Multi-robot synchronous cooperation scheduling method in industrial scene

Abstract

The invention discloses a multi-robot synchronous cooperation scheduling method in an industrial scene, and particularly relates to the technical field of intelligent manufacturing, and the method comprises the following steps: S1, a sensing module collects a machine state, environment information and task data; s2, carrying out primary processing and feature extraction by an edge calculation layer; s3, the cloud decision-making layer carries out global optimization and collaborative strategy making; s4, the system executes cooperative scheduling; the system response speed and efficiency are remarkably improved through the three-layer architecture design, the end-to-end delay is reduced, efficient allocation of resources is achieved through an LSTM + Attention algorithm and a dynamic resource pool mechanism, the resource utilization rate is increased, high-priority task execution is ensured when resources are insufficient, and the system performance is improved. The digital twin technology enhances the adaptability of the system to a complex dynamic environment, the edge computing architecture reduces the dependence on a centralized server, and the problems that a traditional system is uneven in resource allocation, high in delay and poor in adaptability are solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and more specifically, to a scheduling method for synchronous collaboration of multiple robots in industrial scenarios. Background Technology

[0002] With the rapid development of industrial automation and intelligent manufacturing technologies, the collaborative model between machines and human robots on production lines is becoming a key trend for improving production efficiency. Currently, many production lines still rely on preset scheduling strategies, such as first-come, first-served or priority scheduling. This approach is ill-suited to increasingly complex production demands. To address these issues, the industry is exploring new technological solutions. For example, some research attempts to apply digital twin technology to machine collaborative systems to achieve deep collaboration between virtual and physical spaces. Other work focuses on developing deep learning-based prediction algorithms to improve the accuracy and efficiency of task execution. However, the above technologies have the following drawbacks in industrial scenarios: the machine control system lacks the ability to adapt to complex dynamic environments and has difficulty handling both sudden and normal situations at the same time, resulting in high task processing delays and low efficiency; the resource scheduling system fails to fully consider task characteristics and resource status, and cannot achieve efficient allocation and scheduling of resources, resulting in low system resource utilization and affecting production efficiency.

[0003] Therefore, a scheduling method for multi-robot synchronous collaboration in industrial scenarios is proposed to address the above problems. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a scheduling method for multi-robot synchronous cooperation in industrial scenarios to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a scheduling method for multi-robot synchronous cooperation in industrial scenarios, comprising the following steps: S1. A sensing module is deployed on a robot in an industrial setting. The sensing module is a sensor device used to collect the robot's status, environmental information and task data. S2. Deploy an edge server on one side of the robot. The edge server is equipped with a predictive scheduling module. The predictive scheduling module, combined with a dynamic resource pool mechanism, realizes intelligent scheduling of tasks and resource allocation. S3. Build a cloud-based decision-making layer. The cloud-based decision-making layer communicates with the edge server. The cloud-based decision-making layer is located in the data center and is used to optimize computing and global resource scheduling. The cloud-based decision-making layer constructs a digital model of the physical entity through an internal digital twin mapping module. S4. Calculate resource priority based on the urgency and resource weight of the task, and break down complex tasks into appropriate sub-tasks through task decomposition and reasonable scheduling, and make reasonable arrangements according to resource conditions and task urgency. S5. An anomaly detection and feedback mechanism is set up during the operation of the robot. When the anomaly detection and feedback mechanism detects an anomaly, it will automatically generate a detailed anomaly report and send it to the cloud decision-making layer through the communication network for analysis and optimization.

[0006] Preferably, in step S1, the sensing module includes, but is not limited to, a high-definition camera, a lidar, an ultrasonic sensor, and a magnetometer, and the sensing module is connected to the edge server via an industrial Ethernet.

[0007] Preferably, in step S2, the prediction scheduling module, based on the LSTM+Attention mechanism and combined with the dynamic resource pool mechanism, realizes intelligent task scheduling and resource allocation. The specific implementation of the dynamic resource pool mechanism includes the following steps: A1. Resource tokenization: Assigning corresponding virtual tokens to various computing resources, communication resources and actuator resources, wherein the token represents a unit of usage right of the resource; A2. Priority Calculation: Dynamically calculate the resource acquisition priority for each task based on the task urgency and resource weight; A3. Token Allocation and Exchange: Based on the priority, the resource tokens are allocated to high-priority tasks; A4. Resource Recycling and Redistribution: The resource tokens occupied by completed tasks or tasks in an idle state are recycled in real time and added back to the resource pool for subsequent task scheduling.

