A multi-robot intelligent manufacturing cooperation system

Abstract

The application discloses a kind of multi-robot intelligent manufacturing cooperation systems, it is related to industrial robot control and intelligent manufacturing technical field, solve the technical problems of the heterogeneous compatibility difficult of prior art multi-robot system, dynamic response delay is high, collaborative obstacle avoidance precision is insufficient, lack of quality closed-loop management and control and weak fault-tolerant capability.This system uses cloud edge end distributed four-layer architecture, including edge collaborative control layer, multi-robot cluster execution layer, whole-process sensing layer and digital twin closed-loop optimization layer, supporting TSN+5G double-redundancy communication bus module.The application realizes the plug-and-play adaptation of heterogeneous robots, millisecond-level dynamic task scheduling and fault takeover, high-precision collaborative motion obstacle avoidance, and quality closed-loop optimization of whole-process processing, and can widely adapt to the flexible production needs of discrete manufacturing scene, greatly improve the efficiency of production line operation, processing yield and continuous operation stability.

Description

Technical Field

[0001] This invention relates to the field of industrial robot control and intelligent manufacturing technology, and in particular to a multi-robot intelligent manufacturing collaborative system. Background Technology

[0002] With the rapid development of intelligent manufacturing technology, multi-robot collaborative systems have become core equipment in discrete manufacturing scenarios (automobile manufacturing, 3C electronics processing, aerospace component processing, etc.), enabling automated collaborative operations across multiple processes and significantly improving production efficiency. Existing multi-robot collaborative systems mostly adopt a centralized control architecture, pre-planning the work paths and task sequences of each robot via a host computer. This only allows for collaborative operations with fixed workflows, and has the following core drawbacks in practical industrial applications: The centralized control architecture has a computing power bottleneck. When faced with dynamic operating conditions on the production line (such as material orientation deviation, sudden equipment failure, and emergency order insertion), the response delay of task redistribution is high, which cannot meet the real-time requirements of flexible manufacturing. The compatibility and adaptation of heterogeneous robots (from different manufacturers, with different communication protocols and different degrees of freedom) is difficult. Customized secondary development is required for different robots, resulting in long system deployment cycles, high modification costs, and the inability to achieve plug-and-play functionality. When multiple robots work together, they can only achieve pre-planned obstacle avoidance of static paths. They cannot perform dynamic obstacle avoidance and motion coordination based on real-time operation status. In overlapping areas of the work space, motion interference is likely to occur, posing safety hazards. They cannot adapt to the flexible operation requirements of mixed-line production. The lack of a closed-loop quality feedback mechanism throughout the entire processing process makes it impossible to dynamically adjust the collaborative operation parameters of robots based on online detection processing quality data, which easily leads to batches of defective products and makes it difficult to improve product yield. The system has weak fault tolerance. When a single robot fails, it can easily cause the entire production line to shut down. It cannot achieve seamless task takeover under fault conditions, and the stability of continuous production line operation is insufficient. Summary of the Invention

