Collaborative training control method for intelligent manufacturing collaborative robot edge system

Abstract

The application relates to a collaborative training control method of an intelligent manufacturing collaborative robot edge system, which comprises the following steps: S1, establishing a collaborative robot edge system under an intelligent manufacturing scene, which comprises a collaborative robot module, an edge computing module and a cloud server module; S2, distributing tasks to a set number of edge computing modules by the cloud server module, and uploading equipment data to the connected edge computing modules by the collaborative robot module; S3, deploying a public data set in the cloud server module, and distributing a set proportion of the public data set to all the edge computing modules; S4, training a local model by each selected edge computing module, and calculating corresponding aggregation coefficients; S5, detecting the existence of a malicious client, and performing a semi-asynchronous training mode; and S6, obtaining parameters of a global model through multiple rounds of training, and completing a collaborative training task. Compared with the prior art, the application has the advantages of efficient communication, safety, high training efficiency and the like.

Description

Technical Field

[0001] This invention relates to a control method for a collaborative robot in intelligent manufacturing, and more particularly to a collaborative training control method for an edge system of a collaborative robot in intelligent manufacturing. Background Technology

[0002] With the rapid development of information technology, breakthroughs in deep learning and the Industrial Internet of Things (IIoT) have accelerated the promotion of intelligent manufacturing systems. Specifically, human-machine collaborative tasks in intelligent manufacturing can leverage machine learning, big data, and other related technologies to perform sensing, communication, and computing tasks using measured historical industrial data, optimizing manufacturing decisions. This has become a mainstream trend in the era of intelligent manufacturing, significantly improving the productivity and quality of various products. However, in real-world industrial environments, most collaborative robots (such as robotic arms, AGVs, and other collaborative robots) have limited storage capacity, poor communication quality, and intensive and complex computational tasks. This results in manufacturing systems requiring more time to analyze collected data and build task models, and throughout the entire manufacturing process lifecycle, communication latency is high and network congestion is frequent.

[0003] The emergence of edge computing can effectively improve the collaboration efficiency between humans and collaborative robots, providing a high-level integrated intelligent manufacturing system solution. Edge computing can offload data from mobile or stationary collaborative robots to edge servers via wired and wireless networks, thereby expanding computing power, communication resources, and storage capacity. Based on edge computing technology, different collaborative robots can collaborate better to support sensing, classification, and detection tasks in manufacturing events. This allows industrial data to be transmitted rapidly to edge servers. Edge servers will provide powerful computing capabilities, enabling joint adaptive collaborative task scheduling and intelligent decision-making. Furthermore, edge servers can support different communication protocols to solve communication compatibility issues between different collaborative robots and further support the deployment of new types of collaborative robots. Therefore, it is necessary to design a multi-layered collaborative robot edge system to improve productivity and manufacturing performance in various manufacturing scenarios. However, while edge servers can process large amounts of historical data collected by collaborative robots in real time, issues such as industrial data privacy and collaborative learning efficiency between different edge servers during human-robot collaboration cannot be well resolved. To improve productivity and privacy protection, and considering the challenges encountered in collaborative tasks in intelligent manufacturing systems, designing an efficient distributed training method is essential. Currently, solutions for human-machine collaboration tasks in intelligent manufacturing systems are relatively few. Some research has introduced edge computing into manufacturing scenarios, such as production equipment monitoring, preventative maintenance, and quality management. Furthermore, some works have proposed IoT data analysis models based on fog computing and edge computing, and have effectively applied them to dynamic scenarios such as intelligent manufacturing. However, these works have not focused on collaborative robots in intelligent manufacturing, nor have they established a more generalizable edge computing framework. At the same time, issues such as industrial data privacy and the efficiency of collaborative learning between different devices have not been fully addressed.

