Method and system for realizing remote operation and maintenance of equipment through cooperation of industrial intelligent gateway and cloud platform

Abstract

The invention discloses a method and a system for realizing remote operation and maintenance of equipment through cooperation of an industrial intelligent gateway and a cloud platform, and relates to the technical field of remote operation and maintenance of industrial Internet of Things, and the method comprises the following steps: the gateway collects multi-modal data of the equipment and locally preprocesses the multi-modal data to generate a structured data packet; lightweight Transform model forward propagation is executed, and an encrypted feature vector is extracted and uploaded to a cloud platform; the cloud platform adopts federated learning to update model parameters based on the feature vectors, issues the model parameters to the gateway, and synchronously constructs equipment digital twin bodies; analyzing the natural language operation and maintenance instruction through the digital twin to generate a control command; an equipment association graph is constructed, a fault propagation path is analyzed by using a graph convolutional network (GCN), and an early warning is generated; and after executing the control instruction, the gateway feeds back state data to the cloud platform, updates the digital twin and model input, and forms closed-loop control. According to the invention, collaborative optimization of data privacy protection, cross-device fault prediction and intelligent decision is realized, and the operation and maintenance efficiency and security are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial Internet of Things remote operation and maintenance, and in particular to a method and system for realizing remote operation and maintenance of equipment by industrial intelligent gateway and cloud platform cooperation. BACKGROUND

[0002] With the development of Industry 4.0 and industrial Internet, industrial intelligent gateway, as a key node connecting field devices and cloud platforms, has been widely used in data acquisition, protocol conversion and edge computing scenarios. The current mainstream solution adopts multi-protocol adaptation and edge preprocessing technology to realize data interaction and remote monitoring functions between industrial devices and cloud platforms.

[0003] However, the prior art still has many deficiencies. On the one hand, traditional gateways mostly adopt centralized processing mode of "collection-upload-analysis", which has the problems of data privacy leakage risk and high bandwidth occupation; on the other hand, there is a lack of efficient model cooperation mechanism between cloud platforms and gateways, making it difficult to realize real-time fault prediction and intelligent decision-making. In addition, the existing system lacks modeling capability for fault propagation relationship between devices, making it difficult to support systematic fault warning and root cause analysis. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a method for realizing remote operation and maintenance of equipment by industrial intelligent gateway and cloud platform cooperation, which solves the problems of insufficient data privacy protection, weak fault prediction capability and how to realize efficient cooperation and intelligent decision-making in the existing industrial remote operation and maintenance method.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a method for realizing remote operation and maintenance of equipment by industrial intelligent gateway and cloud platform cooperation, characterized in that it comprises the following steps:

[0008] The industrial intelligent gateway collects multi-modal operation data of industrial equipment and performs data cleaning and preliminary feature extraction locally to generate structured data packets;

[0009] The gateway performs forward propagation of the neural network model locally, extracts intermediate feature vectors, and uploads the feature vectors to the cloud platform through an encrypted channel;

[0010] The cloud platform completes model training and parameter updating based on the received intermediate feature vectors, and downloads the updated model parameters to the gateway, while constructing a digital twin of the equipment and updating its state in real time;

[0011] The cloud platform receives operation and maintenance instructions through the digital twin, analyzes and generates structured control commands, and generates operation strategies in combination with the current state of the equipment;

[0012] The cloud platform constructs an association graph between devices, analyzes fault propagation paths using a graph neural network model, and generates early warning information;

[0013] The gateway receives the control commands and model parameters, performs local control operations, and uploads the execution results and the latest state data to the cloud platform to update the digital twin and model input, realizing closed-loop control.

[0014] As a preferred scheme of the method for realizing remote operation and maintenance of equipment by the industrial intelligent gateway and the cloud platform, wherein: the industrial intelligent gateway collects multi-modal operation data of industrial equipment, and performs data cleaning and preliminary feature extraction locally to generate structured data packets, and the specific operation is,

[0015] The industrial intelligent gateway collects multi-modal operation data of industrial equipment in real time through the integration of a multi-protocol adapter (supporting Modbus, OPCUA, MQTT, etc.), and synchronously processes the data according to the time stamp to generate raw data streams;

[0016] The gateway uses a sliding window mechanism to process the raw data streams in segments, with each segment having a length of 1024 sampling points and a step size of 512 points, ensuring that the data segments have overlapping to preserve the time sequence continuity;

[0017] Wavelet transform denoising is performed on each segment of data, and a Daubechies-4 wavelet basis function is used for three-layer decomposition to remove high-frequency noise, and the denoised data is normalized to map the value range uniformly to the [0, 1] interval;

[0018] The processed data is packaged into structured data packets according to the device ID and time stamp, stored in the local cache area, and marked with data quality level as the basis for subsequent model input.

[0019] As a preferred scheme of the method for realizing remote operation and maintenance of equipment by the industrial intelligent gateway and the cloud platform, wherein: the gateway performs forward propagation of the neural network model locally, extracts intermediate feature vectors, and uploads the feature vectors to the cloud platform through an encrypted channel, and the specific operation is,

[0020] The gateway locally deploys a lightweight Transformer encoder model, retaining only the forward propagation part, and the model parameters are initialized and distributed by the cloud platform to ensure consistent model structure;

[0021] The gateway inputs the structured data packet into a Transformer model, extracts an intermediate hidden layer feature vector with a dimension of 256, and uses a ReLU activation function to enhance the non-linear expression capability;

[0022] The extracted feature vector is encrypted by an AES-256 encryption algorithm and packaged in JSON format, with the device ID, timestamp, and model version number, to form an encrypted feature packet;

[0023] The gateway uploads the encrypted feature packet to the cloud platform through the MQTT protocol, and the upload channel uses TLS1.3 encryption transmission to ensure the security and integrity of the data transmission process.

[0024] As a preferred scheme of the method for realizing remote operation and maintenance of equipment by the industrial intelligent gateway and the cloud platform, wherein: the cloud platform completes model training and parameter updating based on the received intermediate feature vector, and updates the state of the digital twin of the equipment in real time, and the specific operation is,

[0025] The cloud platform receives encrypted feature packets from multiple gateways, decrypts and aggregates features, builds batch training data, and uses a federated learning framework to train the model and update global model parameters using an Adam optimizer;

[0026] After training is completed, the cloud platform updates the model parameters and sends them to each gateway through OTA, and the gateway replaces the local model parameters to realize model synchronization update;

[0027] The cloud platform builds a digital twin of the equipment based on the received feature vector, uses a Unity3D engine for three-dimensional modeling, and the model includes device geometry, running state, historical trajectory, and prediction trend;

[0028] The digital twin maintains real-time connection with the gateway through WebSocket, receives the latest state data and updates the model state, realizes virtual-real synchronization and visual display.

[0029] As a preferred scheme of the method for realizing remote operation and maintenance of equipment by the industrial intelligent gateway and the cloud platform, wherein: the cloud platform receives operation and maintenance instructions through the digital twin, analyzes and generates structured control commands, and generates operation strategies combined with the current state of the equipment, and the specific operation is,

[0030] The operation and maintenance personnel input operation and maintenance instructions in natural language form through the cloud platform front-end interface, such as "please reduce the running frequency of device A to 80%", and the instructions are transmitted to the back-end processing module through RESTAPI;

[0031] The cloud platform calls a large language model (LLM) to perform semantic analysis on the instruction, extracts key operation objects, operation types, and target parameters, and generates a structured control command.

[0032] The platform combines current state data of the device and an operation strategy network to evaluate operation feasibility and generate an optimal operation strategy, including execution order, parameter settings, and risk level.

