Digital twin and ai collaborative cloud platform system for fab chemical supply system

Abstract

The application discloses a kind of digital twin and AI collaborative cloud platform systems for FAB chemical supply system.The system uses cloud edge-end collaborative architecture, including: perception unit, the physical parameters, acoustic waves and image data of conveying system are collected in real time by multimodal sensor network;Digital twin unit, a virtual model synchronized with the physical system is constructed in the cloud, driving state update and short-time interval deduction;Cognition unit, call AI model to analyze real-time data and twin simulation results in parallel, realize leakage risk identification and equipment fault prediction;Decision unit, through intelligent collaborative engine, fusion deduction and analysis conclusion, generate optimized collaborative decision scheme;Execution unit, scheme is issued to edge controller or AR terminal, execute automatic control or provide job guidance.The application realizes real-time perception, intelligent early warning, collaborative decision and precise execution in the process of chemical conveying, significantly improves the safety, reliability and operation efficiency of FAB plant system.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing and industrial safety technology, specifically relating to a digital twin and AI collaborative cloud platform system for FAB chemical supply systems. Background Technology

[0002] The chemical liquid manufacturing industry is currently undergoing a digital and intelligent transformation, with new technologies such as the Industrial Internet, digital twins, and cloud-edge collaboration gradually penetrating the market. However, chemical production is characterized by high risks, complex processes, and strong process continuity, leading to a contradiction between traditional plant management systems and the difficulty in balancing safety control and data utilization—a fundamental conflict between functional safety and data openness.

[0003] Traditional safety instrumented systems use fixed logic, and interlock thresholds and response strategies need to be manually modified offline. They cannot dynamically adjust the safety operation window according to real-time environmental conditions (such as temperature, humidity, and changes in material batches).

[0004] The field equipment comes from various brands (PLC, DCS, sensors, etc.), uses different communication protocols (Modbus, Profibus, OPCUA, etc.), has inconsistent data formats, and lacks global clock synchronization.

[0005] This results in the need to deploy a large number of customized gateways for data acquisition, leading to high integration costs; data lacks timestamps and quality bit tags, resulting in low reliability; and it is impossible to build a unified process view, making it difficult to identify risks associated with cross-system cascading.

[0006] Security control systems require physical isolation and logical segregation, while data migration to the cloud and open interfaces require interconnectivity; these two are architecturally mutually exclusive. Traditional solutions often sacrifice data value to ensure security, or open interfaces lead to blurred security boundaries, creating "information silos" or "security funnels" that cannot meet the integration needs of ERP, supply chain, and regulatory systems under digital transformation.

[0007] Traditional safety event logs are simplistic, lacking encrypted snapshots with timestamps and checksums, failing to meet the stringent requirements of safety supervision departments for accident tracing. Data is easily tampered with during accident investigations, making liability determination difficult and posing high compliance risks. Safety systems, process systems, and quality inspection systems operate independently, with data not being correlated for analysis, making it difficult to identify cascading risks across systems. This results in incomplete consequence risk assessments and an inability to predict complex accidents. Summary of the Invention