[0004] Therefore, there is an urgent need to develop a collaborative training and control method that improves the efficiency of human-machine collaboration, while also enhancing communication efficiency and security, using intelligent manufacturing collaborative robots as the research object. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a collaborative training control method for the edge system of intelligent manufacturing collaborative robots.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A collaborative training control method for an edge system of a smart manufacturing collaborative robot, the method comprising the following steps:

[0008] S1. Establish a collaborative robot edge system in intelligent manufacturing scenarios, including a collaborative robot module, an edge computing module, and a cloud server module;

[0009] S2, the cloud server module assigns tasks to a set number of edge computing modules, and the collaborative robot module uploads device data to the connected edge computing modules;

[0010] S3. Deploy a public dataset in the cloud server module and distribute a set proportion of the public dataset to all edge computing modules;

[0011] S4. Train the local model for each selected edge computing module and calculate the corresponding aggregation coefficient;

[0012] S5. Detect the presence of malicious clients and perform semi-asynchronous training.

[0013] S6. After multiple rounds of training, the parameters of the global model are obtained, and the collaborative training task of the intelligent manufacturing collaborative robot is completed.

[0014] Furthermore, the method described is a method for integrating edge computing and federated learning.

[0015] Furthermore, the edge computing described herein is a technology that offloads computing tasks to a nearby edge server, thereby expanding the storage, computing, and communication capabilities of connected devices.

[0016] Furthermore, the federated learning specifically involves: conducting distributed model training among multiple edge computing modules that possess device data, and constructing a global model under the edge system of the intelligent manufacturing collaborative robot by uploading or downloading local model parameters without needing to transmit device sample data, in order to achieve a balance between data privacy protection and data sharing computing.

[0017] Furthermore, the collaborative robot module, edge computing module, and cloud server module exchange data and transmit control information between different layers via wireless or wired networks.

[0018] Furthermore, the public dataset includes sample data for all categories and covers data samples generated by all collaborative robots.

[0019] Furthermore, the aggregation coefficient is calculated using the model error rate of the edge computing module, and the calculation formula is as follows:

[0020]

[0021]

[0022] Among them, e kζ represents the model's error rate, λ is a positive integer factor with a value of 5, and ζ represents the error rate. k c is the aggregation coefficient of the calculated k-th local model. k This is the normalized polymerization coefficient.

[0023] Furthermore, the detection of the existence of a malicious client is specifically achieved by calculating and comparing the difference between the accuracy of the client's local model and global model. If the difference between the two exceeds a certain value, the client is judged to be a malicious client; otherwise, it is a normal client.

[0024] Furthermore, the specified value is set to 0.2.

[0025] Furthermore, the semi-asynchronous training mode is specifically as follows:

[0026] S501, classify all edge computing modules into participating clients, crash clients, and delayed clients;

[0027] S502 requires a certain number of clients to train a local model;

[0028] S503: After receiving local model parameters from a specific number of participating clients and delayed clients, the cloud server module aggregates the global model using the following calculation formula:

[0029]

[0030] Where ω represents the global model parameters, t represents the number of training iterations, α represents the proportion of client participation, M represents the number of clients, and p k,t V represents the client's participation coefficient. If client k is a participating client in the t-th global aggregation, its value is 1; otherwise, it is 0. k Let D be the sample type size for client k. p ω represents the sample type size for all clients participating in global training. k These are the local model parameters for client k.

[0031] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described above.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] I. In the collaborative control of collaborative robots in intelligent manufacturing, this invention integrates edge computing and federated learning. By designing a semi-asynchronous training process to complete the privacy computation of distributed data, it can both ensure the data privacy of different collaborative robots and improve the learning performance of the intelligent manufacturing system. It has the advantages of efficient communication, security and high training efficiency.

[0034] Second, by covering all data samples generated by collaborative robots and distributing a certain proportion of public data to all edge computing modules, this invention effectively alleviates the problem of poor global model performance caused by imbalance in device sample categories. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0036] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0037] Example

[0038] This invention uses the MNIST dataset to perform an image classification task and verify its performance, which is similar to component classification in the manufacturing and packaging process. The MNIST dataset contains 60,000 training samples and 10,000 test samples, and the sample data is similar to batch numbers and category information of different industrial products.

[0039] A collaborative training control method for an edge system of a smart manufacturing collaborative robot, the method comprising the following steps:

[0040] S1. Establish a collaborative robot edge system in intelligent manufacturing scenarios, including a collaborative robot module, an edge computing module, and a cloud server module.

