A fully digital production management control system for chemical liquid plants
By integrating a factory cloud platform and a multi-unit chemical liquid factory control system, problems such as purity fluctuations and delayed leak detection in chemical liquid factories have been solved, realizing full-process digital management, improving production efficiency and safety, and reducing operating costs.
Patent Information
- Application Number
- CN202611409131.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-11
- Publication Date
- 2026-11-17
AI Technical Summary
Existing chemical liquid plant control systems suffer from problems such as fluctuations in chemical liquid purity, unstable impurity control, insufficient sensor sensitivity, delayed leak detection, frequent unplanned downtime, information silos, high energy consumption, low supply chain collaboration efficiency, and significant cybersecurity risks, leading to low production efficiency and safety hazards.
By employing a plant cloud platform, intelligent sensor network, digital twin unit, blockchain traceability unit, data training unit, chemical liquid leak alarm unit, predictive maintenance unit, and AR inspection unit, the system achieves dynamic scheduling of plant resources, real-time monitoring, unified data management, equipment fault prediction, rapid leak response, equipment health status monitoring, and closed-loop information management.
It has enabled full-process digital management of chemical liquid production, which has improved production efficiency, reduced the risk of unplanned downtime, ensured product quality and safety, reduced operating costs, and improved process optimization and supply chain collaboration efficiency.
Smart Images

Figure CN123113757A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent manufacturing and industrial safety technology, specifically relating to a fully digital production management control system for chemical liquid plants. Background Technology
[0002] In recent years, the semiconductor industry has developed rapidly, with increasingly complex processes and higher technological requirements. The production process requires the use of large quantities of electronic chemicals (hereinafter referred to as "chemicals"). These chemicals possess certain risks, including corrosiveness, toxicity, flammability, and spontaneous combustion. Even slight negligence can lead to significant personnel casualties and equipment damage. Therefore, ensuring the safe transfer of these chemicals between various chemical processing equipment while maintaining their quality is a crucial issue for production safety in semiconductor factories. This places higher demands on the continuous, trouble-free operation of each process within the factory and the quality of the process-related raw materials provided. Electronic chemicals, also known as process chemicals, refer to chemical reagents with a main component purity greater than 99.99% and impurity ions and particle counts meeting stringent standards. They are widely used in the cleaning, photolithography, development, etching, and doping processes of micro-processing electronic components such as chips, display panels, solar cells, and LEDs. In IC chip manufacturing, wet electronic chemicals can effectively remove residual contaminants from the wafer surface and reduce the content of metal impurities, ensuring the quality of downstream products. Chemicals mainly include acids, alkalis, organic solvents and other products with a main purity greater than 99.99% and impurity content less than ppm; functional chemicals refer to formulations or compound chemicals prepared by compounding processes to meet special process requirements in wet processes, mainly including various etching solutions, cleaning solutions and photoresist-related reagents (stripping solutions, diluents, edge removers, developers, etc.).
[0003] Currently, most domestic enterprises still rely primarily on traditional automation control (PLC / DCS), while some leading companies have begun piloting industrial internet platforms and MES systems. Most enterprises still depend on traditional industrial control systems to automate the control of basic parameters such as chemical liquid pressure and flow. However, chemical liquid plant equipment often has many proprietary protocols, which can easily lead to data silos. Furthermore, these systems primarily rely on passive defense (firewalls + manual inspection), posing security risks.
[0004] The domestic specialty gas supply system still faces numerous problems, as detailed below: 1) Fluctuations in the purity of chemical liquids and unstable impurity control lead to a decrease in wafer yield.
[0005] 2) Traditional sensors are not sensitive enough, leak location takes a lot of time, and leak detection is delayed.
[0006] 3) Unplanned downtime occupies part of production time, affecting product production and increasing maintenance costs. Excessive spare parts storage leads to high inventory costs.
[0007] 4) The lack of a unified data interface between the Factory Management System (FMCS), production equipment and MES / ERP leads to information fragmentation, inability to link and optimize, and the phenomenon of data silos.
[0008] 5) The chemical liquid transport system consumes too much energy, resulting in energy waste. Energy management is rudimentary and lacks dynamic optimization driven by digital twins.
[0009] 6) Over-reliance on personnel experience; process parameter adjustments depend on engineers' experience, and closed-loop optimization has not been formed.
[0010] 7) The disconnect between specialty gas inventory management, supplier data (purity, batch) and production plans can easily lead to supply interruptions or redundancy, and the efficiency of supply chain collaboration is too low.
[0011] 8) With the convergence of OT networks (operational technology) and IT networks (information technology), vulnerabilities in industrial protocols may be exploited, increasing cybersecurity risks and threatening production continuity. Summary of the Invention
[0012] To address the aforementioned problems in the existing technology, this invention provides a fully digital production management control system for chemical liquid plants.
[0013] The objective of this invention can be achieved through the following technical solutions: A fully digital production management control system for a chemical liquid plant includes: a plant cloud platform, an intelligent sensor network and data acquisition unit, a digital twin unit, a blockchain traceability unit, a data training unit, a chemical liquid leakage alarm unit, a predictive maintenance unit, a model deployment unit, and an AR inspection unit. The plant cloud platform is used to integrate all data, all plant equipment and business processes, and to dynamically schedule and coordinate plant resources across departments through multi-terminal access and remote control. The intelligent sensor network and data acquisition unit are connected to the plant cloud platform to monitor the pressure, flow rate, purity, temperature and humidity parameters of the chemical liquid in real time, and collect and transmit data through the data acquisition unit. The digital twin unit is used to construct a virtual image of the physical factory to achieve dynamic simulation and optimization; The blockchain traceability unit stores production process data and quality data on the blockchain, writes key data hash values into the permissioned chain, and provides equipment lifecycle traceability in the form of QR codes. The data training unit is deployed in the plant cloud platform for AI training. It continuously optimizes the leak identification, quality prediction and equipment fault diagnosis algorithms through active learning and small sample transfer strategies, thereby shortening the model iteration cycle. The chemical liquid leakage alarm unit is linked with the intelligent sensor network to perform graded early warning and emergency alarm based on the fusion results of laser spectrum, ultrasound, and visual multimodal data, and drives the ventilation, shut-off valve and sprinkler system to perform one-button emergency shutdown. The predictive maintenance unit extracts features from vibration and temperature time-series data based on the neural network and knowledge graph output by the data training unit, calculates the remaining service life (RUL) of the equipment in real time, and pushes maintenance suggestions to the AR inspection terminal. The model deployment unit is used to distribute the algorithm model verified by the data training unit to the field edge computing gateway through containerization and edge acceleration technology, so as to ensure that the model can maintain stable inference accuracy when the working conditions change. The AR inspection unit is used to assist on-site operators in automatically identifying equipment and retrieving maintenance history through AR devices, and in completing data recording and report generation using voice commands.
