A system and method for supporting real-time control with artificial intelligence participation

Abstract

The application belongs to the technical field of industrial automation control and artificial intelligence, and relates to a system and method for supporting real-time control with the participation of artificial intelligence. The system comprises a general AI reasoning module of an AI reasoning computer and an industrial real-time Ethernet slave station module which interact with each other through shared memory, and are connected with a PLC through an industrial real-time Ethernet bus, are used for receiving real-time data on site and executing artificial intelligence reasoning, and the PLC cooperates with the AI reasoning computer to complete a real-time control loop according to the reasoning result. The industrial real-time Ethernet slave station module is configured with a hardware industrial real-time Ethernet slave station protocol stack, the AI reasoning computer is connected to the industrial real-time Ethernet as a native slave station node, and real-time communication is completed in the same real-time control cycle as the PLC. The application sinks AI reasoning computing power from the edge computing layer to the terminal device layer, and realizes data interaction, cooperative decision-making and closed-loop control in the same real-time control cycle as the PLC.

Description

Technical Field

[0001] This invention belongs to the field of industrial automation control and artificial intelligence technology, specifically relating to a system and method that supports artificial intelligence in real-time control. Background Technology

[0002] Current industrial control systems are developing towards intelligence, real-time operation, and autonomy, with artificial intelligence technology being widely applied in predictive maintenance, quality inspection, and process optimization. Traditional intelligent control architectures often adopt a layered model of "cloud training, edge inference, and field execution": the artificial intelligence model is deployed on cloud servers or edge computing nodes, and interacts with field programmable logic controllers (PLCs), sensors, and actuators through general Ethernet or non-real-time networks.

[0003] Real-time control tasks in industrial settings (such as motion control, closed-loop regulation, and fast logic control) are typically performed by PLCs via industrial real-time Ethernet networks such as Profinet, EtherCAT, and EtherNet / IP, and are subject to strict timing constraints, low latency, and high determinism requirements. In existing solutions, AI inference nodes are mostly located at the edge computing layer, independent of the field control layer, and can only perform non-real-time functions such as offline analysis, status monitoring, and decision support, making it difficult to deeply integrate them into the real-time control closed loop.

[0004] The existing layered model of "cloud training - edge inference - field execution" typically deploys AI inference units at the edge computing layer or cloud servers, interacting with PLCs, actuators, and sensors via general Ethernet or non-real-time buses. While this architecture is widely used in non-real-time scenarios such as data preprocessing, status monitoring, and offline optimization, it has significant shortcomings in industrial field control tasks requiring millisecond-level closed-loop control, high dynamic response, and strong real-time constraints. First, the separation of inference location from the control link leads to excessively high end-to-end latency. Multiple forwarding, protocol conversion, and data queuing exist between edge computing nodes and field device layers, preventing AI inference results from reaching the PLC and actuators within the control cycle, thus failing to meet the timing requirements of real-time control such as motion control and process closed-loop regulation.

[0005] Second, the existing network architecture does not support native access of AI nodes to the real-time control domain. Industrial real-time Ethernet such as Profinet, EtherCAT, and EtherNet / IP are designed with PLCs as the core and are geared towards deterministic communication and logic control. AI inference devices are usually connected as non-real-time slaves or external nodes, and cannot interact synchronously with the PLC within the same real-time cycle, making it difficult to guarantee the determinism and stability of the control closed loop.

[0006] Third, edge computing is not sufficiently deployed, and the terminal layer lacks real-time AI processing capabilities. The terminal device layer mainly consists of I / O, drivers, and sensors, with limited computing power and model deployment capabilities. A large amount of data needs to be uploaded to the edge layer for processing, resulting in bandwidth consumption, data redundancy, and response lag. In field environments with weak networks, high disturbances, and high reliability requirements, the system's availability and anti-interference capabilities are insufficient.

[0007] Fourth, the system architecture is redundant and highly dependent. The existing three-layer architecture relies heavily on the computing power and network support of edge computing nodes and the cloud. Failures at the edge or in the cloud will directly affect on-site AI applications, and the system's ability to operate autonomously in weak network or highly disturbed environments is limited.

[0008] In summary, how to extend AI inference computing power from the edge computing layer to the terminal device layer, enabling data interaction, collaborative decision-making, and real-time closed-loop control with PLC within the same real-time control cycle, and thus fundamentally addressing issues such as latency, resource waste, non-real-time operation, and insufficient collaboration, has become a critical technical challenge that urgently needs to be solved in the fields of industrial automation control and artificial intelligence. Summary of the Invention

[0009] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a system that supports the participation of artificial intelligence in real-time control, which sinks the AI ​​inference computing power from the edge computing layer to the terminal device layer, and realizes data interaction, collaborative decision-making and real-time closed-loop control with PLC within the same real-time control cycle, thereby solving the problems of latency, resource waste, non-real-time and insufficient collaboration from the architecture.

[0010] This invention provides a system that supports real-time control involving artificial intelligence, comprising: Programmable logic controllers (PLCs) are deployed at the industrial field control layer as the master station of industrial real-time Ethernet, used to execute control programs and manage real-time communication with slave devices. The AI ​​inference computer is deployed at the industrial field device layer as a native slave device of the industrial real-time Ethernet. The programmable logic controller and the AI ​​inference computer are connected via an industrial real-time Ethernet bus. The AI ​​inference computer includes a general AI inference module and an industrial real-time Ethernet slave module. The general AI inference module and the industrial real-time Ethernet slave module interact with each other through shared memory. The general AI reasoning module is used to receive real-time data from the site and perform artificial intelligence reasoning to generate reasoning results; The industrial real-time Ethernet slave module is configured with a hardware-based industrial real-time Ethernet slave protocol stack, which is used to connect the AI ​​inference computer as a native slave node to the industrial real-time Ethernet, so as to achieve real-time communication with the programmable logic controller within the same real-time control cycle. The programmable logic controller is used to work with the AI ​​inference computer to complete the real-time control closed loop of the field equipment based on the inference results transmitted by the AI ​​inference computer.

[0011] In one embodiment of the present invention, the industrial real-time Ethernet slave module is implemented using a dedicated communication chip, which is used to implement the hardware deployment of the hardware-based industrial real-time Ethernet slave protocol stack. The general-purpose AI inference module and the industrial real-time Ethernet slave module are designed with hardware separation, and the industrial real-time Ethernet slave module operates independently of the general-purpose AI inference module.

[0012] In one embodiment of the present invention, the industrial real-time Ethernet is selected from Profinet, EtherCAT or EtherNet / IP; the hardware-based industrial real-time Ethernet slave protocol stack is a hardware-based slave protocol stack of the corresponding protocol.