[0008] To address the aforementioned problems in the existing technology, this invention provides a chemical liquid plant management cloud platform system oriented towards safety control and data coexistence; The objective of this invention can be achieved through the following technical solutions: A digital twin and AI collaborative cloud platform system for FAB chemical supply systems adopts a cloud-edge-device collaborative architecture, including a sensing unit, a digital twin unit, a cognitive unit, a decision-making unit, and an execution unit; The sensing unit collects physical parameter data, acoustic signals, and visual image data in real time through a multimodal sensor network deployed at key nodes of the chemical delivery system. The raw data is then aggregated and preprocessed by an edge data acquisition gateway deployed near the sensor nodes before being uploaded. The multimodal sensor network includes industrial sensors for monitoring chemical pressure, flow rate, purity, temperature, and liquid level; a laser spectrometer and a distributed ultrasonic sensor array for leak detection; and industrial cameras for monitoring on-site conditions and personnel operations. The digital twin unit constructs and runs a digital twin in the cloud. The digital twin is a virtual model obtained by digitally modeling the physical entities and their operating logic in the chemical transportation system. Its virtual structure completely maps the topology and component connection relationships of the physical system, and the operating parameters and interaction logic of the virtual components are defined and configured according to the actual mechanism and behavior rules of the physical system. The digital twin unit injects the real-time data uploaded by the sensing unit into the digital twin, and drives the synchronous refresh of the state parameters of the corresponding virtual components by fusing multi-source real-time data to keep the operating state, location information and interaction relationships of the virtual components consistent with the actual entities in the physical system. Based on the current state and preset component behavior logic and system operation rules, it performs dynamic simulation and short-term extrapolation of the overall behavior and key performance indicators of the system within a preset time to predict the development process of potential faults, assess the evolutionary impact of risk events, or test the expected effects of different control strategies. The cognitive unit invokes an AI analysis model deployed on a cloud platform or edge to perform parallel analysis on the real-time data of the perception unit and the simulation results of the digital twin. The AI ​​analysis model includes a time-series prediction model and a multi-modal fusion diagnostic model. The time-series prediction model acquires historical and real-time vibration, temperature, and current time-series data of key equipment, predicts the degradation trajectory of equipment status and the evolution trend of key performance indicators, and matches the prediction results with the equipment maintenance history knowledge graph to output remaining service life assessment and high-probability fault modes. The multi-modal fusion diagnostic model receives acoustic signals from the distributed ultrasonic sensor array, uses a time difference positioning algorithm to calculate and generate the first location information of the suspected leak area based on the time difference of abnormal acoustic signals received by each sensor node, synchronously receives gas concentration data from the laser spectrometer, and infers the second location information of the leak source through spatial interpolation or source tracing algorithm based on the concentration reading distribution and gradient changes of multiple spectrometer monitoring points. The first location information and the second location information are spatially correlated and fused with confidence to output the leak location information of the corresponding digital twin. The decision-making unit receives short-term extrapolation results from the digital twin unit and analysis conclusions from the cognitive unit via an intelligent collaborative engine. Based on a preset risk assessment matrix and an optimization objective function, it performs multi-source information fusion and quantitative evaluation. The optimization objective function comprehensively considers production continuity loss, safety impact range, and emergency resource consumption. Based on the quantitative evaluation results, it matches or generates collaborative decision-making schemes in real-time from a preset strategy library. The collaborative decision-making scheme is a structured set of executable instructions. For leakage events, it includes a list of specific valve numbers to be closed, suggested isolation areas, emergency ventilation equipment numbers to be activated, and recommended personnel evacuation routes. For predictive maintenance events, it includes suggested maintenance time windows, a list of required spare parts, estimated downtime, and temporary production scheduling suggestions. For abnormal process parameter events, it includes adjustment settings for target parameters, the sequence of equipment to perform the adjustment, and monitoring indicators to verify the adjustment effect. The execution unit sends the collaborative decision-making scheme to the edge controller or personnel terminal to perform automated valve adjustment, pump control, and emergency linkage, or provides visual operation guidance to inspection personnel through an augmented reality interface. The augmented reality interface runs on the mobile terminal and overlays the operation guidance and navigation arrows issued by the decision-making unit, the real-time status information of the virtual equipment mapped by the digital twin unit and aligned with the real equipment location, and the abnormal equipment or risk areas marked in the real-time video screen by the cognitive unit.

[0009] Specifically, the edge data acquisition gateway is deployed at nodes close to the sensors in the factory area to collect and preprocess raw data from the multimodal sensor network, and upload the processed data to the cloud digital twin unit and cognitive unit through the industrial communication network.

[0010] Specifically, the state update refers to the digital twin receiving and fusing multi-source real-time data from the sensing unit, driving the state parameters of the corresponding virtual components in the virtual model to be refreshed synchronously, so as to keep the operating state, location information and interaction relationship of the virtual components consistent with the actual entities in the physical system.

[0011] Specifically, the short-term simulation is based on the current state of the digital twin and the preset component behavior logic and system operation rules to dynamically simulate and predict the overall behavior and key performance indicators of the system within a preset time period. The simulation is used to predict the development process of potential faults, assess the evolutionary impact of risk events, or test the expected effects of different control strategies.