[0014] Specifically, the factory cloud platform adopts a microservice + cloud-native architecture, establishing an industrial dual-ring redundant network between the factory private cloud and the edge computing gateway; the platform has a built-in unified data bus, business flow engine, digital twin service bus and blockchain light node, and provides interfaces in a multi-tenant manner; it achieves cross-domain data aggregation and closed-loop remote equipment control through global clock synchronization and distributed message queues, and uniformly orchestrates and elastically expands the intelligent sensor network, digital twin unit, blockchain traceability unit, data training unit, model deployment unit and AR inspection unit.
[0015] Specifically, the intelligent sensor network and data acquisition unit consists of multimodal redundant nodes composed of laser spectral sensors, ultrasonic sensors, MEMS pressure / flow / temperature and humidity composite sensors, and industrial cameras. These nodes are densely deployed at the base of storage tanks, inflection points of pipelines, valve flanges, and key areas of process equipment interfaces according to a dual-node-cross-verification rule. Each node has a built-in edge AI chip, which is used to complete concentration identification, leak location, and abnormal data filtering locally, and then upload time-synchronized data to the plant cloud platform.
[0016] Specifically, the digital twin unit is based on a system-level multi-physics domain simulation platform, imports the three-dimensional scan point cloud data and equipment engineering data of the factory, and constructs a three-dimensional model of the chemical liquid pipeline system that includes geometric structure, physical properties and behavioral logic; An integrated computational fluid dynamics simulation module is used to simulate the velocity field, pressure distribution, and heat conduction effects of chemical liquids in complex pipe networks. Mark key monitoring points in the virtual model and establish a one-to-one mapping relationship with the actual deployment locations of the smart sensor network; Through the unified data bus of the plant cloud platform, multimodal monitoring data from the intelligent sensor network is received in real time; the sensor data and simulation model output data are fused, and the model state variables are continuously corrected to achieve synchronization between model parameters and physical entity state.
[0017] Specifically, the blockchain traceability unit deploys a permission-required blockchain light node within the factory cloud platform. This node performs a hash operation on the production batch number, process parameters, quality indicators, equipment number, timestamp, and operator digital certificate to obtain a fixed-length digest. This digest, along with the encrypted storage address of the original data, is packaged into a single production block, which is then divided into three sub-chains based on business nature: process, quality, and equipment. After a block is successfully written to the chain, the light node sends the block height and transaction number back to the platform. The platform then generates a QR code with a digital signature, which is affixed to the chemical drum or equipment nameplate. When a user scans a QR code, the system first verifies the validity of the QR code signature, then retrieves the original data hash through the on-chain read-only interface and compares it with the hash recalculated in the cloud. If the two values match, the system can display the full lifecycle report of the bucket or device.
[0018] Meanwhile, the system deploys two sets of smart contract templates: once the hash value of the quality indicator falls within the preset non-conforming range, the contract immediately issues a stop liquid supply instruction to the production control system and simultaneously pushes the audit event to the predictive maintenance module and the augmented reality inspection terminal, realizing automatic sealing and information closure of quality anomalies.
[0019] Specifically, the operation method of the data training unit includes: Through the unified data bus of the factory cloud platform, it continuously receives real-time monitoring data from the intelligent sensor network, simulation data from the digital twin unit, quality event data from the blockchain traceability unit, and equipment timing data from the predictive maintenance unit; it cleans, aligns, and standardizes the aggregated multi-source heterogeneous data to build a dynamic feature engineering pool containing time domain, frequency domain, and time-frequency domain features. Construct an initial, small-scale, high-quality labeled dataset; continuously input unlabeled data into the currently deployed leak identification, quality prediction, and fault diagnosis models for inference and prediction; calculate the uncertainty metric for each unlabeled sample, which includes the entropy of the model prediction probability and the model prediction divergence based on committee queries; the system automatically selects the samples with the highest uncertainty and pushes them to the AR inspection terminal to request manual confirmation and labeling; add the confirmed and labeled samples to the labeled dataset. A meta-learning framework is used to pre-train on source domain tasks that include equipment operating conditions and chemical liquid types. When new failure modes, new chemical production processes, or historical data for specific equipment are extremely scarce, the pre-trained model is used as the base model. Construct a multi-task neural network architecture that shares underlying features and has multiple dedicated output heads; the tasks include leak identification and classification, quality index regression, and equipment fault classification. The multi-task network is jointly trained using the constructed labeled dataset, and the generalization ability of each task in the case of limited data is improved by learning shared representations. The validated algorithm model, its weight file, and model metadata will be versioned and encapsulated. Through the service bus of the factory cloud platform, the packaged model version and related configuration parameters are pushed to the model deployment unit to trigger the subsequent containerization and edge deployment process.
[0020] Specifically, the chemical liquid leakage alarm unit is linked with the intelligent sensor network and receives gas concentration spectral data from the laser spectral sensor, acoustic wave spectrum data from the ultrasonic sensor, and video stream data from the industrial camera in parallel. Feature extraction is performed on laser spectral data to identify changes in the intensity and shape of absorption peaks of specific chemical molecules; Time-frequency analysis was performed on ultrasonic data to extract specific frequency band energy characteristics related to leakage cavitation effects; A lightweight computer vision model is applied to video stream data to detect the dynamic diffusion area of smoke-like aerosols and liquid droplets in real time. An information fusion model is constructed, which uses the extracted concentration features, acoustic features, and visual features as independent sources of evidence; a basic probability assignment is given to each source of evidence, and the probability assignment is obtained by training based on historical leakage data. When a single modality of evidence indicates a potential anomaly, but the confidence level after fusion does not exceed the high-risk threshold, the suspected area is highlighted in the digital twin model, and a notice is sent to the central control room. When the confidence level of the fusion of at least two modal evidences exceeds the medium-risk threshold, it is determined to be a confirmed leak; the system automatically activates the audible and visual alarms in the alarm area, pushes the leak location and diffusion simulation information to the AR inspection terminal, and pre-starts the ventilation system; When the confidence level of multimodal fusion exceeds the emergency threshold, or the instantaneous value of laser spectral concentration exceeds the safety limit, it is determined to be a major leak; the system immediately triggers a plant-wide emergency alarm and locks the precise location of the leak source in the digital twin model.
