Intelligent cooperative control system for multiple welding devices, model training method and storage medium

Abstract

The invention discloses an intelligent cooperative control system for multiple welding devices, a model training method and a storage medium. The intelligent cooperative control system comprises a security authentication module which is deployed at an edge computing gateway, adopts a dynamic password encryption system based on a national cipher SM4 algorithm and comprises a periodically updated device identity certificate; the parameter decision center integrates a local knowledge base and a cloud deep reinforcement learning model, the local knowledge base stores a welding process parameter and material thickness feature mapping table, and the cloud deep reinforcement learning model generates a dynamic tuning strategy based on real-time state data of a welding equipment group; the equipment discovery engine adopts an improved ICMPv6 protocol for scanning and fuses an equipment MAC address white list filtering mechanism; security authentication is achieved through a national secret algorithm and an equipment fingerprint system, the welding quality is improved through intelligent parameter matching and dynamic strategy generation, the model adaptability is enhanced in combination with physical simulation and multi-objective optimization, the resource efficiency is optimized through rapid fault detection and low-power-consumption design, and the industrial safety compliance requirement is met.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent welding, and more specifically, relates to an intelligent collaborative control system for multiple welding devices, a model training method, and a storage medium. Background Art

[0002] With the rapid development of intelligent manufacturing technology, the demand for multi-device collaborative control in the industrial welding field is increasing day by day. The control systems of traditional welding devices mostly adopt a centralized control architecture and rely on a fixed process parameter library for welding operations. However, in actual industrial production, variable factors such as material thickness, environmental disturbances, and equipment status often lead to unstable welding quality, and the existing systems have significant deficiencies in aspects such as dynamic parameter tuning, security authentication, and communication reliability.

[0003] In terms of security authentication, traditional systems mostly adopt static passwords or one-way encryption mechanisms, which are difficult to resist security threats such as man-in-the-middle attacks and device identity forgery. At the same time, in the industrial Internet of Things environment, device access is complex, and the risk of illegal device penetration is high. Existing device discovery mechanisms (such as ARP scanning, basic ICMP protocol) lack active defense capabilities and are vulnerable to attacks such as ARP spoofing, resulting in insufficient overall system security.

[0004] In terms of parameter decision-making, existing technologies mainly rely on manual experience or local database matching and lack the ability to adapt to complex working conditions. For example, when the thickness of the welding material changes or the state of the device group fluctuates, the fixed parameter library is difficult to generate optimal strategies in real time, easily leading to problems such as a decline in weld quality and an increase in energy consumption. In addition, the collaborative ability of traditional control models for multi-objective optimization (such as quality, efficiency, energy consumption) is weak, and the overall process performance is often affected by over-optimization of a single index.

[0005] In terms of communication and collaborative control, the industrial field environment is complex, and problems such as electromagnetic interference and network jitter occur frequently. Traditional single-link communication protocols (such as Modbus, HTTP) are prone to communication interruptions or high delays, resulting in a decline in the accuracy of device collaborative scheduling. Existing redundant communication schemes mostly adopt fixed bandwidth allocation, are difficult to dynamically adapt to changes in network load, and the standby link has high power consumption, unable to meet the dual requirements of low power consumption and high reliability in industrial scenarios.

[0006] In terms of model training and deployment, although the control model based on deep reinforcement learning has the potential to generate dynamic strategies, existing training methods often have problems such as a large difference between the virtual environment and the real scenario and a single reward function design, resulting in insufficient model generalization ability. In addition, complex models (such as deep neural networks) face computational resources and real-time constraints when deployed on edge devices and are difficult to achieve efficient inference.

[0007] Therefore, the present invention provides an intelligent collaborative control system for multiple welding devices, a model training method, and a storage medium. Summary of the Invention

[0008] In view of the above problems existing in the prior art, the purpose of the present invention is to provide an intelligent collaborative control system for multiple welding devices, a model training method, and a storage medium, which achieve security authentication through the national secret algorithm and the device fingerprint system, improve welding quality by using intelligent parameter matching and dynamic strategy generation, enhance the model adaptability by combining physical simulation and multi-objective optimization, and optimize resource efficiency through rapid fault detection and low-power design, meeting the requirements of industrial safety compliance.