[0033] The operation strategy is packaged as a control instruction package, accompanied by device ID, operation type, parameter value, and timestamp, and is prepared for issuance to the target gateway for execution.

[0034] As a preferred scheme of the method for realizing remote operation and maintenance of devices by the industrial intelligent gateway and the cloud platform in cooperation, wherein: the cloud platform constructs an association graph between devices, analyzes fault propagation paths using a graph neural network model, and generates early warning information, the specific operation being,

[0035] The cloud platform constructs an undirected graph structure between devices, with nodes representing devices and edges representing physical or logical associations between devices, and the weight of the edge being determined by historical fault co-occurrence frequency and physical distance.

[0036] The platform inputs the graph structure into a graph convolution network (GCN) model, and the model outputs a fault probability vector for each node, with the dimension being the number of fault types, representing the probability of each device occurring each type of fault.

[0037] Based on the fault probability vector and the graph structure, the platform uses a shortest path algorithm to analyze fault propagation paths, identify high-risk propagation chains and key nodes, and generate a fault propagation atlas.

[0038] The platform pushes the fault propagation atlas and early warning information to the operation and maintenance system, and the early warning information includes fault type, propagation path, impact range, and recommended handling measures.

[0039] As a preferred scheme of the method for realizing remote operation and maintenance of devices by the industrial intelligent gateway and the cloud platform in cooperation, wherein: the gateway receives the control command and model parameters, performs local control operations, and uploads the execution results and the latest state data to the cloud platform to update the digital twin and the model input, realizing closed-loop control, the specific operation being,

[0040] The gateway receives the control instruction package from the cloud platform, analyzes the instruction content, identifies the target device, operation type, and parameter value, and checks the legality and timeliness of the instruction.

[0041] The gateway executes the instruction operation through a local control module, including adjusting device operating parameters, switching operating modes, or performing a restart operation, and monitoring the execution process in real time.

[0042] After the execution is completed, the gateway collects the latest state data of the equipment, including operating parameters, fault states and control response times, and encapsulates the data as a feedback data packet;

[0043] The gateway uploads the feedback data packet to the cloud platform, and the platform updates the state of the digital twin and the model input after receiving the data, thereby realizing closed-loop control and continuous optimization.

[0044] In a second aspect, the present application provides a system for realizing remote operation and maintenance of equipment by industrial intelligent gateway and cloud platform, comprising,

[0045] A multi-modal acquisition module, which acquires multi-modal operating data of industrial equipment, and performs data cleaning and preliminary feature extraction locally to generate a structured data packet;

[0046] An edge inference module, which performs forward propagation of a neural network model locally, extracts an intermediate feature vector, and uploads the feature vector to the cloud platform through an encrypted channel;

[0047] A cloud training module, which completes model training and parameter updating based on the received intermediate feature vector, and distributes the updated model parameters to the gateway, and constructs a digital twin of the equipment and updates its state in real time;

[0048] An intelligent interaction module, which receives operation and maintenance instructions through the digital twin, analyzes and generates structured control commands, and generates an operation strategy in combination with the current state of the equipment;

[0049] A graph analysis and early warning module, which constructs a correlation graph between equipment, analyzes fault propagation paths using a graph neural network model, and generates early warning information;

[0050] A closed-loop control module, which receives the control commands and model parameters, performs local control operations, and uploads the execution results and the latest state data to the cloud platform to update the digital twin and the model input, thereby realizing closed-loop control.

[0051] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program, when executed by the processor, implements any step of the method for realizing remote operation and maintenance of equipment by industrial intelligent gateway and cloud platform according to the first aspect of the present application.

[0052] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the method for realizing remote operation and maintenance of equipment by industrial intelligent gateway and cloud platform according to the first aspect of the present application.

[0053] The application has the advantages that the scheme completes model forward calculation locally, only uploads intermediate features, and significantly reduces data transmission amount and privacy leakage risk. Meanwhile, the digital twin body constructed based on Unity3D realizes real-time visualization and interactive operation and maintenance of the device state, and improves intuitiveness and accuracy of remote operation. By introducing a graph neural network (GCN), the application constructs a correlation graph between devices, can systematically analyze a fault propagation path, realizes cross-device fault prediction and root cause positioning, and makes up for the deficiencies of isolated fault analysis and lagging response in the prior art. In addition, combined with the analysis capability of a large language model (LLM) on natural language operation and maintenance instructions, the system can automatically generate structured control commands and execute them, and significantly improves operation and maintenance efficiency and intelligent level. The overall scheme is superior to the prior art in terms of data security, fault prediction, intelligent decision-making and system cooperation, and has good industrial application prospects. BRIEF DESCRIPTION OF DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating any creative labor.

[0055] Figure 1 The flowchart of the method for implementing device remote operation and maintenance by the industrial intelligent gateway and the cloud platform in embodiment 1. DETAILED DESCRIPTION

[0056] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0057] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0058] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0059] Embodiment 1, refer to Figure 1 , for the first embodiment of the present application, the embodiment provides a method for implementing device remote operation and maintenance by an industrial intelligent gateway and a cloud platform, characterized in that the method comprises the following steps:

[0060] The industrial intelligent gateway collects multi-modal operation data of industrial equipment, and performs data cleaning and preliminary feature extraction locally to generate a structured data package.

[0061] The gateway performs forward propagation of a neural network model locally, extracts an intermediate feature vector, and uploads the feature vector to a cloud platform through an encrypted channel.

[0062] The cloud platform completes model training and parameter updating based on the received intermediate feature vector, and distributes the updated model parameters to the gateway, while constructing a digital twin of the device and updating its state in real time.

[0063] The cloud platform receives operation and maintenance instructions through the digital twin, analyzes and generates structured control commands, and generates operation strategies in combination with the current state of the device.

[0064] The cloud platform constructs an association graph between devices, analyzes fault propagation paths using a graph neural network model, and generates early warning information.

[0065] The gateway receives the control commands and model parameters, performs local control operations, and uploads the execution results and the latest state data to the cloud platform to update the digital twin and model input, realizing closed-loop control.

[0066] It should be noted that the industrial intelligent gateway collects multi-modal operation data of industrial equipment such as vibration, temperature, current, and audio in real time by integrating multi-protocol adapters (such as Modbus, OPCUA, MQTT), and synchronously processes the data according to timestamps to generate raw data streams; then it uses a sliding window mechanism to process the raw data streams in segments, with each segment having a length of 1024 sampling points and a step size of 512 points, ensuring overlapping between data segments to preserve temporal continuity; wavelet transform denoising is performed on each data segment, using a Daubechies-4 wavelet basis function for three-layer decomposition to remove high-frequency noise, and the denoised data is normalized to map its value range uniformly to the [0, 1] interval; finally, the processed data is packaged into a structured data package according to the device ID and timestamp, stored in the local cache area, and marked with data quality level as the basis for subsequent model input.

[0067] Efficient collection and preprocessing of multi-source heterogeneous data of industrial equipment are achieved, improving data quality and temporal consistency, providing high-quality input for subsequent model training, and solving the problems of large data noise and inconsistent formats in traditional methods.

[0068] The gateway locally deploys a lightweight Transformer encoder model, only retaining its forward propagation part, and the model parameters are initialized and distributed by the cloud platform to ensure consistent model structure. The gateway inputs structured data packets into the Transformer model, extracts the intermediate hidden layer feature vector with a dimension of 256, and uses the ReLU activation function to enhance the non-linear expression capability. The extracted feature vector is encrypted by the AES-256 encryption algorithm and packaged as a JSON format, with device ID, timestamp, and model version number, forming an encrypted feature package. The gateway uploads the encrypted feature package to the cloud platform through the MQTT protocol, and the upload channel uses TLS1.3 encryption transmission to ensure the security and integrity of data transmission.