[0021] Specifically, the predictive maintenance unit includes a multi-source time-series data aggregation and edge preprocessing module, a dual-channel feature extraction module, and a RUL joint prediction model module; The multi-source time-series data aggregation and edge preprocessing module acquires in real time simulation parameters such as vibration acceleration, temperature, sound, current, pressure, flow rate, valve opening, bearing load, pump efficiency, and pipeline resistance from the intelligent sensor network via the unified data bus of the plant cloud platform; and completes data cleaning, resampling, time alignment, and outlier removal within the edge computing gateway, generating a device health status vector with a unified timestamp. The dual-channel feature extraction module adopts a dual-parallel processing mechanism of physical feature channel and deep feature channel to perform multi-scale feature mining on the device health status vector; The remaining useful life joint prediction model module adopts a physical feature-deep feature dual-branch network architecture. The physical feature branch inputs the extracted physical features into a long short-term memory network and outputs an initial remaining useful life prediction value. The deep feature branch inputs the extracted deep features into a gated recurrent unit network with an integrated attention mechanism and outputs a deviation correction amount for the initial prediction value. The final remaining useful life prediction value is the result of multiplying the initial prediction value output by the physical feature branch by the coefficient after adding the deviation correction amount.
[0022] Specifically, the physical feature channel extracts the frequency band energy ratio, kurtosis, and margin factor of the vibration signal, and combines them with temperature gradient and current harmonic distortion rate to construct interpretable physical features; the deep feature channel: inputs the multivariate time series within the sliding window into a lightweight temporal convolutional network, captures long-range dependencies through dilated causal convolution, and outputs a deep embedding vector; the two types of features are concatenated to form a hybrid feature vector.
[0023] Specifically, the operation method of the model deployment unit is as follows: After the data training unit completes the multi-task network training, it distributes the model, weights, and configuration together through the service bus of the factory cloud platform. The model deployment unit first performs hash verification on the received compressed package to confirm the complete transmission, and then reads the task type, precision threshold, and hardware affinity tag in the configuration file. It generates a unique version number in the model repository and locks the version to prevent secondary modifications. Containerization employs a multi-stage container build process, separating the compilation and runtime environments in stages. Before the initial distribution, the model deployment unit will perform pre-compiled kernel automatic matching: read the graphics processor computing capability version and system driver number of the edge computing gateway, and search in the kernel repository of the factory cloud to see if there is already a matching hardware fingerprint; if found, the converted engine file will be distributed directly, and the edge end will not need to compile again; if not found, the cloud will trigger real-time compilation and write the new kernel back to the repository for subsequent reuse. For the same model, both old and new versions are retained simultaneously, and the volume is gradually increased through virtual services in increments of 10%, 30%, and 100%. At the same time, online traffic is completely mirrored to the new version, and the latency, error, and memory metrics are compared in real time. If any one of them deteriorates by more than 2%, it is rolled back immediately to ensure that the accuracy remains unchanged when the operating conditions change. All images are signed during the compilation phase, and artifact verification is enabled on edge nodes. Before the inference container starts, the local policy gatekeeper verifies the signature and vulnerability whitelist, and refuses to start if it fails. The results of each canary release, including the version number, release ratio, and whether rollback is required, are written to the blockchain light node of the factory cloud platform to form an immutable model lifecycle record for use in quality auditing and incident tracing.
[0024] Specifically, the AR inspection unit operates as follows: After on-site workers put on augmented reality headbands, the headbands first compare the digital twin model with the visual positioning, automatically identify the equipment in front of them and retrieve the equipment's number, model, maintenance records and remaining life prediction value; If recording is required, personnel can say "start recording" and the system will convert the current voice into text and package it together with the image captured by the headband camera and the equipment number into a maintenance log. When the system detects that a person's gaze lingers on a valve or instrument for more than two seconds, the headband automatically overlays the real-time pressure, temperature, and flow data of that location within the field of vision and uses color to indicate whether the limit has been exceeded. If the predictive maintenance module sends an emergency work order, the headband will display a 3D arrow navigation to guide personnel to the fault point along the shortest route, while simultaneously projecting safe positioning markers on the ground to prevent accidental entry into dangerous areas.
[0025] Specifically, a dual-track link is established between the factory cloud platform and the edge computing gateway, consisting of an industrial security channel and a data flow channel. The industrial security channel is responsible for the encrypted transmission of remote control commands. Before a command is sent, it must undergo a two-way certificate handshake between the platform side and the gateway side. If the handshake fails, the command is automatically discarded to prevent replay attacks. The data stream channel is responsible for uploading large batches of monitoring data, and enables UDP multicast and breakpoint resume mechanism. When the link is interrupted, the data is temporarily stored in the gateway's solid-state storage and resumed according to the timestamp after recovery. The two channels share fiber at the physical layer but are completely isolated at the logical layer, with the control plane and data plane not competing for bandwidth.
[0026] The beneficial effects of this invention are as follows: Through the collaborative work of the aforementioned units, the factory control system achieves digital and intelligent management of the entire chemical liquid production process. The system not only monitors and controls the purity and impurity content of chemical liquids in real time, but also responds rapidly to abnormal situations, minimizing the impact on wafer yield. The combination of intelligent sensor networks and digital twin technology enables precise simulation and dynamic adjustment of key parameters in the production process, thereby optimizing the process flow. Simultaneously, the application of the blockchain traceability unit ensures the immutability and full traceability of production data, providing reliable assurance for quality control. The predictive maintenance unit continuously monitors the health status of equipment, proactively identifying potential faults and significantly reducing the risk of unplanned downtime. The AR inspection unit greatly improves the efficiency and safety of on-site operations, reducing the possibility of human error. Overall, the system demonstrates significant advantages in improving production efficiency, ensuring product quality, and reducing operating costs, providing strong technical support for the modernization of chemical liquid plants. Attached Figure Description
[0027] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.
