Industrial equipment whole life cycle health management system based on digital twinning and AI knowledge graph

Abstract

The present application belongs to the technical field of digital twin and intelligent operation and maintenance management of industrial equipment, and discloses an industrial equipment full life cycle health management system based on digital twin and AI knowledge graph, which comprises a device full life cycle data access module, a digital twin state mapping module, an AI knowledge graph construction module, an intention-driven operation and maintenance analysis module, a multi-agent health diagnosis module, a health prediction and disposal decision module, and a life cycle closed-loop update module. Through full life cycle data access, digital twin state mapping and AI knowledge graph construction, the device structure, operation state, fault mechanism and maintenance knowledge are uniformly expressed. In combination with intention-driven analysis, multi-agent diagnosis, health prediction and closed-loop update, abnormal identification, cause and effect tracing, root cause sorting, disposal recommendation and knowledge continuous sedimentation are realized, and the problems of data fragmentation, difficult root cause search, maintenance lag and insufficient model update are solved.

Description

Technical Field

[0001] This invention belongs to the field of digital twin and intelligent operation and maintenance management technology for industrial equipment, and particularly relates to a full life cycle health management system for industrial equipment based on digital twin and AI knowledge graph. Background Technology

[0002] Industrial equipment continuously generates structural parameters, sensor data, alarm records, maintenance orders, spare parts replacement records, energy consumption data, and quality data during its design, installation, operation, inspection, maintenance, repair, modification, and scrapping processes. As production lines become more automated, equipment failures are no longer just due to the damage of a single component, but are often related to changes in operating conditions, the linkage between upstream and downstream equipment, component degradation, operation records, and historical maintenance results. Therefore, a health management system that can integrate full lifecycle data, digital twin status images, equipment knowledge graphs, and intelligent diagnostic processes is needed to achieve equipment status perception, root cause tracing, maintenance strategy generation, and closed-loop knowledge updates.

[0003] Existing industrial equipment health management systems mainly rely on real-time monitoring and threshold alarms. Data is scattered among ledger systems, monitoring systems, work order systems, spare parts systems, and quality systems. There is a lack of unified equipment coding, component hierarchy, sensor semantics, and lifecycle association. As a result, it is difficult to form a continuous health record of equipment operating status, historical faults, maintenance records, and spare parts replacement information, and it is impossible to accurately reflect the true degradation process of equipment.

[0004] Existing equipment diagnostic methods largely rely on human experience or single predictive models, lacking a collaborative mechanism between anomaly detection, causal analysis, root cause ranking, maintenance recommendations, and manual review. While the system can detect anomalies in temperature, vibration, current, or pressure, it struggles to explain from which component the anomaly propagates, which process variables it is related to, and what appropriate handling strategies to adopt. Furthermore, it struggles to continuously revise the digital twin model and knowledge graph relationships based on maintenance and retesting results, leading to difficulties in tracing the root causes of failures, delayed maintenance decisions, and insufficient updates to the health model. Summary of the Invention

[0005] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an industrial equipment lifecycle health management system based on digital twins and AI knowledge graphs. Through full lifecycle data access, digital twin state mapping, and AI knowledge graph construction, it uniformly expresses equipment structure, operating status, fault mechanisms, and maintenance knowledge. Combined with intent-driven parsing, multi-agent diagnosis, health prediction, and closed-loop updates, it achieves anomaly identification, causal tracing, root cause ranking, treatment recommendations, and continuous knowledge accumulation, solving the problems of data fragmentation, difficulty in identifying root causes, delayed maintenance, and insufficient model updates.

[0006] The technical solution adopted in this invention is as follows: an industrial equipment full life cycle health management system based on digital twin and AI knowledge graph, including an equipment full life cycle data access module, a digital twin state mapping module, an AI knowledge graph construction module, an intent-driven operation and maintenance parsing module, a multi-agent health diagnosis module, a health prediction and treatment decision module, and a life cycle closed-loop update module.