[0013] In one embodiment of the present invention, the shared memory between the general AI inference module and the industrial real-time Ethernet slave module is a dedicated hardware shared memory link; The general AI inference module writes the inference results into the shared memory, and the industrial real-time Ethernet slave module reads the inference results from the shared memory and transmits them as input signals to the programmable logic controller. The industrial real-time Ethernet slave module writes the received output signal from the programmable logic controller into the shared memory, and the general-purpose AI inference module reads the output signal from the shared memory.

[0014] In one embodiment of the present invention, the hardware design of the AI ​​inference computer meets the industrial field equipment design standards, including: wide operating temperature range, passive heat dissipation design, electromagnetic compatibility anti-interference design, dustproof and waterproof protection level, and vibration-resistant installation structure. The AI ​​inference computer is deployed in an industrial field control cabinet using a rail mounting method, arranged adjacent to the remote IO module, servo driver and frequency converter, and directly connected to the programmable logic controller via an industrial real-time Ethernet bus.

[0015] In one embodiment of the present invention, the AI ​​inference computer is configured with a slave device description file corresponding to the hardware-based industrial real-time Ethernet slave protocol stack. The slave device description file includes a GSDML file for the Profinet protocol, an ESI file for the EtherCAT protocol, or an EDS file for the EtherNet / IP protocol. The slave device description file is used to determine the input / output data volume, communication rate, device characteristics, and fault code information supported by the AI ​​inference computer; The programmable logic controller imports the slave device description file through configuration software, and identifies and configures the AI ​​inference computer as a native slave device.

[0016] In one embodiment of the present invention, the communication cycle between the programmable logic controller and the AI ​​inference computer is no greater than 1 millisecond, and the communication jitter is no greater than 1 microsecond. The shared memory enables nanosecond-level data interaction, and the data transmission between the general AI inference module and the industrial real-time Ethernet slave module has no intermediate buffer or protocol conversion loss.

[0017] The present invention also provides a method for supporting artificial intelligence in real-time control, applied to the system supporting artificial intelligence in real-time control, comprising: The AI ​​inference computer is deployed at the industrial field device layer, and the AI ​​inference computer includes a general AI inference module and an industrial real-time Ethernet slave module; Through the hardware-based industrial real-time Ethernet slave protocol stack in the industrial real-time Ethernet slave module, the AI ​​inference computer is connected to the industrial real-time Ethernet as a native slave node, establishing real-time communication with the programmable logic controller located in the industrial field control layer. Data interaction between the general-purpose AI inference module and the industrial real-time Ethernet slave module is achieved through shared memory. The general AI inference module receives real-time data from the site and performs artificial intelligence inference to generate inference results, which are then written into the shared memory. The industrial real-time Ethernet slave module reads the inference result from the shared memory and transmits it to the programmable logic controller via the industrial real-time Ethernet during the real-time control cycle; The programmable logic controller master station, in collaboration with the AI ​​inference computer, completes a real-time control closed loop for the field equipment based on the inference results.

[0018] In one embodiment of the present invention, establishing real-time communication with the programmable logic controller located in the industrial field control layer includes: The AI ​​inference computer provides a slave device description file corresponding to the hardware-based industrial real-time Ethernet slave protocol stack; Import the slave device description file into the configuration software of the programmable logic controller, identify the AI ​​inference computer as a native slave device, and complete the configuration and program writing. According to the configuration and program writing, the programmable logic controller communicates and performs logic control with the AI ​​inference computer through the industrial real-time Ethernet bus in each communication cycle, forming a real-time closed-loop control.

[0019] In one embodiment of the present invention, the step of realizing data interaction between the general AI inference module and the industrial real-time Ethernet slave module through shared memory includes: The programmable logic controller sends out industrial real-time Ethernet data frames. The industrial real-time Ethernet slave module receives industrial real-time Ethernet data frames and writes them into shared memory. The general AI inference module reads the industrial real-time Ethernet data frames from the shared memory and performs real-time inference, and writes the inference results obtained from the inference into the shared memory; The industrial real-time Ethernet slave module reads the inference result from shared memory and writes it into an industrial real-time Ethernet data frame; The programmable logic controller receives the written industrial real-time Ethernet data frames via the industrial real-time Ethernet bus and performs logic calculations and control.

[0020] The beneficial effects of this invention are as follows: This invention brings AI inference computing power down to the device layer, enabling AI inference nodes to access the real-time control domain as native slaves of industrial real-time Ethernet, and achieve deterministic real-time collaboration with PLC, thereby significantly reducing end-to-end control latency, meeting the timing requirements of millisecond-level closed-loop control, and truly integrating AI into the real-time control closed loop.

[0021] By performing AI inference directly at the device layer, a large amount of field data does not need to be uploaded to the edge or cloud, significantly reducing bandwidth consumption and data redundancy, and improving the system's response speed and anti-interference capability in weak network and high-disturbance environments. A flattened architecture of "device-layer AI inference + control-layer PLC collaboration" is reconstructed, reducing dependence on edge nodes and the cloud, and enhancing system robustness and autonomous operation capabilities.

[0022] AI inference computers adhere to industrial field hardware standards and provide standardized equipment description files. They can be directly recognized and configured by mainstream PLC configuration software without modifying existing control systems, making them highly practical and easy to deploy.

[0023] AI-powered inference computers perform real-time reasoning tasks on-site and feed the results back to the PLC for precise control, enabling the system to have adaptive and autonomous decision-making capabilities, significantly improving control accuracy and process optimization in high-end manufacturing scenarios.

[0024] In addition, the hardware-based real-time Ethernet protocol stack and shared memory interaction ensure determinism, low jitter and high stability of communication, which is fully adapted to the stringent real-time control requirements of industrial sites. Attached Figure Description

[0025] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0026] Figure 1 This is a field real-time control topology diagram of a system supporting artificial intelligence in real-time control provided in one embodiment of the present invention; Figure 2 This is a flowchart illustrating a method for supporting real-time control with the participation of artificial intelligence, provided in one embodiment of the present invention. Figure 3 This is a diagram illustrating the architecture and workflow of an AI inference computer provided in one embodiment of the present invention. Figure 4 This is a configuration flowchart for configuring an AI inference computer in a PLC configuration according to one embodiment of the present invention. Detailed Implementation

[0027] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0028] To address the technical shortcomings of existing AI-assisted industrial control systems based on edge computing, such as the AI ​​inference unit being deployed too high, the inability to natively access industrial real-time Ethernet, and the difficulty in achieving real-time collaboration with PLC, resulting in AI's inability to truly participate in real-time control, high control latency, and poor timing, this invention provides a method to support AI participation in real-time control. Through specific hardware configurations, protocol implementations, and communication mechanisms, this method ensures that AI can be deeply integrated into the real-time control closed loop.