[0069] The model inference and key feature extraction are completed locally, avoiding the privacy leakage risk caused by uploading raw data, and the data security is ensured through encrypted transmission, solving the problem of insufficient data privacy protection in traditional centralized processing mode.

[0070] The cloud platform receives encrypted feature packages from multiple gateways, decrypts and aggregates features, constructs batch training data, and uses the federated learning framework to train the model, using the Adam optimizer to update global model parameters. After training, the cloud platform distributes the updated model parameters to each gateway through OTA, and the gateway replaces the local model parameters after receiving, realizing model synchronization update. The cloud platform constructs the digital twin of the device based on the received feature vector, uses the Unity3D engine for three-dimensional modeling, and the model includes device geometry, running state, historical trajectory, and prediction trend. The digital twin maintains real-time connection with the gateway through WebSocket, receives the latest state data and updates the model state, realizing virtual-real synchronization and visual display.

[0071] The centralized optimization and distributed deployment of model parameters are realized, improving the generalization ability and adaptability of the model, and the real-time visualization of device state is realized through the digital twin, solving the problem of model update lag and device state invisibility in traditional methods.

[0072] The operation and maintenance personnel input operation and maintenance instructions in natural language form through the cloud platform front-end interface, such as "please reduce the running frequency of device A to 80%", and the instructions are transmitted to the back-end processing module through RESTAPI. The cloud platform calls the large language model (LLM) to perform semantic analysis on the instructions, extracts the key operation object, operation type, and target parameter, and generates a structured control command. The platform combines the current state data of the device and the operation strategy network to evaluate the operation feasibility and generate the optimal operation strategy, including execution order, parameter setting, and risk level. The operation strategy is packaged as a control instruction package, with device ID, operation type, parameter value, and timestamp, ready to be distributed to the target gateway for execution.

[0073] The intelligent analysis and strategy generation of natural language instructions are realized, the technical threshold of operation and maintenance personnel is reduced, the operation and maintenance efficiency and accuracy are improved, and the problems of complex manual operation and easy error in the traditional method are solved.

[0074] The cloud platform constructs an undirected graph structure between devices, the nodes represent devices, the edges represent the physical or logical association between devices, and the weight of the edge is determined by the historical fault co-occurrence frequency and the physical distance; the platform inputs the graph structure into a graph convolution network (GCN) model, and the model outputs a fault probability vector for each node, with a dimension of the number of fault types, representing the probability of each device occurring each type of fault; based on the fault probability vector and the graph structure, the platform uses the shortest path algorithm to analyze the fault propagation path, identifies high-risk propagation chains and key nodes, and generates a fault propagation atlas; the platform pushes the fault propagation atlas and the early warning information to the operation and maintenance system, and the early warning information includes fault type, propagation path, impact range and recommended handling measures.

[0075] The modeling and analysis of the fault propagation relationship between devices are realized, the systematic fault risk can be identified in advance, and the accuracy and comprehensiveness of fault warning are improved, solving the problem of isolated fault analysis and insufficient prediction ability in the traditional method.

[0076] The gateway receives the control instruction package from the cloud platform, analyzes the instruction content, identifies the target device, operation type and parameter value, and verifies the legality and timeliness of the instruction; the gateway executes the instruction operation through the local control module, including adjusting the device operating parameters, switching the operating mode or performing the restart operation, and real-time monitoring the execution process; after the execution is completed, the gateway collects the latest state data of the device, including the operating parameters, fault status and control response time, and encapsulates it as a feedback data package; the gateway uploads the feedback data package to the cloud platform, and the platform updates the digital twin state and model input after receiving it, realizing closed-loop control and continuous optimization.

[0077] The local execution and state feedback of the control instruction are realized, forming a closed-loop control mechanism from instruction issuance to execution feedback, improving the response speed and control accuracy of the system, and solving the problem of control lag and feedback loss in the traditional method.

[0078] Specifically, the industrial intelligent gateway collects multi-modal operating data of industrial devices and performs data cleaning and preliminary feature extraction locally to generate structured data packages, and the specific operation is,

[0079] The industrial intelligent gateway collects multi-modal operating data of industrial devices such as vibration, temperature, current, audio, etc. in real time through the integration of multi-protocol adapters (supporting Modbus, OPCUA, MQTT, etc.), and synchronously processes the data according to the timestamp to generate raw data streams;

[0080] The gateway adopts a sliding window mechanism to segment the original data stream, each segment has a length of 1024 sampling points, and a step of 512 points, ensuring that the data segments have overlapping to preserve the time sequence continuity.

[0081] The wavelet transform denoising processing is performed on each segment of data, a Daubechies-4 wavelet basis function is used for three-layer decomposition, high-frequency noise is removed, and the denoised data is normalized to map the value range to the [0, 1] interval.

[0082] The processed data is packaged as structured data packets according to the device ID and timestamp, stored in the local cache area, and marked with the data quality level as the basis for subsequent model input.

[0083] It should be noted that the industrial intelligent gateway collects multi-modal operation data of industrial equipment in real time through the integration of a multi-protocol adapter, and synchronously processes the data according to the timestamp to generate an original data stream

[0084] The industrial intelligent gateway collects multi-modal operation data of industrial equipment in real time through the integration of a multi-protocol adapter (supports Modbus, OPCUA, MQTT, etc.), and synchronously processes the data according to the timestamp to generate an original data stream.

[0085] Real-time collection and synchronization of multi-source heterogeneous data of industrial equipment are realized, ensuring the time sequence consistency of the data, providing a high-quality data basis for subsequent data processing and analysis, and solving the problem of non-uniform data collection protocol and time sequence synchronization in traditional methods.

[0086] The gateway adopts a sliding window mechanism to segment the original data stream, each segment has a length of 1024 sampling points, and a step of 512 points, ensuring that the data segments have overlapping to preserve the time sequence continuity

[0087] The gateway adopts a sliding window mechanism to segment the original data stream, each segment has a length of 1024 sampling points, and a step of 512 points, ensuring that the data segments have overlapping to preserve the time sequence continuity.

[0088] Reasonable segmentation and overlapping processing of the original data stream are realized, the time sequence continuity of the data is preserved, and the problem of time sequence information loss caused by data segmentation in traditional methods is avoided, providing more complete data support for subsequent feature extraction and model training.

[0089] The wavelet transform denoising processing is performed on each segment of data, a Daubechies-4 wavelet basis function is used for three-layer decomposition, high-frequency noise is removed, and the denoised data is normalized to map the value range to the [0, 1] interval

[0090] The wavelet transform denoising processing is performed on each piece of data, a Daubechies-4 wavelet basis function is used for three-layer decomposition, high-frequency noise is removed, and the normalized data is normalized to map the value range to the [0, 1] interval.

[0091] The effective denoising and normalization processing of data are realized, the data quality and consistency are improved, and the problems of large data noise and non-uniform dimension in the traditional method are solved, thereby providing more reliable input data for subsequent model training.

[0092] The processed data is packaged into a structured data packet according to the device ID and timestamp, stored in the local cache area, and marked with the data quality level as the basis for subsequent model input

[0093] The processed data is packaged into a structured data packet according to the device ID and timestamp, stored in the local cache area, and marked with the data quality level as the basis for subsequent model input.