[0012] Specifically, the multimodal fusion diagnostic model's identification of leakage risks includes: The system receives acoustic signals from the distributed ultrasonic sensor array and uses a time-difference positioning algorithm to calculate and generate first location information of the suspected leak area based on the time difference between the reception of abnormal acoustic signals by each sensor node. Simultaneously, it receives gas concentration data from the laser spectrometer and, based on the concentration reading distribution and gradient changes of multiple spectrometer monitoring points in the sensor network, infers and generates second location information of the leak source through spatial interpolation or a source tracing algorithm. The system then performs spatial correlation and confidence fusion processing on the first and second location information to output the leak location information of the corresponding digital twin.

[0013] Specifically, the time-series prediction model is used to predict equipment failures: The system acquires historical and real-time vibration, temperature, and current time-series data of key equipment to predict the degradation trajectory of equipment status and the evolution trend of key performance indicators. The prediction results are matched with the equipment maintenance history knowledge graph to output the remaining service life assessment and high-probability failure modes of the key equipment.

[0014] Specifically, the intelligent collaborative engine is configured to receive short-term extrapolation results on the evolution of potential risk events from the digital twin unit, and analytical conclusions from the cognitive unit, including leak location, risk level, and equipment failure prediction. Based on a preset risk assessment matrix and an optimization objective function, information from multiple sources is fused and quantitatively evaluated. The optimization objective function comprehensively considers production continuity loss, safety impact range, and emergency resource consumption. Based on the quantitative evaluation results, the optimal collaborative decision-making scheme is matched from a preset strategy library or generated in real time.

[0015] Specifically, the collaborative decision-making scheme is a structured set of executable instructions; For leak incidents, include a specific list of valve numbers that need to be closed, recommended isolation areas, emergency ventilation equipment numbers that should be activated, and recommended personnel evacuation routes; For predictive maintenance events, the recommended maintenance time window, the required spare parts list, the expected downtime, and temporary production scheduling recommendations are included. For abnormal process parameter events, the adjustment settings of the target parameters, the equipment sequence for performing the adjustment, and the monitoring indicators for verifying the adjustment effect are included.

[0016] Specifically, the augmented reality interface operates on a mobile terminal: The decision-making unit issues operation instructions, navigation arrows, and operation step prompts; the real-time status information of the virtual device mapped from the digital twin unit and aligned with the location of the real device; and the abnormal devices or risk areas identified by the cognitive unit and marked in the real-time video footage.

[0017] Specifically, the system also includes a centralized data management and traceability unit, used to: receive and integrate key monitoring data from the sensing unit, risk or abnormal event reports generated by the cognitive unit, and key operation instructions issued by the decision-making unit; associate and store the integrated data, reports, and instructions with the corresponding physical entities and production periods to form structured traceable electronic records; and generate and output a comprehensive traceability report containing event sequences, key data snapshots, and operation history based on authorization requests and the associated relationships.

[0018] The beneficial effects of this invention are as follows: Through the aforementioned system architecture and functional design, this invention achieves a deep integration of security control and data coexistence, resolving multiple contradictions faced by the traditional chemical liquid manufacturing industry during its digital transformation. The system adopts a cloud-edge-device collaborative architecture, effectively balancing real-time performance and security requirements. The edge computing gateway completes critical data processing and security rule verification locally, avoiding security risks caused by network latency or interruptions. Simultaneously, it uploads non-sensitive data to the cloud via an encrypted tunnel, supporting global analysis and decision optimization.

[0019] The virtual factory, built on digital twin technology, integrates multi-source real-time data, providing end-to-end 3D visualization and monitoring from storage tanks to usage points. Combined with a centralized data management and traceability unit, it can accurately correlate any abnormal event or operational instruction with specific chemical batches, equipment, and production periods, quickly generating comprehensive traceability reports containing complete event sequences and data snapshots. This significantly improves the transparency of production management and provides a reliable data foundation for quality analysis, accident investigation, and compliance audits.

[0020] The application of digital twin technology provides a novel means of visualization and intelligentization for chemical production. By constructing a virtual mirror coupled with multi-physics fields, it is possible not only to accurately characterize the operating status of equipment and changes in process parameters, but also to support online simulation verification and optimization strategy feedback, thereby promoting improved production efficiency and optimized resource allocation. Simultaneously, virtual fault injection and periodic verification functions provide a scientific basis for the reliability assessment of safety instrumented systems. Attached Figure Description

[0021] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0022] Figure 1 This is a system architecture diagram of the chemical liquid plant cloud platform of the present invention. Detailed Implementation

[0023] 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.

