placement schemes that may lead to tipping over.

[0024] Business constraints assess the business rationality of a solution by calculating space utilization, operation time, and material correlation. Space utilization is assessed by calculating the ratio of the total projected area of ​​placed materials to the available area of ​​the pallet. Operation time is based on a predefined motion time model to accumulate the total operation time of the solution. Material correlation is assessed by logically binding related materials in the same production work order and calculating the proximity of related materials in their placement positions.

[0025] By integrating geometric constraints, physical constraints, and business constraints, comprehensive evaluation data on the geometric feasibility, physical stability, and business rationality of each solution are obtained.

[0026] Optionally, step S4 includes:

[0027] Based on the comprehensive evaluation data obtained in step S3, a multi-objective optimization algorithm is used to select the optimal disk allocation scheme from multiple initial disk allocation schemes. The multi-objective optimization algorithm includes the weighted scoring method and the Pareto optimal front analysis method. The weighted scoring method assigns different weight values ​​to multiple evaluation dimensions such as space utilization, operation time, balance and collision-free performance. The comprehensive score of each scheme is calculated by linear weighted summation and sorted according to the score.

[0028] Pareto optimal front analysis identifies non-dominated solutions that do not have an absolute superiority-inferiority relationship among multiple optimization objectives, forming a Pareto optimal solution set containing multiple schemes; the system selects the final optimal scheme from the ranking results obtained by the weighted scoring method or the Pareto optimal solution set according to the optimization preferences set by the user.

[0029] Optionally, step S5 includes: inputting the optimal pallet arrangement scheme selected in step S4 into the instruction compiler, which has a built-in extensible device instruction driver library; the instruction compiler first parses the high-level task description in the optimal pallet arrangement scheme, decomposing high-level operations such as material handling and placement into basic operation units; then, according to the predefined instruction templates in the device instruction driver library, the basic operation units are compiled into a sequence of low-level control instructions that can be directly executed by specific devices, wherein the control instructions of the automated guided vehicle include a sequence of navigation path point coordinates, motion speed parameters, and obstacle avoidance strategies, and the control instructions of the robotic arm include grasping posture parameters, joint motion trajectories, and end effector action instructions; the compiled low-level control instruction sequence is sent to the corresponding physical execution device through the industrial network, and an instruction status tracking mechanism is established to provide an accurate instruction execution benchmark for subsequent closed-loop verification.

[0030] Optionally, step S6 includes: during the execution of the underlying control command sequence, continuously collecting real-time status data of the automated guided vehicle, robotic arm, and auxiliary material tray through a physical sensor network, and comparing these physical sensor feedback data with the expected state of the virtual device in the digital twin in real time; when a deviation exceeding the allowable range is detected between the actual position of the automated guided vehicle and the expected path, or when the device stops for more than a preset time threshold due to obstacles or other reasons, the system triggers a dynamic replanning program; the dynamic replanning program uses the current actual state of the physical world as the new initial condition, quickly re-previews and generates a local adjustment scheme or a global backup scheme in the digital twin, and inputs the adjustment scheme into the instruction compiler to compile it into a new control command sequence and send it to the physical device.

[0031] A digital twin-based auxiliary material batching simulation system includes:

[0032] The physical sensing layer includes ultra-wideband positioning tags and inertial measurement unit sensors deployed on auxiliary material pallets, automated guided vehicles and shelves, as well as industrial network equipment deployed at key nodes in the workshop, for collecting real-time motion data of various entities in the physical environment and transmitting it in real time through the industrial network.

[0033] The digital twin construction module is connected to the physical perception layer, receives the real-time motion data, processes the real-time motion data using a multi-sensor fusion algorithm to obtain accurate pose information, and uses a lightweight 3D engine incremental update algorithm to construct a high-fidelity digital twin that is synchronized with the physical environment.

[0034] The parallel pre-simulation and optimization engine module is connected to the digital twin construction module. It receives the disk allocation task from the production management system and the current state of the digital twin. Based on the improved genetic algorithm, it generates multiple initial disk allocation schemes, pre-simulates the schemes in parallel in the digital twin, and evaluates them through a multi-dimensional constraint rule base.

[0035] The instruction compilation module, connected to the parallel pre-simulation and optimization engine module, has a built-in extensible device instruction driver library that automatically parses and compiles the optimal disk matching scheme into a sequence of low-level control instructions that can be executed by the automated guided vehicle and the robotic arm.

[0036] The closed-loop verification and dynamic correction module is connected to the instruction compilation module and the physical perception layer, respectively. It compares the feedback data from the physical sensors with the expected state of the digital twin in real time, and triggers a dynamic replanning program when an excessive deviation is detected.