[0007] As a preferred technical solution in this plan, the equipment lifecycle data access module accesses data generated during the design, procurement, installation, commissioning, operation, inspection, maintenance, repair, modification, and scrapping stages of industrial equipment; the digital twin status mapping module establishes a digital twin of the industrial equipment based on equipment structural parameters, process parameters, sensor time-series data, and operating conditions, forming a real-time status image of the equipment; the AI ​​knowledge graph construction module constructs an equipment health knowledge graph based on equipment structural hierarchy, component relationships, process flow, sensor semantics, fault modes, maintenance measures, spare parts information, and historical work orders; and the intent-driven operation and maintenance analysis module... The module receives natural language operation and maintenance intentions, converts them into a structured operation and maintenance requirement model, and then extracts the corresponding equipment subgraph from the equipment health knowledge graph. The multi-agent health diagnosis module generates anomaly types, anomaly locations, candidate causes, causal paths, root cause ranking, and handling suggestions based on the equipment subgraph, digital twin status, and historical operation and maintenance data. The health prediction and handling decision module generates health scores, remaining life prediction results, maintenance priorities, and handling strategies. The life cycle closed-loop update module writes the diagnosis results, manual review results, maintenance results, spare parts replacement results, retest results, and handling effects back to the digital twin and the equipment health knowledge graph.

[0008] As a preferred technical solution of this scheme, the equipment life cycle data access module includes a design data access unit, a manufacturing and installation data access unit, an operation monitoring data access unit, an inspection and maintenance data access unit, a quality and energy consumption data access unit, and a data standardization unit; the data standardization unit performs time alignment, unit conversion, missing value filling, outlier marking, and equipment coding standardization on data from different sources.

[0009] As a preferred technical solution of this scheme, the digital twin state mapping module includes a device structure twin unit, an operating state twin unit, an operating condition scenario twin unit, a component degradation twin unit, and a twin deviation correction unit; the twin deviation correction unit corrects the component degradation parameters, operating condition disturbance parameters, and state mapping parameters based on the deviation between the predicted state of the digital twin and the acquired state of the device entity.

[0010] As a preferred technical solution of this scheme, the AI ​​knowledge graph construction module includes an equipment ontology modeling unit, an entity extraction unit, a relation extraction unit, a fault mode fusion unit, a graph update unit, and a subgraph retrieval unit; the graph update unit updates node attributes and relation weights after equipment addition, component replacement, maintenance completion, fault reproduction, and process modification.

[0011] As a preferred technical solution of this scheme, the intent-driven operation and maintenance parsing module includes a natural language receiving unit, an intent parsing unit, a structured requirement generation unit, a constraint extraction unit, a knowledge graph mapping unit, and an operation and maintenance task generation unit; the structured requirement generation unit converts the natural language operation and maintenance intent into a structured operation and maintenance requirement model, and the knowledge graph mapping unit retrieves the device health knowledge graph based on the structured operation and maintenance requirement model.

[0012] As a preferred technical solution of this scheme, the multi-agent health diagnosis module includes a data preprocessing agent, an anomaly detection agent, a causal discovery agent, a root cause analysis agent, a treatment recommendation agent, and a manual review agent; the multi-agent health diagnosis module generates an equipment health score based on anomaly degree score, digital twin bias score, root cause propagation risk score, critical component degradation score, and downtime impact score.

[0013] As a preferred technical solution of this scheme, the health prediction and treatment decision module includes a remaining life prediction unit, a maintenance priority ranking unit, a treatment strategy generation unit, a downtime impact assessment unit, and a decision interpretation output unit; the treatment strategy generation unit generates one or more treatment strategies among continued operation, reduced load operation, planned maintenance, immediate shutdown, spare parts replacement, process parameter adjustment, and equipment modification.

[0014] As a preferred technical solution of this scheme, the life cycle closed-loop update module includes a maintenance result write-back unit, a retest data access unit, a graph relationship correction unit, a twin model correction unit, a knowledge accumulation unit, and a life cycle report generation unit. The graph relationship correction unit corrects the relationship weights between fault nodes, cause nodes, and handling nodes based on abnormal changes before and after maintenance. The twin model correction unit corrects the digital twin parameters based on retest data and equipment operation recovery status.