[0024] To further illustrate the technical means and effects adopted by the present invention to achieve the 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.

[0025] Please see Figure 1 A digital twin and AI collaborative cloud platform system for FAB chemical supply systems adopts a cloud-edge-device collaborative architecture, including a sensing unit, a digital twin unit, a cognitive unit, a decision-making unit, and an execution unit. The sensing unit collects physical parameter data, acoustic signals, and visual image data in real time through a multimodal sensor network deployed at key nodes of the chemical delivery system. The raw data is then aggregated and preprocessed by an edge data acquisition gateway deployed near the sensor nodes before being uploaded. The multimodal sensor network includes industrial sensors for monitoring chemical pressure, flow rate, purity, temperature, and liquid level; a laser spectrometer and a distributed ultrasonic sensor array for leak detection; and industrial cameras for monitoring on-site conditions and personnel operations. The digital twin unit constructs and runs a digital twin in the cloud. The digital twin is a virtual model obtained by digitally modeling the physical entities and their operating logic in the chemical transportation system. Its virtual structure completely maps the topology and component connection relationships of the physical system, and the operating parameters and interaction logic of the virtual components are defined and configured according to the actual mechanism and behavior rules of the physical system. The digital twin unit injects the real-time data uploaded by the sensing unit into the digital twin, and drives the synchronous refresh of the state parameters of the corresponding virtual components by fusing multi-source real-time data to keep the operating state, location information and interaction relationships of the virtual components consistent with the actual entities in the physical system. Based on the current state and preset component behavior logic and system operation rules, it performs dynamic simulation and short-term extrapolation of the overall behavior and key performance indicators of the system within a preset time to predict the development process of potential faults, assess the evolutionary impact of risk events, or test the expected effects of different control strategies. The cognitive unit invokes an AI analysis model deployed on a cloud platform or edge to perform parallel analysis on the real-time data of the perception unit and the simulation results of the digital twin. The AI ​​analysis model includes a time-series prediction model and a multi-modal fusion diagnostic model. The time-series prediction model acquires historical and real-time vibration, temperature, and current time-series data of key equipment, predicts the degradation trajectory of equipment status and the evolution trend of key performance indicators, and matches the prediction results with the equipment maintenance history knowledge graph to output remaining service life assessment and high-probability fault modes. The multi-modal fusion diagnostic model receives acoustic signals from the distributed ultrasonic sensor array, uses a time difference positioning algorithm to calculate and generate the first location information of the suspected leak area based on the time difference of abnormal acoustic signals received by each sensor node, synchronously receives gas concentration data from the laser spectrometer, and infers the second location information of the leak source through spatial interpolation or source tracing algorithm based on the concentration reading distribution and gradient changes of multiple spectrometer monitoring points. The first location information and the second location information are spatially correlated and fused with confidence to output the leak location information of the corresponding digital twin. The decision-making unit receives short-term extrapolation results from the digital twin unit and analysis conclusions from the cognitive unit via an intelligent collaborative engine. Based on a preset risk assessment matrix and an optimization objective function, it performs multi-source information fusion and quantitative evaluation. The optimization objective function comprehensively considers production continuity loss, safety impact range, and emergency resource consumption. Based on the quantitative evaluation results, it matches or generates collaborative decision-making schemes in real-time from a preset strategy library. The collaborative decision-making scheme is a structured set of executable instructions. For leakage events, it includes a list of specific valve numbers to be closed, suggested isolation areas, emergency ventilation equipment numbers to be activated, and recommended personnel evacuation routes. For predictive maintenance events, it includes suggested maintenance time windows, a list of required spare parts, estimated downtime, and temporary production scheduling suggestions. For abnormal process parameter events, it includes adjustment settings for target parameters, the sequence of equipment to perform the adjustment, and monitoring indicators to verify the adjustment effect. The execution unit sends the collaborative decision-making scheme to the edge controller or personnel terminal to perform automated valve adjustment, pump control, and emergency linkage, or provides visual operation guidance to inspection personnel through an augmented reality interface. The augmented reality interface runs on the mobile terminal and overlays the operation guidance and navigation arrows issued by the decision-making unit, the real-time status information of the virtual equipment mapped by the digital twin unit and aligned with the real equipment location, and the abnormal equipment or risk areas marked in the real-time video screen by the cognitive unit.

