A Collaborative Simulation Method for Stamping Production Line Based on Digital Twin

Abstract

This invention relates to a collaborative simulation method for a stamping production line based on digital twins, specifically in the field of digital twins. By constructing a high-fidelity digital twin model, the operation of the stamping production line is fully reproduced in a virtual environment. A comprehensive evaluation value is constructed based on simulation data and a predefined set of key performance indicators. Machine learning optimization algorithms are used to automatically iterate and optimize the adjustable parameter set in the collaborative scheduling rules, generating an optimized adjustable parameter set. This method can pre-simulate, verify, and optimize the collaborative logic of multiple devices in the digital twin model before physical debugging, accurately exposing and avoiding potential strategy conflicts, action interference, and efficiency bottlenecks. Furthermore, robustness testing is completed by simulating preset abnormal operating conditions, ensuring that the final collaborative scheduling rules output are both efficient and reliable. This significantly reduces the trial-and-error costs and cycle of physical debugging, improving the success rate of production line design and final operational efficiency.

Description

Technical Field

[0001] This invention relates to the field of digital twins, and more specifically, to a collaborative simulation method for a stamping production line based on digital twins. Background Technology

[0002] In automated stamping production lines, especially complex systems involving multiple stamping machines and robotic handling units working collaboratively, the design and commissioning of these lines present significant challenges. Such lines typically consist of several stamping machines, associated robotic arms, and multiple conveyor belts responsible for material flow. For example, a typical production line might include four stamping machines and four robotic arms. The conveyor belts need separate blank and finished product columns to separate material flow. The robotic arms, according to precise scheduling rules, must pick up blanks from specific columns and feed them into the stamping machines, and then place the processed workpieces into designated columns. Furthermore, to adapt to the production process… Furthermore, the end of the production line may also be planned to integrate multiple sets of laser marking robots with synchronous left and right gripping functions. This layout of multiple devices operating in parallel and with interwoven operational logics makes the material flow and equipment movement flow highly coupled, resulting in a system with extremely high dynamism and concurrency. At the same time, external factors such as changes in the single-piece cycle time caused by product model changes and adjustments in capacity demand caused by fluctuations in production orders require the production line to have high flexibility and robustness. Before the physical deployment of the production line, how to accurately predict and optimize the collaborative behavior of such complex systems becomes a key prerequisite for ensuring investment efficiency, shortening the delivery cycle, and ensuring operational reliability.

[0003] However, existing technologies have significant limitations in addressing the aforementioned design verification requirements. Currently, the industry mainly relies on discrete event simulation for logistics cycle time analysis or uses programmable logic controller (PLC) simulation software for control logic verification. Discrete event simulation abstracts equipment into nodes with processing time. While it can assess macro-level capacity, it completely ignores the precise kinematics and dynamics of equipment such as robotic arms and conveyors. It cannot simulate potential spatial interference between equipment, such as collision risks, nor can it accurately simulate material grabbing failures or missed grabbing problems caused by minor timing deviations. On the other hand, PLC logic verification focuses on ensuring the correctness of individual control sequences and logic, making it difficult to verify the overall collaborative scheduling strategy of multiple devices. The fundamental flaw in quantitatively evaluating and optimizing physical performance, system resilience under disturbances, and potential production bottlenecks lies in the decoupling between the scheduling strategy model in the design phase and the physical behavior and sensor feedback of the actual equipment. This creates a static, open-loop verification environment. Consequently, existing methods cannot construct a closed-loop simulation platform in a virtual environment that integrates a high-precision equipment model, real control logic, and collaborative scheduling strategies to perform high-fidelity stress tests. This makes it difficult to expose and resolve core collaborative issues such as robot collisions, material misses, and system cycle mismatches caused by strategy conflicts before physical manufacturing and debugging. This creates significant risks of prolonged debugging cycles, cost overruns, and substandard operational efficiency. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing a collaborative simulation method for stamping production lines based on digital twins, thereby solving the problems mentioned in the background art.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: specifically, it includes the following steps: Step S1: Based on the production line layout information including conveyor belts, multiple stamping machines, multiple handling robots, and virtual workpieces to be processed, construct a corresponding digital twin model in the simulation environment. The construction process includes: integrating and fusing the geometric models and physical attribute models of the conveyor belts, stamping machines, and handling robots, as well as the collaborative scheduling rules containing adjustable parameter sets, so that the constructed digital twin model drives each model to perform corresponding physical motion simulations according to virtual control commands; the collaborative scheduling rules define the collaborative logic between devices through adjustable parameter sets and associate a predefined set of key performance indicators for strategy evaluation. Step S2: Load the collaborative scheduling rules into the digital twin model, and drive the digital twin model to run the simulation under the virtual control of the collaborative scheduling rules; during the simulation, synchronously record the pose state transition sequence of the virtual workpiece on the conveyor belt, the action state sequence of each handling robot and each stamping equipment, and the virtual control command sequence generated according to the collaborative scheduling rules, which together serve as the simulation running data; Step S3: Construct a comprehensive evaluation value based on simulation running data and a predefined set of key performance indicators; using the adjustable parameter set as the decision variable and optimizing the comprehensive evaluation value as the objective, automatically iterate and optimize the adjustable parameter set in the simulation environment simulated by the digital twin model through machine learning optimization algorithms to generate an optimized adjustable parameter set. Step S4: Update the optimized adjustable parameter set into the collaborative scheduling rules, and simulate one or more preset abnormal working conditions in the digital twin model to verify the robustness of the updated collaborative scheduling rules; after the robustness test verification is passed, output the final collaborative scheduling rules containing the optimized adjustable parameter set. In a preferred embodiment, the specific operation of integrating and fusing the geometric models and physical property models of the conveyor belt, stamping equipment, and handling robot, as well as the collaborative scheduling rules containing adjustable parameter sets, in step S1 is as follows: First, geometric model instances corresponding one-to-one with the conveyor belt, each stamping machine, and each handling robot in the physical production line are created in the simulation environment. The conveyor belt geometric model instance is assigned geometric dimension parameters of length, width, and height, as well as spatial coordinate position parameters. The stamping machine geometric model instance is assigned geometric dimension parameters of length, width, and height, as well as spatial coordinate position parameters. The handling robot geometric model instance is assigned geometric dimension parameters of link length, rotation angle range, and spatial coordinate position parameters of each joint, thus forming a geometric model. Next, linear velocity and acceleration attribute values ​​representing motion characteristics are assigned to the conveyor belt geometric model instance, stamping cycle attribute value representing the working cycle and stroke curve function describing the motion law of the mold are assigned to each stamping equipment geometric model instance, and kinematic model, maximum velocity attribute value and maximum acceleration attribute value of each moving joint, as well as grasping action time value and placement action time value are assigned to each handling robot geometric model instance, thereby forming a physical attribute model; Then, the collaborative scheduling rules are implemented as decision logic units that can be driven by an adjustable parameter set and output virtual control commands; Finally, the decision logic unit that implements the collaborative scheduling rules is integrated into the physical attribute model through a predefined application programming interface. During simulation, the decision logic unit receives virtual sensing data generated in real time by the simulation environment, performs calculations based on the current adjustable parameter set, generates virtual control commands, drives the corresponding physical attribute model to perform simulated actions according to the assigned physical attributes, completes the closed-loop integration from virtual perception, decision-making to virtual control, and forms a runnable digital twin model.

[0006] In a preferred embodiment, the adjustable parameter set includes a first parameter characterizing the gripping trigger lead deviation of the handling robot based on the position of the conveyor belt encoder, a second parameter being a dynamic priority weighting coefficient when assigning tasks to the k-th handling robot, and a third parameter characterizing the reference tuning coefficient of the conveyor belt speed.

[0007] In a preferred embodiment, the predefined set of key performance indicators includes at least system cycle time, dynamic missed detection risk coefficient, and device cooperative coupling degree, wherein: The calculation process of system cycle time is as follows: obtain the start time when the first virtual workpiece enters the digital twin model and the end time when the last virtual workpiece leaves the digital twin model during the simulation process, calculate the time difference between the start time and the end time, divide the time difference by the total number of virtual workpieces in the simulation, and then multiply the result by a calibration coefficient used to map the simulation time scale to the actual physical time scale to obtain the index value of system cycle time characterizing the production line output efficiency. The calculation process for the dynamic missed grasp risk coefficient is as follows: Accumulate the total number of grasping attempts made by all handling robots during the simulation; for each grasping attempt, calculate the normalized distance deviation between the end effector of the handling robot and the center of the target virtual workpiece at the moment the grasping attempt is triggered, and obtain the relative velocity component related to the conveyor belt and the grasping direction at the moment the grasping attempt is triggered; add the normalized distance deviation to the absolute value of the relative velocity component adjusted by the first weighting coefficient to obtain the median value; take the negative natural exponent of the median value and then map it; sum the mapping results of all grasping attempts, and then divide the summed result by the total number of grasping attempts to obtain a value between zero and one as the index value of the dynamic missed grasp risk coefficient; The calculation process for equipment cooperative coupling degree is as follows: During the total simulation time, the sum of the time when the physical attribute models corresponding to the stamping equipment and the handling robot are in an effective action state is accumulated; the sum of the time is divided by the product of the total number of physical attribute models and the total simulation time to obtain the preliminary cooperative efficiency value; the total blocking time of all physical attribute models waiting for each other due to the cooperative scheduling rules is calculated, and the total blocking time is multiplied by the second weight coefficient to obtain the penalty term; the penalty term is subtracted from the preliminary cooperative efficiency value to obtain the index value of equipment cooperative coupling degree.