[0037] Optionally, the physical sensing layer also includes radio frequency identification (RFID) devices deployed at warehouse entrances and exits, shelving areas, and palletizing stations, used to collect object identification data and associate it with pose information;

[0038] The system adopts an edge-cloud collaborative computing architecture, in which the digital twin building module and instruction compilation module are deployed on the edge server, which are responsible for handling data synchronization and instruction compilation tasks with high real-time requirements, and the parallel pre-drilling and optimization engine module is deployed on the cloud server, which are responsible for executing computationally intensive scheme generation and optimization tasks.

[0039] Optionally, the parallel pre-simulation and optimization engine module includes:

[0040] The dimensional constraint rule library stores evaluation rules for geometric constraints, physical constraints, and business constraints. Geometric constraints integrate continuous collision detection algorithms, physical constraints define centroid stability evaluation criteria, and business constraints include space utilization, operation time, and material correlation calculation logic.

[0041] An improved genetic algorithm unit is implemented to realize a scheme generation mechanism based on adaptive genetic operators. The mating scheme is represented by chromosome encoding, and a diversity preservation strategy is used to initialize the population.

[0042] Parallel computing units utilize multi-core processors to concurrently execute simulation evaluations of multiple schemes within a digital twin;

[0043] The multi-objective optimization unit uses a weighted scoring method and Pareto optimal frontier analysis to process the evaluation results and output the optimal solution set.

[0044] The closed-loop verification and dynamic correction module includes:

[0045] The data comparison unit receives data from physical sensors in real time and compares it with the expected state of the digital twin within tolerance limits.

[0046] The multi-level response unit sequentially activates audible and visual warnings, equipment pause, and dynamic replanning when an excessive deviation is detected.

[0047] The dynamic replanning unit regenerates the adjustment plan based on the current physical state and issues new instructions through the instruction compilation module.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] This application, by constructing a high-fidelity digital twin and establishing a complete virtual pre-simulation and closed-loop control process, achieves comprehensive pre-simulation and optimization of various palletizing schemes in a virtual environment. It can proactively identify and avoid potential physical conflicts and efficiency bottlenecks, effectively solving the problems of unscientific planning, unpredictable physical conflicts, information fragmentation between systems, and disconnect between simulation and real-time control caused by reliance on manual experience in existing palletizing processes. Multi-sensor data fusion and a lightweight incremental update algorithm ensure high synchronization between the digital twin and the physical environment, providing an accurate virtual environment for pre-simulation. An improved genetic algorithm generates diverse initial schemes, providing rich candidates for optimization. A multi-dimensional constraint rule base comprehensively evaluates the feasibility of schemes from geometric, physical, and business perspectives. A multi-objective optimization algorithm ensures that the final scheme achieves the best balance among multiple competing objectives. Automatic instruction compilation technology connects the key links from virtual decision-making to physical execution. Real-time closed-loop verification and dynamic correction mechanisms endow the system with adaptive capabilities to cope with environmental uncertainties. This application establishes a complete closed-loop control system from virtual space to physical space, significantly improving the overall efficiency, reliability, and safety of palletizing operations. Attached Figure Description

[0050] Figure 1 This is a schematic flowchart of an embodiment of the auxiliary material preparation and pre-simulation method based on digital twins according to the present invention.

[0051] Figure 2 This is a schematic diagram of a module of an embodiment of the auxiliary material batching and pre-simulation system based on digital twins according to the present invention. Detailed Implementation

[0052] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0053] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0054] Example 1

[0055] like Figure 1 As shown in Example 1, a method for pre-simulating auxiliary material batching based on digital twins.

[0056] This embodiment is applied to auxiliary material traying operations in discrete manufacturing scenarios, such as the selection, matching, and transportation of auxiliary materials in electronic assembly or automobile manufacturing workshops. It utilizes digital twin technology to achieve virtual pre-simulation of the traying scheme, multi-objective optimization, and closed-loop control of physical execution. The specific implementation process is as follows:

[0057] First, data acquisition of the physical environment and the construction of a high-fidelity digital twin are carried out. Within the physical space involved in the palletizing operation, sensing devices are deployed for all entities requiring tracking: ultra-wideband (UWB) positioning tags and inertial measurement unit (IMU) sensors are installed on pallets of auxiliary materials awaiting palletizing, automated guided vehicles (AGVs) used for transportation, forklifts assisting in handling, and shelves storing auxiliary materials. The UWB positioning tags are used to acquire the absolute position information of the entities, while the IMU sensors are used to collect the motion attitude data of the entities, including acceleration and angular velocity, to achieve real-time perception of the entity's position and attitude. Simultaneously, industrial network equipment is deployed at key nodes in the workshop (such as warehouse entrances and exits, next to the palletizing station, and turns in the AGV's travel path). 5G CPEs or industrial Wi-Fi 6 base stations can be selected to build a low-latency, highly reliable industrial communication network, ensuring that the massive amounts of sensor data collected by UWB and IMU can be transmitted back to the data processing unit in real time.