[0026] Specifically, the edge data acquisition gateway is deployed at nodes close to the sensors in the factory area to collect and preprocess raw data from the multimodal sensor network, and upload the processed data to the cloud digital twin unit and cognitive unit through the industrial communication network.

[0027] Specifically, the state update refers to the digital twin receiving and fusing multi-source real-time data from the sensing unit, driving the state parameters of the corresponding virtual components in the virtual model to be refreshed synchronously, so as to keep the operating state, location information and interaction relationship of the virtual components consistent with the actual entities in the physical system.

[0028] Specifically, the short-term simulation is based on the current state of the digital twin and the preset component behavior logic and system operation rules to dynamically simulate and predict the overall behavior and key performance indicators of the system within a preset time period. The simulation is used to predict the development process of potential faults, assess the evolutionary impact of risk events, or test the expected effects of different control strategies.

[0029] Specifically, the multimodal fusion diagnostic model's identification of leakage risks includes: The system receives acoustic signals from the distributed ultrasonic sensor array and uses a time-difference positioning algorithm to calculate and generate first location information of the suspected leak area based on the time difference between the reception of abnormal acoustic signals by each sensor node. Simultaneously, it receives gas concentration data from the laser spectrometer and, based on the concentration reading distribution and gradient changes of multiple spectrometer monitoring points in the sensor network, infers and generates second location information of the leak source through spatial interpolation or a source tracing algorithm. The system then performs spatial correlation and confidence fusion processing on the first and second location information to output the leak location information of the corresponding digital twin.

[0030] Specifically, the time-series prediction model is used to predict equipment failures: The system acquires historical and real-time vibration, temperature, and current time-series data of key equipment to predict the degradation trajectory of equipment status and the evolution trend of key performance indicators. The prediction results are matched with the equipment maintenance history knowledge graph to output the remaining service life assessment and high-probability failure modes of the key equipment.

[0031] Specifically, the intelligent collaborative engine is configured to receive short-term extrapolation results on the evolution of potential risk events from the digital twin unit, and analytical conclusions from the cognitive unit, including leak location, risk level, and equipment failure prediction. Based on a preset risk assessment matrix and an optimization objective function, information from multiple sources is fused and quantitatively evaluated. The optimization objective function comprehensively considers production continuity loss, safety impact range, and emergency resource consumption. Based on the quantitative evaluation results, the optimal collaborative decision-making scheme is matched from a preset strategy library or generated in real time.

[0032] Specifically, the collaborative decision-making scheme is a structured set of executable instructions; For leak incidents, include a specific list of valve numbers that need to be closed, recommended isolation areas, emergency ventilation equipment numbers that should be activated, and recommended personnel evacuation routes; For predictive maintenance events, the recommended maintenance time window, the required spare parts list, the expected downtime, and temporary production scheduling recommendations are included. For abnormal process parameter events, the adjustment settings of the target parameters, the equipment sequence for performing the adjustment, and the monitoring indicators for verifying the adjustment effect are included.

[0033] Specifically, the augmented reality interface operates on a mobile terminal: The decision-making unit issues operation instructions, navigation arrows, and operation step prompts; the real-time status information of the virtual device mapped from the digital twin unit and aligned with the location of the real device; and the abnormal devices or risk areas identified by the cognitive unit and marked in the real-time video footage.

[0034] Specifically, the system also includes a centralized data management and traceability unit, used to: receive and integrate key monitoring data from the sensing unit, risk or abnormal event reports generated by the cognitive unit, and key operation instructions issued by the decision-making unit; associate and store the integrated data, reports, and instructions with the corresponding physical entities and production periods to form structured traceable electronic records; and generate and output a comprehensive traceability report containing event sequences, key data snapshots, and operation history based on authorization requests and the associated relationships.

[0035] Example 1: Early detection and collaborative handling of high-precision micro-leakage; Near a flange connection of a gas delivery pipeline in the factory workshop, three nodes in a distributed ultrasonic sensor array almost simultaneously acquired anomalous, frequency-specific acoustic signals. An edge data acquisition gateway received these raw signals within 10 milliseconds and performed filtering and preliminary feature extraction.