[0008] Preferably, the edge server adopts a modular design and also includes a data preprocessing module, a feature extraction module, and a resource allocation module. The edge server is deployed next to the production line within 30 meters of the robot, and the edge server has an IP65 protection rating and an active heat dissipation system to ensure stability during high-load operation.

[0009] Preferably, the cloud-based decision-making layer further includes a task priority calculation module, a resource prediction module, and a collaborative strategy optimization module. The cloud-based decision-making layer communicates with the edge server through a 5G or Wi-Fi 6 network, thereby supporting the issuance and feedback of real-time tasks.

[0010] Preferably, the resource priority calculation performed by the prediction and scheduling module adopts a token-based dynamic resource pool mechanism, and the resource priority calculation formula is as follows:

[0011] in, Indicates the urgency of the task; This indicates the resource weight, and resource allocation supports the exchange of usage rights.

[0012] Preferably, the system performance indicators implemented by the scheduling method include: end-to-end latency from the acquisition of data by the perception module to the issuance of execution instructions, system resource utilization, relative position error of multiple groups of robots cooperating, and success rate of cooperative tasks.

[0013] Preferably, in step S1, the sampling rate of the data collected by the sensing module is set to 60fps for visual data, and the sensor data of the sensing module can achieve millisecond-level data acquisition.

[0014] Preferably, the edge server and the cloud decision layer adopt a microservice architecture for Docker containerized deployment, the data stream adopts a publishing mode, the message queue processing capacity is not less than 10,000 messages / second, and the edge server and the cloud decision layer use a hybrid storage mode of time-series database and relational database.

[0015] Preferably, in step S5, the anomaly detection and feedback mechanism further includes support for online upgrades and dynamic collaborative strategy adjustments based on real-time feedback, wherein the dynamic collaborative strategy adjustments are based on the robot's mean time between failures (MTBF) during accelerated life testing and simulated fault injection testing.

[0016] The technical effects and advantages of this invention are as follows: 1. Compared with existing technologies, the scheduling method for multi-robot synchronous collaboration in this industrial scenario introduces a three-layer architecture design, which realizes a complete closed loop from data acquisition to decision execution, significantly improving the system's response speed and efficiency. By deploying lightweight prediction algorithms and collaborative scheduling models at the edge computing layer, it handles real-time computing needs and reduces system response latency.

[0017] 2. Compared with existing technologies, the scheduling method for multi-robot synchronous collaboration in this industrial scenario adopts a predictive scheduling algorithm based on LSTM+Attention and combines it with a dynamic resource pool mechanism to achieve efficient allocation and scheduling of resources. It improves the system resource utilization rate through a resource usage right exchange mechanism and ensures the execution of high-priority tasks when resources are scarce, thus solving the problem of uneven resource allocation in traditional systems.

[0018] 3. Compared with existing technologies, the scheduling method for multi-robot synchronous collaboration in this industrial scenario constructs a digital model of the physical entity through digital twin technology, realizing deep collaboration and pre-demonstration verification in virtual and real spaces, enhancing the system's adaptability to complex dynamic environments, and enabling the system to better cope with emergencies and normal situations.

[0019] 4. Compared with existing technologies, the scheduling method for multi-robot synchronous collaboration in this industrial scenario reduces the dependence on centralized servers by adopting an edge computing architecture, effectively reducing the overall system latency, and improving the system's parallel processing capability by using multi-machine collaborative operations. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention.

