Deterministic Industrial Ethernet Switch Cluster and Method Based on Multi-Agent Collaboration
By using a deterministic industrial Ethernet switch cluster with a multi-agent collaborative architecture, the problems of insufficient determinism, rigid redundancy mechanisms, and weak operation and security of traditional switches in complex industrial scenarios are solved, achieving high reliability, low latency, and multi-layered security for industrial network communication.
Patent Information
- Application Number
- CN202611037287.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-13
- Publication Date
- 2026-11-10
AI Technical Summary
Traditional industrial Ethernet switches suffer from insufficient determinism, rigid redundancy mechanisms, and weak operation and security in complex industrial scenarios, making it difficult to meet the high reliability and security requirements of industrial control systems. Furthermore, existing time-sensitive network implementations are complex and costly.
A deterministic industrial Ethernet switch cluster employing a multi-agent collaborative architecture achieves deterministic link assurance, user-definable redundancy, link-layer encryption, and distributed intelligent management through distributed controllers and agent nodes. Combined with static flow tables and redundancy algorithms, it provides a highly reliable, secure, and easy-to-maintain communication foundation.
It achieves microsecond-level latency protection, 1ms-level fault self-healing, and multi-layered security protection for critical business data, reducing operational complexity and meeting the complex needs of industrial networks.
Abstract
Description
Technical Field
[0001] This invention relates to the field of industrial communication network technology, specifically to an Ethernet switch system for industrial automation scenarios, and more particularly to a multi-agent collaborative industrial Ethernet switch cluster and its control method that supports deterministic communication, high reliability redundancy and intelligent security management. Background Technology
[0002] With the deepening development of Industry 4.0 and intelligent manufacturing, industrial networks are placing unprecedented demands on the determinism, reliability, and security of communication. Traditional industrial Ethernet switches have the following limitations when dealing with complex industrial scenarios:
[0003] 1. Insufficient determinism: Ordinary switches adopt a "first-come, first-served, best-effort" forwarding mechanism based on QoS levels, which makes it difficult to guarantee the consistency of transmission delay for critical control data (such as motion control commands and safety interlock signals). They are prone to packet loss and delay jitter during traffic bursts, and cannot meet the hard real-time requirements of industrial control systems.
[0004] 2. Rigid Redundancy Mechanisms: Although switches adapted for high-reliability industrial control applications support high-reliability redundancy technologies such as Spanning Tree Protocol (STP), PRP, and HSR, the redundancy paths of STP and PRP protocols are automatically calculated by the protocols themselves, and users cannot customize protection strategies for critical links based on service importance; the HSR protocol requires a ring network topology to achieve redundancy, resulting in a fixed topology and high multi-hop latency.
[0005] 3. Weak Operation and Security: Traditional switches have limited functionality and lack intelligent awareness and collaborative management capabilities for network status. Security protection is usually limited to basic port isolation or access control lists, which cannot effectively cope with increasingly complex network attacks. Switches adapted for industrial control applications lack data link layer encryption functions, and sensitive production data and control signaling are at risk of being stolen and tampered with during transmission.
[0006] In the existing technology, although Time-Sensitive Networking (TSN) provides a framework for deterministic communication, its implementation is complex, costly, and requires significant modifications to existing equipment.
[0007] Therefore, there is an urgent need in this field for an industrial network core switching equipment solution that can deeply integrate deterministic communication, extreme redundancy, intelligent security and convenient operation and maintenance, and is easy to deploy domestically. Summary of the Invention
[0008] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a deterministic industrial Ethernet switch cluster and method based on multi-agent collaboration. This cluster, through an innovative "agent" architecture, treats each switch as a node with independent sensing, decision-making, and collaborative capabilities, achieving deterministic link assurance, user-definable redundancy, link-layer encryption, and distributed intelligent management, providing a highly reliable, secure, and easy-to-maintain communication foundation for industrial networks.
[0009] To achieve the above objectives, the present invention adopts the following technical solution:
[0010] In a first aspect, the present invention provides a deterministic industrial Ethernet switch cluster based on multi-agent collaboration, comprising: multiple industrial Ethernet switches, each switch acting as an agent node;
[0011] A distributed controller, wherein the master controller is elected by multiple agents and communicates with the multiple agent nodes;
[0012] Each of the aforementioned intelligent agent nodes includes:
[0013] The switching chip module is used to store and forward data packets;
[0014] The deterministic communication processing module is used to generate and maintain a static flow table based on the configuration from the distributed master controller. The flow table defines at least one time-triggered deterministic link to ensure deterministic and low-latency transmission of data frames of a specified priority.