[0009] The purpose of the present invention can be achieved by the following technical solutions:

[0010] An intelligent collaborative control system for multiple welding devices, comprising:

[0011] A security authentication module, deployed on the edge computing gateway, adopting a dynamic password encryption system based on the national secret SM4 algorithm, and including device identity certificates updated periodically;

[0012] A parameter decision center, which integrates a local knowledge base and a cloud deep reinforcement learning model. The local knowledge base stores a mapping table of welding process parameters and material thickness characteristics, and the cloud deep reinforcement learning model generates a dynamic tuning strategy based on the real-time status data of the welding device group;

[0013] A device discovery engine, which scans using an improved ICMPv6 protocol and integrates a device MAC address whitelist filtering mechanism;

[0014] A data channel management unit, which establishes a dual-link redundant communication. The main link uses MQTT over TLS1.3, and the backup link uses the LoRaWAN protocol;

[0015] A human-machine interaction interface, which implements multi-dimensional parameter visualization, including three-dimensional dynamic simulation of the welding molten pool and a waveform diagram of voiceprint anomaly detection.

[0016] As a further preferred technical solution of the present invention, the security authentication module is provided with a device fingerprint generation unit, an elliptic curve digital signature unit, and a certificate revocation list synchronization unit;

[0017] The device fingerprint generation unit generates a unique identifier based on the welder controller chip ID and the power module ripple characteristics in the welding device group;

[0018] The elliptic curve digital signature unit uses the secp256r1 curve to implement the signature verification of the policy instruction;

[0019] The certificate revocation list synchronization unit synchronizes and updates the blacklisted devices with the CA center every hour.

[0020] As a further preferred technical solution of the present invention, the local knowledge base of the parameter decision center uses an improved B+ tree index structure, and its leaf nodes store: material category code, thickness discretization interval, welding method feature vector, and confidence score of the optimal parameter combination.

[0021] As a further preferred technical solution of the present invention, the cloud deep reinforcement learning model includes a policy network, a value network, and an experience replay buffer pool. The policy network is provided with a 5-layer convolutional neural network, and the input of the policy network is the device group state matrix;

[0022] The value network is used to predict the long-term benefits of different policies and is provided with a 3-layer LSTM network;

[0023] The experience replay buffer pool adopts a priority sampling mechanism, and its storage capacity is at least 1TB;

[0024] The update of the cloud deep reinforcement learning model adopts the double-delay DDPG algorithm, and online incremental learning is performed every 24 hours.

[0025] As a further preferred technical solution of the present invention, the working process of the device discovery engine includes:

[0026] Step A: Send a customized ICMP probe packet, and the payload contains the target subnet mask and protocol version identifier;

[0027] Step B: Parse the device metadata in the response packet, and extract the first 3 bytes of the MAC address for vendor whitelist matching;

[0028] Step C: Implement ARP spoofing attack defense on unauthorized devices and send a false gateway MAC address.

[0029] As a further preferred technical solution of the present invention, the dual-link redundancy mechanism of the data channel management unit is specifically implemented as:

[0030] The dual-link redundancy mechanism of the data channel management unit adopts a main link bandwidth dynamic allocation algorithm, which adjusts the transmission rate in real time based on Welch power spectrum estimation;

[0031] The standby link of the dual-link redundancy mechanism of the data channel management unit adopts an adaptive frequency hopping technology, which divides 16 channels in the 2.4GHz band;

[0032] The link switching trigger condition of the dual-link redundancy mechanism of the data channel management unit is that 3 consecutive RTT values exceed 500ms or the bit error rate > 10^-4.