[0029] Please see Figure 1As shown, in industrial automation, the layers L0 (device layer) and L1 (control layer) in the ISA-95 architecture are truly involved in real-time control, with the Programmable Logic Controller (PLC) at its core. The PLC connects to the device layer's remote I / O, servo drives / motors, frequency converters, and AI inference computers (visual recognition / real-time prediction) via industrial real-time Ethernet, periodically receiving and sending data to the device layer to complete real-time control. Industrial real-time Ethernet (such as Profinet, EtherCAT, etc.) can have a scan cycle of up to 1ms and jitter at the single-microsecond level, meeting the real-time control requirements of the field.

[0030] Among them, ISA-95 (International Standard for Automation, ANSI / ISA-95) Enterprise Control System Integration is one of the most important reference frameworks in the field of industrial automation and Manufacturing Execution System (MES). It defines the integration model between the enterprise layer (ERP / business system) and the field control layer (OT / operational technology). L0 (equipment layer) includes AI inference computers, remote I / O modules, servo drives, frequency converters, encoders, etc., as field execution and sensing nodes. L1 (control layer) includes PLC master station, as the real-time control core, performing logic operations and closed-loop control.

[0031] This invention provides a system that supports real-time control with the participation of artificial intelligence. The AI ​​inference computer, along with I / O modules and servo drives, is deployed at the L0 device layer, directly in the industrial field, eliminating the need for edge computing nodes. The AI ​​inference computer can be used for on-site visual recognition, real-time prediction, or other AI applications. It possesses industrial real-time Ethernet slave functionality, enabling it to exchange data with the PLC in real time, just like other field devices, completing the on-site control closed loop.

[0032] In this embodiment of the invention, the programmable logic controller (PLC) is deployed in the industrial field control layer as the master station of the industrial real-time Ethernet, used to execute control programs and manage real-time communication with slave devices; the AI ​​inference computer is deployed in the industrial field device layer as a native slave device of the industrial real-time Ethernet; the PLC and the AI ​​inference computer are connected via an industrial real-time Ethernet bus.

[0033] The AI ​​inference computer includes a general AI inference module and an industrial real-time Ethernet slave module. The general AI inference module and the industrial real-time Ethernet slave module interact with each other through shared memory. The general AI reasoning module is used to receive real-time data from the site and perform artificial intelligence reasoning to generate reasoning results; The industrial real-time Ethernet slave module is configured with a hardware-based industrial real-time Ethernet slave protocol stack, which is used to connect the AI ​​inference computer as a native slave node to the industrial real-time Ethernet, so as to achieve real-time communication with the programmable logic controller within the same real-time control cycle. The programmable logic controller is used to work with the AI ​​inference computer to complete the real-time control closed loop of the field equipment based on the inference results transmitted by the AI ​​inference computer.

[0034] It should be noted that the method for supporting artificial intelligence in real-time control provided by the embodiments of the present invention can realize real-time deterministic communication between the PLC as the master station and the AI ​​inference computer and other southbound devices (IO slave stations, servo drives, encoders, etc.). Through the collaboration of the two, closed-loop real-time control is completed on site, truly enabling AI inference to participate in real-time control tasks.

[0035] In this embodiment of the invention, the industrial real-time Ethernet slave module is implemented using a dedicated communication chip, which is used to implement the hardware deployment of the hardware-based industrial real-time Ethernet slave protocol stack. In order to ensure low jitter and low latency in industrial Ethernet real-time communication, the general-purpose AI inference module and the industrial real-time Ethernet slave module are designed to be hardware-separated, and the industrial real-time Ethernet slave module operates independently of the general-purpose AI inference module.

[0036] In this embodiment of the invention, the industrial real-time Ethernet is selected from Profinet, EtherCAT, or EtherNet / IP; the hardware-based industrial real-time Ethernet slave protocol stack is a hardware-based slave protocol stack of the corresponding protocol, that is, the hardware-based industrial real-time Ethernet slave protocol stack supports the industrial real-time Ethernet protocol. Different slave protocols (Profinet, EtherCAT, etc.) may require different hardware configurations. A direct real-time communication mechanism is established between the AI ​​inference computer and the PLC. Through industrial real-time Ethernet, a communication cycle as fast as 1 millisecond can be achieved, ensuring the real-time and deterministic nature of data communication. This breaks the traditional model of "AI assistance, PLC independent execution," allowing the AI ​​inference unit to truly participate in the real-time control closed loop as a real-time control node, collaborating with the PLC to complete real-time control tasks, and improving the control accuracy, adaptability, and autonomous decision-making level of the industrial control system.

[0037] In this embodiment of the invention, the shared memory between the general AI inference module and the industrial real-time Ethernet slave module is a dedicated hardware shared memory link; The general AI inference module writes the inference results into the shared memory, and the industrial real-time Ethernet slave module reads the inference results from the shared memory and transmits them as input signals to the programmable logic controller. The industrial real-time Ethernet slave module writes the received output signal from the programmable logic controller into the shared memory, and the general-purpose AI inference module reads the output signal from the shared memory.

[0038] This invention optimizes the internal data interaction method of the AI ​​inference computer, enabling nanosecond-level data exchange through shared memory and an industrial real-time Ethernet protocol stack. This allows the AI ​​inference computer to receive field I / O signals and transmit inference results to the PLC in real time, constructing a complete real-time control closed loop. At the same time, it optimizes the industrial control architecture, reduces the system's dependence on edge computing nodes and the cloud, reduces bandwidth consumption and data redundancy, and improves the system's availability and anti-interference capabilities in industrial field environments with weak networks, high disturbances, and high reliability requirements. This helps industrial control systems upgrade towards high-end intelligence and real-time capabilities.

[0039] In this embodiment of the invention, the AI ​​inference computer is comparable to current AI edge computing, and its hardware design meets the design standards for industrial field equipment, allowing it to be directly deployed in industrial fields. It includes: a wide operating temperature range, passive heat dissipation design, electromagnetic compatibility and anti-interference design, dustproof and waterproof protection level, and vibration-resistant installation structure.

[0040] The AI ​​inference computer is deployed in an industrial field control cabinet using a rail mounting method, arranged adjacent to the remote IO module, servo driver and frequency converter, and directly connected to the programmable logic controller via an industrial real-time Ethernet bus.

[0041] This enables AI inference computing power to be deeply integrated from the edge computing layer to the terminal device layer, giving the terminal device layer real-time AI inference capabilities, reducing latency caused by data uploading and forwarding, and solving the problem of the terminal layer lacking real-time AI processing capabilities.