[0008] In a preferred embodiment, step S2, specifically the operation of loading a cooperative scheduling rule into the digital twin model and driving the digital twin model to perform simulation operation under the virtual control of the cooperative scheduling rule, is as follows: First, initialize a global simulation clock and set its initial value to zero; then read the initial state of all physical property models from the digital twin model, including the initial position of the conveyor belt and the initial poses of each handling robot and each stamping device. Next, a batch of virtual workpieces are generated at the beginning of the conveyor belt at preset intervals, and a unique identifier is assigned to each virtual workpiece; the collaborative scheduling rules already integrated into the digital twin model are activated. Subsequently, the simulation is driven by a main loop, which uses a fixed time step as the basic propagation unit and uses event-driven logic to process state transitions within each fixed time step. The execution process of the main loop includes the following progressive stages: time advancement, physical state calculation, virtual sensor signal generation, cooperative scheduling decision triggering, and instruction application and state update; The main loop continues until the preset simulation termination conditions are met. The simulation termination conditions include the global simulation clock reaching the set maximum simulation duration or all virtual workpieces generated in the last batch leaving the production line.

[0009] In a preferred embodiment, the process of synchronously recording the pose state transition sequence of the virtual workpiece on the conveyor belt, the action state sequence of each handling robot and each stamping device, and the virtual control command sequence generated according to the collaborative scheduling rules is as follows: At the end of each fixed time step, a snapshot of the global state is performed to generate a structured record; Each structured record contains a sequence of virtual workpiece pose state transitions. It records a snapshot of all active virtual workpieces at the end of the current fixed time step. The snapshot content includes the unique identifier of the virtual workpiece, its position coordinates and orientation at the end of the current fixed time step, and the state of the virtual workpiece at the end of the current fixed time step. The state of the virtual workpiece includes conveying, being grabbed, stamping, and completed. Action state sequence, records a snapshot of all handling robots and stamping equipment at the end of the current fixed time step. The snapshot content includes the equipment identifier, position coordinates and attitude at the end of the current fixed time step, the state of the equipment at the end of the current fixed time step, and the instructions being executed by the equipment at the end of the current fixed time step. The state of the equipment includes idle, moving, gripping, and stamping. The virtual control instruction sequence records all new virtual control instructions generated within the current fixed time step. Each virtual control instruction includes the instruction type, target device identifier, and target virtual workpiece identifier. All structured records with fixed time steps are arranged in chronological order according to the global simulation clock to form simulation running data.

[0010] In a preferred embodiment, step S3 involves constructing a comprehensive evaluation value based on simulation data and a predefined set of key performance indicators, specifically including: First, the sequence of virtual workpiece pose state changes, the sequence of motion states of each handling robot and each stamping equipment, and the sequence of virtual control commands recorded in the simulation running data are analyzed, and the index values ​​of system cycle time, dynamic missed capture risk coefficient and equipment cooperative coupling degree are calculated simultaneously. Next, the calculated system cycle time, dynamic missed detection risk coefficient, and equipment coordination coupling degree index values ​​are subjected to nonlinear normalization processing based on objectives and constraints, respectively, and the index value of each index is mapped to the interval between zero and one to obtain a corresponding normalized satisfaction score. Then, based on the normalized satisfaction scores, a multi-objective comprehensive evaluation function is constructed, and the output value of the multi-objective comprehensive evaluation function is used as the comprehensive evaluation value. The multi-objective comprehensive evaluation function is obtained by multiplying two parts. The first part is the value obtained by weighted geometric mean of each normalized satisfaction score. The second part is the balance penalty term. The arithmetic mean of all normalized satisfaction scores is calculated, and then the difference between each normalized satisfaction score and the arithmetic mean is calculated. Each difference is squared and summed. The square root of the sum is taken, and the result is multiplied by the balance penalty factor. Then, the product result is subtracted by one. The calculation process of the weighted geometric mean is as follows: each normalized satisfaction score is assigned a corresponding weight and then multiplied together. The result of the product is then taken as the root of the total number of weights.

[0011] In a preferred embodiment, the process of automatically iteratively optimizing the adjustable parameter set using a machine learning optimization algorithm to generate an optimized adjustable parameter set is as follows: First, an iterative optimization loop is initialized. Within the domain of the adjustable parameter set, a preset number of initial adjustable parameter sets are selected using experimental design methods. For each initial adjustable parameter set, in the simulation environment simulated by the digital twin model, the initial adjustable parameter set is loaded and the digital twin model is driven to perform a complete simulation run, generating corresponding simulation run data and calculating the comprehensive evaluation value corresponding to the initial adjustable parameter set. All initial adjustable parameter sets and their corresponding comprehensive evaluation values ​​are collected to form an initial sample set. Next, based on the initial sample set, a Gaussian process regression model is trained as a surrogate model, and for any new set of adjustable parameters, the mean and variance of its comprehensive evaluation value are predicted. Then, the iterative optimization phase begins. Each iteration includes the following steps: Based on the current Gaussian process regression surrogate model, using the expected improvement acquisition function, the next most valuable adjustable parameter set for evaluation is found within the domain of the adjustable parameter set; the adjustable parameter set with the largest expected improvement acquisition function value is selected as the adjustable parameter set to be evaluated in this iteration; for the selected adjustable parameter set to be evaluated, in the simulation environment simulated by the digital twin model, the adjustable parameter set to be evaluated is loaded and the digital twin model is driven to perform a complete simulation run, generating corresponding simulation run data, and the true comprehensive evaluation value of the adjustable parameter set to be evaluated is calculated; the new adjustable parameter set and its true comprehensive evaluation value are added to the initial sample set, and the Gaussian process regression surrogate model is retrained using the updated initial sample set, updating its parameters; The iterative optimization phase continues for a preset number of rounds, or terminates when the improvement of the comprehensive evaluation value is less than a preset threshold in multiple consecutive iterations. After the iteration terminates, the set of adjustable parameters that results in the highest comprehensive evaluation value is selected from all evaluated adjustable parameter sets in the initial sample set as the optimized set of adjustable parameters.

[0012] In a preferred embodiment, in step S4, a preset abnormal working condition mode library is prepared. The abnormal working condition mode library contains a variety of predefined abnormal working condition modes. Each abnormal working condition mode defines a disturbance type, disturbance intensity, simulation time when the disturbance begins, and duration of disturbance. The abnormal working condition modes include, but are not limited to: temporarily reducing the maximum joint speed attribute value of one or more handling robots within a specified simulation time period; causing the linear speed attribute value of the conveyor belt to fluctuate periodically or randomly around a reference value within a specified simulation time period; and adding random deviations to the generation time interval of virtual workpieces throughout the simulation run. For each abnormal operating condition mode in the abnormal operating condition mode library, an independent stress test simulation is performed. The specific process of each stress test simulation is as follows: In the digital twin model, the cooperative scheduling rules with updated and optimized adjustable parameter sets are loaded, and the digital twin model is driven to start simulation operation according to the normal process. When the global simulation clock reaches the disturbance start time defined by the current abnormal operating condition mode, the parameters of one or more corresponding physical attribute models in the digital twin model are dynamically modified according to the disturbance type and disturbance intensity defined by the current abnormal operating condition mode, thereby injecting the abnormal operating condition mode into the simulation environment. After the abnormal operating condition mode is injected, the simulation continues to run until the preset simulation termination condition is met. During the entire simulation operation including the abnormal injection, the pose state transition sequence of the virtual workpiece on the conveyor belt, the action state sequence of each device, and the virtual control command sequence generated according to the updated cooperative scheduling rules are recorded synchronously, which together serve as the disturbance simulation operation data under the current abnormal operating condition mode.