[0028] Figure 1 This is a schematic diagram illustrating the implementation process of a fully digital production management control system for a chemical liquid plant according to the present invention. Figure 2 This is a structural diagram of the requirements for the factory cloud platform in this invention; Figure 3 This is a diagram of the blockchain continuous correction and automatic quality control technology architecture in this invention; Figure 4 This is a technical architecture diagram of the data training unit in this invention; Figure 5 This is a technical architecture diagram of the chemical liquid leakage alarm unit in this invention; Figure 6 This is a technical architecture diagram of the predictive maintenance unit in this invention; Figure 7 This is a technical architecture diagram of the model deployment unit in this invention. Detailed Implementation
[0029] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of example embodiments to those skilled in the art. Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring aspects of this disclosure. The blocks shown in the drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices. The flowcharts shown in the drawings are merely illustrative and do not necessarily include all contents and operations / steps, nor do they necessarily have to be performed in the order described. For example, some operations / steps can be broken down, while others can be combined or partially combined. Therefore, the actual execution order may change depending on the actual situation.
[0030] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.
[0031] Please see Figure 1-7 A comprehensive digital production management system for chemical liquid plants includes: a plant cloud platform, an intelligent sensor network and data acquisition unit, a digital twin unit, a blockchain traceability unit, a data training unit, a chemical liquid leakage alarm unit, a predictive maintenance unit, a model deployment unit, and an AR inspection unit. The plant cloud platform is used to integrate all data, all plant equipment and business processes, and to dynamically schedule and coordinate plant resources across departments through multi-terminal access and remote control. The intelligent sensor network and data acquisition unit are connected to the plant cloud platform to monitor the pressure, flow rate, purity, temperature and humidity parameters of the chemical liquid in real time, and collect and transmit data through the data acquisition unit. The digital twin unit is used to construct a virtual image of the physical factory to achieve dynamic simulation and optimization; The blockchain traceability unit stores production process data and quality data on the blockchain, writes key data hash values into the permissioned chain, and provides equipment lifecycle traceability in the form of QR codes. The data training unit is deployed in the plant cloud platform for AI training. It continuously optimizes the leak identification, quality prediction and equipment fault diagnosis algorithms through active learning and small sample transfer strategies, thereby shortening the model iteration cycle. The chemical liquid leakage alarm unit is linked with the intelligent sensor network to perform graded early warning and emergency alarm based on the fusion results of laser spectrum, ultrasound, and visual multimodal data, and drives the ventilation, shut-off valve and sprinkler system to perform one-button emergency shutdown. The predictive maintenance unit extracts features from vibration and temperature time-series data based on the neural network and knowledge graph output by the data training unit, calculates the remaining service life (RUL) of the equipment in real time, and pushes maintenance suggestions to the AR inspection terminal. The model deployment unit is used to distribute the algorithm model verified by the data training unit to the field edge computing gateway through containerization and edge acceleration technology, so as to ensure that the model can maintain stable inference accuracy when the working conditions change. The AR inspection unit is used to assist on-site operators in automatically identifying equipment and retrieving maintenance history through AR devices, and in completing data recording and report generation using voice commands.
[0032] Specifically, the factory cloud platform adopts a microservice + cloud-native architecture, establishing an industrial dual-ring redundant network between the factory private cloud and the edge computing gateway; the platform has a built-in unified data bus, business flow engine, digital twin service bus and blockchain light node, and provides interfaces in a multi-tenant manner; it achieves cross-domain data aggregation and closed-loop remote equipment control through global clock synchronization and distributed message queues, and uniformly orchestrates and elastically expands the intelligent sensor network, digital twin unit, blockchain traceability unit, data training unit, model deployment unit and AR inspection unit.
[0033] In this embodiment, the factory cloud platform is deployed using a cloud-native microservice architecture based on Kubernetes; The platform provides dedicated interfaces and data views for different departments and roles in a multi-tenant manner: Production department tenants: Their user interface focuses on process monitoring, production batch scheduling, and equipment operation efficiency dashboards. They do not have access to financial data.
[0034] Equipment Department Tenants: Their interface centrally displays the equipment health status, remaining service life warnings, and pending work orders pushed by the predictive maintenance unit.
[0035] Safety and Environmental Protection Department Tenant: Its interface monitors the status of chemical liquid leak alarm units in real time and has the highest authority to trigger a plant-wide emergency response with one click.
[0036] External audit tenants: Through the read-only interface, they can only access the quality reports and audit events that have been stored on the blockchain traceability unit, and cannot view any core process parameters.
[0037] Specifically, the intelligent sensor network and data acquisition unit consists of multimodal redundant nodes composed of laser spectral sensors, ultrasonic sensors, MEMS pressure / flow / temperature and humidity composite sensors, and industrial cameras. These nodes are densely deployed at the base of storage tanks, inflection points of pipelines, valve flanges, and key areas of process equipment interfaces according to a dual-node-cross-verification rule. Each node has a built-in edge AI chip, which is used to complete concentration identification, leak location, and abnormal data filtering locally, and then upload time-synchronized data to the plant cloud platform.
[0038] Specifically, the digital twin unit is based on a system-level multi-physics domain simulation platform, imports the three-dimensional scan point cloud data and equipment engineering data of the factory, and constructs a three-dimensional model of the chemical liquid pipeline system that includes geometric structure, physical properties and behavioral logic; An integrated computational fluid dynamics simulation module is used to simulate the velocity field, pressure distribution, and heat conduction effects of chemical liquids in complex pipe networks. Mark key monitoring points in the virtual model and establish a one-to-one mapping relationship with the actual deployment locations of the smart sensor network; Through the unified data bus of the plant cloud platform, multimodal monitoring data from the intelligent sensor network is received in real time; the sensor data and simulation model output data are fused, and the model state variables are continuously corrected to achieve synchronization between model parameters and physical entity state.