[0033] As a further preferred technical solution of the present invention, the human-computer interaction interface includes a molten pool dynamic simulation subsystem, a voiceprint feature extraction unit, and an anomaly detection and warning unit;

[0034] The molten pool dynamic simulation subsystem calculates the heat conduction equation in real time based on finite element analysis;

[0035] The voiceprint feature extraction unit analyzes the arc sound signal using Mel Frequency Cepstral Coefficients (MFCC);

[0036] The anomaly detection and warning unit triggers a red stroboscopic warning when the penetration depth volatility exceeds ±15%.

[0037] A method for training an enhanced learning model of a multi-welding equipment intelligent collaborative control system. The cloud deep reinforcement learning model of the above multi-welding equipment intelligent collaborative control system includes the following stage steps:

[0038] Stage 1: Construct a virtual welding environment and inject several typical working condition disturbance modes;

[0039] Stage 2: Design a multi-dimensional reward function:

[0040]

[0041] Stage 3: Adopt a curriculum learning strategy and gradually increase the complexity of the material thickness combination;

[0042] Stage 4: Deploy a model distillation device to transfer the knowledge of the teacher network to the lightweight student network.

[0043] As a further preferred technical solution of the present invention, the construction of the virtual welding environment includes the following steps:

[0044] Sub-step 1.1: Establish an arc physical equation:

[0045]

[0046] Sub-step 1.2: Simulate the turbulent effect of the shielding gas and calculate the flow field distribution using the k-epsilon model;

[0047] Sub-step 1.3: Generate sensor data with noise and add Gaussian white noise and pulse interference.

[0048] A computer-readable storage medium stores a control program for implementing the above multi-welding equipment intelligent collaborative control system. The control program includes a device cooperative scheduling instruction set for achieving nanosecond-level time synchronization accuracy, a real-time kernel patch module that optimizes interrupt response using the RT-Preempt technology, and a memory protection unit configuration strategy for isolating critical algorithm processes from the user space.

[0049] As described above, the intelligent collaborative control system, model training method, and storage medium for multi-welding equipment provided by the present invention have the following beneficial effects:

[0050] 1. By using the above intelligent collaborative control system, model training method, and storage medium for multi-welding equipment, compared with the prior art, the intelligent collaborative control system for multi-welding equipment effectively resists the risks of man-in-the-middle attacks and identity forgery by adopting a dynamic password encryption system (national secret SM4) and a periodic certificate update mechanism; and based on the unique fingerprint generated by the device chip ID and the power supply ripple characteristics, combined with the elliptic curve digital signature (secp256r1), strong device identity verification is achieved; in addition, the certificate revocation list (CRL) mechanism synchronized hourly is also used to isolate abnormal devices in a timely manner, improving the overall security of the system, and thus establishing a secure and trustworthy collaborative authentication mechanism.

[0051] 2. By using the above intelligent collaborative control system, model training method, and storage medium for multi-welding equipment, compared with the prior art, the local knowledge base of the present invention realizes millisecond-level process parameter matching and adapts to the discretized interval of material thickness through an improved B+ tree index structure (supporting confidence score storage); and through the cloud deep reinforcement learning model (double-delay DDPG algorithm) combined with at least 1TB of prioritized experience replay pool, a dynamic optimization strategy is generated to adapt to the complex working conditions of multiple device groups; enabling the intelligent collaborative control system for multi-welding equipment of the present invention to have the ability of intelligent parameter decision-making.

[0052] 3. By using the above intelligent collaborative control system, model training method, and storage medium for multi-welding equipment, compared with the prior art, in the construction of the virtual environment of the model training method, by injecting the arc physical equation (including the turbulent k-epsilon model) and multi-type noise interference, the complex disturbances of the real welding scene are simulated, improving the robustness of the model, and also combining the multi-dimensional reward function (current deviation, weld quality, energy consumption efficiency) to balance the process objectives and avoid over-optimization of a single index; in addition, the curriculum learning mechanism gradually increases the complexity of the material thickness combination, accelerating the model convergence and avoiding the local optimal trap, and the model distillation technology transfers the knowledge of the teacher network to the lightweight student network, reducing the resource requirements for edge device deployment.