[0015] The beneficial effects of this invention are as follows: (1) By combining the principles of full lifecycle data access, digital twin state mapping and AI knowledge graph construction, design data, installation data, operation data, inspection and maintenance data, quality and energy consumption data and historical work orders are unified into the equipment health knowledge graph and digital twin. This achieves unified expression between equipment entities, component levels, sensor semantics, fault modes, maintenance measures and lifecycle stages, thereby solving the technical problem of fragmented ledgers, monitoring, work orders and spare parts data in traditional equipment management, and the difficulty in continuously expressing the health status of equipment.

[0016] (2) By utilizing the principle of intent-driven operation and maintenance parsing and multi-agent health diagnosis collaboration, natural language operation and maintenance intent is converted into a structured operation and maintenance requirement model. The corresponding equipment subgraph is extracted based on the equipment health knowledge graph. Then, through the collaborative processing of data preprocessing agent, anomaly detection agent, causal discovery agent, root cause analysis agent and disposal recommendation agent, abnormal variable identification, causal path construction, candidate root cause ranking and disposal suggestion generation are realized. This solves the technical problem that traditional equipment alarms can only indicate anomalies and are difficult to explain the anomaly propagation path and fault root cause.

[0017] (3) By utilizing the principle of health prediction, disposal decision and life cycle closed-loop update linkage, the equipment health score is generated based on the degree of abnormality, digital twin deviation, root cause propagation risk, key component degradation and downtime impact. The disposal strategy is generated by combining the remaining life, spare parts inventory, production tasks and maintenance window. The maintenance results, retest data and manual review results are then written back to the digital twin and equipment health knowledge graph, realizing health scoring, remaining life prediction, maintenance strategy generation and continuous knowledge accumulation, thereby solving the technical problem that the existing health model cannot be continuously updated with the maintenance, modification and retest process. Attached Figure Description

[0018] The accompanying drawings are provided to further understand the present invention and form part of the specification. They are used together with the embodiments of the present invention to explain the invention and do not constitute a limitation thereof.

[0019] Figure 1 This is a schematic diagram of the industrial equipment full life cycle health management system architecture proposed in this invention; Figure 2 This is a flowchart of the industrial equipment health diagnosis and life cycle closed-loop update process proposed in this invention. Detailed Implementation

[0020] Example 1, see Figures 1-2This invention provides an industrial equipment full lifecycle health management system based on digital twins and AI knowledge graphs, including an equipment full lifecycle data access module, a digital twin state mapping module, an AI knowledge graph construction module, an intent-driven operation and maintenance parsing module, a multi-agent health diagnosis module, a health prediction and treatment decision module, and a lifecycle closed-loop update module.

[0021] The equipment lifecycle data access module accesses data generated during the design, procurement, installation, commissioning, operation, inspection, maintenance, repair, modification and scrapping stages of industrial equipment. The data includes equipment model, structural parameters, process parameters, sensor timing data, alarm records, inspection records, maintenance work orders, spare parts replacement records, energy consumption data, quality inspection data, image data and personnel operation records.

[0022] The digital twin state mapping module establishes a digital twin of industrial equipment based on equipment structural parameters, process parameters, sensor timing data, and operating conditions. It then maps the speed, temperature, vibration, current, pressure, flow rate, energy consumption, load rate, cumulative runtime, and component wear status of the physical equipment to the digital twin, forming a real-time state mirror of the equipment.

[0023] The AI ​​knowledge graph construction module builds an equipment health knowledge graph based on equipment structure hierarchy, component relationships, process flow, sensor semantics, fault modes, maintenance measures, spare parts information, and historical work orders. The equipment health knowledge graph includes equipment nodes, component nodes, sensor nodes, fault nodes, cause nodes, handling nodes, spare parts nodes, and life cycle stage nodes, as well as the subordinate, influence, cause, correspondence, replacement, and handling relationships between the above nodes.

[0024] The intent-driven operation and maintenance parsing module receives natural language operation and maintenance intents and converts them into a structured operation and maintenance requirement model. The structured operation and maintenance requirement model includes the target object, triggering conditions, diagnostic objectives, constraints, execution actions, and output format. The intent-driven operation and maintenance parsing module matches the structured operation and maintenance requirement model with the device health knowledge graph and extracts the device subgraph corresponding to the current operation and maintenance intent.