[0058] The returned sensor data is fused and analyzed to establish an accurate correspondence between the virtual model and the physical entity. A multi-sensor fusion algorithm (such as Kalman filtering) is used to fuse the absolute position data provided by UWB positioning tags with the relative motion data provided by IMU sensors. This data complementarity corrects for measurement errors from individual sensors—for example, when UWB signals drift due to obstruction, the relative displacement data from the IMU is used for compensation; when the IMU accumulates errors due to prolonged operation, it is calibrated using the absolute position data from UWB. Ultimately, the precise three-dimensional coordinates and spatial orientation of each tracked entity in the virtual space are calculated. If RFID devices are deployed in the workshop (e.g., RFID readers are placed at warehouse entrances / exits, shelf areas, and palletizing stations, and RFID tags are affixed to pallets and material boxes), the object identification data collected by RFID (including material type, batch, attributes, etc.) is spatiotemporally mapped with the calculated real-time pose information to ensure that each model in the virtual scene accurately corresponds to a specific entity in the physical world.

[0059] Based on the processed data, a high-fidelity digital twin synchronized with the physical environment is constructed using a lightweight 3D engine incremental update algorithm. First, based on the actual layout of the workshop (e.g., the location and dimensions of walls, columns, shelves, aisles, and workstations), a virtual scene containing all static elements is constructed using a real-time 3D engine. 3D models with dimensions and appearances consistent with the physical entities are created for AGVs, pallets, and various auxiliary materials. During virtual-real synchronization, instead of fully refreshing the entire 3D scene every frame, the currently received physical data is continuously compared with the previous state stored in the virtual world. Only changed target objects and their attributes (e.g., entity position coordinates, rotation angles, and working status indicators) are identified and transmitted. Oriented pose updates are performed on these changed virtual objects, while the original data remains unchanged for virtual objects whose states have not changed. In this way, the synchronization delay between the physical world and the digital twin is controlled within a preset range, ensuring a high degree of consistency between the real-time virtual scene and the real physical environment.

[0060] Upon receiving a palletizing task from a production management system (such as a Manufacturing Execution System (MES) or a Warehouse Management System (WMS), multiple initial palletizing schemes are generated based on the current state of the real-time virtual scene provided by the digital twin. The palletizing task includes the type and quantity of the materials to be palletized, the corresponding production order information, and the current status of available resources (such as the idle status of AGVs, the working status of robotic arms, and the storage location of materials on the shelves). An improved genetic algorithm is used to generate the initial schemes: each palletizing scheme is encoded as a "chromosome" data structure, which includes the picking order of materials (the order in which materials are picked from the shelves), the placement and orientation of materials on the pallet (such as the specific placement point within the X / Y coordinate area of ​​the pallet and the rotation angle of the materials), and the AGV's transportation path (the route from the shelf to the palletizing station). When generating the initial population, a diversity initialization strategy is adopted to avoid excessive concentration of initial schemes. An adaptive genetic operator is introduced to dynamically adjust the crossover probability and mutation probability according to the evolutionary state of the population. When the population diversity is low, the mutation probability is increased to increase the scheme differentiation. When the population tends to converge, the mutation probability is reduced to stabilize the excellent scheme. By iteratively optimizing and balancing the global search and local search capabilities, dozens or even hundreds of initial disk matching schemes with differences are finally output.

[0061] Multiple initial disk configuration schemes are simulated in parallel within a digital twin, and a comprehensive evaluation is performed using a multi-dimensional constraint rule base. Utilizing multi-core computing resources deployed in the workshop, all initial schemes are rapidly simulated simultaneously in a virtual scenario. During the simulation, geometric, physical, and business constraints are rigorously applied to evaluate each scheme in real time. The implementation of geometric constraints involves: establishing accurate 3D models and simplified collision bodies for each material and carrier (pallet, AGV); integrating a continuous collision detection (CCD) algorithm to simulate motion interference between materials, between materials and carriers, and between AGVs and the workshop environment (such as walls, columns, and other equipment) during the execution of the scheme; detecting and recording collision events in real time; and marking the scheme as non-compliant with geometric constraints if a collision occurs. The implementation of physical constraints involves: predefining the weight attributes of each material; setting a center of gravity offset threshold for the pallet; calculating the overall dynamic center of gravity of the pallet after it carries materials in real time during the simulation; and determining whether the projection of the center of gravity on the horizontal plane exceeds the stable area of ​​the pallet (such as the area formed by the inward contraction of the pallet edge); if it does, marking the placement scheme as potentially causing overturning. The implementation of business constraints includes three aspects: first, calculating space utilization by evaluating the utilization efficiency of pallet space through the ratio of the total projected area of ​​placed materials to the usable area of ​​the pallet; second, statistically analyzing operation time based on predefined action time models (such as the standard time for manually picking up an auxiliary material, the action time for a robotic arm to grasp materials, and the time for AGVs to...). The evaluation criteria include: 1) the time taken to travel a unit distance; 2) the total operation time from material picking to placement for each solution; and 3) the assessment of material correlation, which involves logically binding strongly related materials from the same production order (such as screws and washers required to assemble a component) and calculating the proximity of these related materials on the pallet. The lower the proximity, the lower the material correlation score for the solution. By combining the evaluation results of the above three types of constraints, a comprehensive evaluation of each initial palletizing solution in terms of geometric feasibility, physical stability, and business rationality is obtained.