[0021] Figure 2 This is a flowchart of the dynamic resource pool mechanism of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1 As attached Figures 1 to 2 The scheduling method for multi-robot synchronous collaboration in the industrial scenario shown includes the following steps: S1. Deploy a sensing module on the robot in the industrial setting. The sensing module is a sensor device used to collect the robot's status, environmental information and task data. S2. Deploy an edge server on one side of the robot. The edge server is equipped with a predictive scheduling module. The predictive scheduling module, combined with a dynamic resource pool mechanism, realizes intelligent scheduling of tasks and resource allocation. S3. Build a cloud-based decision-making layer. The cloud-based decision-making layer communicates and connects with the edge server. The cloud-based decision-making layer is located in the data center and is used to optimize computing and global resource scheduling. The cloud-based decision-making layer constructs a digital model of the physical entity through its internal digital twin mapping module. S4. Calculate resource priority based on task urgency and resource weight. At the same time, break down complex tasks into appropriate sub-tasks through task decomposition and reasonable scheduling, and make reasonable arrangements based on resource availability and task urgency. S5. Set up an anomaly detection and feedback mechanism during robot operation. When the anomaly detection and feedback mechanism detects an anomaly, it will automatically generate a detailed anomaly report and send it to the cloud decision-making level through the communication network for analysis and optimization.

[0024] Specifically, the sensing module includes a distributed sensor network for collecting machine status, environmental information, and task data. This layer is equipped with various sensor devices, including but not limited to: high-definition cameras, LiDAR, ultrasonic sensors, and magnetometers. These sensors connect to the edge server via Industrial Ethernet or 5G communication networks to achieve real-time data acquisition at the millisecond level.

[0025] The edge server is equipped with an edge computing layer, which is deployed on the edge server closest to the machine. This layer includes lightweight prediction algorithms and a collaborative scheduling model. It adopts a modular design, comprising a data preprocessing module, a feature extraction module, a prediction scheduling module, and a resource allocation module. The prediction scheduling module is based on an LSTM+Attention mechanism combined with a dynamic resource pool mechanism to achieve intelligent task scheduling and resource allocation. The edge computing layer also integrates an active cooling system to ensure the stability of the device under high load. The feature extraction module uses principal component analysis and linear discriminant analysis techniques to extract key features of the task.

[0026] The cloud-based decision-making layer, located in the data center, is responsible for complex optimization calculations and global resource scheduling. This layer includes a task priority calculation module, a resource prediction module, a digital twin mapping module, and a collaborative strategy optimization module. The digital twin mapping module constructs digital models of physical entities, enabling deep collaboration and pre-simulation verification between virtual and physical spaces. The cloud-based decision-making layer communicates with the edge computing layer via 5G or Wi-Fi 6 networks to ensure real-time task distribution and feedback mechanisms.

[0027] The system employs a token-based resource allocation mechanism, calculating resource priority based on task urgency and resource weight. This mechanism supports dynamic resource adjustment and reclamation, ensuring maximum system resource utilization. Simultaneously, the system incorporates a task decomposition and scheduling mechanism, breaking down complex tasks into appropriate sub-tasks and allocating them rationally based on resource availability and task urgency.

[0028] The system also features a comprehensive anomaly monitoring and feedback mechanism. Once an anomaly is detected, the system automatically generates a detailed anomaly report and sends it to the cloud-based decision-making level via the communication network for analysis and optimization. The system supports online upgrades and optimizations, and can dynamically adjust collaborative strategies based on actual task requirements.

[0029] In a preferred embodiment, in step S1, the sensing module includes, but is not limited to, a high-definition camera, a lidar, an ultrasonic sensor, and a magnetometer, and the sensing module is connected to the edge server via an industrial Ethernet.

[0030] In a preferred embodiment, in step S2, the prediction scheduling module, based on the LSTM+Attention mechanism and combined with the dynamic resource pool mechanism, realizes intelligent task scheduling and resource allocation. The specific implementation of the dynamic resource pool mechanism includes the following steps: A1. Resource tokenization: Assigning corresponding virtual tokens to various computing resources, communication resources and actuator resources. Each token represents a unit of usage rights of the resource. A2. Priority Calculation: Dynamically calculate the resource acquisition priority for each task based on task urgency and resource weight; A3. Token Allocation and Exchange: Based on priority, resource tokens are allocated to high-priority tasks; A4. Resource Recycling and Redistribution: Real-time recycling of resource tokens used by completed or idle tasks, and their return to the resource pool for subsequent task scheduling.