[0015] The redundancy control module is used to run the standard spanning tree protocol and execute a centralized distributed redundancy algorithm. This algorithm receives global topology and policy information from the distributed master controller, dynamically calculates and switches to the optimal redundant path.
[0016] The security encryption module is used to encrypt data frames that pass through the time-triggered deterministic link or the user-specified link at the link layer;
[0017] The status awareness and reporting module is used to monitor the status of local ports, link quality and traffic information in real time, and report them to the distributed main controller.
[0018] Secondly, the present invention provides a communication control method applied to the above-mentioned switch cluster, comprising:
[0019] The distributed master controller plans and distributes at least one time-triggered deterministic link in the cluster according to the user configuration.
[0020] Each agent node generates a corresponding static flow table entry in its deterministic communication processing module according to the issued configuration.
[0021] When a data packet enters an agent node, the deterministic communication processing module prioritizes matching the static flow table. If a match is found, the packet is forwarded according to the priority, exit port, and sending time window specified in the flow table.
[0022] If the connection is not found, the switch chip module will handle the process according to the standard Ethernet procedure.
[0023] The distributed master controller periodically collects the status information of each agent node. When a network topology change or link failure is detected, it recalculates the redundant path and sends an update command to the relevant agent node.
[0024] The beneficial effects of this invention are as follows:
[0025] 1. Achieved a balance between determinism and flexibility: By introducing "TT deterministic links" and static flow table mechanisms, a "dedicated channel" with microsecond-level latency guarantees is provided for critical business data, while being compatible with conventional Ethernet communication, meeting the complex needs of mixed transmission of control flow and data flow in industrial scenarios.
[0026] 2. Provides user-definable extreme redundancy: Based on support for STP / RSTP standard protocols, a redundancy algorithm is introduced that allows the distributed master controller to collaborate with other intelligent agents. Users can customize redundancy strategies (such as primary / backup paths, load balancing) for business flows of different importance through the controller interface, achieving 1ms-level fault self-healing. While providing recovery capabilities comparable to traditional protocols such as PRP, it also offers flexibility in redundancy strategies.
[0027] 3. An inherent security protection system has been built: A data link layer encryption module is integrated inside the switch to support encrypted transmission of sensitive control commands and data, effectively preventing attacks such as wiretapping. Combined with functions such as port device identification and abnormal traffic monitoring, a multi-layered security defense is formed from access to transmission.
[0028] 4. Distributed intelligent operation and maintenance is achieved: Each intelligent agent node possesses local state awareness and preliminary decision-making capabilities. Combined with the global perspective of the distributed controller, real-time monitoring, intelligent early warning, and collaborative optimization of network status are realized. The Web-based intelligent operation and management module significantly reduces the complexity of industrial network operation and maintenance. Detailed Implementation
[0029] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail. The following description includes specific details necessary to implement the present invention, but these details should not be considered as limiting the scope of the invention.
[0030] First Implementation Example: Switch Cluster Hardware Architecture and Intelligent Agent Nodes
[0031] This embodiment describes a cluster system comprising four intelligent agent switches (SW1-SW4), wherein SW1 is elected as the master controller.
[0032] 1. Main controller (taking SW1 as an example)
[0033] The master controller is a switch that is elected within the cluster. Its core functions include:
[0034] Topology Management: Automatically learns and maintains the network topology of the entire switch cluster through common data link layer discovery protocols such as LLDP and custom private data link layer discovery protocols.
[0035] Policy Configuration: Provides a graphical web management interface (Web UI) that allows network administrators to perform the following configurations:
[0036] TT Link Configuration: Specify the source device, destination device, service VLAN ID, required bandwidth, and maximum allowable latency. The controller will automatically calculate and plan the TT path across the switch.
[0037] Redundancy policy configuration: Select a redundancy mode (such as "primary path priority" or "load balancing") for the specified service flow or VLAN, and select a redundant path.
[0038] Security policy configuration: Specify the links or VLANs that need to be encrypted, and select the encryption algorithm (such as AES-128, SM4) and encryption / decryption method (such as encrypting at the inlet and decrypting at the outlet).
[0039] Intelligent analysis: Receives status reports from various intelligent agent nodes, performs big data analysis, and realizes functions such as network performance early warning and fault root cause location.
[0040] 2. Intelligent Agent Switch (taking SW2 as an example)
[0041] Each intelligent agent switch has the following internal logical functional modules:
[0042] Switching Processing Module: This is one of the core modules of this invention. It contains a switch core that supports deterministic Ethernet communication, either running on an FPGA or embedded as an ASIC chip, and includes the functions of the aforementioned switching chip module, deterministic communication processing module, and security encryption module. In this example, the switch core supports 12 Gigabit interfaces and 2 10 Gigabit interfaces. Its main responsibilities are:
[0043] Deterministic communication processing module: Integrated into the data path. Based on the deterministic link information issued by the distributed master controller, it performs time-triggered transmission of data frames that match the deterministic link rules.