[0042] In this embodiment of the invention, the AI ​​inference computer is configured with a slave device description file corresponding to the hardware-based industrial real-time Ethernet slave protocol stack. By adopting a hardware implementation scheme, the real-time performance of protocol communication can be guaranteed, enabling the AI ​​inference node to natively access the industrial real-time Ethernet and establishing a strictly synchronized, low-latency, deterministic real-time communication mechanism between the AI ​​inference unit and the PLC, ensuring that the two complete data interaction within the same control cycle.

[0043] The slave device description file includes a GSDML file for the Profinet protocol, an ESI file for the EtherCAT protocol, or an EDS file for the EtherNet / IP protocol. The slave device description file is used to determine the amount of input and output data, communication rate, device characteristics, and fault code information supported by the AI ​​inference computer, so as to facilitate the identification and configuration of the PLC configuration software.

[0044] The programmable logic controller (PLC) imports the slave device description file through configuration software, identifies and configures the AI ​​inference computer as a native slave device; by importing the slave device description file, it is configured as a fieldbus device, and corresponding input and output variables can be bound to it according to different application scenarios, which enables convenient configuration of the AI ​​inference computer by the PLC configuration software.

[0045] The core of this invention lies in realizing the deployment of AI inference computers in industrial settings, native access to industrial real-time Ethernet, and real-time collaborative control with PLCs. The overall implementation focuses on hardware research and development and adaptation, protocol stack development and hardware implementation, software configuration and adaptation, and on-site deployment and debugging, with each stage cooperating with each other and implemented step by step.

[0046] In this embodiment of the invention, regarding the hardware development and industrial environment adaptation of the AI ​​inference computer: The hardware design of the general-purpose AI inference module is benchmarked against the computing power configuration of industrial-grade AI edge computing devices. It selects low-power, high-stability AI chips (such as FPGA, embedded GPU, and dedicated inference chips) adapted to industrial scenarios, and combines them with industrial-grade motherboards, memory and storage modules to design a general-purpose AI inference module that meets the real-time inference requirements of the field, providing hardware computing power support for various field AI applications such as visual recognition and real-time prediction.

[0047] For mainstream industrial real-time Ethernet protocols such as Profinet, EtherCAT, and EtherNet / IP, corresponding hardware slave modules have been developed, equipped with dedicated protocol chips and communication interfaces, to achieve hardware deployment of the protocol stack and ensure low latency and low jitter in communication; the modules reserve standardized industrial interfaces to support hardware docking with general AI inference modules.

[0048] A dedicated shared memory hardware link is designed between the general AI inference module and the industrial real-time Ethernet slave module to achieve nanosecond-level data interaction between the two, ensuring the real-time performance and accuracy of data reading and writing, and eliminating data transmission loss.

[0049] The overall hardware design of the AI ​​inference computer follows industrial field equipment standards: it adopts an industrial-grade shell design to achieve dustproof, waterproof, and vibration-resistant properties; it optimizes the heat dissipation scheme and adopts passive heat dissipation to meet the requirements of fanless installation in industrial fields; it incorporates electromagnetic radiation anti-interference design to meet industrial field electromagnetic compatibility standards; and it is adaptable to a wide temperature operating environment of -40℃ to 70℃ to ensure that it can be directly deployed in the L0 equipment layer of industrial fields.

[0050] Second, regarding the development of industrial real-time Ethernet protocol stacks and the creation of device description files: For Profinet, EtherCAT, and EtherNet / IP protocols, we developed a hardware-based slave protocol stack adapted to the AI ​​inference computer, completing functions such as protocol frame parsing, data interaction, clock synchronization, and fault diagnosis to ensure real-time synchronous communication with the PLC master station; the communication cycle can reach 1 millisecond, and the jitter is controlled at the single microsecond level.

[0051] In accordance with the standard specifications of various industrial real-time Ethernet protocols, corresponding slave device description files are created—GSDML file for Profinet protocol, ESI file for EtherCAT protocol, and EDS file for EtherNet / IP protocol. These files clearly specify key information such as the amount of input / output data supported by the AI ​​inference computer, communication rate, device characteristics, and fault codes, ensuring that mainstream PLC configuration software can directly recognize them.

[0052] The developed hardware protocol stack was tested for protocol compatibility with PLCs from mainstream brands such as Siemens, Beckhoff, Rockwell, and Baoxin Tianxing to resolve communication matching issues and ensure that the AI ​​inference computer can achieve stable real-time communication with PLCs from different brands.

[0053] Third, regarding the adaptation of PLC configuration software and the formulation of control program development specifications: The completed device description file can be imported into mainstream PLC configuration software such as Baoxin Tianxing IDE, Siemens TIA Portal, Beckhoff TwinCAT, and Rockwell RSLogix to complete the configuration, variable binding, and communication parameter setting of the AI ​​inference computer as a slave device, ensuring that the configuration process is consistent with the existing industrial field equipment.

[0054] By combining the workflow of AI inference computers, standardized PLC control program development specifications are formulated, clarifying the input and output variable definitions, data interaction logic, and fault handling mechanisms between PLC and AI inference computers. This enables automation engineers to complete the writing of collaborative control programs according to existing programming habits without having to learn new development languages ​​or logic.

[0055] The system enables the rapid addition of AI inference computers and the batch binding of variables in PLC configuration software; it also develops a simple fault diagnosis interface that displays the communication status and inference operation status of the AI ​​inference computer in real time, facilitating operation and maintenance by field engineers.

[0056] Fourth, regarding on-site deployment, system integration and testing, and industrialization: The AI ​​inference computer adopts a standardized industrial field installation method (such as rail mounting) and is directly deployed in the control cabinet of the L0 equipment layer in the industrial field. It is arranged adjacent to devices such as IO modules and servo drives, and is directly connected to the PLC master station through industrial real-time Ethernet cable. There is no need to deploy edge computing nodes, which simplifies the field cabling.

[0057] In actual industrial settings (such as intelligent manufacturing production lines, precision motion control equipment, and process control production lines), system integration and debugging are carried out to complete the clock synchronization, data interaction, and collaborative control debugging of PLC and AI inference computer; communication cycle and inference process are optimized for different application scenarios (visual inspection + real-time control, state prediction + closed-loop regulation) to solve problems such as electromagnetic interference and data packet loss in the field environment.

[0058] It enables standardized mass production of AI inference computers; provides customized AI inference computing power configurations and protocol stack adaptations for real-time control needs in different industrial fields (automobile manufacturing, 3C electronics, chemical industry, metallurgy); and provides enterprises with integrated technical support for on-site deployment, debugging, and training, reducing the application threshold for enterprises.

[0059] It can also reserve computing power expansion interfaces for AI inference modules and firmware upgrade interfaces for protocol stacks, supporting subsequent computing power upgrades for AI chips and adaptation to new industrial real-time Ethernet protocols, ensuring that the technical solution of this invention can continuously iterate along with the development of industrial control and artificial intelligence technologies.