[0041] The collaborative robot module primarily includes robotic arms, autonomous mobile robots (AMRs), and other collaborative modules (such as sensor modules and monitors). The equipment in the collaborative robot module needs to perform collaborative tasks such as product manufacturing, assembly, measurement, and quality inspection. The massive amounts of data generated by various collaborative robots need to be transmitted to the corresponding edge controllers and preprocessed on the connected edge gateways or edge servers. Collaborative robots also need to automatically execute instructions from different edge controllers. Different types of collaborative robots need to perform related collaborative tasks.

[0042] The edge computing module is the core component of the collaborative robot's edge system. It provides security protection and decision optimization. Depending on the computing power, the edge computing module includes an edge controller layer, an edge gateway layer, and an edge server layer. Each edge computing model comprises an edge controller, an edge gateway, and an edge server.

[0043] Edge Controller Layer: The edge controller layer is deployed close to the collaborative robot module. Depending on the type of built-in control module of the collaborative robot, the edge controller layer mainly includes control units for different collaborative robots, such as motion controllers for AMRs, PLCs for robotic arms, and MCUs for other collaborative robots. The edge controller layer performs industrial data preprocessing or simple logic operations. It can also react immediately to emergencies occurring within the collaborative robots and take measures to switch the working collaborative robot. Furthermore, the edge controller layer must be compatible with different communication protocols and be able to access all sensors or collaborative robots in the manufacturing system. Because multiple collaborative robots perform collaborative tasks together, the coordinated control of collaborative robots becomes very complex and difficult. To simplify collaboration, the edge controller layer must be able to connect to and control the built-in control modules of the collaborative robots in real time. This allows for better control of abnormal manufacturing events across all collaborative robots distributed throughout the layer, reducing labor costs and improving collaborative efficiency.

[0044] Edge Gateway Layer: Edge gateways acquire and store industrial data from edge controllers and perform heterogeneous computing operations. They receive control streams from edge servers and transmit them to edge controllers. Compared to edge controllers, edge gateways have larger storage and computing resources for storing data processing logs and managing multiple modules. Therefore, in smart manufacturing, edge gateways can quickly analyze industrial data, manage collaborating robots that malfunction, and prevent production accidents.

[0045] Edge server layer: Edge servers here possess enhanced computing and storage capabilities. These entities need to connect to the edge gateway via a dedicated network and can perform more complex tasks. Deep learning-based classification or inference models can be trained on edge servers. Simultaneously, edge servers can schedule tasks and optimize operating parameters for collaborative robots across an entire or multiple production lines, optimizing resource allocation and improving the productivity of the manufacturing system.

[0046] The cloud server module primarily collects critical information from edge servers, extracts valuable knowledge, and provides useful feedback to enterprise managers, such as process optimization, collaborative robot management, and industrial production technology solutions. Furthermore, the cloud application layer can provide the edge system layer with the initialization model architecture and parameters for distributed tasks.

[0047] The method described is a fusion design of edge computing and federated learning. Edge computing is a technology that offloads computing tasks to edge servers closer to the end, expanding the storage, computing, and communication capabilities of connected devices. Federated learning specifically involves distributed model training among multiple edge computing modules that possess device data. Without transmitting device sample data, a global model is constructed under the edge system of the intelligent manufacturing collaborative robot by uploading or downloading local model parameters, thereby achieving a balance between data privacy protection and data sharing computing. The collaborative robot module, edge computing module, and cloud server module exchange data and transmit control information between different layers through wireless or wired networks.

[0048] This invention sets up a cloud server module and 100 edge computing modules (clients), with the communication state of each edge computing module randomly configured. A multi-layer perception (MLP) is used as the learning model for optimization in the local model. Furthermore, this patent compares two data settings:

[0049] 1) Independent and identically distributed (IID) distribution: 10,000 samples are randomly selected from the MNIST dataset as public data. Then, the remaining training samples are randomly shuffled and stored on each client, where the sample size of different clients follows a Gaussian distribution with a mean of 500 and a variance of 100.