[0044] Security encryption module: Integrated into the data path. When encryption is configured for a particular link, this module encrypts the payload before the data packet exits and decrypts the ciphertext after the data packet enters. It supports AES and SM4 encryption algorithms and GCM-based message authentication. Keys are centrally distributed and managed by the controller. The encryption process is transparent to upper-layer applications.
[0045] Control Processing Module: This is one of the core modules of this invention. It includes a microprocessor and its firmware that run on an FPGA or are embedded in an ASIC chip, executing the aforementioned redundant control module and status sensing and reporting module. Its main responsibilities are:
[0046] Receive and parse TT link configuration messages from the controller.
[0047] Maintain a static flow table locally. The key fields of this flow table include: matching fields (such as source MAC, destination MAC, VLAN ID, Ethernet type) and action fields (egress port, transmission time window, priority flag).
[0048] In the packet processing pipeline, incoming packets first enter this module for flow table matching. If a TT flow table entry is matched, it is marked with a high priority and placed in a dedicated transmission queue scheduled based on a globally accurate clock (such as through IEEE 1588 synchronization) to ensure that it is transmitted within a specified micro-time window, thereby achieving microsecond-level latency and extremely low jitter.
[0049] Redundancy Control Module: This module runs two protocol stacks in parallel. One is the standard STP / RSTP protocol stack, used for compatibility with third-party switches. The other is an autonomous redundancy algorithm stack, whose workflow is as follows:
[0050] ① Maintain heartbeat with controller 101 and receive global topology and user-defined redundancy policies.
[0051] ② Using algorithms such as Dijkstra, multiple paths to the critical destination are pre-calculated, and the optimal path is set as the "primary forwarding path", while the information of the remaining paths is cached as "backup paths".
[0052] ③ Monitor the link quality of the main path in real time through rapid detection mechanisms such as BFD.
[0053] Once a primary path failure is detected (such as link interruption or sudden latency changes), the traffic forwarding path is switched to a pre-calculated backup path within 1ms, and the controller is notified. The switching action is completed locally without real-time controller intervention, ensuring speed.
[0054] Status Awareness and Reporting Module 206: Collects real-time data such as CPU utilization, memory usage, traffic statistics for each port, packet error rate, and link latency of the local machine. Reports this data to the distributed master controller periodically or via event-triggered events through southbound interfaces (such as NETCONF / gRPC).
[0055] 3. Cluster Collaboration Workflow
[0056] The main controller connects to other agent switches via an independent, reliable in-band or out-of-band management network. The controller issues configurations, the agents execute them, and the agents report back their status, forming a collaborative system of "centralized management and distributed execution." This architecture ensures both policy uniformity and global optimality, while leveraging the local processing capabilities of the agents to achieve rapid response, avoiding the single-point bottlenecks and latency of traditional centralized controllers.
[0057] Second Implementation Example: Establishment and Data Transmission of a TT Deterministic Link
[0058] This embodiment details the establishment and data forwarding process of a TT link from device A (connected to SW1) to device B (connected to SW4). Assume this link is used to transmit real-time control commands for a robotic arm, requiring an end-to-end latency of ≤100μs.
[0059] Step S301: User Configuration. The administrator logs into the controller's web interface and creates a TT service flow. Specify: Name "Robot_Ctrl", MAC address of source device A, MAC address of destination device B, VLAN 10, reserved bandwidth 2Mbps, maximum allowable latency 100μs.
[0060] Step S302: Path Calculation and Resource Reservation. Based on the global topology, the controller uses the constrained shortest path algorithm to calculate a path that meets the latency requirements: SW1 -> SW2 -> SW4. The controller checks the remaining bandwidth resources of each switch along the path and reserves 2Mbps of bandwidth for this TT flow at each hop.
[0061] Step S303: Configuration Distribution. The controller generates configuration commands for SW1, SW2, and SW4, and distributes them to these three switches respectively via a secure channel. The commands include:
[0062] Flow table matching items: source / destination MAC, VLAN 10.
[0063] Action items: Egress port (Port2 of SW1), Transmission time window (a periodic micro-time slice calculated based on the global synchronization clock, such as the 100th-120th μs within each 1ms period).
[0064] For SW2 and SW4, the instructions are similar, but specify their respective exit ports and transmission time windows adjusted for cumulative delay.