[0031] As a preferred implementation, the edge server adopts a modular design and also includes a data preprocessing module, a feature extraction module, and a resource allocation module. The edge server is deployed next to the production line within 30 meters of the robot, and the edge server has an IP65 protection rating and an active heat dissipation system to ensure stability during high-load operation.

[0032] In a preferred implementation, the resource priority calculation performed by the prediction scheduling module adopts a token-based dynamic resource pool mechanism, and the formula for calculating the resource priority is as follows:

[0033] in, Indicates the urgency of the task; This represents resource weights. Resource allocation supports the exchange of usage rights, and in a multi-robot system, the resource allocation problem can be modeled as a multi-dimensional game equilibrium problem.

[0034]

[0035] in, Let i represent the utility function of robot i; This represents the resources allocated to robot i; C represents the total resource constraint. The resource optimization model based on NASH balance and ADMM algorithm reduces the standard deviation of resource utilization and improves the system's resource utilization efficiency.

[0036] As a preferred implementation, a three-layer architecture design ensures the achievement of system performance indicators. The control of end-to-end latency relies on the high-speed data acquisition of the perception layer and the real-time processing capability of the edge computing layer. The perception layer uses a distributed sensor network to achieve millisecond-level data acquisition, and the edge computing layer deploys a lightweight prediction algorithm to handle real-time computing needs, reducing the transmission overhead of data uploading to the cloud, thereby optimizing end-to-end latency. The principle is based on hierarchical computing load distribution and network optimization. Edge nodes process data locally to avoid the latency bottleneck of centralized cloud computing. The implementation method includes using a low-latency communication interface supported by Time Sensitive Network (TSN) and a cooperative scheduling controller with a control cycle of 1 millisecond. As a preferred implementation, a microservice architecture is used for Docker containerized deployment between the edge server and the cloud decision layer, the data stream adopts a publishing mode, the message queue processing capacity is no less than 10,000 messages / second, and a hybrid storage mode of time-series database and relational database is used between the edge server and the cloud decision layer. In the edge computing node processing capacity, assuming the system needs to process n parallel robot tasks, and each task requires an average of m computation cycles, then the total computational requirement is: For typical edge servers such as NVIDIA Jetson AGX Orin, it can provide 120 TOPS of computing power at 45W power consumption, which can meet the collaborative scheduling needs of 10-20 robots. The system end-to-end delay is calculated as follows:

[0037] The delays for each component are: This indicates that sensor acquisition time is 5-10ms. Edge computing processing time is 20-30ms. This indicates a communication delay of 5-10ms. This indicates an execution delay of 10-20ms. Total delay The response time is controlled within 50ms, meeting the real-time control requirements of industrial scenarios.

[0038] As a preferred implementation, based on the closed-loop control principle of edge-cloud collaboration, a lightweight anomaly detection model is deployed at the edge computing layer to monitor the robot's operating status and environmental data in real time. Its design principle is to use the low latency characteristics of edge nodes to achieve initial screening and rapid response of anomalies, while uploading complex data to the cloud decision layer for in-depth analysis and strategy optimization. This includes setting up a rule engine and threshold triggers on the edge side to make millisecond-level judgments on sudden changes or exceeding limits in sensor data and trigger local emergency commands. For example, when the collaborative position error approaches the tolerance or resource utilization fluctuates abnormally, the scheduling strategy is adjusted immediately to prevent the spread of local faults and maintain the basic operation of the system. The cloud-based decision-making layer relies on the digital twin model to perform root cause analysis on the abnormal data reported from the edge. The principle is to simulate the path of failure impact and predict system behavior through a high-precision virtual model, build a failure prediction model by using the historical data learning capability of LSTM plus Attention mechanism, and dynamically adjust the task allocation strategy by combining the dynamic resource pool mechanism. For example, when the performance degradation of a certain robot is predicted, its task is migrated to the backup unit in advance. The online upgrade function is implemented through a microservice architecture and Docker containerized deployment. The principle is that the modular design allows for hot service updates without interrupting system operation. This includes periodically downloading optimized algorithm models and strategy parameters from the cloud to edge nodes, enabling the system to adapt to complex environmental changes and continuously optimize. The verification of dynamic collaborative strategy adjustment is based on accelerated life testing and simulated fault injection methods. The principle is to evaluate the system's extreme reliability through stress acceleration and fault scenario simulation, inject fault modes such as communication delay, node downtime or resource contention in the digital twin environment, observe the system's self-healing ability and iteratively optimize the strategy parameters.