[0050] 2) Non-Independent and Identically Distributed (Non-IID): First, 20% of the training samples are randomly selected as public data. The remaining training samples are divided into 200 modules, each containing only two classes of sample data. Then, two modules of data are randomly assigned to each client.

[0051] S2, the cloud server module assigns tasks to a set number of edge computing modules, and the collaborative robot module uploads device data to the connected edge computing modules.

[0052] When performing certain manufacturing tasks, the cloud server module will randomly select a set number of edge computing modules, denoted as P (|P|=25), based on the communication status of the client. The collaborative robot module connected to these edge computing modules will upload the device data it generates to the edge server layer in the edge computing module.

[0053] S3. Deploy a public dataset in the cloud server module and distribute a set proportion of the public dataset to all edge computing modules.

[0054] The public dataset includes sample data from all categories and covers data samples generated by all collaborative robots. Based on the data type from all devices, the cloud server module randomly selects a set amount of data and stores it on the cloud server as a public dataset for global model parameter initialization. Simultaneously, a set proportion of public data (α = 0.2) is randomly selected and allocated to each edge computing module for local model training and performance verification.

[0055] S4. Train the local model for each selected edge computing module and calculate the corresponding aggregation coefficient.

[0056] The selected edge computing module uses device data generated by the collaborative robot module connected to it to train the local model, and the update formula is:

[0057]

[0058] Where ω represents the global model parameters, E represents the number of training iterations for the local model (valued at 5), α represents the client participation ratio, M represents the number of clients, and V... k Let D be the sample type size for client k. p ω represents the sample type size for all clients participating in global training. k For the local model parameters of client k, η k Let k be the learning rate for the client. This provides the local model gradient information for the client; furthermore, if the batch size B for training the local model is set to 50, then the parallelism u = E * V is calculated. k / B=50.

[0059] Furthermore, the aggregation coefficient is calculated using the model error rate of the edge computing module, and the calculation formula is as follows:

[0060]

[0061]

[0062] Among them, e k ζ represents the model's error rate, λ is a positive integer factor with a value of 5, and ζ represents the error rate. k c is the aggregation coefficient of the calculated k-th local model. k This is the normalized polymerization coefficient.

[0063] S5 detects the presence of malicious clients and performs semi-asynchronous training.

[0064] The detection of malicious clients involves calculating and comparing the difference between the accuracy of the client's local model and global model. If the difference exceeds a certain value, the client is considered malicious; otherwise, it is considered legitimate. The specific detection method is as follows: First, the recognition rate of the local model is calculated based on the device data stored in client k. Second, the recognition rate of the local model is calculated based on the public data. Finally, the difference ε between the two recognition rates is calculated. k Therefore, a precision threshold parameter ε (ε = 0.2) is introduced; if ε k If the value is greater than ε, then client k is considered a malicious client; when a client is identified as malicious, the stored local model parameters will be discarded. Then, a semi-asynchronous training mode is performed, specifically:

[0065] S501 categorizes all edge computing modules into participating clients, crash clients, and delayed clients.

[0066] Participating clients: These clients receive the latest global model parameters from the cloud server module and upload their local models to it. The participating clients possess the computing and communication resources necessary to train their local models effectively. These clients can also establish strong communication connections with the connected edge controller modules.

[0067] Crashing Clients: Due to insufficient communication resources, crashing clients are temporarily unable to communicate with the cloud server module, or these clients are unable to perform local training on their local models. Therefore, the cloud server module cannot receive the latest local models from these clients, nor can it distribute the latest global model to them. In real-world intelligent manufacturing systems, each collaborative robot has a certain probability of crashing.

[0068] Delayed clients: Due to the large amount of data stored on the client or its limited computing power, delayed clients cannot complete the local training task of the local model on time. However, these clients can maintain good communication with the cloud server module, meaning they can upload local model parameters and download global model parameters.

[0069] S502. A certain number of clients (α·M, where α = 0.2, M = 100) are required to train the local model.