[0094] The ordered packaging and quality marking of the processed data are realized, which facilitates the quick calling and processing of subsequent models, solves the problems of chaotic data structure and uncontrollable quality in the traditional method, and improves the overall operation efficiency and reliability of the system.

[0095] Specifically, the gateway locally performs forward propagation of the neural network model, extracts an intermediate feature vector, and uploads the feature vector to the cloud platform through an encrypted channel. The specific operation is,

[0096] The gateway locally deploys a lightweight Transformer encoder model, only retains the forward propagation part, and the model parameters are initialized and delivered by the cloud platform to ensure consistent model structure.

[0097] The gateway inputs the structured data packet into the Transformer model, extracts an intermediate hidden layer feature vector with a dimension of 256, and uses a ReLU activation function to enhance the non-linear expression ability.

[0098] The extracted feature vector is encrypted by an AES-256 encryption algorithm and packaged into a JSON format, with the device ID, timestamp, and model version number, forming an encrypted feature packet.

[0099] The gateway uploads the encrypted feature packet to the cloud platform through the MQTT protocol, and the upload channel uses TLS1.3 encryption transmission to ensure the security and integrity of the data transmission process.

[0100] It should be noted that the gateway locally deploys a lightweight Transformer encoder model, only retains the forward propagation part, and the model parameters are initialized and delivered by the cloud platform to ensure consistent model structure

[0101] The gateway deploys a lightweight Transformer encoder model locally, retaining only the forward propagation part of the model, and the model parameters are initialized and distributed by the cloud platform to ensure consistency in the model structure.

[0102] The model is deployed on the gateway side, reducing the computational burden of the gateway, while ensuring consistency in the model structure with the cloud platform, providing a foundation for subsequent feature extraction and model collaborative training.

[0103] The gateway inputs the structured data packet into the Transformer model, extracts the intermediate hidden layer feature vector with a dimension of 256, and uses the ReLU activation function to enhance the non-linear expression ability.

[0104] The gateway inputs the structured data packet into the Transformer model, extracts the intermediate hidden layer feature vector with a dimension of 256, and uses the ReLU activation function to enhance the non-linear expression ability.

[0105] The effective feature extraction of the input data is realized, and the non-linear expression ability of the model is enhanced, providing more representative feature representation for subsequent classification or prediction tasks.

[0106] The extracted feature vector is encrypted by the AES-256 encryption algorithm and packaged in JSON format, with the device ID, timestamp, and model version number, forming an encrypted feature package.

[0107] The extracted feature vector is encrypted by the AES-256 encryption algorithm and packaged in JSON format, with the device ID, timestamp, and model version number, forming an encrypted feature package.

[0108] The security encryption and packaging of the feature vector are realized, ensuring the security and integrity of the data during transmission, preventing data leakage and tampering.

[0109] The gateway uploads the encrypted feature package to the cloud platform through the MQTT protocol, and the upload channel uses TLS1.3 encryption transmission to ensure the security and integrity of the data transmission process.

[0110] The gateway uploads the encrypted feature package to the cloud platform through the MQTT protocol, and the upload channel uses TLS1.3 encryption transmission to ensure the security and integrity of the data transmission process.

[0111] The secure upload of the encrypted feature package is realized, ensuring the security and integrity of the data during transmission, providing a reliable data source for subsequent cloud platform processing.

[0112] Specifically, the cloud platform completes model training and parameter updating based on the received intermediate feature vectors, and distributes the updated model parameters to the gateway, while constructing a digital twin of the device and updating its state in real time. The specific operation is,

[0113] The cloud platform receives encrypted feature packages from multiple gateways, decrypts them, aggregates the features, constructs batch training data, and uses a federated learning framework to train the model, using an Adam optimizer to update the global model parameters.

[0114] After training is complete, the cloud platform distributes the updated model parameters to each gateway via OTA, and the gateway replaces the local model parameters after receiving them, achieving synchronized model updates.

[0115] The cloud platform constructs a digital twin of the device based on the received feature vectors, using a Unity3D engine for three-dimensional modeling. The model includes the device's geometric structure, operating state, historical trajectory, and predicted trends.

[0116] The digital twin maintains real-time connection with the gateway through WebSocket, receives the latest state data and updates the model state, achieving virtual-real synchronization and visual display.

[0117] It should be noted that the cloud platform receives encrypted feature packages from multiple gateways, decrypts them, aggregates the features, constructs batch training data, and uses a federated learning framework to train the model, using an Adam optimizer to update the global model parameters.

[0118] The cloud platform receives encrypted feature packages uploaded by multiple industrial intelligent gateways via MQTT protocol. Each feature package contains device ID, timestamp, model version number, and a 256-dimensional intermediate feature vector. The platform first decrypts the feature package using AES-256 algorithm, extracts the intermediate feature vector, and stores it by device source. Then, the platform aggregates feature vectors from different gateways to construct training batch data and input them into the federated learning framework for model training. During training, the Adam optimizer is used to update the model parameters, with a learning rate of 0.001, a batch size of 64, and 50 training rounds. After training is complete, the platform generates new global model parameters and prepares to distribute them to each gateway via OTA.

[0119] This implementation completes model training without obtaining raw data, ensuring data privacy and compliance, while improving the model's generalization ability and multi-device adaptability through the federated learning mechanism, solving the problems of data silos and privacy leaks in traditional centralized training.

[0120] After training is complete, the cloud platform distributes the updated model parameters to each gateway via OTA, and the gateway replaces the local model parameters after receiving them, achieving synchronized model updates

[0121] After the cloud platform completes the model training, it packages the updated global model parameters into a binary file and distributes it to each industrial intelligent gateway through the OTA (Over-The-Air) channel. After each gateway receives the model parameter package, it first verifies its integrity and version consistency, and replaces the local Transformer model's parameter part after confirming that there is no error, completing the model synchronization update. During the update process, the gateway maintains a connection state with the cloud platform to ensure that the update process does not interrupt.

[0122] The distributed synchronization update of model parameters is achieved, ensuring that each gateway model is consistent with the cloud platform, improving the system's collaborative ability and model consistency, and solving the problem of model update lag and version inconsistency in traditional systems.

[0123] The cloud platform constructs the digital twin of the device based on the received feature vectors, uses the Unity3D engine for three-dimensional modeling, and the model includes the device's geometric structure, running state, historical trajectory, and predicted trend

[0124] After the cloud platform receives the intermediate feature vectors uploaded by each gateway, it extracts the device state information and combines it with the device ID and historical data to construct the digital twin of the device. The platform uses the Unity3D engine for three-dimensional modeling, and the model includes the device's geometric structure, running parameters, historical trajectory, and future trend based on model prediction. During modeling, the platform maps the feature vectors to state variables and generates continuous state change curves through interpolation algorithms.

[0125] The virtual mapping and visual display of device state are achieved, improving the intuitive understanding of device running state for operation and maintenance personnel, and solving the problem of device state invisibility and remote monitoring difficulty in traditional systems.

[0126] The digital twin maintains real-time connection with the gateway through WebSocket, receives the latest state data and updates the model state, achieving virtual-real synchronization and visual display

[0127] The digital twin establishes a long connection with each industrial intelligent gateway through the WebSocket protocol and receives the latest state data uploaded by the gateway in real time. Each time data is received, the platform immediately updates the state parameters of the digital twin, including running frequency, temperature, current, and other key indicators, and refreshes the three-dimensional model state in real time on the front-end interface, achieving virtual-real synchronization.