[0065] Step S304: Local Flow Table Generation and Synchronization. Upon receiving the instruction, the deterministic communication processing modules of SW1, SW2, and SW4 generate corresponding entries in their local static flow tables. Each switch maintains time synchronization via the IEEE 1588 protocol, achieving an accuracy of 100ns.
[0066] Step S305: Deterministic forwarding. Device A sends a control data packet with VLAN 10 tag, destined for Device B.
[0067] 1. The packet was received on Port1 of SW1. The deterministic communication processing module performed a match and found the "Robot_Ctrl" flow table entry.
[0068] 2. The data packet is marked with the highest priority and sent to a dedicated "TT send queue". The scheduler of this queue strictly follows global time and only sends the data packet out from Port2 when the predetermined time window (100μs) arrives.
[0069] 3. When the ingress port of SW2 receives the packet near the scheduled time, it also hits the flow table. After a very short processing and queuing delay, it is sent out from the designated port in the next scheduled time window.
[0070] 4. Similarly, SW4 ultimately ensures that the control packet arrives at device B at the precisely calculated time. The entire process eliminates interference from other background traffic, achieving deterministic low-latency transmission.
[0071] Third embodiment: User-defined redundancy and fault self-healing process
[0072] This example describes configuring a primary / backup redundancy policy for an important surveillance video stream (VLAN 20) and achieving 1ms self-healing in the event of a primary path failure.
[0073] Step S401: Policy Definition. The administrator configures a redundancy policy for VLAN 20 on the controller: the primary path is SW1->SW3->SW4; the backup path is SW1->SW2->SW4. The fault detection mode is "Fast Link Detection".
[0074] Step S402: Path pre-calculation and distribution. The controller distributes the primary and backup path information to the relevant switches (SW1, SW2, SW3, SW4). After receiving the policy, the redundancy control module of SW1 generates two entries for VLAN 20 in its forwarding table: the primary next hop is the port leading to SW3, with a better metric value; the backup next hop is the port leading to SW2, with a second-best metric value.
[0075] Step S403: Fault Detection. SW1 and SW3 detect the link status by sending probe packets every 100μs via the BFD protocol. Simultaneously, the status awareness modules 206 of each switch monitor the physical status and bit error rate of the ports.
[0076] Step S404: Fault Switching. Assume the fiber optic cable between SW1 and SW3 is accidentally broken. The BFD session on SW1 times out within one detection cycle (approximately 300μs), immediately triggering a local fault event. Upon receiving the event, the redundancy control module switches the next hop for VLAN 20 traffic forwarding from the primary port to the backup port (to SW2) within a very short time (approximately several hundred nanoseconds). The entire switching process is completed locally on SW1, without waiting for controller commands or STP protocol convergence, with the total time controlled within 1ms.
[0077] Step S405: Status Synchronization and Visual Management. After completing the handover, SW1 immediately reports the link fault and handover event to the controller. The controller displays an alarm prominently on the web interface and updates the displayed status of the link in the topology diagram, facilitating timely handling by the administrator.
[0078] Fourth Implementation: Implementation of Link Layer Encryption
[0079] This example describes how to encrypt a link (VLAN 30) that transmits production formula data.
[0080] Step S501: Encryption Policy Distribution. The administrator configures the controller to change the encryption for VLAN 30 data at each hop, selecting AES-128-GCM algorithm for SW1 and SM4-GCM algorithm for SW2. The controller generates encryption keys Key_VLAN30_SW1 and Key_VLAN30_SW2, and distributes them through a secure channel to all member switches (SW1, SW2, SW4) that may transmit VLAN 30 data.
[0081] Step S502: Encryption at the sending end. When SW1 needs to forward a data packet tagged with VLAN 30 from a port connected to the industrial control computer, the security encryption module intercepts the packet. The module extracts the Ethernet frame payload, encrypts it using Key_VLAN30_SW1 and the AES-128-GCM algorithm, and generates a message authentication code appended to the end. Then, it replaces the original payload with the new ciphertext payload, recalculates the frame check sequence, and sends it out from the egress port. The Ethernet header (including MAC address and VLAN tag) remains unchanged, thus being transparent to other non-member switches in the network.
[0082] Step S503: The intermediate switch decrypts and then encrypts. When SW2 receives this encrypted frame destined for the server, its security encryption module 203 identifies the VLAN 30 tag and queries the local configuration to determine that this VLAN requires decryption before encryption. The module uses Key_VLAN30_SW1 and the AES-128-GCM algorithm to decrypt and verify the integrity of the encrypted payload. After successful verification, the decrypted original payload is re-encapsulated into an encrypted frame using Key_VLAN30_SW2 and the SM4-GCM algorithm and sent out from the egress port.