[0013] In a preferred embodiment, the specific process for calculating a corresponding condition-specific robustness score for the disturbance simulation data obtained under each abnormal operating condition mode is as follows: First, based on the disturbance simulation data, the first comprehensive evaluation value of the system under abnormal operating conditions is calculated, and the performance recovery time consumed by the system from the end of the abnormality to its performance recovery to the normal level is calculated; at the same time, the second comprehensive evaluation value of the baseline simulation without abnormality is obtained. Next, the first operating condition-specific robustness score is calculated. The first operating condition-specific robustness score consists of two weighted components: the first component is the performance retention score, which is calculated based on the difference between the first comprehensive evaluation value and the second comprehensive evaluation value; the second component is the recovery speed score, which is obtained by calculating the ratio of performance recovery time to the duration of abnormality, and this ratio is normalized to a maximum of one. The performance retention score and the recovery speed score are weighted and summed to obtain the first operating condition-specific robustness score. Then, based on the first condition-specific robustness scores under all abnormal operating conditions, a global robustness score is calculated. The calculation of the global robustness score uses the minimum value among all first condition-specific robustness scores to characterize the ability of the collaborative scheduling rules to resist the weakest link in the preset abnormal operating condition mode library. Next, the calculated global robustness score is compared with a pre-set robustness pass threshold; if the global robustness score is greater than or equal to the robustness pass threshold, the robustness test is deemed to have passed. Finally, after the robustness test is passed, the output contains the final cooperative scheduling rule with the optimized adjustable parameter set. The output includes the formatting and encapsulation of the optimized adjustable parameter set and the corresponding cooperative scheduling rule logic.

[0014] The beneficial effects of this invention are as follows: By constructing a high-fidelity digital twin model, the operation of the stamping production line is completely reproduced in a virtual environment. A comprehensive evaluation value is constructed based on simulation operation data and a predefined set of key performance indicators. Furthermore, machine learning optimization algorithms are used to automatically iterate and optimize the adjustable parameter set in the collaborative scheduling rules, generating an optimized adjustable parameter set. This method can pre-simulate, verify, and optimize the collaborative logic of multiple devices in the digital twin model before physical debugging, accurately expose and avoid potential strategy conflicts, action interference, and efficiency bottlenecks. In addition, robustness testing is completed by simulating preset abnormal working conditions, ensuring that the final collaborative scheduling rules output are both efficient and reliable. This significantly reduces the trial and error cost and cycle of physical debugging, and improves the success rate of production line design and final operating efficiency. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0018] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0019] Example 1 This embodiment provides, for example Figure 1 The method for collaborative simulation of a stamping production line based on digital twins, as shown, specifically includes the following steps: Step S1: Based on the production line layout information including conveyor belts, multiple stamping machines, multiple handling robots, and virtual workpieces to be processed, construct a corresponding digital twin model in the simulation environment. The construction process includes: integrating and fusing the geometric models and physical attribute models of the conveyor belts, stamping machines, and handling robots, as well as the collaborative scheduling rules containing adjustable parameter sets, so that the constructed digital twin model drives each model to perform corresponding physical motion simulations according to virtual control commands; the collaborative scheduling rules define the collaborative logic between devices through adjustable parameter sets and associate a predefined set of key performance indicators for strategy evaluation. Step S2: Load the collaborative scheduling rules into the digital twin model, and drive the digital twin model to run the simulation under the virtual control of the collaborative scheduling rules; during the simulation, synchronously record the pose state transition sequence of the virtual workpiece on the conveyor belt, the action state sequence of each handling robot and each stamping equipment, and the virtual control command sequence generated according to the collaborative scheduling rules, which together serve as the simulation running data; Step S3: Construct a comprehensive evaluation value based on simulation running data and a predefined set of key performance indicators; using the adjustable parameter set as the decision variable and optimizing the comprehensive evaluation value as the objective, automatically iterate and optimize the adjustable parameter set in the simulation environment simulated by the digital twin model through machine learning optimization algorithms to generate an optimized adjustable parameter set. Step S4: Update the optimized adjustable parameter set into the collaborative scheduling rules, and simulate one or more preset abnormal working conditions in the digital twin model to verify the robustness of the updated collaborative scheduling rules; after the robustness test verification is passed, output the final collaborative scheduling rules containing the optimized adjustable parameter set.

[0020] In this embodiment, it is specifically necessary to explain the following steps in step S1: integrating the geometric models and physical property models of the conveyor belt, stamping equipment, and handling robot, as well as the collaborative scheduling rules containing adjustable parameter sets. First, geometric model instances corresponding one-to-one with the conveyor belt, each stamping machine, and each handling robot in the physical production line are created in the simulation environment. This process is completed using a 3D modeling engine (such as a physically based rendering engine or a professional CAD kernel) to ensure that the geometric dimensions, joint configurations, and spatial poses of the virtual models are strictly consistent with the design drawings or physical measurement data, providing an accurate geometric basis for subsequent interference collision detection and precise motion simulation. The conveyor belt geometric model instance is assigned length, width, and height geometric dimension parameters and spatial coordinate position parameters, the stamping machine geometric model instance is assigned length, width, and height geometric dimension parameters and spatial coordinate position parameters, and the handling robot geometric model instance is assigned link length, rotation angle range geometric dimension parameters, and spatial coordinate position parameters of each joint, thus forming the geometric model. Next, linear velocity and acceleration attribute values ​​representing motion characteristics are assigned to the conveyor belt geometric model instances. A stamping cycle attribute value representing the working cycle and a stroke curve function describing the mold's motion law are assigned to each stamping equipment geometric model instance. A kinematic model, maximum velocity and maximum acceleration attribute values ​​for each moving joint, as well as grasping and placing action time attribute values ​​are assigned to each handling robot geometric model instance, thus forming a physical attribute model. The linear velocity attribute value assigned to the conveyor belt geometric model instances is typically set between 0.1 m / s and 2.0 m / s, and the acceleration attribute value is set between 0.5 m / s² and 5 m / s², to simulate actual motion. The performance range of the frequency converter driver is set between 4 and 12 seconds for the stamping cycle attribute value assigned to the geometric model instance of the stamping equipment, based on the nominal force and stroke. The stroke curve function adopts a sinusoidal-trapezoidal acceleration curve to simulate a smooth stamping dynamic process. The kinematic model assigned to the geometric model instance of the handling robot is established using the DH parameter method. The maximum velocity attribute value and maximum acceleration attribute value of each joint are set according to the servo motor specifications. The gripping action time attribute value and the placement action time attribute value are usually between 0.3 and 1.5 seconds. The injection of these physical attributes enables the virtual model to simulate real inertia, rhythm and motion smoothness, which is the key to high-fidelity timing simulation. Then, the collaborative scheduling rules are implemented as decision logic units that can be driven by an adjustable parameter set and output virtual control commands. This decision logic unit is implemented at the software level as an independent, reusable algorithm module (e.g., encapsulated as a dynamic link library or ROS node), which contains an event listener, a rule parser, and a command sender. The event listener subscribes to virtual sensing data (such as encoder pulse events and workpiece arrival signals) in real time. The rule parser performs decision calculations based on a preset logic framework (such as a finite state machine or behavior tree) and the current adjustable parameter set. The command sender formats the decision results into virtual control commands (such as target coordinates and action commands). This modular implementation makes the collaborative scheduling rules easy to modify, replace, and integrate for testing. Finally, the decision logic unit that implements the collaborative scheduling rules is integrated into the physical attribute model through a predefined application programming interface. During simulation, the decision logic unit receives virtual sensing data generated in real time by the simulation environment, performs calculations based on the current adjustable parameter set, generates virtual control commands, drives the corresponding physical attribute model to perform simulated actions based on the assigned physical attributes, completes the closed-loop integration from virtual perception, decision-making to virtual control, and forms a runnable digital twin model. The application programming interface (API) defines a set of standard service call and topic publish / subscribe protocols. For example, the decision logic unit obtains the real-time pose of the physical property model through service calls and publishes virtual control commands through topics. During simulation, the integrated system forms a closed loop: the physical property model generates virtual sensing data (such as "workpiece has reached the gripping position") and publishes it; the decision logic unit receives this data, calculates based on the current adjustable parameter set (e.g., whether the triggering condition is met), and generates virtual control commands (such as "robot A performs gripping"); this command is received by the physical property model and drives it to perform simulated actions based on the assigned physical properties (such as moving with maximum acceleration). This closed-loop integration fully reproduces the real production line cyber-physical system operation process from virtual perception, decision-making to virtual control, forming an operable, observable, and measurable digital twin model. The adjustable parameter set includes a first parameter characterizing the gripping trigger lead deviation of the handling robot based on the position of the conveyor belt encoder, a second parameter representing the dynamic priority weight coefficient when assigning tasks to the k-th handling robot, and a third parameter representing the reference tuning coefficient of the conveyor belt speed. The typical range of the first parameter (grip trigger lead deviation) is -50 mm to +50 mm. A positive value indicates early triggering, which is used to compensate for the robot's acceleration time and prevent the workpiece from moving out of the range. A negative value indicates delayed triggering, which is used to avoid interference or wait for the workpiece to stabilize. The second parameter (dynamic priority weight coefficient) assigns an initial weight to each handling robot, with a value range of 0.5 to 2.0. It can be dynamically fine-tuned according to the robot's load during operation to achieve load balancing. The third parameter (conveyor belt speed reference tuning coefficient) is a multiplier factor, with a typical value range of 0.8 to 1.2. It is used to fine-tune the nominal speed of the conveyor belt in the simulation to find the best matching point with the robot's cycle time and avoid periodic congestion or idleness caused by a fixed speed. These adjustable parameters together define the behavior pattern of the collaborative scheduling rules and are the direct objects of subsequent automatic optimization. The predefined set of key performance indicators includes at least system cycle time, dynamic missed detection risk coefficient, and device coordination coupling degree, among which: The system cycle time is calculated as follows: The start time of the first virtual workpiece entering the digital twin model and the end time of the last virtual workpiece leaving the digital twin model are obtained during the simulation. The time difference between the start and end times is calculated, divided by the total number of virtual workpieces in the simulation, and then multiplied by a calibration coefficient used to map the simulation time scale to the actual physical time scale. This yields the system cycle time index, which characterizes the production line's output efficiency. The calibration coefficient is determined by the ratio of the simulation engine's time step to the actual time unit. For example, if 1 second in the simulation represents 1 second in reality, the calibration coefficient is 1. If the simulation is accelerated, the coefficient is scaled up proportionally. This index directly reflects the production line's capacity and is a core economic indicator for evaluating the quality of scheduling strategies. The calculation process for the dynamic missed grasp risk coefficient is as follows: Accumulate the total number of grasping attempts made by all handling robots during the simulation; for each grasping attempt, calculate the normalized distance deviation between the end effector of the handling robot and the center of the target virtual workpiece at the moment the grasping attempt is triggered, and obtain the relative velocity component related to the conveyor belt and the grasping direction at the moment the grasping attempt is triggered; add the normalized distance deviation to the absolute value of the relative velocity component adjusted by the first weighting coefficient to obtain an intermediate value; map the intermediate value after taking the negative natural exponent; sum the mapping results of all grasping attempts, and then divide the summed result by the total number of grasping attempts to obtain a value between zero and one as the index value of the dynamic missed grasp risk coefficient. The normalized distance deviation is calculated by dividing the actual three-dimensional distance by the grasping direction. The tolerance threshold (e.g., 20 mm) is used to obtain the value. If the value exceeds the threshold, it is recorded as 1. The first weighting coefficient is used to adjust the importance of the influence of conveyor belt speed. The typical value is set to 0.01 seconds / mm. The operation of negative natural exponent (i.e., calculating e^-(intermediate value)) transforms the linear adverse factors of distance and speed into a "success probability" mapping value between 0 and 1. The closer the value is to 1, the lower the risk of a single grab. The final dynamic missed grab risk coefficient is the arithmetic mean of all these "success probabilities". However, its actual meaning is a risk measure. Therefore, the closer its value is to 1, the higher the average missed grab risk. This indicator is a forward-looking and probabilistic evaluation indicator, which is better than the simple "number of missed grabs" statistics. It can warn of insufficient robustness of the strategy before a missed grab actually occurs. The calculation process for equipment coordination coupling degree is as follows: During the total simulation time, the sum of the time when the physical attribute models corresponding to the stamping equipment and the handling robot are in an effective action state is accumulated; the sum of the time is divided by the product of the total number of physical attribute models and the total simulation time to obtain the preliminary coordination efficiency value; the total blocking time of all physical attribute models waiting for each other due to the coordination scheduling rules is calculated, and the total blocking time is multiplied by the second weighting coefficient to obtain the penalty term; the penalty term is subtracted from the preliminary coordination efficiency value to obtain the equipment coordination coupling degree index value. "Effective action state" is defined as the equipment performing a value-added process action (such as stamping or handling) or a necessary idle action, excluding idle waiting states. The second weighting coefficient is used to quantify the degree of damage to overall efficiency caused by blocking. Its setting is related to the production line cycle value, with a typical value set at 2.0, meaning that one unit of blocking time is considered to cause two units of wasted time. The equipment coordination coupling degree index value comprehensively quantifies equipment utilization and coordination smoothness. The higher the value, the less mutual interference the equipment experiences while working efficiently, and the better the coordination scheduling rules.