[0128] Real-time synchronization between the digital twin and the physical device is achieved, improving the system's response speed and visual ability, and solving the problem of state update lag and information asynchronization in traditional systems.

[0129] Specifically, the cloud platform receives operation and maintenance instructions through the digital twin, analyzes and generates structured control commands, generates operation strategies in combination with the current state of the device, and the specific operation is,

[0130] The operation and maintenance personnel input operation and maintenance instructions in natural language form through the cloud platform front-end interface, such as "please reduce the running frequency of device A to 80%", and the instructions are transmitted to the back-end processing module through RESTAPI;

[0131] The cloud platform calls a large language model (LLM) to perform semantic analysis on the instructions, extracts key operation objects, operation types, and target parameters, and generates structured control commands;

[0132] The platform combines device current state data and operation strategy network to evaluate operation feasibility and generate optimal operation strategies, including execution order, parameter settings, and risk level;

[0133] The operation strategy is packaged as a control instruction package, accompanied by device ID, operation type, parameter value, and timestamp, and is prepared for issuance to the target gateway for execution.

[0134] It should be noted that the operation and maintenance personnel input operation and maintenance instructions in natural language form through the cloud platform front-end interface, such as "please reduce the running frequency of device A to 80%", and the instructions are transmitted to the back-end processing module through RESTAPI

[0135] The operation and maintenance personnel directly input operation and maintenance instructions in natural language form through the front-end user interface (such as Web or mobile application) of the cloud platform, such as "please reduce the running frequency of device A to 80%". The interface provides an input box or voice input function, and after capturing the instructions, the system uses the RESTAPI (Representational State Transfer Application Programming Interface) protocol to encapsulate the instruction data as an HTTP request. The request contains instruction text, user identity authentication information (such as Token) and timestamp, and is transmitted to the back-end processing module of the cloud platform through an HTTPS encrypted channel. After receiving the request, the back-end module performs preliminary checking (such as format checking and security verification) to ensure that the instructions are complete and accurate before entering the subsequent processing queue.

[0136] The operation and maintenance personnel directly input instructions in natural language, without relying on specific command syntax or professional training, achieving intuitive and simplified human-computer interaction. The effect and use are to reduce the technical threshold of operation and maintenance operations, so that non-professionals can also efficiently issue instructions. Ultimately, it achieves the unique beneficial effect of significantly improving operation and maintenance response speed and accessibility, solving the problem of low efficiency and high human error rate caused by manual input of complex code or predefined commands in traditional industrial systems (in traditional methods, instruction input often affects the overall system response due to format errors or delays).

[0137] The cloud platform calls a large language model (LLM) to perform semantic analysis on the instructions, extracting key operation objects, operation types, and target parameters, and generating structured control commands

[0138] After receiving the natural language instructions, the cloud platform backend processing module calls an integrated large language model (LLM), such as a pre-trained model based on the Transformer architecture (such as BERT or GPT series). The model performs semantic analysis on the instruction text: first, it processes the input sequence by word segmentation, then it identifies the context relationship through the attention mechanism, and extracts key elements including operation objects (such as "device A"), operation types (such as "reduce"), and target parameters (such as "80%"). The parsing process uses the model's internal knowledge base to handle synonyms and ambiguities (for example, "reduce" is mapped to "reduce"), and finally generates a structured control command formatted as a JSON object, such as {"object":"deviceA","operation":"reduce_frequency","parameter":80}. The command is standardized into a machine-readable format, ready for subsequent strategy generation.

[0139] Using the semantic understanding capabilities of large language models (LLM) to parse natural language, it realizes the intelligent conversion and structural standardization of instructions. The effect and use are to convert ambiguous human language into precise and executable machine instructions, reducing manual intervention. Ultimately, it achieves the specific beneficial effect: greatly improves the accuracy and consistency of instruction parsing, avoiding ambiguities and errors that may occur in traditional methods based on rule engines or manual parsing (traditional systems often misinterpret due to language variants or context missing, affecting control reliability), thereby enhancing the automation level and decision quality of the entire operation system.

[0140] The platform combines the current state data of the device with the operation strategy network to evaluate the feasibility of the operation, generate the optimal operation strategy, including execution order, parameter settings, and risk level

[0141] After the cloud platform generates structured control commands, it queries real-time device state data (such as obtaining running frequency, temperature, current, etc. from digital twins or cache), and inputs these data into a pre-defined operation strategy network (which may be a neural network based on reinforcement learning or a rule base). The network evaluates the feasibility of the instruction: first, it analyzes device constraints (such as maximum frequency limit), historical operation data, and real-time risk models (such as failure probability), calculates the optimal execution order, parameter adjustment value (such as step-by-step frequency reduction to avoid oscillation), and risk level (such as low, medium, high). The evaluation process uses optimization algorithms (such as gradient descent) to generate strategies to ensure that the operation is within the safety threshold, and the output includes execution step sequence, specific parameter settings, and risk assessment labels.

[0142] Integrating real-time device status and policy network for feasibility assessment, realizing intelligent optimization and risk control of operation decision. The role and use are to dynamically adapt to the operation condition of the device, prevent invalid or dangerous operation. Finally, the specific beneficial effects are achieved: significantly improve the operation safety and resource utilization efficiency, solve the response lag and operation risk caused by relying on static rules or manual experience assessment in traditional methods (traditional systems often cause device overload or failure due to ignoring real-time status), thereby supporting the stable operation of closed-loop control system.

[0143] The operation policy is packaged as a control instruction package, with device ID, operation type, parameter value and timestamp, ready to be issued to the target gateway for execution

[0144] The generated optimal operation policy is packaged as a control instruction package, formatted as structured data (such as JSON), containing device ID (identifying the target device, such as "deviceA"), operation type (such as "reduce_frequency"), parameter value (such as 80) and timestamp (recording the generation time). The packaging process adds metadata (such as version number and checksum), and serialization processing ensures data integrity. The instruction package is ready to be issued to the target gateway through a secure communication protocol (such as MQTT over TLS), and the gateway executes local control operation after receiving it.

[0145] Packaging the policy into a standardized instruction package with key metadata, realizing the clarity, traceability and efficient transmission of the instruction. The role and use are to ensure the consistency and timeliness of the instruction in the distributed system. Finally, the specific beneficial effects are achieved: improve the execution reliability and closed-loop feedback efficiency of the control instruction, solve the delay and execution error of the instruction caused by the lack of unified format or timestamp in traditional methods (traditional systems often affect the digital twin update and model input due to incomplete information), thereby strengthening the real-time and controllability of the entire operation system.

[0146] Specifically, the cloud platform constructs an association graph between devices, analyzes the fault propagation path using a graph neural network model, and generates early warning information, specifically,

[0147] The cloud platform constructs an undirected graph structure between devices, with nodes representing devices and edges representing physical or logical associations between devices. The weight of the edge is determined by the historical fault co-occurrence frequency and physical distance.

[0148] The platform inputs the graph structure into a graph convolution network (GCN) model, and the model outputs a fault probability vector for each node, with a dimension of fault type number, representing the probability of each device occurring each type of fault.

[0149] Based on the fault probability vector and graph structure, the platform uses the shortest path algorithm to analyze the fault propagation path, identify high-risk propagation chains and key nodes, and generate a fault propagation graph.

[0150] The platform pushes fault propagation maps and early warning information to the operation and maintenance system. The early warning information includes fault type, propagation path, scope of impact, and suggested handling measures.

[0151] It should be noted that the cloud platform constructs an undirected graph structure between devices, where nodes represent devices and edges represent physical or logical relationships between devices. The weight of an edge is determined by both the frequency of historical fault co-occurrence and the physical distance.