[0062] Based on comprehensive evaluation data, a multi-objective optimization algorithm is used to select the optimal pallet arrangement scheme from multiple initial schemes. The multi-objective optimization algorithm includes a weighted scoring method and a Pareto optimal front analysis method: In the weighted scoring method, weights are assigned to different evaluation dimensions according to the actual needs of the workshop (such as space utilization, operation time, physical balance, collision-free operation, etc.), and the comprehensive score of each scheme is calculated by linear weighted summation, and the schemes are ranked from highest to lowest score; In the Pareto optimal front analysis method, "non-dominated solutions" are identified where there is no absolute superiority or inferiority relationship among multiple optimization objectives—that is, no other scheme is superior to this one in all objectives. These non-dominated solutions form the Pareto optimal solution set. The system determines the final optimal pallet arrangement scheme from the weighted scoring ranking results or the Pareto optimal solution set based on the user-defined optimization preferences (such as prioritizing the scheme with the shortest operation time during peak production seasons, and prioritizing the scheme with the best physical balance when materials are prone to tipping over).

[0063] The optimal disk configuration is input into the instruction compiler, which automatically parses and compiles it into a sequence of low-level control instructions that can be directly executed by the device. The instruction compiler has a built-in extensible device instruction driver library, which includes the instruction formats and communication protocols of mainstream AGV, robotic arms, and automated equipment brands. The compilation process consists of two steps: First, the high-level task descriptions in the optimal palletizing scheme (such as "moving a certain type of auxiliary material from a designated shelf to a pallet at a specific palletizing station" or "placing the auxiliary material at a designated position on the pallet") are parsed and decomposed into a series of basic operation units (such as "the AGV moves from its current position to the target shelf", "the robotic arm grabs the auxiliary material from the shelf", "the AGV carries the auxiliary material to the palletizing station", or "the robotic arm places the auxiliary material at a designated position on the pallet"). Second, based on the predefined instruction templates in the device instruction driver library, each basic operation unit is converted into the corresponding low-level control instructions for the physical device. For example, for the AGV, navigation path point coordinate sequences, driving speed parameters, and obstacle avoidance trigger conditions are generated; for the robotic arm, grasping pose parameters (such as the three-dimensional coordinates and orientation of the end effector), joint motion trajectories (rotation angle and speed of each joint), and end effector action instructions (such as grasping force and opening / closing time) are generated. The compiled low-level control instruction sequence is sent to the corresponding execution device through the industrial network, and an instruction status tracking mechanism is established to record the issuance time, expected execution duration and execution status of each instruction, providing a benchmark for subsequent closed-loop verification.

[0064] During the execution of the underlying control command sequence, closed-loop verification and dynamic correction are performed in real time. The sensor network of the physical perception layer continuously collects real-time status data of the AGV, robotic arm, and auxiliary material tray (such as the actual position of the AGV, the joint angle of the robotic arm, and the change of the center of gravity of the tray), and compares these physical sensor feedback data with the expected state of the virtual device in the digital twin (the motion trajectory and state changes of the virtual device generated based on the issued commands) in real time. When a deviation exceeding the preset tolerance is detected (e.g., the deviation between the actual position of the AGV and the expected path exceeds a set threshold, or the equipment stagnates for more than a preset time due to dynamic obstacles), the system immediately activates a multi-level response mechanism: First, it triggers an audible and visual alert to notify on-site personnel; if the deviation persists or shows an increasing trend, it automatically suspends the operation of the relevant equipment to prevent collisions or overturning accidents; simultaneously, it initiates a dynamic replanning program—this program uses the current physical world's actual state (e.g., the current position of the AGV, the real-time position of obstacles, and the remaining quantity of materials) as new initial conditions, and quickly re-enacts local adjustment schemes (e.g., adjusting the AGV's travel path to bypass obstacles) or global backup schemes in the digital twin, inputs the generated adjustment scheme into the instruction compiler to compile it into a new sequence of control instructions, and sends it to the physical equipment to correct the execution deviation, ensuring that the palletizing task is ultimately completed smoothly.