[0025] The multi-agent health diagnosis module includes a data preprocessing agent, an anomaly detection agent, a causal discovery agent, a root cause analysis agent, a treatment recommendation agent, and a manual review agent. Each agent collaboratively performs health diagnosis based on the equipment sub-graph, digital twin status, and historical operation and maintenance data, generating anomaly type, anomaly location, candidate causes, causal paths, root cause ranking, and treatment suggestions.

[0026] The health prediction and handling decision module generates a health score, remaining life prediction results, maintenance priority, and handling strategies based on anomaly detection results, root cause ranking results, equipment life stage, spare parts inventory, production tasks, and maintenance windows. The handling strategies include continued operation, reduced load operation, planned maintenance, immediate shutdown, spare parts replacement, process parameter adjustment, and equipment modification.

[0027] The lifecycle closed-loop update module writes diagnostic results, manual review results, maintenance results, spare parts replacement results, retest results, and treatment effects back to the digital twin and the equipment health knowledge graph, so that the equipment health knowledge graph is continuously updated as the equipment is operated, maintained, and modified.

[0028] Through the above structure, this invention links industrial equipment lifecycle data, digital twin state mirrors, AI knowledge graphs, intent-driven parsing, multi-agent health diagnosis, health prediction, and closed-loop updates, enabling industrial equipment health management to move beyond fixed threshold alarms and human experience judgment. Instead, it enables interpretable diagnosis based on equipment lifecycle stages, component relationships, operating condition changes, anomaly propagation paths, and historical maintenance knowledge, solving problems such as fragmented equipment data, difficulty in tracing the root cause of failures, outdated maintenance strategies, and the inability to continuously update the health status throughout the entire lifecycle.

[0029] Example 2: Based on all the above examples, the equipment lifecycle data access module specifically includes a design data access unit, a manufacturing and installation data access unit, an operation monitoring data access unit, an inspection and maintenance data access unit, a quality and energy consumption data access unit, and a data standardization unit.

[0030] The design data access unit accesses the equipment's structural drawings, component list, design parameters, rated operating conditions, life indicators, and safety boundaries; the manufacturing and installation data access unit accesses the equipment's factory serial number, installation location, commissioning parameters, acceptance records, and initial status records; the operation monitoring data access unit accesses the time-series data uploaded by PLCs, DCSs, SCADA systems, edge acquisition gateways, and sensors; the inspection and maintenance data access unit accesses inspection texts, maintenance photos, maintenance work orders, spare parts replacement records, and manual notes; and the quality and energy consumption data access unit accesses product quality testing data, unit energy consumption data, production capacity data, and downtime loss data.

[0031] The data standardization unit performs time alignment, unit conversion, missing value completion, outlier marking, and equipment coding standardization on data from different sources, and establishes an index relationship for data from the same equipment, the same component, the same sensor, and the same work order.

[0032] When there is a time discrepancy between operation monitoring data and inspection and maintenance data, the data standardization unit aligns them based on the equipment event time. When the equipment names in different systems are inconsistent, the data standardization unit merges them according to the equipment code, installation location, and component level. When sensor data is temporarily missing, the data standardization unit generates missing data markers based on adjacent sampling periods and data from similar sensors.

[0033] Through the above processing, the equipment lifecycle data access module can unify the data from the design, installation, operation, maintenance and modification stages into the same equipment health management framework, solving the problem of scattered ledger, monitoring, work order and spare parts data in traditional equipment management systems.

[0034] Example 3: Based on all the above examples, the digital twin state mapping module specifically includes a device structure twin unit, an operating state twin unit, a working condition scenario twin unit, a component degradation twin unit, and a twin deviation correction unit.