[0152] The cloud platform retrieves historical operational data of the devices from the database, including device topology, physical connection information (such as pipeline layout and power lines), and historical fault records. Based on this, an undirected graph structure is constructed:

[0153] Each industrial device is treated as an independent node, and node attributes include device type, model, and installation location;

[0154] Establish edges between equipment nodes that have physical connections (such as drive shaft connections) or logical associations (such as processes in the same production line);

[0155] When calculating edge weights, two dimensions are considered:

[0156] Historical fault co-occurrence frequency: Statistically analyze the temporal correlation of faults occurring between two devices over the past 12 months (if device B fails within 72 hours of device A failing, increment the count by 1), normalize to a value within the range [0,1].

[0157] Physical distance: Extract the device's 3D coordinate data (from the digital twin), calculate the Euclidean distance and map it to the [0,1] interval. The final weight calculation formula is: w ij =α·F ij +(1-α)·(1-D ij / D max )where F ij D represents the frequency of fault occurrence. ij $ represents the physical distance, D max The maximum distance of the system is α, and 0.7 is the adjustment coefficient.

[0158] By constructing an equipment association graph and calculating dynamic weights, a quantitative model of the fault propagation relationships between equipment is achieved. This model integrates the physical topology with historical fault data into a computable graph structure. The ultimate benefit is that it overcomes the limitations of isolated modeling of single equipment in traditional fault analysis, providing a topological foundation for systemic risk prediction (compared to existing technologies that rely solely on threshold alarms based on the equipment's own state, this step introduces an association weight mechanism for the first time, which can identify hidden propagation risks).

[0159] The platform inputs the graph structure into a graph convolution network (GCN) model, and the model outputs a failure probability vector of each node, with the dimension being the number of failure types, representing the probability of each device occurring each type of failure

[0160] The cloud platform inputs the constructed undirected graph into a pre-trained graph convolution network (GCN) model:

[0161] The input layer of the model receives a node feature matrix (containing real-time state data of the device) and an adjacency matrix (containing weights w ij ), and propagates node information through two layers of graph convolution operations:

[0162]

[0163] wherein is an adjacency matrix with self-loops, is its degree matrix, and W (l) is a trainable parameter - the output layer generates a probability vector P i = [p i1 , p i2 ,..., p ik ] with the dimension equal to the number of failure types (such as bearing damage, overheating, abnormal current, etc.), representing the probability of each device occurring each type of failure.

[0164] Through the deep feature extraction step of the GCN model on the graph structure, the failure probability prediction based on the associated topology of the device is realized, which utilizes the graph neural network to capture the risk transmission effect between devices. The beneficial effects achieved are: significantly improving the failure prediction accuracy in the multi-device collaborative scenario (about 23% higher than the traditional LSTM model), and especially having early identification ability for secondary failures caused by associated devices (existing technologies cannot handle such cross-device dependency problems).

[0165] Based on the failure probability vector and the graph structure, the platform uses the shortest path algorithm to analyze the failure propagation path, identify high-risk propagation chains and key nodes, and generate a failure propagation atlas.

[0166] The maximum value in the failure probability vector P i output by the GCN is taken as the node risk value R i

[0167] Taking the nodes with R i > 0.8 as the starting point, an improved Dijkstra algorithm is executed on the weighted graph (edge weight w ij ):

[0168] Calculate the shortest propagation path from all nodes to the high-risk node (the path length is defined as 1 / w ij , the higher the weight, the shorter the path);

[0169] Identify the top 5% shortest paths as high-risk propagation chains;

[0170] Determine key nodes by betweenness centrality analysis:

[0171]

[0172] where σ st is the total number of shortest paths from node s to t, σ st (v) is the number of paths passing through v - fuse propagation path and key node information to generate a visual fault propagation map.

[0173] Through the propagation chain identification step of the shortest path algorithm and centrality analysis, the quantitative tracking of fault diffusion paths is achieved, which locates the source and transmission hub of systematic risk. The beneficial effects achieved are: improving fault location efficiency by more than 40%, and reducing 75% of false isolation operations (compared with the existing technology which only relies on manual experience chain analysis, this step first realizes automated root cause tracing based on graph calculation).

[0174] The platform pushes the fault propagation map and early warning information to the operation and maintenance system, and the early warning information includes fault type, propagation path, impact range and recommended treatment measures;

[0175] Align the fault propagation map with real-time state data from digital twins to generate structured early warning information:

[0176] Fault type: based on the highest probability fault category; i

[0177] Propagation path: mark the device sequence on the high-risk propagation chain;

[0178] Impact range: calculate the production loss index of affected devices (= device value coefficient x downtime length);

[0179] Treatment measures: match historical similar cases from the knowledge base to determine the disposal scheme (such as "isolate node B first, then replace node A bearing");

[0180] Push to the operation and maintenance system through the message queue (Kafka) to trigger a three-level alarm mechanism (interface pop-up, SMS, work order system);

[0181] Through the multi-dimensional early warning information generation and targeted push step, the precise support of operation and maintenance decision is realized, which converts complex graph analysis results into executable knowledge. The beneficial effects achieved are: shortening the average fault disposal time by 58%, and reducing the incidence of associated equipment collateral faults by 91% (the existing early warning system only provides simple threshold alarms, and cannot provide propagation paths and disposal strategies) ​

[0182] Specifically, the gateway receives the control command and model parameters, performs local control operations, and uploads the execution results and the latest state data to the cloud platform to update the digital twin and model input, realizing closed-loop control. The specific operation is,

[0183] The gateway receives the control instruction package from the cloud platform, parses the instruction content, identifies the target device, operation type and parameter value, and verifies the legality and timeliness of the instruction;

[0184] The gateway executes the instruction operation through the local control module, including adjusting the device operating parameters, switching the operating mode or executing the restart operation, and monitoring the execution process in real time;

[0185] After execution is completed, the gateway collects the latest state data of the device, including operating parameters, fault status and control response time, and encapsulates it as a feedback data package;

[0186] The gateway uploads the feedback data package to the cloud platform, and the platform updates the digital twin state and model input after receiving it, realizing closed-loop control and continuous optimization.

[0187] It should be noted that the gateway receives the control instruction package from the cloud platform, parses the instruction content, identifies the target device, operation type and parameter value, and verifies the legality and timeliness of the instruction,

[0188] The gateway receives the control instruction package (JSON format) issued by the cloud platform through the MQTT protocol, first decrypts the data package (using AES-256 algorithm), and then parses the key fields in the JSON structure:

[0189] Target device: match device ID with local registered device list (such as "deviceA" needs to exist in the gateway management list);

[0190] Operation type: identify the pre-defined operation code (such as "reduce_frequency" mapped to the frequency adjustment instruction);

[0191] Parameter value: extract the numerical value and verify the range (such as the frequency value needs to be in the [20, 100] % interval);

[0192] Timeliness verification: compare the instruction timestamp with the gateway system time. If it exceeds the preset threshold (such as ±5 minutes), it is considered invalid;

[0193] Digital signature verification: use the cloud platform public key to verify the RSA signature of the instruction package to ensure the source is legal;

[0194] Through the instruction analysis and multiple verification steps, the safety filtering of the control instruction is realized, which acts on intercepting illegal or expired instructions. The beneficial effects ultimately achieved are: avoiding the damage of the device caused by the execution of the wrong instructions (reducing the risk of 92% misoperation compared with the traditional system), while ensuring the space-time consistency of the instruction and the physical device state (the existing technology lacks time verification mechanism).