[0039] As a preferred implementation, the sampling rate of the perception module is set based on the system's coordinated requirements for the real-time performance and accuracy of multi-source information. It balances the amount of data and the processing load through a hierarchical sampling strategy. Visual data uses a frame rate of 60fps to ensure continuous capture of the robot's motion trajectory and dynamic environment, avoiding the loss of motion details due to insufficient sampling. At the same time, sensor data uses high-frequency sampling of 1kHz to achieve millisecond-level response. The principle is that high-frequency sampling can capture instantaneous state changes and provide a high-timeliness data foundation for the prediction algorithm of the edge layer. For non-visual sensors such as LiDAR and ultrasonic sensors, parameters such as distance, temperature and vibration are collected at a frequency of 1kHz through 32 digital I / O and 8 analog input interfaces. Hardware triggering and timer interrupt mechanisms are used to ensure the accuracy of sampling timing, and parallel acquisition channels are used to reduce signal transmission delay.

[0040] Example 2 The main hardware configurations used in the implementation are as follows for reference: Perception module: standard 19-inch rack mount, interface: supports 16 GigE Vision cameras, 32 digital I / O, 8 analog inputs, sampling rate: 60fps for visual data, 1kHz for sensor data; Edge server: 300mm×250mm×100mm, IP65 protection rating, active cooling system, maximum heat dissipation power 80W, deployed next to the production line, with the maximum distance controlled within 30m; Cooperative scheduling controller: 400mm×350mm×150mm, communication interface supports EtherCAT, PROFINET, OPC UA, real-time performance with a control cycle of 1ms and jitter of less than 100μs; The core of resource management adopts a microservice architecture, Docker containerized deployment, publish or subscribe data flow, message queue processing capacity of 10,000 msg / s, and uses a hybrid storage of time-series database and relational database; The cloud-based decision-making layer is configured with an NVIDIA Jetson AGX Orin (64GB) processor, 32GB LPDDR5 memory, 1TB NVMe SSD storage, and a 10GbE interface supporting TSN. Its employee number is 45W.

[0041] The working process of this invention is as follows: The specific implementation steps are as follows: First, the sensing module collects machine status, environmental information, and task data. Through a distributed sensor network, including high-definition cameras, LiDAR, ultrasonic sensors, and magnetometers, comprehensive monitoring of the production environment is achieved. These sensors connect to the edge computing layer via industrial Ethernet or 5G communication networks to ensure real-time data acquisition and transmission. The edge computing layer performs preliminary processing and feature extraction. The data preprocessing module standardizes the collected data, while the feature extraction module uses principal component analysis and linear discriminant analysis to extract key features from the raw data. The prediction and scheduling module, based on the LSTM+Attention mechanism and combined with a dynamic resource pool mechanism, achieves intelligent task scheduling and resource allocation. The cloud-based decision-making layer performs global optimization and collaborative strategy formulation. The task priority calculation module calculates the priority of each task based on its urgency and resource requirements. The resource prediction module uses time series analysis and recurrent neural networks to predict future resource needs. The digital twin mapping module constructs digital models of physical entities, enabling deep collaboration and pre-simulation verification between virtual and physical spaces. The collaborative strategy optimization module formulates the optimal collaborative strategy based on the overall resource status and task conditions. Then, the system performs collaborative scheduling. Based on task priority and resource prediction results, the cloud decision layer issues scheduling instructions to the edge computing layer. The edge computing layer allocates resources and schedules tasks according to its own resource availability and task characteristics. The system adopts a token-based resource allocation mechanism and a dynamic resource allocation model to ensure efficient resource utilization. Finally, monitoring and feedback. The system has a complete anomaly monitoring mechanism. Once an anomaly is detected, the system automatically generates a detailed anomaly report and sends it to the cloud-based decision-making layer via the communication network for analysis and optimization. The system supports online upgrades and optimizations and can dynamically adjust the collaboration strategy according to the actual task situation. The above describes the working principle of the multi-robot synchronous collaboration scheduling method in this industrial scenario.