[0083] Step S504: Decryption at the receiving end. When SW4 receives this encrypted frame destined for the server, its security encryption module identifies the VLAN 30 tag and queries the local configuration to determine that this VLAN needs decryption. The module uses Key_VLAN30_SW2 and the SM4-GCM algorithm to decrypt and verify the integrity of the encrypted payload. After successful verification, the decrypted original payload is re-encapsulated into a standard Ethernet frame and forwarded to the server.
[0084] Step S505: Security Assurance. The key is updated periodically or as needed by the controller. Even if an attacker eavesdrops on data on the physical line, they will only obtain ciphertext that cannot be cracked, effectively protecting sensitive information such as production processes. The entire process is completely transparent to the terminal devices (industrial control computers and servers) without requiring any software modification.
[0085] Fifth Implementation Example: Application Scenarios and Effect Verification
[0086] The switch cluster of this invention is deployed on an intelligent assembly line for automotive parts. This production line includes 20 industrial robots, 50 vision sensors, multiple conveyor lines, and an MES system.
[0087] In summary, this invention provides a complete and autonomous high-end industrial network solution through a multi-agent collaborative architecture that deeply integrates deterministic communication, definable redundancy, link encryption, and intelligent operation and maintenance. This solution effectively supports the upgrade needs of key fields such as intelligent manufacturing, smart grids, and rail transit.
Claims
1. A deterministic industrial Ethernet switch cluster based on multi-agent cooperation, characterized in that, include: Multiple industrial Ethernet switches, each acting as an intelligent agent node; A distributed controller, with the main controller elected by multiple agent nodes, is used for global policy configuration and state management; Each of the aforementioned intelligent agent nodes includes: The switching chip module is used to store and forward data packets; A deterministic communication processing module is used to generate and maintain a static flow table based on the configuration from the distributed master controller, wherein the static flow table defines at least one time-triggered deterministic link; A redundancy control module is used to execute a distributed redundancy algorithm and dynamically switch forwarding paths according to the strategy from the distributed controller.
2. The switch cluster according to claim 1, characterized in that, The deterministic communication processing module is configured to schedule and forward data packets that match the static flow table according to the sending time window specified by the flow table, so as to achieve microsecond-level transmission delay.
3. The switch cluster according to claim 1 or 2, characterized in that, The static flow table supports the definition of no fewer than 4096 concurrent time-triggered deterministic links.
4. The switch cluster according to claim 1, characterized in that, The redundancy control module also supports the operation of standard spanning tree protocol, fast spanning tree protocol and custom high-speed redundancy protocol, and the fault self-healing time implemented by the distributed redundancy algorithm is less than 1 millisecond.
5. The switch cluster according to claim 1, characterized in that, Each of the aforementioned agent nodes also includes: The security encryption module is used to encrypt and decrypt data frames of a specified service flow at the link layer.
6. The switch cluster according to claim 5, characterized in that, The encryption and decryption processes support symmetric encryption algorithms including but not limited to AES and SM4, and message verification codes including but not limited to AES-GCM and SM4-GCM. The encryption keys are uniformly distributed and managed by the distributed master controller.
7. The switch cluster according to claim 1, characterized in that, Each of the aforementioned agent nodes also includes: The status awareness and reporting module is used to monitor the local status in real time and report it to the distributed main controller. The distributed master controller also includes a Web-based intelligent operation and management module, which provides a graphical management interface to display network topology, service flow status, and security alarms.
8. A communication control method applied to a switch cluster according to any one of claims 1-7, characterized in that, include: The distributed master controller plans and distributes configuration information for at least one time-triggered deterministic link in the cluster according to user configuration. Each agent node generates a corresponding static flow table entry in its deterministic communication processing module based on the configuration information it receives. When a data packet enters an agent node, it is preferentially matched against the static flow table. If a match is found, it is deterministically forwarded according to the flow table entry. If no match is found, it is forwarded according to the normal process.
9. The method according to claim 8, characterized in that, The method also includes a redundancy control step: The distributed master controller calculates the primary path and at least one backup path for triggering deterministic links at a specified time according to the user policy, and sends the path information to the relevant intelligent agent nodes. When an agent node detects a failure in the primary path, it locally switches the traffic of the deterministic link that was triggered at the specified time to the backup path.
10. The method according to claim 8, characterized in that, The method also includes a security control step: The distributed master controller triggers the deterministic link and distributes the encryption key at the time when encryption is required. When the relevant intelligent agent nodes forward data frames that match the time-triggered deterministic link, they encrypt and / or decrypt them at the link layer according to the rules issued by the distributed controller.