[0021] In this embodiment, the specific operation of loading the cooperative scheduling rule into the digital twin model and driving the digital twin model to perform simulation operation under the virtual control of the cooperative scheduling rule in step S2 is as follows: First, initialize a global simulation clock, setting its initial value to zero. Then, read the initial states of all physical property models from the digital twin model. These initial states include the initial position of the conveyor belt, and the initial poses of each handling robot and each stamping device. The global simulation clock provides a unique and continuous time reference for all virtual events and state records. Its initial value of zero ensures the determinism of the simulation time start. The initial poses of the physical property models are determined based on the 3D layout drawings of the production line. For example, each handling robot is located at a designated "origin" or "waiting" position, and the stamping device slider is located at the top dead center. Next, a batch of virtual workpieces are generated at the beginning of the conveyor belt at preset intervals. The preset intervals are set according to the production line design cycle time. Each virtual workpiece is assigned a unique identifier. The collaborative scheduling rules integrated into the digital twin model are activated, and the collaborative scheduling rules are put into a ready listening state. The generation of virtual workpieces simulates the material loading process in real production. The preset interval can be set to 1.0 to 1.2 times the design cycle time to test the performance of the scheduling strategy under different material densities. The unique identifier is usually an incrementing integer or a globally unique identifier (GUID) to unambiguously track each virtual workpiece throughout the simulation cycle. Subsequently, the simulation is driven by a main loop, which uses a fixed time step as the basic propagation unit. The fixed time step is set according to the simulation accuracy requirements, and is usually set between 1 millisecond and 10 milliseconds. The choice of this time range balances simulation accuracy and computational efficiency: if the step is too large, the details of the device acceleration process will be lost, resulting in inaccurate motion simulation; if the step is too small, the amount of computation will increase sharply and the simulation speed will be too slow. This step is on the same order of magnitude as the scan cycle of the real programmable logic controller, ensuring the time comparability between the virtual control loop and the physical control logic. Within each fixed time step, event-driven logic is used to process state transitions. The execution process of the main loop includes the following progressive stages: The first stage is the main loop's time progression, which adds a fixed time step to the global simulation clock to determine the current simulation moment. The second stage is physical state calculation. Based on the virtual control commands at the current moment, it drives all physical attribute models to calculate and update their instantaneous states at the current moment according to their own dynamic models. The dynamic models include the uniform motion model of the conveyor belt, the kinematic model of the handling robot, and the stroke curve model of the stamping equipment. When calculating and updating the instantaneous states, for the uniform motion model of the conveyor belt, the linear velocity and acceleration attribute values ​​of the conveyor belt are used, combined with the position at the previous moment, to calculate the new coordinates of each virtual workpiece on the conveyor belt at the current moment. This calculation is based on the formula for uniform acceleration or uniform motion, ensuring that the movement of the workpiece on the conveyor belt is continuous and the position is determined, providing accurate input for position-based triggering decisions. For the kinematic model of the handling robot, the maximum velocity attribute value, maximum acceleration attribute value of each moving joint, and the preceding virtual control commands are used to solve the angles of each joint at the current moment. The calculation process employs inverse kinematics algorithms and performs trajectory planning based on physical constraints (maximum speed, acceleration) to generate smooth joint motion curves. This realistically simulates the robot's motion inertia, avoiding unrealistic movements with theoretically unlimited acceleration. For the stamping equipment stroke curve model, based on the stamping cycle attribute values, it determines whether the die is currently in the downward, holding, or return phase, and calculates the current displacement of the stamping slide. The stroke curve model simulates the actual working cycle of the stamping machine; its displacement and speed changes directly affect the safe time window for the robot to place and retrieve parts, making it crucial for high-fidelity simulation. The third stage involves virtual sensor signal generation. Based on the updated physical state, virtual sensor data for the current moment is generated, including virtual photoelectric sensor signals and virtual encoder pulse signals. The generation of virtual photoelectric sensor signals is achieved by determining whether the virtual workpiece or the robot's end effector enters the geometric detection area of ​​the virtual sensor model, and a small random delay (such as ±0) can be added.The virtual encoder pulse signal is generated periodically at a set density of several pulses per millimeter, based on the precise displacement of the conveyor belt, using a 5-millisecond time interval to simulate the response jitter of a real sensor. This virtual sensing data, containing simulated noise, makes the testing environment of the decision logic unit closer to reality, improving the robustness of the optimized strategy. The fourth stage is collaborative scheduling decision triggering, where the generated virtual sensing data is input to the decision logic unit in real time. The decision logic unit judges the input event based on its internal logic and the current adjustable parameter set. If the decision conditions are met, a new virtual control command is immediately generated and placed in an instant command queue. The decision conditions include the virtual workpiece reaching the gripping position of the handling robot and the handling robot being idle. The condition "reaching the gripping position" has a position tolerance threshold, for example, set to 5 millimeters. That is, when the distance between the center of the virtual workpiece and the theoretical gripping point is less than this threshold, it is judged as "reaching". This simulates... The fifth stage involves instruction application and state re-updating. Within the current fixed time step, newly generated virtual control instructions are immediately applied to fine-tune the target state of the affected physical property model within this fixed time step. When fine-tuning the target state, if the physical property model receives a grasping instruction, the target coordinates of the end effector of the handling robot are immediately updated to the coordinates of the current virtual workpiece, and the action state of the handling robot is marked as moving towards the target. If a placement or stamping instruction is received, the target coordinates and action state are updated accordingly. The "immediate application" mechanism simulates the real-time issuance of control instructions, ensuring zero delay in event response (within the simulation step accuracy). The "fine-tuning" of the target state is a dynamic replanning process. For example, after receiving a new grasping instruction, the robot will immediately calculate a new trajectory that smoothly transitions to the new target point based on the current motion state. This avoids abrupt changes in motion and reflects the physical rationality of the simulation. The main loop continues until the preset simulation termination conditions are met. The simulation termination conditions include the global simulation clock reaching the set maximum simulation duration or all virtual workpieces generated in the last batch leaving the production line. The maximum simulation duration is usually set to a simulation time sufficient to produce hundreds to thousands of virtual workpieces, such as the simulation duration corresponding to several hours in actual time, to ensure that the collected simulation operation data can cover the complete dynamic process of production line startup, stable operation, possible congestion and decline, so that the statistically obtained key performance index values ​​are statistically significant. The process of synchronously recording the pose state transition sequence of the virtual workpiece on the conveyor belt, the action state sequence of each handling robot and each stamping device, and the virtual control command sequence generated according to the collaborative scheduling rules is as follows: At the end of each fixed time step, a global state snapshot is performed, generating a structured record. The structured record contains three time-synchronized sequences with the same timestamp, which is the global simulation clock value at the end of the current fixed time step. This "synchronous snapshot" mechanism is an important feature of this method. It forces the alignment of the virtual workpiece, equipment actions, and controller commands at the same simulation moment. This solves the causal logic confusion and time matching problems that may be caused by asynchronous recording, and provides a strict data foundation for subsequent accurate analysis of event chains (such as "after a certain command is issued, after a certain delay, the robot arm reaches a certain position and grabs a certain workpiece"). Each structured record contains a sequence of virtual workpiece pose state transitions. It records a snapshot of all active virtual workpieces at the end of the current fixed time step. The snapshot content includes the unique identifier of the virtual workpiece, its position coordinates and orientation at the end of the current fixed time step, and the state of the virtual workpiece at the end of the current fixed time step. The state of the virtual workpiece includes conveying, being grabbed, stamping, and completed. Action state sequence, records a snapshot of all handling robots and stamping equipment at the end of the current fixed time step. The snapshot content includes the equipment identifier, position coordinates and attitude at the end of the current fixed time step, the state of the equipment at the end of the current fixed time step, and the instructions being executed by the equipment at the end of the current fixed time step. The state of the equipment includes idle, moving, gripping, and stamping. The virtual control instruction sequence records all new virtual control instructions generated within the current fixed time step. Each virtual control instruction includes the instruction type, target device identifier, and target virtual workpiece identifier. Recording the complete virtual control instruction content allows for the retrospective analysis of the specific content and triggering context of each scheduling decision when analyzing simulation data. This facilitates the diagnosis of strategy defects and the understanding of behavioral changes after the optimization algorithm adjusts its parameters. All structured records with fixed time steps are arranged in chronological order according to the global simulation clock to form simulation running data. This data comprehensively encompasses the complete spatiotemporal evolution history of all virtual workpieces, all equipment, and newly generated virtual control commands in the digital twin model at each discrete time point under the control of a specified set of adjustable parameters. The specific organization of the simulation running data is a series of structured records arranged sequentially in chronological order, including records at the initial moment, records after one fixed time step, records after two fixed time steps, and so on, until the last record at the simulation's end. This high-density... High-dimensional simulation operation data constitutes a complete "data field" about production line collaborative behavior. It is not only used to calculate several aggregated key performance index values, but it is also a rich database that can be directly used to train deep learning-based scheduling models or to perform root cause analysis on complex abnormal working conditions. For example, by querying simulation operation data, the start time, end time, waiting time, and direct cause of each "carrying robot waiting" event can be accurately extracted (such as "target stamping machine is busy" or "no available workpiece on the conveyor belt"). This is a structured insight that traditional simulation logs cannot provide.