Claims

1. A scheduling method for multi-robot synchronous cooperation in an industrial scenario, characterized in that, The method comprises the following steps: S1, laying a perception module on a robot in an industrial scene, the perception module being a sensor device for collecting state, environmental information and task data of the robot; S2, deploying an edge server on one side of the robot, the edge server being provided with a prediction scheduling module, the prediction scheduling module realizing intelligent scheduling and resource allocation of tasks in combination with a dynamic resource pool mechanism; S3, building a cloud decision layer, the cloud decision layer being in communication connection with the edge server, the cloud decision layer being located in a data center for optimized calculation and global resource scheduling, the cloud decision layer constructing a digital model of a physical entity through an internal digital twin mapping module; S4, calculating resource priority according to the urgency of the task and resource weight, and splitting a complex task into moderate subtasks through decomposition and reasonable scheduling of the task, and reasonably arranging according to resource conditions and task urgency; S5, setting an abnormality detection and feedback mechanism in the robot operation, the abnormality detection and feedback mechanism generating a detailed abnormality report automatically when detecting an abnormality and sending the report to the cloud decision layer through a communication network for analysis and optimization.

2. The method of Claim 1, wherein: In step S1, the perception module includes but is not limited to a high-definition camera, a laser radar, an ultrasonic sensor and a magnetometer, and the perception module is connected to the edge server through an industrial Ethernet.

3. The method of Claim 1, wherein: In step S2, the prediction scheduling module is based on an LSTM+Attention mechanism and realizes intelligent scheduling and resource allocation of tasks in combination with a dynamic resource pool mechanism, and the specific implementation of the dynamic resource pool mechanism comprises the following steps: A1, resource tokenization: allocating corresponding virtual tokens to various computing resources, communication resources and executor resources, the token representing the unit usage right of the resource; A2, priority calculation: dynamically calculating the resource acquisition priority of each task according to the task urgency and resource weight; A3, token allocation and exchange: allocating the resource tokens to tasks with high priority based on the priority; A4, resource recycling and redistribution: recycling the resource tokens occupied by completed tasks or tasks in an idle state in real time and re-introducing them into the resource pool for subsequent task scheduling.

4. The method of claim 1, wherein: The edge server adopts a modular design and further comprises a data preprocessing module, a feature extraction module and a resource allocation module, the edge server being deployed beside a production line within a range of 30 meters from the robot, and the edge server having an IP65 protection level and an active cooling system to ensure stability during high-load operation.

5. The method of claim 1, wherein: The cloud decision layer further comprises a task priority calculation module, a resource prediction module and a collaborative strategy optimization module, the cloud decision layer being in communication with the edge server through a 5G or Wi-Fi 6 network to support real-time task issuing and feedback.

6. The method of Claim 3, wherein: The resource priority calculation performed by the prediction scheduling module adopts a token-based dynamic resource pool mechanism, and the resource priority calculation formula is: wherein, represents a task urgency degree; represents a resource weight, and the resource allocation supports the use of right exchange.

7. The method of claim 1, wherein: The system performance indicators realized by the scheduling method include end-to-end delay from the perception module to the issuing of execution instructions, system resource utilization rate, relative position error of multiple groups of robots in cooperation and success rate of collaborative tasks.

8. The method of claim 1, wherein: In step S1, the sampling rate of the perception module is set to 60fps for visual data, and the sensor data of the perception module can achieve millisecond-level data acquisition.

9. The method of claim 1, wherein: The edge server and the cloud decision layer use a micro-service architecture for Docker containerized deployment, and the data stream uses a publish mode, the message queue processing capacity is not less than 10000 / s, and the edge server and the cloud decision layer use a mixed storage mode of a time series database and a relational database.

10. The method of claim 1, wherein: In step S5, the anomaly detection and feedback mechanism further includes supporting online upgrading and dynamic cooperative strategy adjustment based on real-time feedback, and the dynamic cooperative strategy adjustment is based on the average failure interval time of the robot through accelerated life testing and simulated fault injection testing.

Images