[0060] Fifth, regarding the continuous verification of reliability and stability: Long-term reliability testing can be conducted in industrial settings, simulating extreme environments such as high temperature, high humidity, strong electromagnetic interference, and network fluctuations to verify the continuous operation capability and communication stability of the AI ​​inference computer. A fault early warning and handling mechanism can be established, and functions such as fault self-checking and redundant communication can be added at the hardware and software levels to further improve the long-term operational reliability of the system in industrial settings.

[0061] Please see Figure 2 As shown, an embodiment of the present invention provides a method for supporting real-time control with artificial intelligence, applied to the system supporting real-time control with artificial intelligence, comprising the following steps: S1. Deploy the AI ​​inference computer at the industrial field device layer. The AI ​​inference computer includes a general AI inference module and an industrial real-time Ethernet slave module. S2. Through the hardware-based industrial real-time Ethernet slave protocol stack in the industrial real-time Ethernet slave module, the AI ​​inference computer is connected to the industrial real-time Ethernet as a native slave node, and real-time communication is established with the programmable logic controller located in the industrial field control layer. S3. Data interaction between the general AI inference module and the industrial real-time Ethernet slave module is achieved through shared memory; S4. The general AI inference module receives real-time data from the site and performs artificial intelligence inference to generate inference results, and writes the inference results into the shared memory. S5. The industrial real-time Ethernet slave module reads the inference result from the shared memory and transmits it to the programmable logic controller via the industrial real-time Ethernet during the real-time control cycle. S6. The programmable logic controller master station, in collaboration with the AI ​​inference computer, completes the real-time control closed loop of the field equipment based on the inference result.

[0062] In this embodiment of the invention, establishing real-time communication with the programmable logic controller located in the industrial field control layer includes: The AI ​​inference computer provides a slave device description file corresponding to the hardware-based industrial real-time Ethernet slave protocol stack; Import the slave device description file into the configuration software of the programmable logic controller, identify the AI ​​inference computer as a native slave device, and complete the configuration and program writing. According to the configuration and program writing, the programmable logic controller communicates and performs logic control with the AI ​​inference computer through the industrial real-time Ethernet bus in each communication cycle, forming a real-time closed-loop control.

[0063] Specifically, each slave device (remote I / O coupler, servo drive, frequency converter, etc.) has a device description file (such as Profinet's GSDML, EtherCAT's ESI, EtherNet / IP's EDS, etc.) that describes the communication capabilities and characteristics of the field device. The AI ​​inference computer also provides a device description file, describing its characteristics as an industrial real-time Ethernet slave.

[0064] See Figure 3 As shown, the working principle of configuring the AI ​​inference computer into the PLC program is described as follows: The AI ​​inference computer provides a device description file, which is imported into the PLC configuration software. For the PLC, the AI ​​inference computer, like other L0 devices, is a slave device. Automation control engineers can write automation control programs normally, which are then downloaded to the PLC via the PLC configuration software. The PLC performs slave communication and logic control based on the configuration and program, completing the control loop.

[0065] In this embodiment of the invention, the step of enabling data interaction between the general AI inference module and the industrial real-time Ethernet slave module through shared memory includes: The programmable logic controller sends out industrial real-time Ethernet data frames. The industrial real-time Ethernet slave module receives industrial real-time Ethernet data frames and writes them into shared memory. The general AI inference module reads the industrial real-time Ethernet data frames from the shared memory and performs real-time inference, and writes the inference results obtained from the inference into the shared memory; The industrial real-time Ethernet slave module reads the inference result from shared memory and writes it into an industrial real-time Ethernet data frame; The programmable logic controller receives the written industrial real-time Ethernet data frames via the industrial real-time Ethernet bus and performs logic calculations and control.

[0066] Specifically, see Figure 4 As shown, the PLC transmits data to the AI ​​inference computer as its output signal. The industrial real-time Ethernet slave module receives the PLC's output signal and copies it to the shared memory. The general-purpose AI inference module reads real-time data transmitted by the PLC from the shared memory and receives other data for inference, such as image signals for machine vision. The general-purpose AI inference module performs on-site inference, such as visual AI recognition, measurement, or real-time prediction, and then writes the inference results back to the shared memory. The industrial real-time Ethernet slave module reads the shared memory as the PLC input signal. The PLC receives the inference results through the real-time Ethernet bus and performs logic calculations and control.

[0067] In summary, this invention directly deploys the AI ​​inference computer at the L0 device layer of the ISA-95 model, employing a nanosecond-level data interaction method combining a hardware-based industrial real-time Ethernet protocol stack and shared memory. This enables the communication cycle between the AI ​​inference unit and the PLC to reach as fast as 1 millisecond, with communication jitter controlled at the microsecond level. This design eliminates the uncertain delays caused by multi-hop forwarding, protocol conversion, and data queuing at the edge computing layer, ensuring that AI inference results can be accurately delivered to the PLC within the real-time control cycle. It fully meets the timing constraints of millisecond-level real-time control tasks such as motion control and closed-loop regulation, effectively solving the problem of excessive end-to-end latency caused by the separation of the inference link and the control link in existing technologies.

[0068] By enabling AI inference nodes to connect to the real-time control domain as native slaves of industrial real-time Ethernet, and to synchronously interact with field devices such as PLCs, I / O modules, and servo drives within the same real-time cycle, the traditional control mode of "AI-assisted decision-making and PLC independent execution" is broken. The AI ​​inference unit thus becomes a crucial component of the real-time control node, collaborating with the PLC to complete real-time control tasks such as logic calculations, motion control, and process adjustment. This propels artificial intelligence from offline analysis and auxiliary suggestions to online real-time closed-loop control, achieving deterministic real-time collaboration between AI and PLC.

[0069] This invention endows industrial field terminal devices with real-time AI inference capabilities. Large amounts of field data do not need to be uploaded to the edge layer or cloud for processing; AI inference can be completed directly at the device layer, significantly reducing network bandwidth consumption and data redundancy, and effectively solving the problem of insufficient computing power at the terminal layer. In industrial field environments with weak networks, high disturbances, and high reliability requirements, the system does not rely on edge nodes or cloud computing power support, significantly improving system availability, anti-interference capabilities, and dynamic response speed.

[0070] This invention reconstructs a novel industrial control architecture of "AI inference at the device layer + PLC collaboration at the control layer," reducing excessive reliance on edge computing nodes and the cloud, simplifying data transmission links, and making the entire control system more hierarchical and data flow more efficient. Simultaneously, this architecture reduces the risks to on-site control caused by edge node or cloud failures, and enhances the robustness and autonomous operation capabilities of the industrial control system.