[0003] To address the aforementioned shortcomings of existing technologies, the present invention aims to provide a multi-robot intelligent manufacturing collaborative system that solves the technical problems of heterogeneous compatibility difficulties, high dynamic response delays, insufficient collaborative obstacle avoidance accuracy, lack of quality closed-loop control, and weak fault tolerance in existing technologies, thereby achieving efficient, stable, and high-precision collaborative operation of multiple robots in flexible manufacturing scenarios.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a multi-robot intelligent manufacturing collaborative system, comprising an edge collaborative control layer, a multi-robot cluster execution layer, a full-process perception layer, a digital twin closed-loop optimization layer, and a supporting dual-redundant communication bus module. The dual-redundant communication bus module adopts a dual-redundant architecture of TSN (Time-Sensitive Network) and 5G industrial private network to achieve low-latency deterministic data transmission between each layer, ensuring the synchronization of control commands and perception data. The full-process perception layer is signal-connected to the edge collaborative control layer and the digital twin closed-loop optimization layer, and is used to collect data on the material arrival pose, operating environment, robot operating status, and processing quality in real time. The collected data is synchronously uploaded to the edge collaborative control layer and the digital twin closed-loop optimization layer, providing data support for dynamic collaborative control and quality closed-loop optimization. The digital twin closed-loop optimization layer, connected to the edge collaborative control layer, is used to construct a 1:1 digital twin model of the production line, complete the simulation pre-verification of collaborative operations, and generate process parameter optimization instructions based on the quality data uploaded from the full-process perception layer. These instructions are then sent to the edge collaborative control layer, achieving a complete closed-loop control of virtual simulation, physical execution, quality feedback, and parameter optimization. The edge collaborative control layer, the core control hub of the system, is connected to the multi-robot cluster execution layer. It receives production orders and optimization instructions from the digital twin closed-loop optimization layer, performs protocol adaptation for heterogeneous robots, dynamic allocation of global tasks, multi-robot collaborative motion control, and seamless task takeover under fault conditions, overcoming the computing power bottleneck of centralized control. The multi-robot cluster execution layer includes several heterogeneous robot units that receive control instructions from the edge collaborative control layer, complete collaborative operations for corresponding processing steps, and adapt to the processing requirements of different processes.

[0005] Furthermore, the edge collaborative control layer includes a global task scheduling module, a heterogeneous robot protocol adaptation module, a dynamic collaborative motion control module, and a fault tolerance and task takeover module. The global task scheduling module is equipped with a task scheduling algorithm based on DQN deep reinforcement learning. With production line delivery cycle, robot load balancing rate, and production efficiency as optimization objectives, it decomposes production orders into executable sub-tasks and completes the initial allocation of sub-tasks and task reassignment under dynamic working conditions, achieving multi-objective optimized dynamic task scheduling. The heterogeneous robot protocol adaptation module has a built-in unified robot equipment abstract model and protocol conversion unit, compatible with mainstream industrial bus protocols such as Profinet, EtherCAT, Modbus, and DeviceNet, while also adapting to the private control protocols of multiple robot brands, achieving automatic identification and plug-and-play adaptation of heterogeneous robot units without the need for customized secondary development. The dynamic cooperative motion control module employs a multi-robot cooperative motion planning algorithm based on Model Predictive Control (MPC), combined with real-time data collected from the full-process perception layer, to dynamically plan the motion paths of each robot unit in real time. This enables temporal synchronization and cooperative obstacle avoidance among multiple robots within overlapping workspaces, mitigating the risk of motion interference. The fault tolerance and task takeover module is signal-connected to the robot status monitoring module of the full-process perception layer. When a robot unit malfunction is detected, it triggers the global task scheduling module to reassign tasks, transferring the tasks of the malfunctioning robot to normally operating robot units, achieving seamless task takeover and preventing a complete production line shutdown.

[0006] Furthermore, each heterogeneous robot unit in the multi-robot cluster execution layer is equipped with a local edge controller, an end effector, a six-dimensional force control sensor, and a status acquisition unit. The local edge controller has the ability to make autonomous decisions for single-machine operation, and at the same time receives collaborative control instructions from the edge collaborative control layer through a dual-redundant communication bus module, realizing a control mode of distributed autonomous decision-making + centralized collaborative management and control.

[0007] Furthermore, the full-process perception layer includes an incoming material pose perception module, an online quality inspection module, a workspace environment perception module, and a robot status monitoring module. The incoming material pose perception module uses a 3D structured light vision camera to collect real-time workpiece pose data, providing data support for path adjustment of the handling and processing robot and improving incoming material positioning accuracy. The online quality inspection module uses a 3D laser vision inspection unit and high-precision dimensional inspection equipment to collect real-time quality data of the processed workpiece, identify processing defects, and synchronously upload the quality data to the digital twin closed-loop optimization layer, providing data support for quality closed-loop optimization. The workspace environment perception module uses a combination of lidar and depth cameras to collect real-time dynamic obstacle data within the production line workspace, providing environmental data support for collaborative obstacle avoidance and improving operational safety under dynamic conditions. The robot status monitoring module, through the robot's built-in encoder, torque sensor, vibration sensor, and temperature sensor, collects real-time robot operating status data, enabling fault warning and real-time monitoring, providing data support for fault tolerance.