[0039] Specifically, the blockchain traceability unit deploys a permission-required blockchain light node within the factory cloud platform. This node performs a hash operation on the production batch number, process parameters, quality indicators, equipment number, timestamp, and operator digital certificate to obtain a fixed-length digest. This digest, along with the encrypted storage address of the original data, is packaged into a single production block, which is then divided into three sub-chains based on business nature: process, quality, and equipment. After a block is successfully written to the chain, the light node sends the block height and transaction number back to the platform. The platform then generates a QR code with a digital signature, which is affixed to the chemical drum or equipment nameplate. When a user scans a QR code, the system first verifies the validity of the QR code signature, then retrieves the original data hash through the on-chain read-only interface and compares it with the hash recalculated in the cloud. If the two values match, the system can display the full lifecycle report of the bucket or device.
[0040] Meanwhile, the system deploys two sets of smart contract templates: once the hash value of the quality indicator falls within the preset non-conforming range, the contract immediately issues a stop liquid supply instruction to the production control system and simultaneously pushes the audit event to the predictive maintenance module and the augmented reality inspection terminal, realizing automatic sealing and information closure of quality anomalies.
[0041] In this embodiment, a permissioned blockchain (Hyperledger Fabric 2.5) is deployed within the factory cloud platform, consisting of three organizations, three channels, and nine nodes: - Production organization (4 nodes): responsible for writing process parameters; - Quality organization (3 nodes): responsible for writing quality inspection results; - Equipment organization (2 nodes): Responsible for writing maintenance and calibration records; Raft consensus is adopted.
[0042] 1. Data upload process (taking a batch of PO products as an example): 1) Hash generation: Business fields: Production batch number PO-230724-B, process parameters (temperature 90.2 ℃, pressure 0.42 MPa), quality indicators (purity 99.52%, color 5, acid value 0.015 mg KOH / g), equipment number P-305, timestamp 2023-07-24 14:36:58, operator digital certificate UID-12045.
[0043] Calculation method: SM3 hash, output 256-bit digest → 0x3a7f...6e9c (64-character hexadecimal).
[0044] 2) Block Packaging: The original data is encrypted using AES-256 and stored in object storage, returning the ciphertext of the storage address; the transaction is composed of "digest + ciphertext address + business type", the chaincode WriteHash() is called, and it is written to the quality channel.
[0045] 3) Block height and transaction number: Block height: Quality-Channel / BlockNo-185739.
[0046] Transaction ID: tx0a8f...c12d (unique on the blockchain).
[0047] 2. QR code generation and affixing: The platform calls the QRCode service to generate a QR code (Version 10, 177×177 module) with a national cryptographic SM2 signature (block height + transaction number + digest + signature), measuring 20 mm × 20 mm. The code is laser-engraved on a stainless steel nameplate and riveted to the chemical drum lid.
[0048] 3. Client-side QR code verification: The user scans the QR code using WeChat / enterprise APP → obtains a Base64 encoded string → front-end verifies the signature (public key is stored in a CA certificate store in the cloud) → after successful verification, an HTTPS request is sent to the read-only interface on the chain → the node returns the ciphertext at the storage address → the front-end automatically decrypts the ciphertext → obtains the original JSON → the hash is calculated again locally using SM3 and compared with the digest in the QR code.
[0049] Once the comparison is successful, "On-chain verification successful" will be displayed, showing the full lifecycle report: raw material source, process curve, quality inspection report, equipment maintenance record, transportation temperature, and arrival time.
[0050] 4. Automatic lockdown via smart contracts: Contract Name: QualityGate.sol / QualityGate.wasm (Dual Templates)
[0051] Triggering condition: The chain detects a hash range with "purity < 99.45% or acid value > 0.02 mg KOH / g".
[0052] Perform the following actions: ① Send a "stop liquid supply" command to SCADA (Modbus-TCP write register 40001=1), valve action time <1.2 s; ② Generate an AuditEvent and push it to the predictive maintenance module and the augmented reality inspection terminal. The event fields include: batch number, non-conforming item, occurrence time, responsible person's digital certificate, and block height.
[0053] Specifically, the operation method of the data training unit includes: Through the unified data bus of the factory cloud platform, it continuously receives real-time monitoring data from the intelligent sensor network, simulation data from the digital twin unit, quality event data from the blockchain traceability unit, and equipment timing data from the predictive maintenance unit; it cleans, aligns, and standardizes the aggregated multi-source heterogeneous data to build a dynamic feature engineering pool containing time domain, frequency domain, and time-frequency domain features. Construct an initial, small-scale, high-quality labeled dataset; continuously input unlabeled data into the currently deployed leak identification, quality prediction, and fault diagnosis models for inference and prediction; calculate the uncertainty metric for each unlabeled sample, which includes the entropy of the model prediction probability and the model prediction divergence based on committee queries; the system automatically selects the samples with the highest uncertainty and pushes them to the AR inspection terminal to request manual confirmation and labeling; add the confirmed and labeled samples to the labeled dataset. A meta-learning framework is used to pre-train on source domain tasks that include equipment operating conditions and chemical liquid types. When new failure modes, new chemical production processes, or historical data for specific equipment are extremely scarce, the pre-trained model is used as the base model. Construct a multi-task neural network architecture that shares underlying features and has multiple dedicated output heads; the tasks include leak identification and classification, quality index regression, and equipment fault classification. The multi-task network is jointly trained using the constructed labeled dataset, and the generalization ability of each task in the case of limited data is improved by learning shared representations. The validated algorithm model, its weight file, and model metadata will be versioned and encapsulated. Through the service bus of the factory cloud platform, the packaged model version and related configuration parameters are pushed to the model deployment unit to trigger the subsequent containerization and edge deployment process.
[0054] Specifically, the chemical liquid leakage alarm unit is linked with the intelligent sensor network and receives gas concentration spectral data from the laser spectral sensor, acoustic wave spectrum data from the ultrasonic sensor, and video stream data from the industrial camera in parallel. Feature extraction is performed on laser spectral data to identify changes in the intensity and shape of absorption peaks of specific chemical molecules; Time-frequency analysis was performed on ultrasonic data to extract specific frequency band energy characteristics related to leakage cavitation effects; A lightweight computer vision model is applied to video stream data to detect the dynamic diffusion area of smoke-like aerosols and liquid droplets in real time. An information fusion model is constructed, which uses the extracted concentration features, acoustic features, and visual features as independent sources of evidence; a basic probability assignment is given to each source of evidence, and the probability assignment is obtained by training based on historical leakage data. When a single modality of evidence indicates a potential anomaly, but the confidence level after fusion does not exceed the high-risk threshold, the suspected area is highlighted in the digital twin model, and a notice is sent to the central control room. When the confidence level of the fusion of at least two modal evidences exceeds the medium-risk threshold, it is determined to be a confirmed leak; the system automatically activates the audible and visual alarms in the alarm area, pushes the leak location and diffusion simulation information to the AR inspection terminal, and pre-starts the ventilation system; When the confidence level of multimodal fusion exceeds the emergency threshold, or the instantaneous value of laser spectral concentration exceeds the safety limit, it is determined to be a major leak; the system immediately triggers a plant-wide emergency alarm and locks the precise location of the leak source in the digital twin model.