[0053] 4. The present invention utilizes the above-mentioned intelligent collaborative control system for multi-welding equipment, model training method, and storage medium. Compared with the prior art, when implementing the present invention, its dynamic parameter tuning improves the welding qualified rate, enhances the collaborative scheduling efficiency, and its voiceprint anomaly detection shortens the equipment fault warning time to within 5 seconds, improving the welding production efficiency. Moreover, the dual-link redundancy reduces the communication interruption loss, the power consumption of the LoRaWAN backup link is lower than that of traditional protocols, and the use of the lightweight student network model reduces the occupancy of edge computing resources, resulting in significant overall resource optimization. In addition, the national cryptographic algorithm and the device fingerprint system meet the requirements of the third level of the equal protection 2.0 for industrial control systems, ensuring the security and compliance of the system.

[0054] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0056] Figure 1 It is the architecture topology diagram of the intelligent collaborative control system for multi-welding equipment of the present invention application;

[0057] Figure 2 It is the structural framework diagram of the intelligent collaborative control system for multi-welding equipment of the present invention application;

[0058] Figure 3 It is the work flow chart of the device discovery engine of the intelligent collaborative control system for multi-welding equipment of the present invention application;

[0059] Figure 4 It is the stage step flow chart of the training method of the cloud deep reinforcement learning model of the device discovery engine of the intelligent collaborative control system for multi-welding equipment of the present invention application;

[0060] Figure 5 It is the step flow chart of constructing the virtual welding environment in the training method of the cloud deep reinforcement learning model of the device discovery engine of the intelligent collaborative control system for multi-welding equipment of the present invention application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] The following specific embodiments illustrate the implementation manners of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification.

[0062] It should be noted that the structures, proportions, sizes, etc. shown in the attached drawings of this specification are only used to cooperate with the content disclosed in the specification for those familiar with this technology to understand and read, and are not used to limit the implementation conditions of the present invention. Therefore, they do not have substantial technical significance. Any modification of the structure, change of the proportional relationship, or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope covered by the technical content disclosed in the present invention. At the same time, the terms such as "upper", "lower", "left", "right", "middle", and "one" cited in this specification are only for the convenience of narration and are not used to limit the scope of implementation of the present invention. The change or adjustment of their relative relationships, without substantial change in the technical content, should also be regarded as the scope of implementation of the present invention. The specific structure can be described with reference to the attached drawings of the patent application.

[0063] The present invention provides a multi-welding equipment intelligent collaborative control system, a model training method, and a storage medium. Please refer to Figures 1 to 5 As shown in the figure, a multi-welding equipment intelligent collaborative control system includes:

[0064] A security authentication module, deployed on the edge computing gateway, adopts a dynamic password encryption system based on the national cryptographic SM4 algorithm and includes device identity certificates updated periodically;

[0065] A parameter decision center, which integrates a local knowledge base and a cloud deep reinforcement learning model. The local knowledge base stores a mapping table of welding process parameters and material thickness characteristics, and the cloud deep reinforcement learning model generates a dynamic optimization strategy based on the real-time status data of the welding equipment group;

[0066] A device discovery engine, which uses an improved ICMPv6 protocol scan and integrates a device MAC address white list filtering mechanism;

[0067] A data channel management unit, which establishes a dual-link redundant communication. The main link uses MQTT over TLS1.3, and the backup link uses the LoRaWAN protocol;

[0068] A human-machine interaction interface, which implements multi-dimensional parameter visualization, including three-dimensional dynamic simulation of the welding molten pool and a sound wave anomaly detection waveform diagram.

[0069] The security authentication module is provided with a device fingerprint generation unit, an elliptic curve digital signature unit, and a certificate revocation list synchronization unit;

[0070] The device fingerprint generation unit generates a unique identifier based on the ID of the welding machine controller chip and the ripple characteristics of the power module in the welding equipment group;

[0071] The elliptic curve digital signature unit uses the secp256r1 curve to implement the signature verification of the policy instruction;

[0072] The described book revocation list synchronization unit synchronizes and updates the blacklisted devices with the CA center every hour.