[0065] Example 2

[0066] like Figure 2 As shown, this embodiment provides a digital twin-based auxiliary material batching pre-simulation system to implement the batching pre-simulation method in Embodiment 1 above. This system is deployed in a discrete manufacturing workshop and works collaboratively with the workshop's existing production management system (MES / WMS). Specifically, it includes a physical sensing layer, a digital twin construction module, a parallel pre-simulation and optimization engine module, an instruction compilation module, and a closed-loop verification and dynamic correction module. The composition and function of each module are as follows:

[0067] The physical sensing layer is the foundation for the system's interaction with the physical world, responsible for collecting real-time data on entities related to palletizing operations and transmitting it to upper-layer modules. Its components include sensing devices and industrial network equipment: Sensing devices include UWB positioning tags and IMU sensors installed on auxiliary material pallets, AGVs, forklifts, and shelves, as well as RFID readers deployed at warehouse entrances / exits, shelf areas, and palletizing workstations. UWB positioning tags and IMU sensors work together to achieve real-time perception of entity positions and orientations. RFID readers are used to collect identification data (including material attributes, batch information, etc.) from pallets and material boxes. Some workshops may also add industrial cameras as needed to assist in verifying the placement of materials and the operational status of equipment. Industrial network equipment includes 5G CPEs or industrial Wi-Fi 6 base stations deployed at key nodes in the workshop, building a low-latency, highly reliable wireless communication network to ensure that the massive amounts of data collected by sensing devices (such as UWB location data, IMU orientation data, and RFID identification data) can be transmitted in real-time and stably to subsequent data processing modules, avoiding impacts on system real-time performance due to data transmission delays or losses.

[0068] The digital twin construction module connects to the physical sensing layer. Its core function is to map physical entities and states into a virtual scene, generating a high-fidelity digital twin synchronized with the physical environment. This module includes a data fusion unit and a lightweight incremental update unit: The data fusion unit receives raw sensor data transmitted from the physical sensing layer and uses multi-sensor fusion algorithms (such as Kalman filtering) to fuse the absolute position data of the UWB and the relative motion data of the IMU. By using data complementarity to correct the measurement errors of a single sensor, it calculates the precise three-dimensional coordinates and spatial orientation of each tracked entity in the virtual space. At the same time, this unit performs spatiotemporal correlation mapping between the identification data collected by RFID and the calculated real-time pose information to ensure that the virtual model corresponds one-to-one with the physical entity, avoiding identity mismatch or pose deviation. The lightweight incremental update unit, based on the real-time pose information output by the data fusion unit and the correspondence between the entities, uses a 3D engine to construct a virtual scene. First, it creates 3D models of static elements (walls, columns, shelves, aisles) according to the actual layout of the workshop, and creates dynamic models for AGVs, pallets, and auxiliary materials that are consistent with the physical entities. During the virtual-real synchronization process, this unit does not use a full scene refresh method, but continuously compares the current physical data with the historical state of the virtual world. It only identifies and updates target objects whose states have changed and their attributes (such as position coordinates, rotation angle, and working status), while keeping the data unchanged for objects whose states have not changed. This reduces the computational and communication burden, controls the virtual-real synchronization delay within a preset range, and ensures that the virtual scene can accurately reflect the real-time state of the physical environment.

[0069] The parallel pre-simulation and optimization engine module is the core of the system's intelligent decision-making, connecting the digital twin construction module and the production management system. It is responsible for generating initial pallet allocation schemes, conducting parallel pre-simulation evaluations, and selecting the optimal solution. This module includes a multi-dimensional constraint rule library, an improved genetic algorithm unit, a parallel computing unit, and a multi-objective optimization unit. The multi-dimensional constraint rule library stores evaluation rules for geometric, physical, and business constraints. The geometric constraint rules integrate a continuous collision detection (CCD) algorithm, defining collision judgment criteria between materials, between materials and carriers, and between equipment and the environment. The physical constraint rules define the weight attribute data of materials and the center-of-gravity offset threshold of the pallet, clarifying the judgment conditions for the center-of-gravity projection exceeding the stable area. The business constraint rules include the calculation logic for space utilization (the ratio of the projected area of ​​placed materials to the usable area of ​​the pallet), the statistical logic for operation time consumption (a total time accumulation method based on the action time model), and the evaluation logic for material correlation (the calculation method for the placement proximity of related materials in the same work order). An improved genetic algorithm unit is used to generate initial palletizing schemes: the palletizing scheme is encoded as a "chromosome" (containing picking order, placement position, and transportation path), an initial population is generated using a diversity initialization strategy, and the population is iteratively optimized using adaptive genetic operators (dynamically adjusting crossover and mutation probabilities) to output multiple differentiated initial schemes. A parallel computing unit utilizes multi-core processor resources to simultaneously perform rapid simulations of multiple initial schemes in a virtual scene generated by the digital twin construction module. During the simulation, a multi-dimensional constraint rule library is invoked to evaluate each scheme in real time, recording the evaluation results of geometric feasibility, physical stability, and business rationality. A multi-objective optimization unit receives the evaluation data from the parallel computing unit, processes the data using a weighted scoring method (assigning weights to evaluation dimensions and calculating a comprehensive score) and Pareto optimal front analysis (identifying non-dominated solutions to form an optimal solution set), and outputs a set of optimal palletizing schemes based on user optimization preferences for the user to choose the final scheme.