[0022] In this embodiment, it is particularly important to explain step S3, which involves constructing a comprehensive evaluation value based on simulation data and a predefined set of key performance indicators. This includes: First, the sequence of virtual workpiece pose state changes, the sequence of motion states of each handling robot and each stamping equipment, and the sequence of virtual control commands recorded in the simulation running data are analyzed. Based on the calculation process of the predefined set of key performance indicators, the index values ​​of system cycle time, dynamic missed capture risk coefficient and equipment cooperative coupling degree are calculated simultaneously. Next, the calculated system cycle time, dynamic missed detection risk coefficient, and equipment coordination coupling degree are subjected to nonlinear normalization based on targets and constraints. Each indicator value is mapped to a range between zero and one, resulting in a corresponding normalized satisfaction score. In the normalization of the system cycle time indicator, the target cycle time is determined based on the production line's designed capacity, for example, set to 15 seconds / piece. The tolerance coefficient is used to control the rate of satisfaction decay when the actual cycle time exceeds the target, typically set to 5 seconds. This means that when the actual cycle time is 5 seconds slower than the target, satisfaction drops to approximately 37% (1 / e) of the initial value. This exponential decay severely punishes poor cycles far exceeding the target, while being relatively mild for cycles close to the target. The normalization of the dynamic missed detection risk coefficient indicator uses a linear function that subtracts the risk coefficient value from the target value. The linear mapping function directly yields the risk inhibition score. Since the dynamic missed detection risk coefficient is already between 0 and 1, this linear mapping concisely transforms "risk" into "inhibition degree," facilitating unified measurement with other indicators. For the normalization of the equipment coordination coupling index value, an S-shaped growth curve mapping function is used to map the actual coupling value to a coupling health score. This score exhibits a smooth but sharp increase near a preset health threshold. The health threshold is typically set to 0.6, and the slope coefficient is set to 10. When the equipment coordination coupling increases from 0.5 to 0.7, the coupling health score will rise sharply from near 0 to near 1. This effectively distinguishes between "inefficient coordination" and "efficient coordination" states, guiding the optimization algorithm to quickly cross the performance plateau. The smoothness of the S-shaped function also avoids abrupt changes in the evaluation value at the threshold, which is beneficial to the stable convergence of the optimization algorithm. Then, based on the normalized satisfaction scores, a multi-objective comprehensive evaluation function is constructed, and the output value of the multi-objective comprehensive evaluation function is used as the comprehensive evaluation value. The multi-objective comprehensive evaluation function is obtained by multiplying two parts. The first part is the value obtained by weighted geometric mean of each normalized satisfaction score. The weighted geometric mean emphasizes the balanced improvement of each indicator. If the normalized satisfaction score of any indicator is too low, it will significantly lower the comprehensive result. The weight coefficients of each indicator are set according to the production management strategy. For example, in the scenario of prioritizing production capacity, the weight of system cycle time satisfaction can be set to 0.5, the weight of dynamic omission risk suppression can be set to 0.3, and the weight of equipment collaborative coupling health can be set to 0.2. Compared with the arithmetic mean, the weighted geometric mean is more sensitive to any "weak" indicator and can effectively prevent the imbalance of one performance item being severely degraded while other items are excellent in the optimization result. The second part is the balance penalty term, which calculates the arithmetic mean of all normalized satisfaction scores, and then calculates the weight of each normalized score. The difference between the normalized satisfaction score and the arithmetic mean is calculated by squaring each difference and summing the results. The square root of the sum is then multiplied by a balance penalty factor, and the result is subtracted from the sum. The larger the difference between the normalized satisfaction scores, the smaller the value of the balance penalty term. The balance penalty factor is used to adjust the intensity of the requirement for balance among the indicators. A typical value is set to 0.3. For example, if the three normalized satisfaction scores are 0.9, 0.9, and 0.9, the balance penalty term is 1. If they are 1.0, 0.5, and 0.9, the balance penalty term will be significantly less than 1, thereby reducing the overall evaluation value. This design forces the optimization process not only to pursue a high total score but also to pursue relatively balanced development of the performance of each sub-item, which meets the actual engineering requirements for stable and reliable operation of the production line. The calculation process of the weighted geometric mean is as follows: each normalized satisfaction score is assigned a corresponding weight and then multiplied together. The result of the product is then taken as the root of the total number of weights. The process of automatically iteratively optimizing the adjustable parameter set using machine learning optimization algorithms to generate an optimized adjustable parameter set is as follows: First, an iterative optimization loop is initialized. Within the domain of the adjustable parameter set, a predetermined number of initial adjustable parameter sets are selected using experimental design methods. The preferred experimental design method is Latin hypercube sampling to ensure that the initial samples are evenly distributed within the multidimensional domain of the adjustable parameter set, providing good initial global information for the subsequent surrogate model. The predetermined number is typically 5 to 10 times the dimension of the adjustable parameter set; for example, for an adjustable parameter set containing 3 parameters, the initial sample size is set to 15 to 30. For each initial adjustable parameter set, in the simulation environment simulated by the digital twin model, the initial adjustable parameter set is loaded, and the digital twin model is driven to perform a complete simulation run, generating corresponding simulation data and calculating the comprehensive evaluation value corresponding to the initial adjustable parameter set. All initial adjustable parameter sets and their corresponding comprehensive evaluation values ​​are collected to form an initial sample set. Next, based on the initial sample set, a Gaussian process regression model is trained as a surrogate model. The Gaussian process regression model selects the squared exponential kernel function as the covariance function, which can smoothly fit the complex nonlinear relationship between the adjustable parameter set and the comprehensive evaluation value. This surrogate model not only provides the predicted mean but also the predicted variance, quantifying the uncertainty of the prediction. This is the key to driving active learning (balancing exploration and utilization). The Gaussian process regression model can learn the complex mapping relationship between the adjustable parameter set and the comprehensive evaluation value, and for any new set of adjustable parameters, predict the mean and variance of its comprehensive evaluation value. Then, the iterative optimization phase begins. Each iteration includes the following steps: Based on the current Gaussian process regression surrogate model, the expected improvement acquisition function is used to find the next most valuable adjustable parameter set within the domain of the adjustable parameter set; the expected improvement acquisition function calculates the expected improvement brought by the new parameter set based on the current highest observed value. It automatically balances "utilization" (searching in regions with high predicted mean of the surrogate model) and "exploration" (searching in regions with high predicted variance and uncertainty), which is the core of the Bayesian optimization framework; the expected improvement acquisition function considers both the mean and variance of the surrogate model's predictions, achieving a balance between utilizing known high-performance regions and exploring regions with high uncertainty; the adjustable parameter set with the largest expected improvement acquisition function value is selected. The selected set of adjustable parameters is used as the evaluation set in this iteration. For the selected set of adjustable parameters, the set of adjustable parameters is loaded into the simulation environment simulated by the digital twin model, and the digital twin model is driven to perform a complete simulation run to generate corresponding simulation run data and calculate the true comprehensive evaluation value of the set of adjustable parameters. The new set of adjustable parameters and its true comprehensive evaluation value are added to the initial sample set, and the Gaussian process regression surrogate model is retrained using the updated initial sample set to update its parameters. The surrogate model is updated every time a new sample is added, so that the model's understanding of the parameter-performance relationship continues to evolve, guiding the search direction more and more accurately and greatly reducing the number of high-cost simulations required to find the optimal solution. The iterative optimization phase continues for a preset number of rounds, or terminates when the improvement in the comprehensive evaluation value is less than a preset threshold in multiple consecutive iterations. The preset number of rounds is usually set to 50 to 200 rounds to ensure sufficient search. The improvement threshold is set to 0.001 to 0.005. When the improvement in the comprehensive evaluation value is less than this threshold in 5 to 10 consecutive iterations, the optimization is considered to have converged to a plateau and can be terminated early to save computational resources. After the iteration terminates, the set of adjustable parameters that results in the highest comprehensive evaluation value is selected from all evaluated adjustable parameters in the initial sample set as the optimized set of adjustable parameters. This optimized set of adjustable parameters is the result of extensive virtual simulation experiments and data-driven intelligent optimization. Its performance is usually significantly better than the initial parameter settings based on experience, and it balances multiple key performance indicators.