[0008] Furthermore, the digital twin closed-loop optimization layer includes a production line digital twin modeling module, a collaborative operation simulation verification module, a process parameter optimization module, and an operation data visualization module. The production line digital twin modeling module is used to construct a 1:1 digital twin model of the physical production line, achieving real-time bidirectional data mapping with the physical production line, with a data update frequency ≥100Hz, ensuring the synchronization between the virtual model and the physical production line. The collaborative operation simulation verification module is used to perform simulation pre-runs of path planning, task allocation, and timing synchronization of multi-robot collaborative operations in a virtual environment before production tasks are issued, identifying and avoiding motion interference risks in advance, generating initial operation plans, and reducing on-site debugging costs and operational risks. The process parameter optimization module incorporates a process parameter optimization model based on a BP neural network, used to analyze quality data uploaded by the online quality inspection module, identify the root causes of quality deviations, generate corresponding process parameter optimization instructions, and send them to the edge collaborative control layer to achieve dynamic optimization of the processing technology and closed-loop quality control. The operational data visualization module is used to display production line operating efficiency, product yield, robot operating status, and fault warning information in real time, realizing visualized management and control of the entire production line process, and facilitating real-time monitoring of the production line status by operation and maintenance personnel.

[0009] Furthermore, in the dual-redundant communication bus module, the end-to-end communication latency of the TSN time-sensitive network is ≤1ms and the jitter is ≤100ns, which is used for the deterministic control command transmission of fixed equipment on the production line; the communication latency of the 5G industrial private network is ≤5ms, which is used for seamless roaming access and data transmission of mobile robot units, ensuring communication reliability in all scenarios.

[0010] Compared with the prior art, the present invention has the following outstanding advantages: This invention adopts a cloud-edge-device distributed collaborative control architecture, breaking through the computing power bottleneck of centralized control. Combined with a dynamic task allocation algorithm based on DQN deep reinforcement learning, the task reassignment response latency is ≤5ms when facing dynamic working conditions, which greatly improves the real-time performance and responsiveness of flexible manufacturing and adapts to the needs of mixed-line production of multiple varieties and small batches.

[0011] This invention designs a universal heterogeneous robot protocol adaptation module. Through a unified device abstraction interface and protocol conversion unit, it is compatible with mainstream industrial bus protocols and vendor proprietary protocols, enabling plug-and-play heterogeneous robots without the need for customized secondary development, significantly shortening the system deployment cycle and reducing production line transformation costs.

[0012] This invention constructs a full-process, multi-dimensional perception system, combined with TSN low-latency deterministic communication, to realize real-time dynamic planning and collaborative obstacle avoidance of multi-robot motion paths. It achieves precise temporal synchronization in overlapping work spaces, with a 100% success rate in avoiding motion interference and a motion control accuracy of ≥±0.02mm, significantly improving the accuracy and safety of collaborative operations.

[0013] This invention establishes a digital twin-driven quality closed-loop optimization system, realizing real-time bidirectional mapping between physical production lines and virtual models. Based on online quality inspection data, it achieves dynamic optimization of process parameters, which can improve product yield by more than 15%, while realizing traceability of the entire processing process and improving product quality control capabilities.

[0014] This invention designs a fault tolerance and seamless task takeover mechanism based on the Raft consensus election mechanism. When a single robot fails, it can achieve millisecond-level task redistribution and seamless takeover with a takeover time of ≤10ms, avoiding a complete production line shutdown and increasing the system's continuous operation rate to over 99.95%, thus significantly improving the operational stability of the production line. Detailed Implementation

[0015] Specific embodiments of the present invention provide a detailed description of a multi-robot intelligent manufacturing collaborative system. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments; however, those skilled in the art can use other alternative methods to implement some well-known technologies.