[0073] It should be noted that: The security authentication module uses national cryptography SM4 dynamic password encryption and elliptic curve digital signature (secp256r1 curve), combines device fingerprints (based on chip ID and power ripple characteristics) with the certificate revocation list synchronized every hour, effectively resists illegal device access and network attacks, and ensures the security of system identity authentication and data transmission.

[0074] The local knowledge base of the described parameter decision center uses an improved B+ tree index structure, and its leaf nodes store: material category encoding (e.g., 0: carbon steel, 1: stainless steel, 2: aluminum alloy), thickness discretization interval (e.g., 0.5mm step), welding method feature vector (including current type, shielding gas composition encoding, etc.) and the confidence score of the optimal parameter combination.

[0075] The described cloud deep reinforcement learning model includes a policy network, a value network, and an experience replay buffer pool. The policy network is set with a 5-layer convolutional neural network, and the input of the policy network is the device group state matrix;

[0076] The value network is used to predict the long-term benefits of different policies and is set with a 3-layer LSTM network;

[0077] The experience replay buffer pool adopts a priority sampling mechanism, and its storage capacity is at least 1TB;

[0078] The update of the described cloud deep reinforcement learning model uses the double-delay DDPG algorithm and performs online incremental learning every 24 hours.

[0079] It should be noted that: The parameter decision center integrates the local knowledge base (improved B+ tree index, quickly matching material thickness and process parameters) and the cloud deep reinforcement learning model (double-delay DDPG algorithm), and generates dynamic tuning strategies in real time, significantly improving welding quality and process stability.

[0080] The working process of the described device discovery engine includes:

[0081] Step A: Send a customized ICMP probe packet, and the payload contains the target subnet mask and protocol version identifier;

[0082] Step B: Analyze the device metadata in the response packet and extract the first 3 bytes of the MAC address for vendor whitelist matching;

[0083] Step C: Implement ARP spoofing attack defense on unauthorized devices and send a false gateway MAC address.

[0084] The improved ICMPv6 protocol of the device discovery engine integrates MAC whitelist filtering to achieve rapid device discovery and active defense against illegal devices (ARP spoofing suppression), further blocking the penetration risk of unauthorized devices.

[0085] The dual-link redundancy mechanism of the data channel management unit is specifically implemented as follows:

[0086] The dual-link redundancy mechanism of the data channel management unit adopts a dynamic bandwidth allocation algorithm for the main link, which adjusts the transmission rate in real time based on Welch power spectrum estimation;

[0087] The standby link of the dual-link redundancy mechanism of the data channel management unit adopts adaptive frequency hopping technology, which divides 16 channels in the 2.4GHz frequency band;

[0088] The link switching trigger condition of the dual-link redundancy mechanism of the data channel management unit is that 3 consecutive RTT values exceed 500ms or the bit error rate > 10^-4; The dual-link redundancy communication (MQTT over TLS1.3 main link + LoRaWAN standby link) combines dynamic bandwidth allocation and adaptive frequency hopping technology to ensure the continuity and low latency of communication in harsh industrial environments (switching condition: RTT > 500ms or bit error rate > 10^-4), ensuring high-reliability communication and redundancy fault tolerance of the system.

[0089] The human-machine interaction interface includes a molten pool dynamic simulation subsystem, a voiceprint feature extraction unit, and an anomaly detection and warning unit;

[0090] The molten pool dynamic simulation subsystem calculates the heat conduction equation in real time based on finite element analysis;

[0091] The voiceprint feature extraction unit analyzes the arc sound signal using Mel Frequency Cepstral Coefficients (MFCC);

[0092] The anomaly detection and warning unit triggers a red stroboscopic warning when the penetration depth volatility exceeds ±15%.

[0093] Three-dimensional molten pool dynamic simulation and voiceprint MFCC analysis provide multi-dimensional anomaly detection, and when the penetration depth volatility exceeds ±15%, it triggers real-time warnings, reducing the welding defect rate.