[0023] In this embodiment, it is specifically necessary to explain step S4, which involves preparing a preset abnormal operating condition mode library. This library contains various predefined abnormal operating condition modes, each clearly defining a disturbance type, disturbance intensity, the simulation time at which the disturbance begins, and the duration of the disturbance. The construction of the abnormal operating condition mode library is based on the analysis of historical fault records of the physical stamping production line, equipment performance degradation data, and common interference sources. It aims to cover typical non-ideal scenarios ranging from equipment-level soft performance degradation to random fluctuations in material flow. For example, a 30% decrease in the maximum joint speed attribute value of a handling robot simulates a decrease in servo drive performance or slight mechanical jamming; periodic fluctuations in the conveyor belt linear speed. (e.g., ±10%) simulates unstable drive motor speed or belt slippage; random deviation of virtual workpiece generation interval (±20%) simulates the unevenness of upstream process production cycle. The disturbance start time is usually set after the simulation enters the stable production stage (e.g., at the 50th second), and the disturbance duration is set between 20 and 60 seconds to simulate brief equipment abnormalities or continuous interference; abnormal working condition modes include but are not limited to: temporarily reducing the maximum joint speed attribute value of one or more handling robots within a specified simulation time period, causing the linear speed attribute value of the conveyor belt to fluctuate periodically or randomly around the reference value within a specified simulation time period, and adding random deviation to the virtual workpiece generation time interval throughout the simulation run; For each abnormal operating condition mode in the abnormal operating condition mode library, an independent stress test simulation is executed. The specific process of each stress test simulation is as follows: In the digital twin model, the cooperative scheduling rules with the updated and optimized adjustable parameter set are loaded, and the digital twin model is driven to start simulation operation according to the conventional process. The "conventional process" refers to the simulation operation and data recording main loop defined in claims 5 and 6. The core difference between stress test simulation and conventional simulation is that preset abnormalities are dynamically injected during the operation. When the global simulation clock reaches the disturbance start time defined by the current abnormal operating condition mode, the parameters of one or more corresponding physical property models in the digital twin model are dynamically modified according to the disturbance type and disturbance intensity defined by the current abnormal operating condition mode, thereby injecting the abnormal operating condition mode into the simulation environment. "Dynamic modification" is implemented through the runtime application programming interface provided by the simulation engine, for example, in At a specified time, a function is called to multiply the "maximum speed attribute value" of a specific handling robot model instance by a coefficient (1 - disturbance intensity). This modification takes effect instantaneously, simulating the suddenness of a fault. After the abnormal working condition mode is injected, the simulation continues to run until the preset simulation termination condition is met. During the entire simulation operation including the abnormal injection, the pose state transition sequence of the virtual workpiece on the conveyor belt, the action state sequence of each device, and the virtual control command sequence generated according to the updated collaborative scheduling rules are recorded synchronously, which together serve as the disturbance simulation operation data under the current abnormal working condition mode. The recording format of this disturbance simulation operation data is completely consistent with that of the regular simulation operation data, ensuring the uniformity of the subsequent analysis and processing interface. Its value lies in fully capturing the full-time and space dynamic response of the digital twin system to a specific disturbance under the optimized adjustable parameter set control, and it is the only data source for performing robustness quantitative evaluation. The specific process for calculating a corresponding condition-specific robustness score for the disturbance simulation data obtained under each abnormal operating condition mode is as follows: First, based on the disturbance simulation data, the first comprehensive evaluation value of the system under abnormal operating conditions is calculated, and the performance recovery time from the end of the disturbance to the recovery of its performance to normal levels is also calculated. The calculation of performance recovery time requires defining a metric. Typically, the second comprehensive evaluation value from the baseline simulation without anomalies is used as the "normal level" benchmark. The performance recovery time is defined as the time elapsed from the end of the disturbance defined in the abnormal operating conditions until the average of the first comprehensive evaluation value sequence (calculated based on the disturbance simulation data and sliding in fixed time windows, e.g., 10 seconds) reaches or exceeds 90% of the second comprehensive evaluation value for the first three consecutive windows. If this condition is not met by the end of the simulation, the performance recovery time is set to a penalty value much larger than the duration of the disturbance (e.g., ten times the duration of the disturbance). This method quantifies the speed at which the system "bounces" back from the disturbance. Simultaneously, the second comprehensive evaluation value from the baseline simulation without anomalies is obtained. The second comprehensive evaluation value is typically obtained during the optimization process in step S3 by performing at least one complete simulation under abnormal conditions using the optimized set of adjustable parameters, serving as the "gold standard" for performance comparison. Next, the robustness score specific to the first operating condition is calculated. This score is a value between zero and one, with a higher value indicating better robustness of the collaborative scheduling rules to the current abnormal operating condition. The robustness score consists of two weighted components: the first is the performance retention score, calculated based on the difference between the first comprehensive evaluation value and the second comprehensive evaluation value. When the first comprehensive evaluation value is less than or equal to the second comprehensive evaluation value, the performance retention score is one. When the first comprehensive evaluation value is greater than the second comprehensive evaluation value, the performance retention score decays exponentially, with the decay rate controlled by a preset performance degradation tolerance coefficient. The specific calculation for exponential decay is: Performance retention score = exp(-(first comprehensive evaluation value - second comprehensive evaluation value) / (performance degradation tolerance coefficient * second comprehensive evaluation value)), where the performance degradation tolerance coefficient is a key parameter used to define the acceptable performance degradation ratio, typically ranging from 0.1 to 0.3. For example, it can be set to 0. The score of .2 means that if performance drops by 20%, the performance retention score is approximately exp(-1)≈0.37; if performance does not drop, the score is 1; if it drops by 40%, the score is approximately 0.14. This exponential form severely penalizes significant performance degradation. The second part is the recovery speed score, which is obtained by calculating the ratio of performance recovery time to the duration of the anomaly. This ratio is normalized to a maximum of one. That is, recovery speed score = min(performance recovery time / duration of anomaly, 1.0). The closer the score is to 1, the faster the recovery (recovery time is less than or equal to the duration of the disturbance). If the recovery time exceeds the duration of the disturbance, the score will be reduced proportionally, with a minimum of 0. The performance retention score and the recovery speed score are weighted and summed to obtain the first condition-specific robustness score. The weighting coefficients used in the weighted summation reflect the emphasis on "anti-interference capability" and "self-recovery capability". For example, they can be set to 0.6 and 0.4, which means that performance retention during the disturbance is given more importance. The final score value is between 0 and 1. Then, based on the first condition-specific robustness scores under all abnormal operating conditions, a global robustness score is calculated. The calculation of the global robustness score adopts the minimum value among all first condition-specific robustness scores to characterize the ability of the collaborative scheduling rule to resist the weakest link in the preset abnormal operating condition mode library. Adopting the minimum value (barrel principle) is a conservative but engineering-robust strategy. It ensures that the optimized collaborative scheduling rule is not too vulnerable to any type of preset typical anomaly, thereby ensuring its overall reliability in complex real-world environments. Next, the calculated global robustness score is compared with a pre-set robustness pass threshold. If the global robustness score is greater than or equal to the robustness pass threshold, the robustness test is deemed to have passed. The robustness pass threshold needs to be set according to the production line reliability requirements, with a typical value of 0.7. This means that for various preset anomalies, the comprehensive robustness performance of the collaborative scheduling rules (considering performance preservation and recovery) must reach a good level (70 points) or above in order to be considered to have sufficient engineering application resilience. Finally, after the robustness test verification is passed, the output includes the final cooperative scheduling rules containing the optimized adjustable parameter set. The output includes formatting and encapsulating the optimized adjustable parameter set and the corresponding cooperative scheduling rule logic. Formatting and encapsulation usually produce a structured data file (such as JSON, XML, or proprietary binary format) or a software module (such as a dynamic link library or container image). This deliverable contains the specific values ​​of the optimized adjustable parameter set, the version identifier of the cooperative scheduling rule decision logic, and necessary metadata (such as applicable equipment models and simulation environment versions). This deliverable can be directly imported by the programmable logic controller programming software, manufacturing execution system, or edge computing platform of the physical production line to update the parameters in the actual control program or replace the scheduling algorithm module, thereby realizing a digital continuous process from virtual simulation, optimization verification to physical deployment.