[0070] The instruction compilation module connects to the parallel pre-simulation and optimization engine module, responsible for converting the optimal palletizing scheme into low-level instructions executable by physical devices. The core of this module is a built-in extensible device instruction driver library, which stores instruction formats, communication protocols, and control parameters for mainstream AGVs, robotic arms, and automated equipment. The library's content can be expanded based on new equipment types added to the workshop. Upon receiving the finalized optimal palletizing scheme, the instruction compilation module first parses the high-level task description in the scheme, decomposing it into a series of independent basic operation units (such as AGV movement, robotic arm grasping, and material placement). Then, based on predefined templates in the device instruction driver library, it compiles each basic operation unit into the corresponding low-level control instructions for the device—for example, generating navigation point coordinates, travel speed, and obstacle avoidance parameters for the AGV, and grasping pose, joint motion trajectory, and end effector action instructions for the robotic arm. After compilation, the low-level instruction sequence is sent to the corresponding physical execution devices via the industrial network, and an instruction status tracking mechanism is established to record the instruction issuance time, expected execution status, and progress, providing benchmark data for closed-loop verification.

[0071] The closed-loop verification and dynamic correction modules are connected to the instruction compilation module and the physical perception layer, respectively, to achieve closed-loop control of physical execution and virtual pre-simulation, ensuring stable execution of the palletizing task. This module includes a data comparison unit, a multi-level response unit, and a dynamic replanning unit: The data comparison unit receives real-time equipment and material status data (such as the actual position of the AGV, the joint angle of the robotic arm, and the center of gravity of the pallet) transmitted from the physical perception layer, and simultaneously obtains the expected state of the instructions issued by the instruction compilation module (such as the expected path of the AGV and the expected trajectory of the robotic arm). It compares the two in real time, calculates the deviation value, and compares it with preset tolerances (such as position deviation thresholds and dwell time thresholds). The multi-level response unit triggers corresponding responses based on the data comparison results: if the deviation is within the tolerance range, the equipment continues to operate normally; if the deviation exceeds the tolerance range, an audible and visual warning is first activated to alert on-site personnel; if the deviation continues or shows an increasing trend, a pause command is automatically sent to the relevant equipment to prevent accidents. The dynamic replanning unit is activated when the equipment is paused or the deviation exceeds the limit. It uses the current physical world state (such as the current position of the equipment, the position of obstacles, and the amount of remaining material) as the initial conditions, calls the real-time virtual scene of the digital twin construction module, and quickly re-previews to generate local adjustment plans (such as adjusting the AGV path to bypass obstacles) or global backup plans. The new plan is then compiled into new low-level instructions by the input instruction compilation module and sent to the physical equipment to correct the deviation, ensuring that the palletizing task can continue to be executed until it is completed.

[0072] Furthermore, the system adopts an edge-cloud collaborative computing architecture to balance real-time performance and computing power requirements: edge servers are deployed near the workshop, running digital twin building modules and instruction compilation modules, responsible for handling tasks with high real-time requirements (such as virtual-real synchronization and instruction compilation), reducing data transmission latency; cloud servers or local central servers run parallel pre-simulation and optimization engine modules, responsible for handling computationally intensive tasks (such as generating a large number of initial schemes, parallel pre-simulation evaluation, and multi-objective optimization), utilizing the powerful computing resources of the cloud or central servers to improve the efficiency of scheme generation and optimization, ensuring that the system can still quickly output the optimal scheme when facing complex disk configuration tasks.