[0071] The AI ​​inference computer adheres to industrial field device design standards, adapting to industrial environment requirements such as wide operating temperature range, electromagnetic compatibility, dust and water resistance, passive heat dissipation, and vibration resistance, and can be directly deployed in existing industrial sites. Simultaneously, this AI inference computer provides standardized industrial real-time Ethernet slave device description files (GSDML files for Profinet protocol, ESI files for EtherCAT protocol, and EDS files for EtherNet / IP protocol), which can be directly recognized and configured by mainstream PLC configuration software. Automation engineers can complete device configuration and programming according to existing operating habits, without the need for large-scale modifications to existing PLC control systems and fieldbus architectures, significantly reducing the transformation costs and technical barriers for enterprise intelligent upgrades.

[0072] AI-powered inference computers can perform real-time AI applications such as visual recognition, state prediction, and process parameter optimization on-site, and feed the inference results back to the PLC for precise control, enabling industrial control systems to possess real-time adaptive and autonomous decision-making capabilities. Compared to traditional PLC pure logic control, this invention can dynamically adjust control strategies based on real-time on-site operating conditions, effectively improving control accuracy and process optimization levels in high-end manufacturing, precision control, and autonomous collaboration scenarios, and helping industrial control systems upgrade towards high-end intelligence.

[0073] This invention employs a hardware-based industrial real-time Ethernet slave protocol stack to replace the traditional software protocol stack, effectively reducing communication jitter (down to the microsecond level) and ensuring the determinism and stability of communication between the AI ​​inference unit and the PLC. Simultaneously, nanosecond-level data exchange between the AI ​​inference module and the slave module is achieved through shared memory, further improving the real-time performance and accuracy of data interaction. The above design is fully adapted to the stringent requirements of industrial environments for control stability and determinism.

[0074] This invention applies the system and method for supporting real-time control with artificial intelligence described in this invention to the visual positioning scenario of crescent weld seams in a cold-rolled galvanized production line. The system uses a Baoxin Tianxing T3 PLC as the industrial real-time Ethernet master station, deployed at the L1 control layer of the ISA-95 model; it uses a Rockchip RK3588 chip as the core to build an AI inference computer, deployed at the L0 device layer; the communication protocol used is Profinet real-time Ethernet, with a fixed communication period of 1ms and jitter ≤5μs.

[0075] In this embodiment of the invention, the general-purpose AI inference module is designed based on the industrial-grade Rockchip RK3588 chip. This chip integrates a 6TOPS high-performance NPU specifically for the visual positioning and inference task of crescent weld seams. The module is equipped with 8GB of industrial-grade DDR4 memory, 64GB of eMMC storage, and a customized industrial-grade motherboard; it integrates a MIPI / CameraLink industrial camera interface and a gigabit Ethernet port, which can be directly connected to linear or area scan industrial cameras, meeting the full-process computing power requirements for high-speed acquisition of weld seam images, real-time AI positioning, and coordinate data output.

[0076] The Profinet slave module is developed using the Renesas RZ / N2L communication chip and integrates an isolated RJ45 industrial communication interface to achieve hardware deployment of the Profinet RT / IRT protocol stack. This module directly connects to the RK3588 inference module via a high-speed PCIe interface, ensuring hardware-level interconnectivity.

[0077] Furthermore, a dedicated hardware shared memory link is designed between the general AI inference module and the Profinet slave module to achieve nanosecond-level data interaction. Key data output by AI inference, such as weld center coordinates, weld width, and effective positioning signals, are directly written to the protocol module through the shared memory, without data transfer or transmission loss.

[0078] Furthermore, the entire machine's hardware is adapted to industrial site requirements, adopting an aluminum alloy fanless passive heat dissipation structure, DIN rail mounting, and IP40 protection rating; through EMC anti-electromagnetic interference design, it adapts to the strong electromagnetic, high dust, and high vibration environment of the cold rolling workshop; it supports wide temperature operation from -40℃ to 70℃ and can be deployed in the L0 equipment layer control cabinet of the cold rolling galvanizing production line, adjacent to the Baoxin Tianxing PLC master station.

[0079] In this embodiment of the invention, a hardware-based Profinet slave protocol stack adapted to Renesas RZ / N2L is developed to realize frame parsing, distributed clock synchronization, data interaction, and fault diagnosis functions. The communication cycle is fixed at 1ms, achieving precise clock synchronization with the Baoxin Tianxing PLC master station.

[0080] In detail, a GSDML device description file is generated according to the Profinet standard specification. The file defines dedicated communication variables for weld seam positioning: input variables (such as strip running speed, camera trigger signal, production line running command, weld seam search command), output variables (such as crescent effective positioning signal, weld seam X-axis center coordinate, weld seam Y-axis offset, weld seam width, and positioning reliability), and also specifies communication parameters and fault codes, supporting direct recognition and automatic parsing by Baoxin Tianxing PLC configuration software.

[0081] Furthermore, the hardware protocol stack was integrated with the Baoxin Tianxing PLC for specialized joint debugging and testing. This completed full-function verification of slave station online, data transmission and reception, clock synchronization, and fault diagnosis, ensuring stable real-time communication with no packet loss and no delay.

[0082] In this embodiment of the invention, the GSDML device description file is imported into the Baoxin Tianxing IDE configuration software to complete the addition of the Profinet slave station, IP configuration, variable mapping, and communication cycle setting of the AI ​​inference computer. The configuration process is completely consistent with that of standard field I / O modules and servo devices.

[0083] In detail, a dedicated PLC control specification for cold-rolled galvanized weld seam positioning was formulated, clarifying the collaborative logic between the PLC and the AI ​​computer. Variables are defined as the weld seam coordinates output by the AI, with valid signals directly mapped to PLC process control variables. The core logic is that upon receiving a valid crescent-shaped positioning signal, the PLC immediately triggers adjustments to the galvanizing process parameters and the weld seam tracking mechanism. Fault mechanisms include defined alarm logic and safety protection mechanisms for communication interruption, AI positioning failure, and camera malfunction.

[0084] In the Baoxin Tianxing IDE, we can realize one-click configuration of AI computer and batch binding of weld variables, develop a dedicated diagnostic interface, and display the camera status, AI weld positioning frame rate, Profinet communication status and real-time coordinates of weld in real time.

[0085] In this embodiment of the invention, the RK3588 AI inference computer is mounted on a DIN rail and directly deployed in the L0 layer control cabinet of the cold-rolled galvanizing production line; the industrial camera is installed at the galvanizing weld inspection station and directly connected to the AI ​​computer via an Ethernet interface; the Profinet bus is directly connected to the Baoxin Tianxing PLC master station, eliminating the need for an edge server.