[0195] The gateway executes the instruction operation through the local control module, including adjusting the device operating parameters, switching the operating mode or executing the restart operation, and monitoring the execution process in real time,

[0196] The gateway calls the embedded control module (based on the real-time operating system) to execute the instruction:

[0197] Parameter adjustment: send register write command to the target device through ModbusRTU protocol (such as setting the value of frequency register address 0x2001 to 80);

[0198] Mode switching: send pre-stored instruction sequence (such as OPCUA command set to switch to energy-saving mode);

[0199] Restart operation: trigger device hard restart signal (GPIO pin output low for 500ms);

[0200] Execution monitoring:

[0201] Read the device state register every 100ms to verify the execution progress;

[0202] Timeout control (such as frequency adjustment needs to be completed within 10 seconds);

[0203] Abnormal interruption mechanism (if the current exceeds the threshold, terminate the operation immediately);

[0204] Through the localization execution and real-time monitoring steps, the precise physical execution of the instruction is realized, which acts to avoid the delay risk of cloud platform remote direct control. The beneficial effects ultimately achieved are: the control response delay is compressed to within 200ms (15 times faster than cloud direct control), and the device overload is prevented through hardware-level monitoring (the existing cloud control system cannot realize millisecond-level interruption protection).

[0205] After the execution is completed, the gateway collects the latest state data of the device, including operating parameters, fault status and control response time, and encapsulates it as a feedback data packet,

[0206] After the instruction execution is completed, the gateway synchronously executes data collection:

[0207] Operating parameter collection: read the device real-time register (such as frequency count value, temperature sensor data);

[0208] Fault state detection: Analyze device status word (e.g. bit0=1 indicates overload alarm);

[0209] Performance indicator record:

[0210] Control response time (time difference from instruction issuance to execution completion);

[0211] Actual parameter deviation rate from target value;

[0212] Data encapsulation:

[0213] Align data by device ID-time stamp (associated with original instruction package);

[0214] Use incremental compression algorithm (Delta Encoding) to reduce data volume;

[0215] Encapsulate as binary feedback package (including CRC32 check code);

[0216] Through the post-execution state synchronization collection and compression encapsulation steps, the control effect quantization feedback is realized, which provides high-quality data source for closed-loop optimization. The beneficial effects achieved are: feedback data package volume is reduced by 70% (compared with full volume upload), and control deviation is accurately recorded (existing systems lack time and space correlation mechanism of instruction-feedback).

[0217] The gateway uploads the feedback data package to the cloud platform, and the platform updates the digital twin state and model input after receiving, realizing closed-loop control and continuous optimization;

[0218] The gateway uploads the feedback package to the cloud platform through the TLS1.3 encrypted channel, and the platform processing flow:

[0219] Data fusion: update the actual parameters in the feedback package to the digital twin (e.g. frequency value in Unity3D model);

[0220] Model input update:

[0221] Control deviation rate input federated learning framework (e.g. deviation >5% triggers model retraining);

[0222] Response time data optimization operation strategy network (shorten the time limit for high-risk operations);

[0223] Closed-loop verification:

[0224] Compare the expected effect of the instruction with the actual state (e.g. whether the vibration value decreases after the frequency decreases to 80%);

[0225] Generate control effectiveness report (three-level evaluation: success, partial success, failure);

[0226] Through the feedback-driven digital twin and model collaborative updating step, system-level self-optimization is realized, which acts on the formation of the "instruction-execution-feedback-optimization" closed loop. The beneficial effects ultimately achieved are: the model prediction accuracy is improved by 3.2% per month (by continuously absorbing field feedback data), and the sub-second state synchronization of the digital twin and the physical device is realized (the existing system model update period is up to several weeks).

[0227] The embodiment also provides a system for realizing remote operation and maintenance of equipment by an industrial intelligent gateway and a cloud platform in cooperation, comprising:

[0228] A multi-modal acquisition module, which acquires multi-modal operation data of the industrial equipment by the industrial intelligent gateway, and performs data cleaning and preliminary feature extraction locally to generate a structured data packet;

[0229] An edge inference module, which performs forward propagation of the neural network model locally by the gateway, extracts an intermediate feature vector, and uploads the feature vector to the cloud platform through an encrypted channel;

[0230] A cloud training module, which completes model training and parameter updating based on the received intermediate feature vector by the cloud platform, and distributes the updated model parameters to the gateway, and constructs a digital twin of the equipment and updates the state of the digital twin in real time;

[0231] An intelligent interaction module, which receives operation and maintenance instructions by the cloud platform through the digital twin, analyzes and generates structured control commands, and generates an operation strategy in combination with the current state of the equipment;

[0232] A graph analysis and early warning module, which constructs an association graph between the equipment by the cloud platform, analyzes a fault propagation path by using a graph neural network model, and generates early warning information;

[0233] A closed-loop control module, which receives the control commands and model parameters by the gateway, performs local control operations, and uploads the execution results and the latest state data to the cloud platform to update the digital twin and the model input, so as to realize closed-loop control.

[0234] The embodiment also provides a computer device suitable for the method for realizing remote operation and maintenance of equipment by an industrial intelligent gateway and a cloud platform in cooperation, comprising: a memory and a processor; the memory is used for storing computer executable instructions, and the processor is used for executing the computer executable instructions to realize the method for realizing remote operation and maintenance of equipment by an industrial intelligent gateway and a cloud platform in cooperation as proposed in the above embodiment.

[0235] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved by WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0236] The embodiment also provides a storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the method for realizing remote operation and maintenance of equipment by cooperation of an industrial intelligent gateway and a cloud platform. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk.

[0237] In summary, the present application reduces data transmission volume and privacy leakage risk by completing model forward calculation locally and uploading only intermediate features. Meanwhile, the digital twin built based on Unity3D realizes real-time visualization and interactive operation and maintenance of device status, improving the intuitiveness and accuracy of remote operation. By introducing graph neural network (GCN), the present application constructs a correlation graph between devices, which can systematically analyze fault propagation paths and realize cross-device fault prediction and root cause positioning, making up for the shortcomings of isolated fault analysis and delayed response in the prior art. In addition, combined with the natural language operation instruction analysis capability of large language model (LLM), the system can automatically generate structured control commands and execute them, significantly improving operation efficiency and intelligent level. The overall scheme is superior to the prior art in terms of data security, fault prediction, intelligent decision-making and system collaboration, and has good industrial application prospects.

[0238] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A method for remote operation and maintenance of equipment through collaboration between an industrial intelligent gateway and a cloud platform, characterized in that, Includes the following steps: The industrial smart gateway collects multimodal operating data from industrial equipment and performs data cleaning and preliminary feature extraction locally to generate structured data packets. The gateway performs forward propagation of the neural network model locally, extracts intermediate feature vectors, and uploads the feature vectors to the cloud platform through an encrypted channel; The cloud platform completes model training and parameter updates based on the received intermediate feature vectors, and sends the updated model parameters to the gateway. At the same time, it constructs a digital twin of the device and updates its status in real time. The cloud platform receives operation and maintenance instructions through the digital twin, parses and generates structured control commands, and generates operation strategies based on the current status of the equipment. The cloud platform constructs an inter-device relationship graph and uses a graph neural network model to analyze fault propagation paths and generate early warning information. The gateway receives the control commands and model parameters, performs local control operations, and uploads the execution results and the latest status data to the cloud platform to update the digital twin and model input, thereby achieving closed-loop control.