[0094] Through the collaborative innovation of multiple modules, this system realizes the efficient collaborative control of the welding equipment group, has significant advantages in terms of safety, decision-making accuracy, communication reliability, human-machine interaction, and model training efficiency, and can be widely applied in the field of high-precision welding to promote the upgrading of intelligent manufacturing.

[0095] A method for training a reinforcement learning model of an intelligent collaborative control system for multiple welding devices. The cloud deep reinforcement learning model of the above-mentioned intelligent collaborative control system for multiple welding devices includes the following stage steps:

[0096] Stage 1: Construct a virtual welding environment and inject several typical working condition disturbance modes;

[0097] Stage 2: Design a multi-dimensional reward function:

[0098]

[0099] Stage 3: Adopt a curriculum learning strategy and gradually increase the complexity of the material thickness combination;

[0100] Stage 4: Deploy a model distillation device to transfer the knowledge of the teacher network to the lightweight student network.

[0101] The construction of the virtual welding environment includes the following steps:

[0102] Sub-step 1.1: Establish an arc physical equation:

[0103]

[0104] Sub-step 1.2: Simulate the turbulent effect of the shielding gas and calculate the flow field distribution using the k-epsilon model;

[0105] Sub-step 1.3: Generate sensor data with noise and add Gaussian white noise and pulse interference.

[0106] It should be noted that: The virtual welding environment integrates the arc physical equation, turbulent effect simulation (k-epsilon model) and noise injection to accelerate the model's learning of real working conditions; and the model distillation technology transfers the knowledge of the teacher network to the lightweight student network to reduce the resource occupancy of edge devices; in addition, the design of the multi-dimensional reward function (current deviation, weld quality, energy consumption efficiency) and the curriculum learning strategy enable the model to quickly adapt to complex working conditions in the virtual welding environment and enhance the generalization ability.

[0107] A computer-readable storage medium stores a control program for implementing the above-mentioned intelligent collaborative control system for multiple welding devices. The control program includes a device collaborative scheduling instruction set for achieving nanosecond-level time synchronization accuracy, a real-time kernel patch module for optimizing interrupt response using the RT-Preempt technology, and a memory protection unit configuration strategy for isolating critical algorithm processes from the user space; the present invention ensures the microsecond-level response accuracy of multi-device collaborative scheduling through the nanosecond-level time synchronization instruction set and the RT-Preempt real-time kernel patch, and the memory protection unit isolates critical algorithm processes to prevent user space operations from interfering with the core control logic, providing a guarantee for high-precision real-time control.

[0108] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced within the present invention. Any reference signs in the claims should not be construed as limiting the claims involved.

Claims

1. An intelligent collaborative control system for multiple welding devices, characterized in that, Including: A security authentication module, deployed on the edge computing gateway, adopting a dynamic password encryption system based on the national cryptographic SM4 algorithm, and including a device identity certificate updated periodically; A parameter decision-making center, which integrates a local knowledge base and a cloud deep reinforcement learning model. The local knowledge base stores a mapping table of welding process parameters and material thickness characteristics, and the cloud deep reinforcement learning model generates a dynamic optimization strategy based on the real-time status data of the welding equipment group; A device discovery engine, adopting an improved ICMPv6 protocol scan and integrating a device MAC address white list filtering mechanism; A data channel management unit, establishing dual-link redundant communication. The main link uses MQTT over TLS1.3, and the backup link uses the LoRaWAN protocol; A human-machine interaction interface, implementing multi-dimensional parameter visualization, including three-dimensional dynamic simulation of the welding molten pool and a waveform diagram of voiceprint anomaly detection.

2. The intelligent collaborative control system for multi-welding equipment according to claim 1, characterized in that The security authentication module is provided with a device fingerprint generation unit, an elliptic curve digital signature unit, and a certificate revocation list synchronization unit; The device fingerprint generation unit generates a unique identifier based on the welder controller chip ID and the power module ripple characteristics in the welding equipment group; The elliptic curve digital signature unit uses the secp256r1 curve to implement policy instruction signature verification; The book revocation list synchronization unit synchronizes and updates the blacklisted devices with the CA center every hour.