[0035] The equipment structure twin unit generates the equipment structure model based on equipment hierarchy, component assembly relationships, transmission paths, pipeline connection relationships, and control loop relationships; the operating status twin unit maps temperature, vibration, current, pressure, flow rate, speed, load rate, and energy consumption to the corresponding nodes of the equipment structure model; the operating condition twin unit generates the equipment operating condition scenario based on production tasks, process formulas, shift status, ambient temperature and humidity, and upstream and downstream equipment status; the component degradation twin unit generates the component degradation status based on cumulative running time, number of start-stop cycles, load fluctuations, historical fault records, and number of maintenance operations.

[0036] The twin deviation correction unit compares the predicted state of the digital twin with the acquired state of the actual equipment, and corrects the component degradation parameters, operating condition disturbance parameters, and state mapping parameters based on the comparison results. When the deviation between the vibration trend predicted by the digital twin and the vibration trend of the actual equipment exceeds the deviation threshold, the twin deviation correction unit increases the degradation weight of the corresponding component. After the equipment is repaired or the spare parts are replaced, the twin deviation correction unit resets the initial health state of the corresponding component.

[0037] Through the above processing, the digital twin status mapping module can establish a dynamic correspondence between the device entity and the digital twin, solving the problem that static equipment ledgers cannot reflect real-time operating status and degradation trends.

[0038] Example 4: Based on all the above examples, the AI ​​knowledge graph construction module specifically includes a device ontology modeling unit, an entity extraction unit, a relationship extraction unit, a fault mode fusion unit, a graph update unit, and a subgraph retrieval unit.

[0039] The equipment body modeling unit establishes ontology templates for equipment, production lines, processes, components, sensors, faults, causes, handling, spare parts, and lifecycle stages; the entity extraction unit extracts equipment entities, component entities, fault entities, cause entities, and handling entities from equipment manuals, maintenance manuals, historical work orders, alarm texts, and inspection records; the relationship extraction unit generates graph edges based on component assembly relationships, fault propagation relationships, sensor monitoring relationships, and maintenance handling relationships; and the fault mode fusion unit integrates FMEA, maintenance experience, alarm rules, and historical fault samples into the equipment health knowledge graph.

[0040] The graph update unit updates node attributes and relationship weights after equipment addition, component replacement, maintenance completion, fault reproduction, and process modification; the subgraph retrieval unit extracts local equipment health subgraphs based on equipment name, abnormal variables, fault phenomena, production stage, and maintenance intention.

[0041] When a fault recurs in historical work orders, the graph update unit increases the weight of the relationship between the fault node and the cause node; when the anomaly is eliminated after the maintenance measures are implemented, the graph update unit increases the corresponding weight between the cause node and the handling node; when the anomaly is not eliminated after the maintenance measures are implemented, the graph update unit decreases the weight of the corresponding handling relationship and retains the review mark.

[0042] Through the above processing, the AI ​​knowledge graph construction module can transform equipment structure, fault mechanism, sensor semantics and maintenance experience into a searchable, reasonable and updatable knowledge structure, solving the problems of inconsistent semantics and difficulty in reusing experiential knowledge in equipment health diagnosis.

[0043] Example 5: Based on all the above examples, the intent-driven operation and maintenance parsing module specifically includes a natural language receiving unit, an intent parsing unit, a structured requirement generation unit, a constraint extraction unit, a knowledge graph mapping unit, and an operation and maintenance task generation unit.

[0044] The natural language receiving unit receives natural language operation and maintenance (O&M) intentions input by operators, equipment engineers, or managers. These intentions include querying equipment health, analyzing the causes of anomalies, generating maintenance plans, comparing spare parts risks, assessing downtime impacts, and tracing maintenance effectiveness. The intent parsing unit identifies the equipment object, anomaly, time range, diagnostic goals, and output requirements within the natural language O&M intentions. The structured requirements generation unit converts the natural language O&M intentions into a structured O&M requirement model.

[0045] The constraint extraction unit extracts maintenance window, production priority, downtime limit, spare parts inventory, energy consumption constraints, quality constraints, and safety constraints; the knowledge graph mapping unit retrieves equipment health knowledge graphs based on the structured operation and maintenance requirement model to obtain equipment nodes, component nodes, sensor nodes, fault nodes, cause nodes, and handling nodes related to the current intent; the operation and maintenance task generation unit generates diagnostic tasks, prediction tasks, root cause analysis tasks, and handling recommendation tasks based on the retrieved equipment subgraphs.