[0016] This embodiment discloses a multi-robot intelligent manufacturing collaboration system, which is applied to an intelligent manufacturing production line for welding automotive chassis subframes. It enables multi-robot collaborative operation throughout the entire process of subframe loading, positioning, welding, inspection, and unloading, solving the problems of difficult adaptation of heterogeneous robots, unstable welding quality, long downtime due to malfunctions, and low flexibility in traditional welding production lines.

[0017] Specifically, the production line in this embodiment is equipped with 6 heterogeneous welding robots (4 ABBIRB1410 six-axis welding robots, 2 KUKA KR6R900 six-axis welding robots), 2 Siasun SR10C six-axis handling robots, and 1 Siasun AGV mobile robot, for a total of 9 heterogeneous robot units.

[0018] The multi-robot intelligent manufacturing collaboration system in this embodiment has an overall architecture including an edge collaborative control layer, a multi-robot cluster execution layer, a full-process perception layer, a digital twin closed-loop optimization layer, and a TSN+5G dual-redundant communication bus module.

[0019] The dual-redundant communication bus module connects the fixed robots and testing equipment on the production line via a time-sensitive network (TSN), ensuring a deterministic transmission of welding control commands with an end-to-end communication latency of ≤1ms and jitter of ≤100ns. The AGV mobile robot accesses the system via a 5G industrial private network, with a communication latency of ≤5ms, enabling seamless roaming and data transmission within the production line.

[0020] The multi-robot cluster execution layer consists of the aforementioned nine heterogeneous robot units. Each robot is equipped with a local edge controller, a corresponding end effector (welding gun / clamping fixture), a six-dimensional force control sensor, and a status acquisition unit. The local edge controller adopts an STM32H743 industrial-grade controller, which has the ability to make autonomous decisions for single-machine operation, and at the same time receives collaborative control commands from the edge collaborative control layer through a dual redundant communication bus module.

[0021] The entire process perception layer is deployed at each workstation on the production line, specifically including: Material Incoming Position Perception Module: Equipped with two sets of 3D structured light vision cameras, installed at the loading station, to collect the position and orientation data of the subframe blank in real time, with a positioning accuracy of ±0.05mm. The data is transmitted to the edge collaborative control layer for adjusting the loading and positioning path of the handling robot. Online quality inspection module: Equipped with one 3D laser vision inspection camera, installed at the welding completion station, to detect defects such as weld width, height, excess height, incomplete penetration, and porosity in real time. The inspection data is uploaded to the digital twin closed-loop optimization layer in real time. Workspace Environment Perception Module: Equipped with 4 sets of 2D LiDAR, deployed at the four corners of the production line workspace, to collect dynamic obstacle data in the workspace in real time. At the same time, combined with the real-time movement position of each robot, it provides environmental data support for collaborative obstacle avoidance. Robot Status Monitoring Module: Through the encoders, torque sensors, vibration sensors and temperature sensors built into each robot, the module collects real-time operating status data such as motor speed, joint torque, vibration amplitude and winding temperature of the robot to achieve fault early warning and real-time monitoring.

[0022] The digital twin closed-loop optimization layer, deployed in the industrial control computer of the production line, specifically includes: Production line digital twin modeling module: Uses Unity3D to build a 1:1 digital twin model of the production line, including 3D models of all robots, tooling fixtures, and testing equipment, to achieve real-time bidirectional data mapping with the physical production line, with a data update frequency of 100Hz; Collaborative operation simulation verification module: After receiving the production order, the module first conducts simulation verification of multi-robot welding operations in a virtual environment, completes the pre-rehearsal of path planning, task allocation, and timing synchronization, identifies and avoids motion interference risks in advance, and generates an initial welding operation plan. Process parameter optimization module: It has a built-in welding process parameter optimization model based on BP neural network. It takes weld defect data from online quality inspection as input, outputs optimized values ​​of parameters such as welding current, voltage, welding speed, and wire feeding speed, and sends them down to the edge collaborative control layer to dynamically adjust the operating parameters of the welding robot. Operational data visualization module: Using configuration software, the system displays the production line's operating efficiency, product welding yield, robot operating status, and fault warning information in real time, enabling visualized control of the entire production line process.