[0036] Acoustic localization: Preprocessed acoustic data is uploaded to the cloud-based cognitive unit. The acoustic submodule in the multimodal fusion diagnostic model uses a time-difference localization algorithm to calculate the first position coordinates of the abnormal sound source (suspected leak point) in three-dimensional space, initially locating it within the pipeline.

[0037] Spectral confirmation: Data from the laser spectrometer network deployed in the same area showed that the C4F6 concentration at monitoring point A, the closest to the first location coordinates, increased by 15 ppb from the background value within 30 seconds, and monitoring point B downwind also began to show an increasing concentration trend. The spectroscopic submodule used a spatial interpolation algorithm to deduce the second location coordinates of the leak source.

[0038] Fusion decision: The model performs confidence-weighted fusion of the two location information and finally highlights the precise leak point (virtual coordinates) on the virtual pipeline model of the digital twin and judges it as a "Level 1 micro-leakage (early stage)" risk.

[0039] Simulation: Based on current wind speed, pipeline pressure, and leak location information, the digital twin unit immediately initiated a short-term simulation. The simulation results showed that without intervention, the leaked gas would spread to the nearby main power cable trough area within 5 minutes, and the risk level would rise to "Level 2".

[0040] Decision-making: The intelligent collaborative engine receives a "Level 1 Micro Leakage" alarm and a propagation prediction from twin simulation. Combining the real-time production plan (informing that the pipeline is scheduled to supply gas to a batch of critical wafers in 15 minutes), and based on the optimization objective function (minimizing production impact and safety risks), the engine generates a collaborative decision within milliseconds: automatically shutting off the valves upstream and downstream of the leak point to isolate the leak section; activating the emergency ventilation system in the isolated area; pushing an alarm to the AR terminal and mobile control screen of the plant maintenance engineer, and suggesting the activation of backup pipeline L-102B to ensure production continuity.

[0041] The execution unit sends the instructions to close the valve and start the exhaust to the edge PLC controller, and the relevant equipment completes the execution within 2 seconds.

[0042] AR-assisted inspection: The AR glasses (mobile terminal) worn by the on-duty engineer receive navigation guidance, with arrows superimposed on the real field of view, guiding them to the leak point. Upon arrival, the glasses screen automatically displays the maintenance history of the pipe section, the current pressure (now reduced to zero), and the standard emergency response procedures. After confirming the on-site situation, the engineer generates a maintenance work order via voice commands.

[0043] Traceability Records: All data, instructions, and timestamps from the generation of ultrasonic abnormal signals, AI diagnosis, decision generation, valve action, and manual confirmation are stored in a centralized data management and traceability unit, forming a complete auditable traceability chain.

[0044] Example 2: Predictive maintenance of critical delivery pumps; The system continuously monitors the vibration, bearing temperature, and operating current data of the core magnetic pump that supplies pressure to the sulfuric acid delivery system. The edge gateway collects data once per second and performs local characteristic calculations (such as the energy value of a specific high-frequency band in the vibration spectrum).

[0045] The time-series prediction model analyzes these time-series characteristics daily. The model found that the high-frequency energy values ​​associated with bearing defects in the vibration spectrum showed a slow but steady upward trend over the past week, and its evolution trajectory highly matched the "early bearing wear" pattern in the historical failure case library.

[0046] The intelligent collaborative engine receives a prediction: the pump's bearing's remaining service life (RUL) is expected to be at a certain time. The engine then queries the spare parts inventory system (finding the required bearing in stock) and compares it with the production plan for the next month (finding an 8-hour maintenance window with low load operation). Based on this, the engine generates a predictive maintenance decision plan: recommending preventative replacement during the low-load period at night, and automatically generating a scheduling instruction that includes a spare parts requisition form, maintenance work instructions, and temporary activation of the standby pump.

[0047] The plant supervisor reviewed and confirmed the plan on a mobile device. The system notified the relevant maintenance team one day in advance and automatically switched to the standby pump on the day of maintenance. After maintenance was completed, the new vibration data was entered into the system for model self-learning optimization.

[0048] Example 3: Adaptive optimization of process parameters and energy management (energy efficiency scenario) The digital twin runs simulations periodically to explore the optimal combination of valve openings for each branch under different total flow demands, in order to minimize the total system energy consumption. Meanwhile, the process optimization model analysis in the cognitive unit reveals that the constant pressure setpoint for a certain etching fluid delivery line has room for adjustment during low-flow periods at night.