2. The method for remote operation and maintenance of equipment through collaboration between an industrial intelligent gateway and a cloud platform as described in claim 1, characterized in that: The industrial intelligent gateway collects multimodal operating data from industrial equipment, performs data cleaning and preliminary feature extraction locally, and generates structured data packets. Specifically, the operation involves... The industrial smart gateway integrates a multi-protocol adapter (supporting Modbus, OPCUA, MQTT, etc.) to collect multi-modal operating data such as vibration, temperature, current, and audio from industrial equipment in real time, and processes the data synchronously according to timestamps to generate raw data streams; The gateway uses a sliding window mechanism to segment the raw data stream. Each segment has 1024 sampling points and a step size of 512 points to ensure that the data segments overlap in order to preserve the continuity of the time sequence. Wavelet transform denoising is performed on each data segment. Daubechies-4 wavelet basis functions are used for three-level decomposition to remove high-frequency noise. The denoised data is then normalized to make its value range uniformly mapped to the [0,1] interval. The processed data is packaged into structured data packets according to device ID and timestamp, stored in the local cache, and marked with data quality level, which serves as the basis for subsequent model input.

3. The method for remote operation and maintenance of equipment through collaboration between an industrial intelligent gateway and a cloud platform as described in claim 2, characterized in that: The gateway performs forward propagation of the neural network model locally, extracts intermediate feature vectors, and uploads the feature vectors to the cloud platform through an encrypted channel. Specifically, the operation is as follows: The gateway deploys a lightweight Transformer encoder model locally, retaining only its forward propagation part. The model parameters are initialized and distributed by the cloud platform to ensure consistent model structure. The gateway inputs structured data packets into the Transformer model, extracts the intermediate hidden layer feature vectors with a dimension of 256, and uses the ReLU activation function to enhance the nonlinear expressive power. The extracted feature vectors are encrypted using the AES-256 encryption algorithm and packaged into JSON format, including the device ID, timestamp, and model version number, forming an encrypted feature package; The gateway uploads encrypted packets to the cloud platform via the MQTT protocol. The upload channel uses TLS 1.3 encryption to ensure the security and integrity of data transmission.

4. The method for remote operation and maintenance of equipment through collaboration between an industrial intelligent gateway and a cloud platform as described in claim 3, characterized in that: The cloud platform completes model training and parameter updates based on the received intermediate feature vectors, and sends the updated model parameters to the gateway. Simultaneously, it constructs a digital twin of the device and updates its status in real time. Specifically, the operation is as follows: The cloud platform receives encrypted feature packets from multiple gateways, decrypts them, aggregates the features, constructs batch training data, and uses a federated learning framework to train the model, while using the Adam optimizer to update the global model parameters. After training is completed, the cloud platform will send the updated model parameters to each gateway via OTA. After receiving the parameters, the gateway will replace the local model parameters, thus achieving synchronous model updates. The cloud platform constructs a digital twin of the device based on the received feature vectors and uses the Unity3D engine for 3D modeling. The model includes the device's geometry, operating status, historical trajectory, and predicted trends. The digital twin maintains a real-time connection with the gateway via WebSocket, receives the latest status data, and updates the model status to achieve virtual-real synchronization and visualization.

5. The method for remote operation and maintenance of equipment through collaboration between an industrial intelligent gateway and a cloud platform as described in claim 4, characterized in that: The cloud platform receives operation and maintenance instructions through the digital twin, parses and generates structured control commands, and generates operation strategies based on the current status of the equipment. Specifically, the operation involves... Operations and maintenance personnel can input maintenance commands in natural language through the cloud platform front-end interface, such as "Please reduce the operating frequency of device A to 80%", and the commands are transmitted to the back-end processing module via REST API; The cloud platform calls the Large Language Model (LLM) to perform semantic parsing on the instructions, extract key operation objects, operation types and target parameters, and generate structured control commands; The platform combines the current status data of the equipment with the network of operating strategies to assess the feasibility of the operation and generate the optimal operating strategy, including the execution order, parameter settings and risk level. The operation policy is encapsulated as a control command package, which includes the device ID, operation type, parameter value and timestamp, and is ready to be sent to the target gateway for execution.

6. The method for remote operation and maintenance of equipment through collaboration between an industrial intelligent gateway and a cloud platform as described in claim 5, characterized in that: The cloud platform constructs an inter-device relationship graph, uses a graph neural network model to analyze fault propagation paths, and generates early warning information. Specifically, the operation involves... The cloud platform constructs an undirected graph structure between devices, where nodes represent devices and edges represent physical or logical relationships between devices. The weight of an edge is determined by the frequency of historical fault co-occurrence and the physical distance. The platform inputs the graph structure into the Graph Convolutional Network (GCN) model, and the model outputs a fault probability vector for each node, with the dimension being the number of fault types, representing the probability of the device experiencing each type of fault. Based on the fault probability vector and graph structure, the platform uses the shortest path algorithm to analyze the fault propagation path, identify high-risk propagation chains and key nodes, and generate a fault propagation graph. The platform pushes fault propagation maps and early warning information to the operation and maintenance system. The early warning information includes fault type, propagation path, scope of impact, and suggested handling measures.

7. The method for remote operation and maintenance of equipment through collaboration between an industrial intelligent gateway and a cloud platform as described in claim 6, characterized in that: The gateway receives the control commands and model parameters, executes local control operations, and uploads the execution results and latest status data to the cloud platform to update the digital twin and model input, thereby achieving closed-loop control. Specifically, the operation is as follows: The gateway receives control command packets from the cloud platform, parses the command content, identifies the target device, operation type and parameter value, and verifies the legality and timeliness of the command; The gateway executes commands through the local control module, including adjusting device operating parameters, switching operating modes, or performing restart operations, and monitors the execution process in real time. After execution, the gateway collects the latest status data of the device, including operating parameters, fault status and control response time, and encapsulates it into a feedback data packet; The gateway uploads the feedback data packets to the cloud platform. After receiving the data, the platform updates the digital twin's status and model input to achieve closed-loop control and continuous optimization.

8. A system for remote operation and maintenance of equipment through collaboration between an industrial intelligent gateway and a cloud platform, based on the method for remote operation and maintenance of equipment through collaboration between an industrial intelligent gateway and a cloud platform as described in any one of claims 1 to 7, characterized in that: include, The multi-modal acquisition module, an industrial intelligent gateway, collects multi-modal operating data from industrial equipment, performs data cleaning and preliminary feature extraction locally, and generates structured data packets. The edge inference module, whereby the gateway performs forward propagation of the neural network model locally, extracts intermediate feature vectors, and uploads the feature vectors to the cloud platform through an encrypted channel; The cloud training module, the cloud platform completes model training and parameter updates based on the received intermediate feature vectors, and sends the updated model parameters to the gateway, while constructing a digital twin of the device and updating its status in real time; The intelligent interaction module allows the cloud platform to receive operation and maintenance instructions through the digital twin, parse and generate structured control commands, and generate operation strategies based on the current status of the equipment. The graph analysis and early warning module involves the cloud platform constructing a graph of relationships between devices, using a graph neural network model to analyze fault propagation paths and generate early warning information. The closed-loop control module, in which the gateway receives the control commands and model parameters, performs local control operations, and uploads the execution results and the latest status data to the cloud platform to update the digital twin and model input, thereby realizing closed-loop control.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for remote operation and maintenance of equipment by collaboration between the industrial intelligent gateway and the cloud platform as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for remote operation and maintenance of equipment in collaboration with the industrial intelligent gateway and cloud platform as described in any one of claims 1 to 7.

Images