[0023] The edge collaborative control layer is deployed in an industrial edge control cabinet and uses an edge controller equipped with an NVIDIA Jetson AGX Orin edge computing module. Specifically, it includes a global task scheduling module, a heterogeneous robot protocol adaptation module, a dynamic collaborative motion control module, and a fault tolerance and task takeover module. Global Task Scheduling Module: Equipped with a task scheduling algorithm based on DQN deep reinforcement learning, this module optimizes production order delivery cycles, robot load rates, and production line welding efficiency. It breaks down welding production orders into seven sub-tasks: loading, positioning, root pass welding, fill pass welding, cover pass welding, inspection, and unloading. Initial task allocation is completed based on each robot's capabilities, load, and position. In case of urgent orders or equipment malfunctions, task reassignment can be completed within 5ms. Heterogeneous robot protocol adaptation module: It has a built-in unified robot equipment abstract model and protocol conversion unit, which is compatible with Profinet and EtherCAT industrial bus protocols, and also adapts to ABRapid, KUKA KRL and Siasun SRSP robot control languages, enabling automatic identification and plug-and-play of 9 heterogeneous robots without the need for customized secondary development. Dynamic Cooperative Motion Control Module: Employing a multi-robot cooperative motion planning algorithm based on Model Predictive Control (MPC), combined with real-time material arrival pose and environmental data from the full-process perception layer, the module dynamically adjusts the motion paths of each robot in real time. Within the overlapping space of welding operations, it achieves temporal synchronization and cooperative obstacle avoidance among multiple robots, with a welding path control accuracy of ±0.02mm and a 100% success rate in avoiding motion interference. Fault tolerance and task takeover module: Based on the Raft consensus election mechanism, it receives real-time operation data from the robot status monitoring module. When a fault is detected in a single welding robot, the global task scheduling module is immediately triggered to complete task reassignment, assigning the welding station tasks of the faulty robot to adjacent welding robots of the same type. At the same time, the operation sequence of each robot is adjusted to achieve seamless task takeover with a takeover time of ≤10ms and no production line downtime required.

[0024] The workflow of the multi-robot intelligent manufacturing collaboration system in this embodiment is as follows: S1, Digital twin modeling and pre-simulation of the production line: The digital twin closed-loop optimization layer constructs a 1:1 digital twin model of the automotive chassis welding production line. After receiving the subframe welding production order, the simulation verification of the collaborative welding operation of 9 robots is completed in the virtual environment to avoid the risk of welding path interference and generate the initial welding operation plan. S2. Heterogeneous robot adaptation and task assignment: The heterogeneous robot protocol adaptation module of the edge collaborative control layer automatically identifies the 9 heterogeneous robots connected and completes protocol adaptation and interconnection; the global task scheduling module breaks down the initial welding operation plan into 7 sub-tasks, completes the initial task allocation, and assigns the tasks to the corresponding robot units. S3. Real-time perception and dynamic collaborative control: The full-process perception layer collects data on material pose, working environment, and robot operating status in real time and transmits it to the edge collaborative control layer; the dynamic collaborative motion control module performs real-time dynamic planning of the welding path and handling path of each robot based on real-time data, realizes time synchronization and collaborative obstacle avoidance in the overlapping work space, and completes the collaborative welding operation of the subframe. S4. Quality Closed-Loop Feedback and Process Optimization: The online quality inspection module uploads real-time weld quality data to the digital twin closed-loop optimization layer. The process parameter optimization module analyzes the weld defect data, generates welding process parameter optimization instructions, and sends them to the edge collaborative control layer to dynamically adjust parameters such as welding current, voltage, and welding speed of the welding robot, thereby achieving closed-loop control of welding quality. S5. Fault Tolerance and Task Takeover: When the robot status monitoring module detects a fault in a single welding robot, it immediately uploads the fault information to the edge collaborative control layer. The fault tolerance and task takeover module immediately triggers task reassignment, assigning the faulty task to an adjacent normal welding robot, adjusting the operation sequence, achieving seamless task takeover, and ensuring continuous operation of the production line. S6. Full-process data iterative optimization: The digital twin closed-loop optimization layer stores and analyzes welding data, quality data, and fault data throughout the entire production line process. Through machine learning algorithms, it continuously optimizes the models of task scheduling, motion planning, and welding process parameters to achieve continuous iterative optimization of the system.