[0086] Furthermore, the YOLO / NCNN model for locating the crescent-shaped weld seam in cold rolling can be deployed to the RK3588 NPU, with an inference frame rate of ≥30fps and a weld seam positioning accuracy of ≤1mm. After the AI ​​inference computer identifies the crescent-shaped weld seam, it sends the positioning data to the PLC via Profinet within 10ms. The PLC adjusts the strip steel correction mechanism, the height of the galvanizing scraper, and the annealing process parameters in real time according to the weld seam coordinates to achieve adaptive control of the weld seam. The inference timing is optimized for high-speed strip steel operation (≤400m / min), and the communication shielding is optimized for strong electromagnetic environments, thereby achieving system integration and debugging.

[0087] The embodiments of this invention verify the effectiveness of the system and method described in this invention in real-time visual positioning and closed-loop control scenarios in the metallurgical industry, realizing the decentralization of AI inference computing power, hardware-based real-time communication, and deterministic collaboration with PLC.

[0088] This invention applies to the scenario of visual inspection and closed-loop control of strip misalignment across all stands in hot-rolled strip mills. It employs an industrial-grade Moore's Threads ABE300 chip as the AI ​​inference core, equipped with an EtherCAT industrial real-time Ethernet hardware slave module, and is compatible with the Baoxin Tianxing T4 PLC master station. Through real-time AI visual detection of the lateral offset of the strip across all stands, and utilizing the EtherCAT bus to achieve sub-millisecond data synchronization, a distributed detection-centralized collaborative control architecture is constructed to complete the closed-loop measurement and control of strip misalignment between hot continuous rolling mill stands.

[0089] In this embodiment of the invention, the general-purpose AI inference module is designed based on the Moore Threads ABE300 chip, providing 120 TOPS of high-performance AI computing power. It is equipped with a dedicated industrial-grade inference acceleration engine to meet the high-concurrency, low-latency inference requirements of parallel visual inspection across 7 racks of hot rolling mills. The module features 16GB of industrial-grade DDR5 memory, 128GB of industrial-grade NVMe storage, and a customized high-temperature resistant industrial motherboard. It integrates multiple GMSL2 / PoE industrial camera interfaces, supporting synchronous direct connection to industrial cameras across all racks from F1 to F7.

[0090] The EtherCAT slave module is developed using the Renesas RZ / N2L communication chip, integrating dual-channel isolated RJ45 industrial communication interfaces and cascading ports to achieve hardware deployment of the EtherCAT protocol stack. The module supports distributed clock synchronization, with a communication cycle ≤1ms and jitter ≤1μs. It directly connects to the ABE300 inference module via a high-speed PCIe interface, supporting rack-wide parallel data transmission.

[0091] Furthermore, a dedicated hardware shared memory link is designed between the ABE300 inference module and the EtherCAT slave module to achieve nanosecond-level data interaction. Key data such as the lateral offset of the strip steel in the F1-F7 full rack, offset direction, and effective alignment signal are directly written to the protocol module through the shared memory.

[0092] Furthermore, the entire machine is customized for the extreme environment of the hot rolling workshop, with fully sealed aluminum alloy passive heat dissipation, supporting an ultra-wide operating temperature range of -40℃ to 85℃; IP54 protection rating, DIN rail reinforced installation; through a three-level EMC electromagnetic interference anti-interference design, it resists strong electromagnetic interference from the rolling mill motor and hydraulic system; the seismic design meets the GB / T 2423 industrial vibration standard.

[0093] In this embodiment of the invention, a hardware-based EtherCAT slave protocol stack adapted to Renesas RZ / N2L is developed to realize frame parsing, distributed clock synchronization, real-time interaction of process data objects, and fault diagnosis functions. The communication cycle is fixed at 1ms, achieving sub-microsecond clock synchronization with the Baoxin Tianxing PLC master station.

[0094] In detail, an ESI equipment description file is generated according to the EtherCAT standard specification. The file defines dedicated communication variables for hot rolling mill full-stand correction: input variables (such as finishing mill running speed, F1-F7 stand enable signals, camera synchronization trigger signals, and correction mode commands), output variables (such as strip lateral offset of each F1-F7 stand, offset valid signal, correction recommendation coefficient, and equipment fault status), and also specifies the communication parameters and equipment topology.

[0095] Furthermore, the protocol stack was integrated with the Baoxin Tianxing PLC for EtherCAT bus-specific joint testing, completing slave scanning, topology identification, clock synchronization, and full rack data transmission and reception verification, thus resolving issues related to bus cascading and multi-data channel matching.

[0096] In this embodiment of the invention, the ESI device description file is imported into the Baoxin Tianxing IDE configuration software to complete the addition of EtherCAT slaves of the AI ​​inference computer, node address allocation, batch binding of F1-F7 rack variables and communication cycle configuration, and to build a system configuration architecture of distributed vision inspection + centralized PLC control.

[0097] In detail, a standardized control procedure was developed based on the strip deviation mechanism between hot strip mill stands, binding variables such as the strip deviation of the entire stand and the effective deviation signal output by ABE300; the lateral deviation detected in real time by AI vision is used as the core feedback basis, following the logic of feedforward compensation + feedback closed-loop composite control, inter-stand linkage and coordination, and tension-correction coupling adjustment; and an inter-stand linkage, graded early warning and interlocking protection mechanism was developed.

[0098] Develop a dedicated configuration panel in Tianxing IDE to integrate visual inspection data (offset, detection accuracy), AI vision operation status (inference frame rate, camera status), deviation monitoring, and fault diagnosis functions.

[0099] In this embodiment of the invention, the Moore Threads ABE300 AI inference computer is reinforced with DIN rails and deployed in the main control cabinet of the L0 layer of the hot rolling mill; the F1-F7 full-frame high-temperature resistant industrial line array cameras are installed on the mill inlet side and directly connected to the AI ​​computer via the GMSL2 bus; the AI ​​computer, PLC master station and full-frame hydraulic correction servo system are cascaded via EtherCAT bus.

[0100] Furthermore, a lightweight model for hot-rolled strip steel offset detection is deployed to the ABE300AI inference computer acceleration engine, supporting parallel inference from 7 cameras, a frame rate ≥60fps, and offset detection accuracy ≤1mm. After the AI ​​inference computer completes the offset detection of the entire stand, it sends the data to the PLC via EtherCAT within 1ms. The PLC adjusts the tension between the hydraulic correction rollers of stands F1-F7 and the stands according to the composite control strategy to achieve fully automatic alignment of the strip steel at high speed (≤800m / min). The linkage parameters between stands are optimized to solve the deviation measurement and control errors caused by high temperature, vibration, and tension coupling, thereby achieving system integration and debugging.

[0101] The embodiments of this invention verify the high real-time performance, high reliability, and multi-rack collaborative control capabilities of this invention in extreme industrial environments, enabling the AI ​​inference computer to move from the edge layer to the device layer and complete sub-millisecond closed-loop control with the PLC via the EtherCAT bus.