[0070] S503: After receiving local model parameters from a specific number of participating clients and delayed clients, the cloud server module aggregates the global model using the following calculation formula:

[0071]

[0072] Where ω represents the global model parameters, t represents the number of training iterations, α represents the proportion of client participation, M represents the number of clients, and pk,t V represents the client's participation coefficient. If client k is a participating client in the t-th global aggregation, its value is 1; otherwise, it is 0. k Let D be the sample type size for client k. p ω represents the sample type size for all clients participating in global training. k These are the local model parameters for client k.

[0073] S6. After multiple rounds of training, the parameters of the global model are obtained, and the collaborative training task of the intelligent manufacturing collaborative robot is completed. Table 1 shows the global model results obtained using different model parameters, and Table 2 shows the global model results obtained using different federated learning methods.

[0074] Table 1 shows the global model results obtained using different model parameters.

[0075]

[0076] Table 2 shows the global model results obtained using different federated learning methods.

[0077]

[0078] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A collaborative training control method for an edge system of an intelligent manufacturing collaborative robot, characterized in that, The method described is a fusion design of edge computing and federated learning; the method includes the following steps: S1. Establish a collaborative robot edge system in intelligent manufacturing scenarios, including a collaborative robot module, an edge computing module, and a cloud server module; S2, the cloud server module assigns tasks to a set number of edge computing modules, and the collaborative robot module uploads device data to the connected edge computing modules; S3. Deploy a public dataset in the cloud server module and distribute a set proportion of the public dataset to all edge computing modules; S4. Train the local model for each selected edge computing module and calculate the corresponding aggregation coefficient; S5. Detect the presence of malicious clients and perform semi-asynchronous training. S6. After multiple rounds of training, the parameters of the global model are obtained, and the collaborative training task of the intelligent manufacturing collaborative robot is completed. The aggregation coefficient is calculated using the model error rate of the edge computing module, and the calculation formula is as follows: in, The error rate of the model, A positive integer factor, with a value of 5. For the calculated first k The aggregation coefficients of a local model Normalized polymerization coefficient; The semi-asynchronous training mode is specifically as follows: S501, classify all edge computing modules into participating clients, crash clients, and delayed clients; S502 requires a certain number of clients to train a local model; After receiving a certain number of local model parameters from participating clients and delayed clients, the S503 cloud server module aggregates the global model using the following calculation formula: in These are global model parameters. t To determine the number of training iterations, For the client's participation rate, M For the number of clients, For the client's participation coefficient, the client k In the t In the next global aggregation, if it is a participating client, its value is 1; otherwise, it is 0. For the client k Sample type size, The sample type size for all clients participating in global training. For the client k Local model parameters.

2. The collaborative training and control method for an edge system of an intelligent manufacturing collaborative robot according to claim 1, characterized in that, Edge computing is a technology that offloads computing tasks to edge servers closer to the end, thereby expanding the storage, computing, and communication capabilities of connected devices.

3. The collaborative training and control method for an edge system of an intelligent manufacturing collaborative robot according to claim 1, characterized in that, The federated learning specifically refers to: by conducting distributed model training among multiple edge computing modules that possess device data, a global model under the edge system of intelligent manufacturing collaborative robots is constructed by uploading or downloading local model parameters without the need to transfer device sample data, in order to achieve a balance between data privacy protection and data sharing computing.

4. The collaborative training and control method for an edge system of an intelligent manufacturing collaborative robot according to claim 1, characterized in that, The collaborative robot module, edge computing module, and cloud server module exchange data and transmit control information between different layers via wireless or wired networks.

5. The collaborative training control method for an edge system of an intelligent manufacturing collaborative robot according to claim 1, characterized in that, The public dataset includes sample data for all categories and covers all data samples generated by collaborative robots.

6. The collaborative training control method for an edge system of an intelligent manufacturing collaborative robot according to claim 1, characterized in that, The detection of the existence of malicious clients is specifically achieved by calculating and comparing the difference between the accuracy of the client's local model and global model. If the difference between the two exceeds a certain value, the client is judged to be a malicious client; otherwise, it is a normal client.

7. The collaborative training control method for an edge system of an intelligent manufacturing collaborative robot according to claim 6, characterized in that, The specified value is set to 0.

2.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

Images