[0025] This embodiment, through the aforementioned system, realizes multi-robot collaborative operation throughout the entire process of automotive chassis subframe welding production line. The heterogeneous robots can be plugged and used without customized development, shortening the production line deployment cycle by more than 60%. The task response latency is ≤5ms when facing dynamic working conditions, and it can adapt to mixed-line production of multiple subframe models. The yield rate of welded products has been increased from 82% in traditional production lines to over 98%. The production line does not need to stop when a single robot fails, and the system's continuous operation rate has been increased to over 99.95%, achieving significant technical results.

[0026] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0027] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A multi-robot intelligent manufacturing collaborative system, characterized in that, The system includes an edge collaborative control layer, a multi-robot cluster execution layer, a full-process perception layer, a digital twin closed-loop optimization layer, and a supporting dual-redundant communication bus module. The dual-redundant communication bus module adopts a dual-redundant architecture of TSN (Time-Sensitive Network) and 5G industrial private network to achieve low-latency deterministic data transmission between each layer. The full-process perception layer is signal-connected to the edge collaborative control layer and the digital twin closed-loop optimization layer, and is used to collect real-time data on material arrival pose, operating environment, robot operating status, and processing quality, and synchronously upload the collected data to the edge collaborative control layer and the digital twin closed-loop optimization layer. The digital twin closed-loop optimization layer is signal-connected to the edge collaborative control layer and is used for... A 1:1 digital twin model of the production line is constructed to complete the simulation pre-verification of collaborative operations. Based on the quality data uploaded by the full-process perception layer, process parameter optimization instructions are generated and sent to the edge collaborative control layer. The edge collaborative control layer is the core control center of the system, connected to the multi-robot cluster execution layer. It is used to receive production orders and optimization instructions from the digital twin closed-loop optimization layer, complete the protocol adaptation of heterogeneous robots, dynamic allocation of global tasks, multi-robot collaborative motion control, and seamless task takeover under fault conditions. The multi-robot cluster execution layer includes several heterogeneous robot units, which are used to receive control instructions from the edge collaborative control layer and complete the collaborative operation of the corresponding processing steps.

2. The multi-robot intelligent manufacturing collaborative system according to claim 1, characterized in that, The edge collaborative control layer includes a global task scheduling module, a heterogeneous robot protocol adaptation module, a dynamic collaborative motion control module, and a fault tolerance and task takeover module. The global task scheduling module is equipped with a task scheduling algorithm based on DQN deep reinforcement learning. With production line delivery cycle, robot load balancing rate, and production efficiency as optimization objectives, it decomposes production orders into executable sub-tasks and completes the initial allocation of sub-tasks and task reassignment under dynamic working conditions. The heterogeneous robot protocol adaptation module has a built-in unified robot equipment abstract model and protocol conversion unit, compatible with mainstream industrial bus protocols such as Profinet, EtherCAT, Modbus, and DeviceNet, and adapts to multiple product types. The proprietary control protocol of the robot enables automatic identification and plug-and-play adaptation of heterogeneous robot units. The dynamic cooperative motion control module adopts a multi-robot cooperative motion planning algorithm based on model predictive control (MPC), combined with real-time data collected from the full-process perception layer, to perform real-time dynamic planning of the motion paths of each robot unit, achieving temporal synchronization and cooperative obstacle avoidance of multiple robots in overlapping work spaces. The fault tolerance and task takeover module is signal-connected to the robot state monitoring module of the full-process perception layer. When a robot unit fault is detected, it triggers the global task scheduling module to complete task reassignment, allocating the tasks to be executed by the faulty robot to the normally operating robot unit, achieving seamless task takeover.