[0055] Specifically, the predictive maintenance unit includes a multi-source time-series data aggregation and edge preprocessing module, a dual-channel feature extraction module, and a RUL joint prediction model module; The multi-source time-series data aggregation and edge preprocessing module acquires in real time simulation parameters such as vibration acceleration, temperature, sound, current, pressure, flow rate, valve opening, bearing load, pump efficiency, and pipeline resistance from the intelligent sensor network via the unified data bus of the plant cloud platform; and completes data cleaning, resampling, time alignment, and outlier removal within the edge computing gateway, generating a device health status vector with a unified timestamp. The dual-channel feature extraction module adopts a dual-parallel processing mechanism of physical feature channel and deep feature channel to perform multi-scale feature mining on the device health status vector; The remaining useful life joint prediction model module adopts a physical feature-deep feature dual-branch network architecture. The physical feature branch inputs the extracted physical features into a long short-term memory network and outputs an initial remaining useful life prediction value. The deep feature branch inputs the extracted deep features into a gated recurrent unit network with an integrated attention mechanism and outputs a deviation correction amount for the initial prediction value. The final remaining useful life prediction value is the result of multiplying the initial prediction value output by the physical feature branch with the coefficient after adding the deviation correction amount.
[0056] Specifically, the physical feature channel extracts the frequency band energy ratio, kurtosis, and margin factor of the vibration signal, and combines them with temperature gradient and current harmonic distortion rate to construct interpretable physical features; the deep feature channel: inputs the multivariate time series within the sliding window into a lightweight temporal convolutional network, captures long-range dependencies through dilated causal convolution, and outputs a deep embedding vector; the two types of features are concatenated to form a hybrid feature vector.
[0057] Specifically, the operation method of the model deployment unit is as follows: After the data training unit completes the multi-task network training, it distributes the model, weights, and configuration together through the service bus of the factory cloud platform. The model deployment unit first performs hash verification on the received compressed package to confirm the complete transmission, and then reads the task type, precision threshold, and hardware affinity tag in the configuration file. It generates a unique version number in the model repository and locks the version to prevent secondary modifications. Containerization employs a multi-stage container build process, separating the compilation and runtime environments in stages. Before the initial distribution, the model deployment unit will perform pre-compiled kernel automatic matching: read the graphics processor computing capability version and system driver number of the edge computing gateway, and search in the kernel repository of the factory cloud to see if there is already a matching hardware fingerprint; if found, the converted engine file will be distributed directly, and the edge end will not need to compile again; if not found, the cloud will trigger real-time compilation and write the new kernel back to the repository for subsequent reuse. For the same model, both old and new versions are retained simultaneously, and the volume is gradually increased through virtual services in increments of 10%, 30%, and 100%. At the same time, online traffic is completely mirrored to the new version, and the latency, error, and memory metrics are compared in real time. If any one of them deteriorates by more than 2%, it is rolled back immediately to ensure that the accuracy remains unchanged when the operating conditions change. All images are signed during the compilation phase, and artifact verification is enabled on edge nodes. Before the inference container starts, the local policy gatekeeper verifies the signature and vulnerability whitelist, and refuses to start if it fails. The results of each canary release, including the version number, release ratio, and whether rollback is required, are written to the blockchain light node of the factory cloud platform to form an immutable model lifecycle record for use in quality auditing and incident tracing.
[0058] Specifically, the AR inspection unit operates as follows: After on-site workers put on augmented reality headbands, the headbands first compare the digital twin model with the visual positioning, automatically identify the equipment in front of them and retrieve the equipment's number, model, maintenance records and remaining life prediction value; If recording is required, personnel can say "start recording" and the system will convert the current voice into text and package it together with the image captured by the headband camera and the equipment number into a maintenance log. When the system detects that a person's gaze lingers on a valve or instrument for more than two seconds, the headband automatically overlays the real-time pressure, temperature, and flow data of that location within the field of vision and uses color to indicate whether the limit has been exceeded. If the predictive maintenance module sends an emergency work order, the headband will display a 3D arrow navigation to guide personnel to the fault point along the shortest route, while simultaneously projecting safe positioning markers on the ground to prevent accidental entry into dangerous areas.
[0059] Specifically, a dual-track link is established between the factory cloud platform and the edge computing gateway, consisting of an industrial security channel and a data flow channel. The industrial security channel is responsible for the encrypted transmission of remote control commands. Before a command is sent, it must undergo a two-way certificate handshake between the platform side and the gateway side. If the handshake fails, the command is automatically discarded to prevent replay attacks. The data stream channel is responsible for uploading large batches of monitoring data, and enables UDP multicast and breakpoint resume mechanism. When the link is interrupted, the data is temporarily stored in the gateway's solid-state storage and resumed according to the timestamp after recovery. The two channels share fiber at the physical layer but are completely isolated at the logical layer, with the control plane and data plane not competing for bandwidth.
[0060] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.