[0046] When the input intent is "Analyze the reasons for the increased vibration of compressor No. 1 in the past three days", the intent-driven operation and maintenance analysis module generates a structured operation and maintenance requirement model with compressor No. 1 as the target object, the time range as the past three days, the abnormal variable as vibration, and the diagnostic goal as root cause analysis. When the input intent is "Arrange maintenance of pump No. 2 under the condition of no production stoppage", the intent-driven operation and maintenance analysis module generates a structured operation and maintenance requirement model with pump No. 2 as the target object, the constraint condition as no production stoppage, and the output goal as maintenance plan.

[0047] Through the above processing, the intent-driven operation and maintenance parsing module can transform the personnel's colloquial operation and maintenance needs into executable structured tasks, and align them with the equipment, components, sensors and fault knowledge in the equipment health knowledge graph, solving the problem that traditional systems require manual menu selection and manual data splicing.

[0048] Example 6: Based on all the above examples, the multi-agent health diagnosis module specifically includes a data preprocessing agent, an anomaly detection agent, a causal discovery agent, a root cause analysis agent, a treatment recommendation agent, and a manual review agent.

[0049] The data preprocessing agent invokes the equipment lifecycle data access module based on the structured operation and maintenance requirements model to complete data cleaning, time alignment, operating condition segmentation, variable description supplementation, and feature generation. The anomaly detection agent identifies anomalous variables, anomalous time periods, and anomalous severity based on threshold rules, trend deviations, operating condition comparisons, and historical samples. The causal discovery agent constructs a causal graph of equipment anomaly propagation based on component relationships, process sequences, control variables, and observed variable constraints in the equipment health knowledge graph. The root cause analysis agent generates a candidate root cause ranking based on anomalous variables, the causal graph, historical work orders, maintenance records, and digital twin deviation results. The action recommendation agent generates action suggestions based on the candidate root cause ranking, spare parts inventory, maintenance window, downtime losses, and safety level. The manual review agent receives engineer confirmations, rejections, supplementary evidence, and re-analysis operations, and writes the manual review results back to the multi-agent health diagnosis module.

[0050] The equipment health score is calculated using the following formula:

[0051] In the formula, Indicates the device health score. Indicates the degree of abnormality score. Indicates the upper limit of the degree of abnormality. Indicates the digital twin bias score. Indicates the upper limit of digital twin deviation. Indicates the risk score of root cause transmission. Indicates the upper limit of the risk of root cause transmission. This indicates the degradation score of critical components. Indicates the upper limit of degradation of critical components. This indicates that downtime affects the score. This indicates the maximum impact of the shutdown. , , , and These represent the weight coefficients of the corresponding indicators.

[0052] When the device health score is not lower than the first health threshold, the multi-agent health diagnosis module generates a health level; when the device health score is lower than the first health threshold but not lower than the second health threshold, the multi-agent health diagnosis module generates a concern level; when the device health score is lower than the second health threshold but not lower than the third health threshold, the multi-agent health diagnosis module generates a warning level; when the device health score is lower than the third health threshold, the multi-agent health diagnosis module generates a fault risk level.

[0053] Regarding parameter adjustment: When the equipment is in full-load production, the multi-agent health diagnosis module improves... When the abnormal variable is a safety protection variable, the multi-agent health diagnosis module improves... and When the equipment has just undergone major overhaul or component replacement, the multi-agent health diagnosis module reduces... The degradation scores of key components were corrected using maintenance and testing data.

[0054] Through the above processing, the multi-agent health diagnosis module can organize anomaly detection, cause discovery, root cause analysis, treatment recommendation, and manual review into a continuous diagnostic process, solving the problem of the separation between anomaly alarm, root cause judgment, and maintenance suggestions in traditional equipment diagnosis.

[0055] Example 7: Based on all the above examples, the health prediction and treatment decision module specifically includes a remaining life prediction unit, a maintenance priority ranking unit, a treatment strategy generation unit, a downtime impact assessment unit, and a decision interpretation output unit.