3. The multi-robot intelligent manufacturing collaborative system according to claim 2, characterized in that, The task reassignment response delay of the global task scheduling module is ≤5ms, and the motion control accuracy of the dynamic cooperative motion control module is ≥±0.02mm.

4. The multi-robot intelligent manufacturing collaborative system according to claim 1, characterized in that, Each heterogeneous robot unit in the multi-robot cluster execution layer is equipped with a local edge controller, an end effector, a six-dimensional force control sensor, and a status acquisition unit. The local edge controller has the ability to make autonomous decisions for single-machine operation, and at the same time receives collaborative control commands from the edge collaborative control layer through a dual-redundant communication bus module.

5. The multi-robot intelligent manufacturing collaborative system according to claim 1, characterized in that, The entire process perception layer includes an incoming material pose perception module, an online quality inspection module, a workspace environment perception module, and a robot status monitoring module. The incoming material pose perception module uses a 3D structured light vision camera to collect real-time pose data of the incoming workpiece, providing data support for path adjustment of the handling and processing robots. The online quality inspection module uses a 3D laser vision inspection unit and a high-precision dimensional inspection device to collect real-time quality data of the processed workpiece, identify processing defects, and synchronously upload the quality data to the digital twin closed-loop optimization layer. The workspace environment perception module uses a combination of lidar and a depth camera to collect real-time dynamic obstacle data within the production line workspace, providing environmental data support for collaborative obstacle avoidance. The robot status monitoring module uses the robot's built-in encoder, torque sensor, vibration sensor, and temperature sensor to collect real-time operating status data of the robot, enabling fault warning and real-time monitoring.

6. The multi-robot intelligent manufacturing collaborative system according to claim 1, characterized in that, The digital twin closed-loop optimization layer includes a production line digital twin modeling module, a collaborative operation simulation verification module, a process parameter optimization module, and an operation data visualization module. The production line digital twin modeling module is used to construct a 1:1 digital twin model of the physical production line, realize real-time bidirectional data mapping with the physical production line, and the data update frequency is ≥100Hz. The collaborative operation simulation verification module is used to complete the simulation pre-run of path planning, task allocation and timing synchronization of multi-robot collaborative operation in a virtual environment before the production task is issued, identify and avoid motion interference risks in advance, and generate an initial operation plan. The process parameter optimization module has a built-in process parameter optimization model based on a BP neural network. It is used to analyze the quality data uploaded by the online quality inspection module, identify the root causes of quality deviations, generate corresponding process parameter optimization instructions, and send them to the edge collaborative control layer. The operation data visualization module is used to display the production line operation efficiency, product yield, robot operation status, and fault warning information in real time, so as to realize the visualized management and control of the entire production line process.

7. The multi-robot intelligent manufacturing collaborative system according to claim 1, characterized in that, In the dual-redundant communication bus module, the end-to-end communication latency of the TSN (Time-Sensitive Network) is ≤1ms, and the jitter is ≤100ns, which is used for the deterministic control command transmission of fixed equipment on the production line; the communication latency of the 5G industrial private network is ≤5ms, which is used for seamless roaming access and data transmission of mobile robot units.

8. The multi-robot intelligent manufacturing collaborative system according to claim 2, characterized in that, The fault tolerance and task takeover module is based on the Raft consensus election mechanism to achieve rapid identification and task takeover of faulty nodes. The seamless task takeover time when a single robot fails is ≤10ms.