[0049] The intelligent collaborative engine submitted "nighttime pressure reduction operation" and "optimized valve combination" as two candidate strategies to the digital twin for non-real-time simulation testing. The simulation results showed that the new strategies can reduce the overall energy consumption of the system while ensuring stable terminal pressure, and have no negative impact on pipeline safety.

[0050] The decision-making unit approves the optimization plan and translates it into automatic control commands for the nighttime period. The execution unit automatically adjusts the frequency of the variable frequency pump and the corresponding valve openings at preset time points. The system continuously monitors key parameters after adjustment (flow stability, pressure, energy consumption), confirms the optimization effect, and records and archives the data, forming a continuous improvement closed loop of "analysis-optimization-verification".

[0051] 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 digital twin and AI collaborative cloud platform system for FAB chemical supply systems, characterized in that, It adopts a cloud-edge-device collaborative architecture, including a perception unit, a digital twin unit, a cognition unit, a decision-making unit, and an execution unit; The sensing unit collects physical parameter data, acoustic signals and visual image data in real time through a multimodal sensor network deployed at key nodes of the chemical transport system. The raw data is then aggregated and preprocessed by an edge data acquisition gateway deployed near the sensor nodes before being uploaded. The digital twin unit constructs and runs a digital twin in the cloud. The digital twin is a virtual model obtained by digitally modeling the physical entities and their operating logic in the chemical transportation system. Its virtual structure completely maps the topology and component connection relationships of the physical system, and the operating parameters and interaction logic of the virtual components are defined and configured according to the actual mechanism and behavior rules of the physical system. The digital twin unit injects the real-time data uploaded by the sensing unit into the digital twin, and drives the synchronous refresh of the status parameters of the corresponding virtual components by fusing multi-source real-time data. Based on the current state and the preset component behavior logic and system operation rules, dynamic simulation and short-term extrapolation of the overall behavior and key performance indicators of the system within a preset time period are performed. The cognitive unit invokes an AI analysis model deployed on a cloud platform or edge to perform parallel analysis on the real-time data of the perception unit and the simulation results of the digital twin. The AI ​​analysis model includes a time-series prediction model and a multi-modal fusion diagnostic model. The time-series prediction model acquires historical and real-time vibration, temperature, and current time-series data of key equipment, predicts the degradation trajectory of equipment status and the evolution trend of key performance indicators, and matches the prediction results with the equipment maintenance history knowledge graph to output remaining service life assessment and high-probability fault modes. The multi-modal fusion diagnostic model receives acoustic signals from the distributed ultrasonic sensor array, uses a time difference positioning algorithm to calculate and generate the first location information of the suspected leak area based on the time difference of abnormal acoustic signals received by each sensor node, synchronously receives gas concentration data from the laser spectrometer, and infers the second location information of the leak source through spatial interpolation or source tracing algorithm based on the concentration reading distribution and gradient changes of multiple spectrometer monitoring points. The first location information and the second location information are spatially correlated and fused with confidence to output the leak location information of the corresponding digital twin. The decision-making unit receives short-term extrapolation results from the digital twin unit and analysis conclusions from the cognitive unit via an intelligent collaborative engine. Based on a preset risk assessment matrix and an optimization objective function, it performs multi-source information fusion and quantitative evaluation. The optimization objective function comprehensively considers production continuity loss, safety impact range, and emergency resource consumption. Based on the quantitative evaluation results, it matches or generates collaborative decision-making schemes in real time from a preset strategy library. The collaborative decision-making scheme is a structured set of executable instructions. For leakage events, it includes a list of valve numbers to be closed, suggested isolation areas, emergency ventilation equipment numbers to be activated, and recommended personnel evacuation routes. For predictive maintenance events, it includes suggested maintenance time windows, a list of required spare parts, estimated downtime, and temporary production scheduling suggestions. For abnormal process parameter events, the adjustment settings of the target parameters, the sequence of equipment to perform the adjustment, and the monitoring indicators to verify the adjustment effect are included. The execution unit sends the collaborative decision-making scheme to the edge controller or personnel terminal to perform automated valve adjustment, pump control, and emergency linkage, or provides visual operation guidance to inspection personnel through an augmented reality interface. The augmented reality interface runs on the mobile terminal and overlays the operation guidance and navigation arrows issued by the decision-making unit, the real-time status information of the virtual equipment mapped by the digital twin unit and aligned with the real equipment location, and the abnormal equipment or risk areas marked in the real-time video screen by the cognitive unit.