[0073] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

Claims

1. A method for pre-simulating auxiliary material batching based on digital twins, characterized in that, Includes the following steps: S1. Real-time data of auxiliary material pallets, automated guided vehicles and shelves are collected through a sensor network deployed in the physical environment. A high-fidelity digital twin synchronized with the physical environment is constructed based on a lightweight incremental update algorithm to obtain a real-time virtual scene. S2. Receive the palletizing task from the production management system using the current state of the real-time virtual scene, and generate multiple initial palletizing schemes encoded with material picking order, placement position and transportation path based on the improved genetic algorithm; wherein the improved genetic algorithm encodes the material picking order, placement position and orientation on the pallet, and transportation path of the automated guided vehicle into chromosome data structure. S3. In the digital twin, the multiple initial disk configuration schemes are pre-performed in parallel. A comprehensive evaluation is performed by integrating a multi-dimensional constraint rule base of geometric constraints, physical constraints and business constraints to obtain comprehensive evaluation data for each scheme. The step S3 includes: In the digital twin, multiple initial disk configuration schemes are simultaneously simulated using a parallel computing architecture, and each scheme is comprehensively evaluated through a multi-dimensional constraint rule base; Among them, geometric constraints simulate the interference between materials, between materials and vehicles, and between automated guided vehicles and the environment in a virtual environment by integrating continuous collision detection algorithms, and detect and record collision events in real time during the execution of the scheme. Physical constraints, by defining the weight attributes of materials and setting a center of gravity offset threshold, calculate the dynamic center of gravity position of the pallet in real time in the virtual environment, and determine whether the center of gravity projection exceeds the stable area of ​​the pallet, thus marking placement schemes that may lead to tipping over. Business constraints assess the business rationality of a solution by calculating space utilization, operation time, and material correlation. Space utilization is assessed by calculating the ratio of the total projected area of ​​placed materials to the available area of ​​the pallet. Operation time is based on a predefined motion time model to accumulate the total operation time of the solution. Material correlation is assessed by logically binding related materials in the same production work order and calculating the proximity of related materials in their placement positions. By integrating geometric constraints, physical constraints, and business constraints, comprehensive evaluation data on the geometric feasibility, physical stability, and business rationality of each solution are obtained. S4. Based on the comprehensive evaluation data, a multi-objective optimization algorithm is used to select the optimal disk allocation scheme from the multiple initial disk allocation schemes; S5. Input the optimal disk configuration scheme into the instruction compiler, which will automatically parse and compile it into a sequence of low-level control instructions that can be directly executed by the execution device; S6. During the execution of the underlying control command sequence, the feedback data from the physical sensors is compared with the expected state of the digital twin in real time. Once a deviation exceeding the tolerance is detected, a dynamic replanning program is triggered to generate an adjustment scheme, thereby achieving closed-loop verification and correction. The step S6 includes: during the execution of the underlying control command sequence, continuously collecting real-time status data of the automated guided vehicle, robotic arm, and auxiliary material tray through a physical sensor network, and comparing these physical sensor feedback data with the expected state of the virtual device in the digital twin in real time; when a deviation exceeding the allowable range is detected between the actual position of the automated guided vehicle and the expected path, or when the device stops due to an obstacle for more than a preset time threshold, the system triggers a dynamic replanning program; the dynamic replanning program uses the current actual state of the physical world as the new initial condition, quickly re-previews and generates a local adjustment scheme or a global backup scheme in the digital twin, and inputs the adjustment scheme into the instruction compiler to compile it into a new control command sequence and send it to the physical device.

2. The auxiliary material preparation method based on digital twins according to claim 1, characterized in that, Step S1 specifically includes: By collecting real-time position and attitude data of auxiliary material pallets, automated guided vehicles and shelves through ultra-wideband positioning tags and inertial measurement unit sensors deployed in the physical environment, a multi-sensor fusion algorithm is used to fuse the absolute position data provided by the ultra-wideband positioning tags and the relative motion data provided by the inertial measurement unit sensors. By correcting the measurement error of individual sensors through data complementarity, the precise three-dimensional coordinates and spatial orientation of each tracked entity in virtual space are calculated to obtain high-precision real-time pose information. Meanwhile, by collecting the identification data of objects through radio frequency identification devices deployed at key nodes in the warehouse, the identification data is spatiotemporally correlated and mapped with real-time pose information to establish an accurate correspondence between the virtual model and the physical entity. Based on real-time pose information and correspondence, a lightweight 3D engine incremental update algorithm is used to continuously monitor the state changes of physical data. By setting a state change recognition mechanism, the difference between physical data and the stored state of the virtual world is detected. Only the target virtual objects whose state has changed are updated in a directional pose, while the virtual objects whose state has not changed remain unchanged. A high-fidelity digital twin that is highly consistent with the real environment is constructed, resulting in a virtual scene that can accurately reflect the real-time state of the physical environment.

3. The auxiliary material preparation method based on digital twins according to claim 1, characterized in that, Step S2 specifically includes: Utilizing the current state of the real-time virtual scene provided by the digital twin, receive palletizing tasks from the manufacturing execution system or warehouse management system, and obtain material type, quantity information, and current available resource status data from the task list; Multiple initial palletizing schemes are generated based on an improved genetic algorithm. The improved genetic algorithm encodes the picking order of materials, their placement and orientation on the pallet, and the transportation path of the automated guided vehicle into a chromosome data structure. It adopts a diversity initialization strategy to generate an initial population and introduces an adaptive genetic operator to dynamically adjust the crossover and mutation probabilities according to the evolutionary state of the population in order to balance global search and local search capabilities. The improved genetic algorithm is used to iteratively optimize the initial population. In each iteration, excellent individuals are selected based on the fitness evaluation results, and a new generation of population is generated through adaptive genetic operators. Finally, multiple initial mating schemes with differences are output.