[0102] The embodiments provided in this invention are merely preferred embodiments and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

[0103] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0104] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0105] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A system supporting real-time control with the participation of artificial intelligence, characterized in that, include: Programmable logic controllers (PLCs) are deployed at the industrial field control layer as the master station of industrial real-time Ethernet, used to execute control programs and manage real-time communication with slave devices. The AI ​​inference computer is deployed at the industrial field device layer as a native slave device of the industrial real-time Ethernet. The programmable logic controller and the AI ​​inference computer are connected via an industrial real-time Ethernet bus. The AI ​​inference computer includes a general AI inference module and an industrial real-time Ethernet slave module. The general AI inference module and the industrial real-time Ethernet slave module interact with each other through shared memory. The general AI reasoning module is used to receive real-time data from the site and perform artificial intelligence reasoning to generate reasoning results; The industrial real-time Ethernet slave module is configured with a hardware-based industrial real-time Ethernet slave protocol stack, which is used to connect the AI ​​inference computer as a native slave node to the industrial real-time Ethernet, so as to achieve real-time communication with the programmable logic controller within the same real-time control cycle. The programmable logic controller is used to work with the AI ​​inference computer to complete the real-time control closed loop of the field device based on the inference results transmitted by the AI ​​inference computer. The AI ​​inference computer is deployed directly on the L0 device layer of the ISA-95 model.

2. The system supporting real-time control with the participation of artificial intelligence according to claim 1, characterized in that, The industrial real-time Ethernet slave module is implemented using a dedicated communication chip, which is used to implement the hardware deployment of the hardware-based industrial real-time Ethernet slave protocol stack. The general-purpose AI inference module and the industrial real-time Ethernet slave module are designed with hardware separation, and the industrial real-time Ethernet slave module operates independently of the general-purpose AI inference module.

3. The system supporting real-time control with the participation of artificial intelligence according to claim 1, characterized in that, The industrial real-time Ethernet is selected from Profinet, EtherCAT, or EtherNet / IP; the hardware-based industrial real-time Ethernet slave protocol stack is the corresponding hardware-based slave protocol stack.

4. The system supporting real-time control with the participation of artificial intelligence according to claim 1, characterized in that, The shared memory between the general AI inference module and the industrial real-time Ethernet slave module is a dedicated hardware shared memory link. The general AI inference module writes the inference results into the shared memory, and the industrial real-time Ethernet slave module reads the inference results from the shared memory and transmits them as input signals to the programmable logic controller. The industrial real-time Ethernet slave module writes the received output signal from the programmable logic controller into the shared memory, and the general-purpose AI inference module reads the output signal from the shared memory.

5. The system supporting real-time control with the participation of artificial intelligence according to claim 1, characterized in that, The hardware design of the AI ​​inference computer meets the industrial field equipment design standards, including: wide operating temperature range, passive heat dissipation design, electromagnetic compatibility and anti-interference design, dustproof and waterproof protection level, and vibration-resistant installation structure. The AI ​​inference computer is deployed in an industrial field control cabinet using a rail mounting method, arranged adjacent to the remote IO module, servo driver and frequency converter, and directly connected to the programmable logic controller via an industrial real-time Ethernet bus.

6. The system supporting real-time control with the participation of artificial intelligence according to claim 1, characterized in that, The AI ​​inference computer is configured with a slave device description file corresponding to the hardware-based industrial real-time Ethernet slave protocol stack. The slave device description file includes a GSDML file for the Profinet protocol, an ESI file for the EtherCAT protocol, or an EDS file for the EtherNet / IP protocol. The slave device description file is used to determine the input / output data volume, communication rate, device characteristics, and fault code information supported by the AI ​​inference computer; The programmable logic controller imports the slave device description file through configuration software, and identifies and configures the AI ​​inference computer as a native slave device.

7. The system supporting real-time control with the participation of artificial intelligence according to claim 1, characterized in that, The communication cycle between the programmable logic controller and the AI ​​inference computer is no greater than 1 millisecond, and the communication jitter is no greater than 1 microsecond. The shared memory enables nanosecond-level data interaction, and the data transmission between the general AI inference module and the industrial real-time Ethernet slave module has no intermediate buffer or protocol conversion loss.

8. A method for supporting real-time control with artificial intelligence, applied to the system for supporting real-time control with artificial intelligence as described in any one of claims 1 to 7, characterized in that, include: The AI ​​inference computer is deployed at the industrial field device layer, and the AI ​​inference computer includes a general AI inference module and an industrial real-time Ethernet slave module; Through the hardware-based industrial real-time Ethernet slave protocol stack in the industrial real-time Ethernet slave module, the AI ​​inference computer is connected to the industrial real-time Ethernet as a native slave node, establishing real-time communication with the programmable logic controller located in the industrial field control layer. Data interaction between the general-purpose AI inference module and the industrial real-time Ethernet slave module is achieved through shared memory. The general AI inference module receives real-time data from the site and performs artificial intelligence inference to generate inference results, which are then written into the shared memory. The industrial real-time Ethernet slave module reads the inference result from the shared memory and transmits it to the programmable logic controller via the industrial real-time Ethernet during the real-time control cycle; The programmable logic controller master station, in collaboration with the AI ​​inference computer, completes a real-time control closed loop for the field equipment based on the inference results.

9. The method for supporting real-time control with artificial intelligence according to claim 8, characterized in that, The establishment of real-time communication with the programmable logic controller located in the industrial field control layer includes: The AI ​​inference computer provides a slave device description file corresponding to the hardware-based industrial real-time Ethernet slave protocol stack; Import the slave device description file into the configuration software of the programmable logic controller, identify the AI ​​inference computer as a native slave device, and complete the configuration and program writing. According to the configuration and program writing, the programmable logic controller communicates and performs logic control with the AI ​​inference computer through the industrial real-time Ethernet bus in each communication cycle, forming a real-time closed-loop control.

10. The method for supporting real-time control with artificial intelligence according to claim 8, characterized in that, The method of enabling data interaction between the general AI inference module and the industrial real-time Ethernet slave module through shared memory includes: The programmable logic controller sends out industrial real-time Ethernet data frames. The industrial real-time Ethernet slave module receives industrial real-time Ethernet data frames and writes them into shared memory. The general AI inference module reads the industrial real-time Ethernet data frames from the shared memory and performs real-time inference, and writes the inference results obtained from the inference into the shared memory; The industrial real-time Ethernet slave module reads the inference result from shared memory and writes it into an industrial real-time Ethernet data frame; The programmable logic controller receives the written industrial real-time Ethernet data frames via the industrial real-time Ethernet bus and performs logic calculations and control.

Images