[0024] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0025] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0026] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0027] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0028] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0029] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0030] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A collaborative simulation method for a stamping production line based on digital twins, characterized in that, Specifically, the steps include the following: Step S1: Based on the production line layout information including conveyor belts, multiple stamping machines, multiple handling robots, and virtual workpieces to be processed, construct a corresponding digital twin model in the simulation environment. The construction process includes: integrating and fusing the geometric models and physical attribute models of the conveyor belts, stamping machines, and handling robots, as well as the collaborative scheduling rules containing adjustable parameter sets, so that the constructed digital twin model drives each model to perform corresponding physical motion simulations according to virtual control commands; the collaborative scheduling rules define the collaborative logic between devices through adjustable parameter sets and associate a predefined set of key performance indicators for strategy evaluation. Step S2: Load the collaborative scheduling rules into the digital twin model, and drive the digital twin model to run the simulation under the virtual control of the collaborative scheduling rules; during the simulation, synchronously record the pose state transition sequence of the virtual workpiece on the conveyor belt, the action state sequence of each handling robot and each stamping equipment, and the virtual control command sequence generated according to the collaborative scheduling rules, which together serve as the simulation running data; Step S3: Construct a comprehensive evaluation value based on simulation running data and a predefined set of key performance indicators; using the adjustable parameter set as the decision variable and optimizing the comprehensive evaluation value as the objective, automatically iterate and optimize the adjustable parameter set in the simulation environment simulated by the digital twin model through machine learning optimization algorithms to generate an optimized adjustable parameter set. Step S4: Update the optimized adjustable parameter set into the collaborative scheduling rules, and simulate one or more preset abnormal working conditions in the digital twin model to verify the robustness of the updated collaborative scheduling rules; after the robustness test verification is passed, output the final collaborative scheduling rules containing the optimized adjustable parameter set.

2. The collaborative simulation method for a stamping production line based on digital twins according to claim 1, characterized in that: In step S1, the specific operation of integrating and fusing the geometric models and physical property models of the conveyor belt, stamping equipment, and handling robot, as well as the collaborative scheduling rules containing adjustable parameter sets, is as follows: First, geometric model instances corresponding one-to-one with the conveyor belt, each stamping machine, and each handling robot in the physical production line are created in the simulation environment. The conveyor belt geometric model instance is assigned geometric dimension parameters of length, width, and height, as well as spatial coordinate position parameters. The stamping machine geometric model instance is assigned geometric dimension parameters of length, width, and height, as well as spatial coordinate position parameters. The handling robot geometric model instance is assigned geometric dimension parameters of link length, rotation angle range, and spatial coordinate position parameters of each joint, thus forming a geometric model. Next, linear velocity and acceleration attribute values ​​representing motion characteristics are assigned to the conveyor belt geometric model instance, stamping cycle attribute value representing the working cycle and stroke curve function describing the motion law of the mold are assigned to each stamping equipment geometric model instance, and kinematic model, maximum velocity attribute value and maximum acceleration attribute value of each moving joint, as well as grasping action time value and placement action time value are assigned to each handling robot geometric model instance, thereby forming a physical attribute model; Then, the collaborative scheduling rules are implemented as decision logic units that can be driven by an adjustable parameter set and output virtual control commands; Finally, the decision logic unit that implements the collaborative scheduling rules is integrated into the physical attribute model through a predefined application programming interface. During simulation, the decision logic unit receives virtual sensing data generated in real time by the simulation environment, performs calculations based on the current adjustable parameter set, generates virtual control commands, drives the corresponding physical attribute model to perform simulated actions according to the assigned physical attributes, completes the closed-loop integration from virtual perception, decision-making to virtual control, and forms a runnable digital twin model.

3. The collaborative simulation method for a stamping production line based on digital twins according to claim 2, characterized in that: The adjustable parameter set includes a first parameter characterizing the gripping trigger lead deviation of the handling robot based on the position of the conveyor belt encoder, a second parameter being the dynamic priority weight coefficient when assigning tasks to the k-th handling robot, and a third parameter characterizing the reference tuning coefficient of the conveyor belt speed.

4. The collaborative simulation method for a stamping production line based on digital twins according to claim 3, characterized in that: The predefined set of key performance indicators includes at least system cycle time, dynamic missed detection risk coefficient, and device coordination coupling degree, wherein: The calculation process of system cycle time is as follows: obtain the start time when the first virtual workpiece enters the digital twin model and the end time when the last virtual workpiece leaves the digital twin model during the simulation process, calculate the time difference between the start time and the end time, divide the time difference by the total number of virtual workpieces in the simulation, and then multiply the result by a calibration coefficient used to map the simulation time scale to the actual physical time scale to obtain the index value of system cycle time characterizing the production line output efficiency. The calculation process for the dynamic missed grasp risk coefficient is as follows: Accumulate the total number of grasping attempts made by all handling robots during the simulation; for each grasping attempt, calculate the normalized distance deviation between the end effector of the handling robot and the center of the target virtual workpiece at the moment the grasping attempt is triggered, and obtain the relative velocity component related to the conveyor belt and the grasping direction at the moment the grasping attempt is triggered; add the normalized distance deviation to the absolute value of the relative velocity component adjusted by the first weighting coefficient to obtain the median value; take the negative natural exponent of the median value and then map it; sum the mapping results of all grasping attempts, and then divide the summed result by the total number of grasping attempts to obtain a value between zero and one as the index value of the dynamic missed grasp risk coefficient; The calculation process for equipment cooperative coupling degree is as follows: During the total simulation time, the sum of the time when the physical attribute models corresponding to the stamping equipment and the handling robot are in an effective action state is accumulated; the sum of the time is divided by the product of the total number of physical attribute models and the total simulation time to obtain the preliminary cooperative efficiency value; the total blocking time of all physical attribute models waiting for each other due to the cooperative scheduling rules is calculated, and the total blocking time is multiplied by the second weight coefficient to obtain the penalty term; the penalty term is subtracted from the preliminary cooperative efficiency value to obtain the index value of equipment cooperative coupling degree.

5. The collaborative simulation method for a stamping production line based on digital twins according to claim 4, characterized in that: In step S2, the specific operation of loading cooperative scheduling rules into the digital twin model and driving the digital twin model to perform simulation operation under the virtual control of the cooperative scheduling rules is as follows: First, initialize a global simulation clock and set its initial value to zero; then read the initial state of all physical property models from the digital twin model, including the initial position of the conveyor belt and the initial poses of each handling robot and each stamping device. Next, a batch of virtual workpieces are generated at the beginning of the conveyor belt at preset intervals, and a unique identifier is assigned to each virtual workpiece; the collaborative scheduling rules already integrated into the digital twin model are activated. Subsequently, the simulation is driven by a main loop, which uses a fixed time step as the basic propagation unit and uses event-driven logic to process state transitions within each fixed time step. The execution process of the main loop includes the following progressive stages: time advancement, physical state calculation, virtual sensor signal generation, cooperative scheduling decision triggering, and instruction application and state update; The main loop continues until the preset simulation termination conditions are met. The simulation termination conditions include the global simulation clock reaching the set maximum simulation duration or all virtual workpieces generated in the last batch leaving the production line.