3. The intelligent collaborative control system for multi-welding equipment according to claim 1, wherein The local knowledge base of the parameter decision-making center adopts an improved B+ tree index structure, and its leaf nodes store: material category encoding, thickness discretization interval, welding method feature vector, and confidence score of the optimal parameter combination.

4. The intelligent collaborative control system for multi-welding equipment according to claim 1, characterized in that, The cloud deep reinforcement learning model includes a policy network, a value network, and an experience replay buffer pool. The policy network is provided with a 5-layer convolutional neural network, and the input of the policy network is the device group state matrix; The value network is used to predict the long-term benefits of different strategies and is provided with a 3-layer LSTM network; The experience replay buffer pool adopts a priority sampling mechanism, and its storage capacity is at least 1TB; The update of the cloud deep reinforcement learning model adopts the double-delay DDPG algorithm, and online incremental learning is performed every 24 hours.

5. The intelligent collaborative control system for multi-welding equipment according to claim 1, wherein The working process of the device discovery engine includes: Step A: Send a customized ICMP probe packet, and the payload includes the target subnet mask and protocol version identifier; Step B: Analyze the device metadata in the response packet, and extract the first 3 bytes of the MAC address for manufacturer white list matching; Step C: Implement ARP spoofing attack defense on unauthorized devices and send a false gateway MAC address.

6. The intelligent collaborative control system for multi-welding equipment according to claim 1, characterized in that, The specific implementation of the dual-link redundant mechanism of the data channel management unit is as follows: The dual-link redundant mechanism of the data channel management unit adopts a main link bandwidth dynamic allocation algorithm, which adjusts the transmission rate in real time based on Welch power spectrum estimation; The backup link of the dual-link redundant mechanism of the data channel management unit adopts an adaptive frequency hopping technology, which divides 16 channels in the 2.4GHz frequency band; The link switching trigger condition of the dual-link redundancy mechanism of the data channel management unit is that three consecutive RTT values exceed 500 ms or the bit error rate > 10^-4.

7. The intelligent cooperative control system for the multi-welding device according to any one of claims 1-6, characterized in that, The human-computer interaction interface includes a molten pool dynamic simulation subsystem, a voiceprint feature extraction unit, and an anomaly detection and warning unit; The molten pool dynamic simulation subsystem calculates the heat conduction equation in real time based on finite element analysis; The voiceprint feature extraction unit analyzes the arc sound signal using Mel Frequency Cepstral Coefficients (MFCC); The anomaly detection and warning unit triggers a red stroboscopic warning when the penetration depth volatility exceeds ±15%; 8. A method for training a reinforcement learning model of an intelligent collaborative control system for multiple welding devices, based on the cloud deep reinforcement learning model of the intelligent collaborative control system for multiple welding devices according to any one of claims 1-7, characterized in that, It includes the following stage steps: Stage 1: Construct a virtual welding environment and inject several typical working condition disturbance modes; Stage 2: Design a multi-dimensional reward function: Stage 3: Adopt a curriculum learning strategy to gradually increase the complexity of the material thickness combination; Stage 4: Deploy a model distillation device to transfer the knowledge of the teacher network to the lightweight student network.

9. According to the reinforcement learning model training method of a multi-welding equipment intelligent collaborative control system described in claim 8, the construction of the virtual welding environment includes the following steps: Sub-step 1.1: Establish an arc physical equation: Sub-step 1.2: Simulate the turbulent effect of the shielding gas and calculate the flow field distribution using the k-epsilon model; Sub-step 1.3: Generate sensor data with noise, adding Gaussian white noise and pulse interference.

10. A computer-readable storage medium storing a control program for implementing the multi-welding equipment intelligent collaborative control system according to any one of claims 1-7, characterized in that, The control program includes a device collaborative scheduling instruction set that achieves nanosecond-level time synchronization accuracy, a real-time kernel patch module that optimizes interrupt response using RT-Preempt technology, and a memory protection unit configuration strategy for isolating critical algorithm processes from the user space.

Images