Claims
1. A fully digital production management control system for a chemical liquid plant, characterized in that, include: The system includes a factory cloud platform, an intelligent sensor network and data acquisition unit, a digital twin unit, a blockchain traceability unit, a data training unit, a chemical liquid leak alarm unit, a predictive maintenance unit, a model deployment unit, and an AR inspection unit. The plant cloud platform is used to integrate all data, all plant equipment and business processes, and to dynamically schedule and coordinate plant resources across departments through multi-terminal access and remote control. The intelligent sensor network and data acquisition unit are connected to the plant cloud platform to monitor the pressure, flow rate, purity, temperature and humidity parameters of the chemical liquid in real time, and collect and transmit data through the data acquisition unit. The digital twin unit is used to construct a virtual image of the physical factory to achieve dynamic simulation and optimization; The blockchain traceability unit stores production process data and quality data on the blockchain, writes key data hash values into the permissioned chain, and provides equipment lifecycle traceability in the form of QR codes. The data training unit is deployed in the plant cloud platform for AI training. It continuously optimizes the leak identification, quality prediction and equipment fault diagnosis algorithms through active learning and small sample transfer strategies, thereby shortening the model iteration cycle. The chemical liquid leakage alarm unit is linked with the intelligent sensor network to perform graded early warning and emergency alarm based on the fusion results of laser spectrum, ultrasound, and visual multimodal data, and drives the ventilation, shut-off valve and sprinkler system to perform one-button emergency shutdown. The predictive maintenance unit extracts features from vibration and temperature time-series data based on the neural network and knowledge graph output by the data training unit, calculates the remaining service life (RUL) of the equipment in real time, and pushes maintenance suggestions to the AR inspection terminal. The model deployment unit is used to distribute the algorithm model verified by the data training unit to the field edge computing gateway through containerization and edge acceleration technology, so as to ensure that the model can maintain stable inference accuracy when the working conditions change. The AR inspection unit is used to assist on-site operators in automatically identifying equipment and retrieving maintenance history through AR devices, and in completing data recording and report generation using voice commands.
2. The method according to claim 1, characterized in that, The factory cloud platform adopts a microservice + cloud-native architecture, and establishes an industrial dual-ring redundant network between the factory private cloud and the edge computing gateway. The platform has a built-in unified data bus, business flow engine, digital twin service bus and blockchain light node, and provides interfaces in a multi-tenant manner. By using global clock synchronization and distributed message queues to achieve cross-domain data aggregation and closed-loop remote device control, the system can uniformly orchestrate and flexibly expand intelligent sensor networks, digital twin units, blockchain traceability units, data training units, model deployment units, and AR inspection units.
3. The method according to claim 1, characterized in that, The intelligent sensor network and data acquisition unit consists of multimodal redundant nodes composed of laser spectral sensors, ultrasonic sensors, MEMS pressure / flow / temperature and humidity composite sensors, and industrial cameras. These nodes are densely deployed at the base of storage tanks, bends in pipelines, valve flanges, and key areas of process equipment interfaces according to a dual-node-cross-verification rule. Each node has a built-in edge AI chip, which is used to complete concentration identification, leak location, and abnormal data filtering locally, and then upload time-synchronized data to the plant cloud platform.
4. The method according to claim 1, characterized in that, The digital twin unit is based on a system-level multi-physics domain simulation platform. It imports the 3D scan point cloud data and equipment engineering data of the factory to construct a 3D model of the chemical liquid pipeline system, which includes geometric structure, physical properties and behavioral logic. An integrated computational fluid dynamics simulation module is used to simulate the velocity field, pressure distribution, and heat conduction effects of chemical liquids in complex pipe networks. Mark key monitoring points in the virtual model and establish a one-to-one mapping relationship with the actual deployment locations of the smart sensor network; Through the unified data bus of the plant cloud platform, multimodal monitoring data from the intelligent sensor network is received in real time; the sensor data and simulation model output data are fused, and the model state variables are continuously corrected to achieve synchronization between model parameters and physical entity state.
5. The method according to claim 1, characterized in that, The blockchain traceability unit deploys a permission-required blockchain light node within the factory cloud platform. It performs a hash operation on the production batch number, process parameters, quality indicators, equipment number, timestamp, and operator digital certificate to obtain a fixed-length digest. This digest, along with the encrypted storage address of the original data, is packaged into a single block as the production block, and is further divided into three sub-chains based on business nature: process, quality, and equipment. After the block is successfully written to the chain, the light node sends the block height and transaction number back to the platform, which then generates a QR code with a digital signature, which is affixed to the chemical drum or equipment nameplate. When a user scans a QR code, the system first verifies the validity of the QR code signature, then retrieves the original data hash through the on-chain read-only interface and compares it with the hash recalculated in the cloud. If the two values match, the system can display the full lifecycle report of the bucket or device. Meanwhile, the system deploys two sets of smart contract templates: once the hash value of the quality indicator falls within the preset non-conforming range, the contract immediately issues a stop liquid supply instruction to the production control system and simultaneously pushes the audit event to the predictive maintenance module and the augmented reality inspection terminal, realizing automatic sealing and information closure of quality anomalies.
6. The method according to claim 1, characterized in that, The operation method of the data training unit includes: Through the unified data bus of the factory cloud platform, it continuously receives real-time monitoring data from the intelligent sensor network, simulation data from the digital twin unit, quality event data from the blockchain traceability unit, and equipment timing data from the predictive maintenance unit; it cleans, aligns, and standardizes the aggregated multi-source heterogeneous data to build a dynamic feature engineering pool containing time domain, frequency domain, and time-frequency domain features. Construct an initial, small-scale, high-quality labeled dataset; continuously input unlabeled data into the currently deployed leak identification, quality prediction, and fault diagnosis models for inference and prediction; calculate the uncertainty metric for each unlabeled sample, which includes the entropy of the model prediction probability and the model prediction divergence based on committee queries; the system automatically selects the samples with the highest uncertainty and pushes them to the AR inspection terminal to request manual confirmation and labeling; add the confirmed and labeled samples to the labeled dataset. A meta-learning framework is used to pre-train on source domain tasks that include equipment operating conditions and chemical liquid types. When new failure modes, new chemical production processes, or historical data for specific equipment are extremely scarce, the pre-trained model is used as the base model. Construct a multi-task neural network architecture that shares underlying features and has multiple dedicated output heads; the tasks include leak identification and classification, quality index regression, and equipment fault classification. The multi-task network is jointly trained using the constructed labeled dataset, and the generalization ability of each task in the case of limited data is improved by learning shared representations. The validated algorithm model, its weight file, and model metadata will be versioned and encapsulated. Through the service bus of the factory cloud platform, the packaged model version and related configuration parameters are pushed to the model deployment unit to trigger the subsequent containerization and edge deployment process.