[0056] The remaining life prediction unit predicts the remaining life of equipment based on equipment health score, component degradation status, historical failure intervals, operating load, and operating condition fluctuations; the maintenance priority ranking unit generates maintenance priorities based on equipment importance, health level, downtime impact, spare parts inventory, and maintenance window; the disposal strategy generation unit generates one or more disposal strategies based on maintenance priorities, including continue operation, reduced load operation, planned maintenance, immediate shutdown, spare parts replacement, process parameter adjustment, and equipment modification; the downtime impact assessment unit assesses the impact of disposal strategies on output, energy consumption, quality, and delivery time based on production plan, upstream and downstream equipment relationships, inventory buffer, and order delivery time; and the decision interpretation output unit outputs the components of the health score, abnormal variables, causal paths, candidate root causes, disposal basis, and expected effects.

[0057] When the remaining life prediction result is lower than the lower life limit and spare parts inventory is sufficient, the disposal strategy generation unit generates planned maintenance and spare parts replacement strategies; when the equipment health level is the fault risk level and the downtime impact is lower than the downtime threshold, the disposal strategy generation unit generates an immediate shutdown strategy; when the equipment health level is the warning level and the production task is urgent, the disposal strategy generation unit generates reduced load operation and delayed maintenance strategies.

[0058] Through the above processing, the health prediction and treatment decision module can transform diagnostic results into actionable decisions for production, maintenance, and spare parts, solving the problem that traditional health management systems only issue alarms without providing explanations or linking them to maintenance plans.

[0059] Example 8: Based on all the above examples, the lifecycle closed-loop update module specifically includes a maintenance result write-back unit, a retest data access unit, a graph relationship correction unit, a twin model correction unit, a knowledge accumulation unit, and a lifecycle report generation unit.

[0060] The maintenance result write-back unit receives maintenance work orders, spare parts replacements, process adjustments, downtime, and maintenance personnel confirmation results; the retest data access unit receives vibration, temperature, current, pressure, energy consumption, and quality retest data after maintenance; the graph relationship correction unit corrects the relationship weights between fault nodes, cause nodes, and handling nodes based on abnormal changes before and after maintenance; the twin model correction unit corrects the digital twin parameters based on retest data and equipment operation recovery status; the knowledge accumulation unit writes successful handling cases, failed handling cases, manual review opinions, and newly added fault modes into the equipment health knowledge graph; and the life cycle report generation unit generates a life cycle health report based on changes in the equipment's health status from installation to scrapping, maintenance costs, downtime losses, spare parts consumption, fault frequency, and remaining lifespan.

[0061] When the abnormal variables return to normal after maintenance, the graph relationship correction unit increases the weight of the corresponding treatment relationship; when the abnormal variables do not return to normal after maintenance, the graph relationship correction unit decreases the weight of the corresponding treatment relationship and triggers the root cause analysis agent to re-analyze; when the equipment is upgraded, the life cycle closed-loop update module updates the equipment structure nodes, process nodes and component nodes to keep the digital twin and the equipment health knowledge graph consistent.

[0062] Through the above processing, the lifecycle closed-loop update module can accumulate each diagnosis, maintenance, retest, and review into subsequent health management knowledge, transforming industrial equipment health management from one-time alarm handling into a continuous learning and optimization process throughout the entire lifecycle, solving the problem that the equipment health model cannot be updated with the actual maintenance and modification process.