2. The system according to claim 1, characterized in that, The edge data acquisition gateway is deployed at nodes close to the sensors in the factory area. It is used to aggregate and preprocess the raw data from the multimodal sensor network, and upload the processed data to the cloud digital twin unit and cognitive unit through the industrial communication network.

3. The system according to claim 1, characterized in that, The state update refers to the digital twin receiving and fusing multi-source real-time data from the sensing unit, driving the state parameters of the corresponding virtual components in the virtual model to be refreshed synchronously, so as to keep the running state, position information and interaction relationship of the virtual components consistent with the actual entities in the physical system.

4. The system according to claim 1, characterized in that, The short-term simulation is based on the current state of the digital twin and the preset component behavior logic and system operation rules. It dynamically simulates and predicts the overall behavior and key performance indicators of the system within a preset time period. The simulation is used to predict the development process of potential faults, assess the evolutionary impact of risk events, or test the expected effects of different control strategies.

5. The system according to claim 1, characterized in that, The multimodal fusion diagnostic model specifically identifies leakage risks by including: The system receives acoustic signals from the distributed ultrasonic sensor array and uses a time-difference positioning algorithm to calculate and generate first location information of the suspected leak area based on the time difference between the reception of abnormal acoustic signals by each sensor node. Simultaneously, it receives gas concentration data from the laser spectrometer and, based on the concentration reading distribution and gradient changes of multiple spectrometer monitoring points in the sensor network, infers and generates second location information of the leak source through spatial interpolation or a source tracing algorithm. The first location information and the second location information are then spatially correlated and fused with confidence to output the leak location information of the corresponding digital twin.

6. The system according to claim 1, characterized in that, The time-series prediction model is used to predict equipment failures: Acquire historical and real-time vibration, temperature, and current time-series data of key equipment to predict the degradation trajectory of equipment status and the evolution trend of key performance indicators; match the prediction results with the equipment maintenance history knowledge graph to output the remaining service life assessment and high-probability failure modes of the key equipment.

7. The system according to claim 1, characterized in that, The intelligent collaborative engine is configured to receive short-term extrapolation results on the evolution of potential risk events from the digital twin unit, and analytical conclusions from the cognitive unit, including leak location, risk level, and equipment failure prediction. Based on a preset risk assessment matrix and an optimization objective function, information from multiple sources is fused and quantitatively evaluated. The optimization objective function comprehensively considers production continuity loss, safety impact range, and emergency resource consumption. Based on the quantitative evaluation results, the optimal collaborative decision-making scheme is matched from a preset strategy library or generated in real time.

8. The system according to claim 1, characterized in that, The collaborative decision-making scheme is a structured set of executable instructions; For leak incidents, include a specific list of valve numbers that need to be closed, recommended isolation areas, emergency ventilation equipment numbers that should be activated, and recommended personnel evacuation routes; For predictive maintenance events, the recommended maintenance time window, the required spare parts list, the expected downtime, and temporary production scheduling recommendations are included. For abnormal process parameter events, the adjustment settings of the target parameters, the equipment sequence for performing the adjustment, and the monitoring indicators for verifying the adjustment effect are included.

9. The system according to claim 1, characterized in that, The augmented reality interface runs on a mobile terminal: The decision-making unit issues operation instructions, navigation arrows, and operation step prompts; the real-time status information of the virtual device mapped from the digital twin unit and aligned with the location of the real device; and the abnormal devices or risk areas identified by the cognitive unit and marked in the real-time video footage.

10. The system according to claim 1, characterized in that, The system also includes a centralized data management and traceability unit, used to: receive and integrate key monitoring data from the sensing unit, risk or abnormal event reports generated by the cognitive unit, and key operation instructions issued by the decision-making unit; The integrated data, reports, and instructions are associated with the corresponding physical entities and production periods for storage, forming structured, traceable electronic records; Based on the authorization request and the associated relationships, a comprehensive traceability report is generated and output, which includes event sequences, key data snapshots, and operation history.

Images