7. The method according to claim 1, characterized in that, The chemical liquid leakage alarm unit is linked with the intelligent sensor network and receives gas concentration spectral data from the laser spectral sensor, acoustic wave spectrum data from the ultrasonic sensor, and video stream data from the industrial camera in parallel. Feature extraction is performed on laser spectral data to identify changes in the intensity and shape of absorption peaks of specific chemical molecules; Time-frequency analysis was performed on ultrasonic data to extract specific frequency band energy characteristics related to leakage cavitation effects; A lightweight computer vision model is applied to video stream data to detect the dynamic diffusion area of smoke-like aerosols and liquid droplets in real time. An information fusion model is constructed, which uses the extracted concentration features, acoustic features, and visual features as independent sources of evidence; a basic probability assignment is given to each source of evidence, and the probability assignment is obtained by training based on historical leakage data. When a single modality of evidence indicates a potential anomaly, but the confidence level after fusion does not exceed the high-risk threshold, the suspected area is highlighted in the digital twin model, and a notice is sent to the central control room. When the confidence level of the fusion of at least two modal evidences exceeds the medium-risk threshold, it is determined to be a confirmed leak; the system automatically activates the audible and visual alarms in the alarm area, pushes the leak location and diffusion simulation information to the AR inspection terminal, and pre-starts the ventilation system; When the confidence level of multimodal fusion exceeds the emergency threshold, or the instantaneous value of laser spectral concentration exceeds the safety limit, it is determined to be a major leak; the system immediately triggers a plant-wide emergency alarm and locks the precise location of the leak source in the digital twin model.
8. The method according to claim 1, characterized in that, The predictive maintenance unit includes a multi-source time-series data aggregation and edge preprocessing module, a dual-channel feature extraction module, and a RUL joint prediction model module; The multi-source time-series data aggregation and edge preprocessing module acquires in real time simulation parameters such as vibration acceleration, temperature, sound, current, pressure, flow rate, valve opening, bearing load, pump efficiency, and pipeline resistance from the intelligent sensor network via the unified data bus of the plant cloud platform. Within the edge computing gateway, data cleaning, resampling, time alignment, and outlier removal are performed, and a device health status vector with a unified timestamp is generated. The dual-channel feature extraction module adopts a dual-parallel processing mechanism of physical feature channel and deep feature channel to perform multi-scale feature mining on the device health status vector; The remaining useful life joint prediction model module adopts a physical feature-deep feature dual-branch network architecture, in which the physical feature branch inputs the extracted physical features into the long short-term memory network and outputs the initial remaining useful life prediction value. The deep feature branch inputs the extracted deep features into a gated recurrent unit network with an integrated attention mechanism and outputs a deviation correction for the initial prediction value. The final remaining device life prediction value is the result of multiplying the initial prediction value output by the physical feature branch with the coefficient after adding the deviation correction.
9. The method according to claim 8, characterized in that, The physical feature channel extracts the frequency band energy ratio, kurtosis, and margin factor of the vibration signal, and combines them with temperature gradient and current harmonic distortion rate to construct interpretable physical features; the deep feature channel: inputs the multivariate temporal sequence within the sliding window into a lightweight temporal convolutional network, captures long-range dependencies through dilated causal convolution, and outputs a deep embedding vector. The two types of features are concatenated to form a hybrid feature vector.
10. The method according to claim 1, characterized in that, The operation method of the model deployment unit is as follows: After the data training unit completes the multi-task network training, it distributes the model, weights, and configuration together through the service bus of the factory cloud platform. The model deployment unit first performs hash verification on the received compressed package to confirm the complete transmission, and then reads the task type, precision threshold, and hardware affinity tag in the configuration file. It generates a unique version number in the model repository and locks the version to prevent secondary modifications. Containerization employs multi-stage container build files, separating the compilation and runtime environments in stages. Before the initial distribution, the model deployment unit will perform pre-compiled kernel automatic matching: read the graphics processor computing capability version and system driver number of the edge computing gateway, and search in the kernel repository of the factory cloud to see if there is already a matching hardware fingerprint; If found, the converted engine file is directly distributed, and no further compilation is needed at the edge; if not found, real-time compilation is triggered in the cloud, and the new kernel is written back to the repository for future reuse. For the same model, both old and new versions are retained simultaneously, and the volume is gradually increased through virtual services in increments of 10%, 30%, and 100%. At the same time, online traffic is completely mirrored to the new version, and the latency, error, and memory metrics are compared in real time. If any one of them deteriorates by more than 2%, it is rolled back immediately to ensure that the accuracy remains unchanged when the operating conditions change. All images are signed during the compilation phase, and artifact verification is enabled on edge nodes. Before the inference container starts, the local policy gatekeeper verifies the signature and vulnerability whitelist, and refuses to start if it fails. The results of each canary release, including the version number, release ratio, and whether rollback is required, are written to the blockchain light node of the factory cloud platform to form an immutable model lifecycle record for use in quality auditing and incident tracing.
11. The method according to claim 1, characterized in that, The operation method of the AR inspection unit is as follows: After on-site workers put on augmented reality headbands, the headbands first compare the digital twin model with the visual positioning, automatically identify the equipment in front of them and retrieve the equipment's number, model, maintenance records and remaining life prediction value; If recording is required, personnel can say "start recording" and the system will convert the current voice into text and package it together with the image captured by the headband camera and the equipment number into a maintenance log. When the system detects that a person's gaze lingers on a valve or instrument for more than two seconds, the headband automatically overlays the real-time pressure, temperature, and flow data of that location within the field of vision and uses color to indicate whether the limit has been exceeded. If the predictive maintenance module sends an emergency work order, the headband will display a 3D arrow navigation to guide personnel to the fault point along the shortest route, while simultaneously projecting safe positioning markers on the ground to prevent accidental entry into dangerous areas.
12. The method according to claim 1, characterized in that, The factory cloud platform and the edge computing gateway are connected by a dual-track link: an industrial security channel and a data flow channel. The industrial security channel is responsible for the encrypted transmission of remote control commands. Before a command is sent, it must undergo a two-way certificate handshake between the platform side and the gateway side. If the handshake fails, the command is automatically discarded to prevent replay attacks. The data stream channel is responsible for uploading large batches of monitoring data, and enables UDP multicast and breakpoint resume mechanism. When the link is interrupted, the data is temporarily stored in the gateway's solid-state storage and resumed according to the timestamp after recovery. The two channels share fiber at the physical layer but are completely isolated at the logical layer, with the control plane and data plane not competing for bandwidth.