6. The collaborative simulation method for a stamping production line based on digital twins according to claim 5, characterized in that: The process of synchronously recording the pose state transition sequence of the virtual workpiece on the conveyor belt, the action state sequence of each handling robot and each stamping device, and the virtual control command sequence generated according to the collaborative scheduling rules is as follows: At the end of each fixed time step, a snapshot of the global state is performed to generate a structured record; Each structured record contains a sequence of virtual workpiece pose state transitions. It records a snapshot of all active virtual workpieces at the end of the current fixed time step. The snapshot content includes the unique identifier of the virtual workpiece, its position coordinates and orientation at the end of the current fixed time step, and the state of the virtual workpiece at the end of the current fixed time step. The state of the virtual workpiece includes conveying, being grabbed, stamping, and completed. Action state sequence, records a snapshot of all handling robots and stamping equipment at the end of the current fixed time step. The snapshot content includes the equipment identifier, position coordinates and attitude at the end of the current fixed time step, the state of the equipment at the end of the current fixed time step, and the instructions being executed by the equipment at the end of the current fixed time step. The state of the equipment includes idle, moving, gripping, and stamping. The virtual control instruction sequence records all new virtual control instructions generated within the current fixed time step. Each virtual control instruction includes the instruction type, target device identifier, and target virtual workpiece identifier. All structured records with fixed time steps are arranged in chronological order according to the global simulation clock to form simulation running data.

7. The collaborative simulation method for a stamping production line based on digital twins according to claim 6, characterized in that: In step S3, a comprehensive evaluation value is constructed based on simulation running data and a predefined set of key performance indicators, specifically including: First, the sequence of virtual workpiece pose state changes, the sequence of motion states of each handling robot and each stamping equipment, and the sequence of virtual control commands recorded in the simulation running data are analyzed, and the index values ​​of system cycle time, dynamic missed capture risk coefficient and equipment cooperative coupling degree are calculated simultaneously. Next, the calculated system cycle time, dynamic missed detection risk coefficient, and equipment coordination coupling degree index values ​​are subjected to nonlinear normalization processing based on objectives and constraints, respectively, and the index value of each index is mapped to the interval between zero and one to obtain a corresponding normalized satisfaction score. Then, based on the normalized satisfaction scores, a multi-objective comprehensive evaluation function is constructed, and the output value of the multi-objective comprehensive evaluation function is used as the comprehensive evaluation value. The multi-objective comprehensive evaluation function is obtained by multiplying two parts. The first part is the value obtained by weighted geometric mean of each normalized satisfaction score. The second part is the balance penalty term. The arithmetic mean of all normalized satisfaction scores is calculated, and then the difference between each normalized satisfaction score and the arithmetic mean is calculated. Each difference is squared and summed. The square root of the sum is taken, and the result is multiplied by the balance penalty factor. Then, the product result is subtracted by one. The calculation process of the weighted geometric mean is as follows: each normalized satisfaction score is assigned a corresponding weight and then multiplied together. The result of the product is then taken as the root of the total number of weights.

8. The collaborative simulation method for a stamping production line based on digital twins according to claim 7, characterized in that: The process of automatically iteratively optimizing the adjustable parameter set using machine learning optimization algorithms to generate an optimized adjustable parameter set is as follows: First, an iterative optimization loop is initialized. Within the domain of the adjustable parameter set, a preset number of initial adjustable parameter sets are selected using experimental design methods. For each initial adjustable parameter set, in the simulation environment simulated by the digital twin model, the initial adjustable parameter set is loaded and the digital twin model is driven to perform a complete simulation run, generating corresponding simulation run data and calculating the comprehensive evaluation value corresponding to the initial adjustable parameter set. Collect all initial adjustable parameter sets and their corresponding comprehensive evaluation values ​​to form an initial sample set; Next, based on the initial sample set, a Gaussian process regression model is trained as a surrogate model, and for any new set of adjustable parameters, the mean and variance of its comprehensive evaluation value are predicted. Then, the iterative optimization phase begins. Each iteration includes the following steps: Based on the current Gaussian process regression surrogate model, using the expected improvement acquisition function, the next most valuable adjustable parameter set for evaluation is found within the domain of the adjustable parameter set; the adjustable parameter set with the largest expected improvement acquisition function value is selected as the adjustable parameter set to be evaluated in this iteration; for the selected adjustable parameter set to be evaluated, in the simulation environment simulated by the digital twin model, the adjustable parameter set to be evaluated is loaded and the digital twin model is driven to perform a complete simulation run, generating corresponding simulation run data, and the true comprehensive evaluation value of the adjustable parameter set to be evaluated is calculated; the new adjustable parameter set and its true comprehensive evaluation value are added to the initial sample set, and the Gaussian process regression surrogate model is retrained using the updated initial sample set, updating its parameters; The iterative optimization phase continues for a preset number of rounds, or terminates when the improvement of the comprehensive evaluation value is less than a preset threshold in multiple consecutive iterations. After the iteration terminates, select the set of adjustable parameters that gives the highest overall evaluation value from all evaluated adjustable parameter sets in the initial sample set, and use it as the optimized set of adjustable parameters.

9. The collaborative simulation method for a stamping production line based on digital twins according to claim 8, characterized in that: In step S4, a preset abnormal working condition mode library is prepared. The abnormal working condition mode library contains a variety of predefined abnormal working condition modes. Each abnormal working condition mode defines a disturbance type, disturbance intensity, simulation time when the disturbance begins, and duration of disturbance. Abnormal working condition modes include, but are not limited to: temporarily reducing the maximum joint speed attribute value of one or more handling robots within a specified simulation time period; causing the linear speed attribute value of the conveyor belt to fluctuate periodically or randomly around a reference value within a specified simulation time period; and adding random deviations to the generation time interval of virtual workpieces throughout the simulation run. For each abnormal operating condition mode in the abnormal operating condition mode library, an independent stress test simulation is performed. The specific process of each stress test simulation is as follows: In the digital twin model, the cooperative scheduling rules with the updated and optimized adjustable parameter set are loaded, and the digital twin model is driven to start the simulation operation according to the conventional process. When the global simulation clock reaches the disturbance start time defined by the current abnormal operating condition mode, the parameters of one or more physical property models in the digital twin model are dynamically modified according to the disturbance type and disturbance intensity defined by the current abnormal operating condition mode, thereby injecting the abnormal operating condition mode into the simulation environment; after the abnormal operating condition mode is injected, the simulation continues to run until the preset simulation termination condition is met. Throughout the entire simulation process, including the anomaly injection, the sequence of pose state changes of the virtual workpiece on the conveyor belt, the sequence of action states of each device, and the sequence of virtual control commands generated according to the updated collaborative scheduling rules are recorded synchronously, which together serve as the disturbance simulation data under the current abnormal working condition mode.

10. A collaborative simulation method for a stamping production line based on digital twins according to claim 9, characterized in that: The specific process for calculating a corresponding condition-specific robustness score for the disturbance simulation data obtained under each abnormal operating condition mode is as follows: First, based on the disturbance simulation data, the first comprehensive evaluation value of the system under abnormal operating conditions is calculated, and the performance recovery time consumed by the system from the end of the abnormality to its performance recovery to the normal level is calculated; at the same time, the second comprehensive evaluation value of the baseline simulation without abnormality is obtained. Next, the robustness score specific to the first operating condition is calculated; The first operating condition-specific robustness score is composed of two weighted parts: the first part is the performance retention score, which is calculated based on the difference between the first comprehensive assessment value and the second comprehensive assessment value; the second part is the recovery speed score, which is obtained by calculating the ratio of performance recovery time to the duration of abnormality, and this ratio is normalized to a maximum of one. The performance retention score and the recovery speed score are weighted and summed to obtain the first operating condition-specific robustness score. Then, based on the first condition-specific robustness scores under all abnormal operating conditions, a global robustness score is calculated. The calculation of the global robustness score uses the minimum value among all first condition-specific robustness scores to characterize the ability of the collaborative scheduling rules to resist the weakest link in the preset abnormal operating condition mode library. Next, the calculated global robustness score is compared with a pre-set robustness pass threshold; if the global robustness score is greater than or equal to the robustness pass threshold, the robustness test is deemed to have passed. Finally, after the robustness test is passed, the output contains the final cooperative scheduling rule with the optimized adjustable parameter set. The output includes the formatting and encapsulation of the optimized adjustable parameter set and the corresponding cooperative scheduling rule logic.

Images