4. The auxiliary material preparation method based on digital twins according to claim 1, characterized in that, Step S4 includes: Based on the comprehensive evaluation data obtained in step S3, a multi-objective optimization algorithm is used to select the optimal disk allocation scheme from multiple initial disk allocation schemes. The multi-objective optimization algorithm includes the weighted scoring method and the Pareto optimal front analysis method. The weighted scoring method assigns different weight values ​​to multiple evaluation dimensions such as space utilization, operation time, balance and collision-free performance. The comprehensive score of each scheme is calculated by linear weighted summation and sorted according to the score. Pareto optimal front analysis identifies non-dominated solutions that do not have an absolute superiority-inferiority relationship among multiple optimization objectives, forming a Pareto optimal solution set containing multiple schemes; the system selects the final optimal scheme from the ranking results obtained by the weighted scoring method or the Pareto optimal solution set according to the optimization preferences set by the user.

5. The auxiliary material preparation method based on digital twins according to claim 1, characterized in that, Step S5 includes: inputting the optimal pallet arrangement scheme selected in step S4 into the instruction compiler, which has a built-in extensible device instruction driver library; the instruction compiler first parses the high-level task description in the optimal pallet arrangement scheme, decomposing high-level operations such as material handling and placement into basic operation units; then, according to the predefined instruction templates in the device instruction driver library, the basic operation units are compiled into a sequence of low-level control instructions that can be directly executed by specific devices. The control instructions for the automated guided vehicle include a sequence of navigation path point coordinates, motion speed parameters, and obstacle avoidance strategies, while the control instructions for the robotic arm include grasping posture parameters, joint motion trajectories, and end effector action instructions; the compiled low-level control instruction sequence is distributed to the corresponding physical execution device through the industrial network, and an instruction status tracking mechanism is established to provide an accurate instruction execution benchmark for subsequent closed-loop verification.

6. A digital twin-based auxiliary material batching simulation system, used to implement the digital twin-based auxiliary material batching simulation method according to any one of claims 1-5, characterized in that, include: The physical sensing layer includes ultra-wideband positioning tags and inertial measurement unit sensors deployed on auxiliary material pallets, automated guided vehicles and shelves, as well as industrial network equipment deployed at key nodes in the workshop, for collecting real-time motion data of various entities in the physical environment and transmitting it in real time through the industrial network. The digital twin construction module is connected to the physical perception layer, receives the real-time motion data, processes the real-time motion data using a multi-sensor fusion algorithm to obtain accurate pose information, and uses a lightweight 3D engine incremental update algorithm to construct a high-fidelity digital twin that is synchronized with the physical environment. The parallel pre-simulation and optimization engine module is connected to the digital twin construction module. It receives the disk allocation task from the production management system and the current state of the digital twin. Based on the improved genetic algorithm, it generates multiple initial disk allocation schemes, pre-simulates the schemes in parallel in the digital twin, and evaluates them through a multi-dimensional constraint rule base. The instruction compilation module, connected to the parallel pre-simulation and optimization engine module, has a built-in extensible device instruction driver library that automatically parses and compiles the optimal disk matching scheme into a sequence of low-level control instructions that can be executed by the automated guided vehicle and the robotic arm. The closed-loop verification and dynamic correction module is connected to the instruction compilation module and the physical perception layer, respectively. It compares the feedback data from the physical sensors with the expected state of the digital twin in real time, and triggers a dynamic replanning program when an excessive deviation is detected.

7. The auxiliary material batching simulation system based on digital twins according to claim 6, characterized in that, The physical sensing layer also includes radio frequency identification (RFID) devices deployed at warehouse entrances and exits, shelving areas, and palletizing stations, used to collect object identification data and associate it with pose information; The system adopts an edge-cloud collaborative computing architecture, in which the digital twin building module and instruction compilation module are deployed on the edge server, which are responsible for handling data synchronization and instruction compilation tasks with high real-time requirements, and the parallel pre-drilling and optimization engine module is deployed on the cloud server, which are responsible for executing computationally intensive scheme generation and optimization tasks.

8. The auxiliary material batching simulation system based on digital twins according to claim 6, characterized in that, The parallel pre-simulation and optimization engine module includes: The dimensional constraint rule library stores evaluation rules for geometric constraints, physical constraints, and business constraints. Geometric constraints integrate continuous collision detection algorithms, physical constraints define centroid stability evaluation criteria, and business constraints include space utilization, operation time, and material correlation calculation logic. An improved genetic algorithm unit is implemented to realize a scheme generation mechanism based on adaptive genetic operators. The mating scheme is represented by chromosome encoding, and a diversity preservation strategy is used to initialize the population. Parallel computing units utilize multi-core processors to concurrently execute simulation evaluations of multiple schemes within a digital twin; The multi-objective optimization unit uses a weighted scoring method and Pareto optimal frontier analysis to process the evaluation results and output the optimal solution set. The closed-loop verification and dynamic correction module includes: The data comparison unit receives data from physical sensors in real time and compares it with the expected state of the digital twin within tolerance limits. The multi-level response unit sequentially activates audible and visual warnings, equipment pause, and dynamic replanning when an excessive deviation is detected. The dynamic replanning unit regenerates the adjustment plan based on the current physical state and issues new instructions through the instruction compilation module.

Images