Claims

1. An industrial equipment lifecycle health management system based on digital twins and AI knowledge graphs, characterized in that: It includes a device lifecycle data access module, a digital twin state mapping module, an AI knowledge graph construction module, an intent-driven operation and maintenance analysis module, a multi-agent health diagnosis module, a health prediction and treatment decision module, and a lifecycle closed-loop update module; The equipment lifecycle data access module accesses data generated during the design, procurement, installation, commissioning, operation, inspection, maintenance, repair, modification and scrapping stages of industrial equipment; The digital twin status mapping module establishes a digital twin of industrial equipment based on equipment structural parameters, process parameters, sensor timing data, and operating conditions, and forms a real-time status image of the equipment. The AI ​​knowledge graph construction module builds an equipment health knowledge graph based on equipment structure hierarchy, component relationships, process flow, sensor semantics, failure modes, maintenance measures, spare parts information, and historical work orders. The intent-driven operation and maintenance parsing module receives natural language operation and maintenance intents, converts them into structured operation and maintenance requirement models, and extracts corresponding device subgraphs from the device health knowledge graph based on the structured operation and maintenance requirement models. The multi-agent health diagnosis module generates anomaly types, anomaly locations, candidate causes, causal paths, root cause ranking, and handling suggestions based on equipment sub-graphs, digital twin status, and historical operation and maintenance data. The health prediction and handling decision-making module generates health scores, remaining life prediction results, maintenance priorities and handling strategies based on anomaly detection results, root cause ranking results, equipment life stage, spare parts inventory, production tasks and maintenance windows. The lifecycle closed-loop update module writes diagnostic results, manual review results, maintenance results, spare parts replacement results, retest results, and handling effects back to the digital twin and the equipment health knowledge graph.

2. The industrial equipment full lifecycle health management system based on digital twins and AI knowledge graphs according to claim 1, characterized in that: The equipment lifecycle data access module includes a design data access unit, a manufacturing and installation data access unit, an operation monitoring data access unit, an inspection and maintenance data access unit, a quality and energy consumption data access unit, and a data standardization unit. The data standardization unit performs time alignment, unit conversion, missing value completion, outlier marking, and equipment coding standardization on data from different sources, and establishes an index relationship for data from the same equipment, the same component, the same sensor, and the same work order.

3. The industrial equipment full lifecycle health management system based on digital twins and AI knowledge graphs according to claim 1, characterized in that: The digital twin state mapping module includes a device structure twin unit, an operating state twin unit, an operating condition twin unit, a component degradation twin unit, and a twin deviation correction unit. The device structure twin unit generates a device structure model. The operating state twin unit maps temperature, vibration, current, pressure, flow rate, speed, load rate, and energy consumption to the corresponding nodes of the device structure model. The twin deviation correction unit corrects the component degradation parameters, operating condition disturbance parameters, and state mapping parameters based on the deviation between the digital twin's predicted state and the acquired state of the device entity.

4. The industrial equipment full lifecycle health management system based on digital twins and AI knowledge graphs as described in claim 1, characterized in that: The AI ​​knowledge graph construction module includes an equipment ontology modeling unit, an entity extraction unit, a relation extraction unit, a fault mode fusion unit, a graph update unit, and a subgraph retrieval unit. The equipment ontology modeling unit establishes ontology templates for equipment, production lines, processes, components, sensors, faults, causes, handling, spare parts, and lifecycle stages. The graph update unit updates node attributes and relation weights after equipment addition, component replacement, maintenance completion, fault reproduction, and process modification. The subgraph retrieval unit extracts local equipment health subgraphs based on equipment name, abnormal variables, fault phenomena, production stage, and maintenance intention.

5. The industrial equipment full lifecycle health management system based on digital twins and AI knowledge graphs according to claim 1, characterized in that: The intent-driven operation and maintenance parsing module includes a natural language receiving unit, an intent parsing unit, a structured requirement generation unit, a constraint extraction unit, a knowledge graph mapping unit, and an operation and maintenance task generation unit. The intent parsing unit identifies the device object, abnormal phenomenon, time range, diagnostic target, and output requirements in the natural language operation and maintenance intent; The structured requirements generation unit generates a structured operation and maintenance requirements model; the knowledge graph mapping unit retrieves the equipment health knowledge graph based on the structured operation and maintenance requirements model to obtain the corresponding equipment nodes, component nodes, sensor nodes, fault nodes, cause nodes, and handling nodes.

6. The industrial equipment full lifecycle health management system based on digital twins and AI knowledge graphs according to claim 1, characterized in that: The multi-agent health diagnosis module includes a data preprocessing agent, an anomaly detection agent, a causal discovery agent, a root cause analysis agent, a treatment recommendation agent, and a manual review agent. The data preprocessing agent performs data cleaning, time alignment, condition segmentation, variable description supplementation, and feature generation; anomalies... Detect abnormal variables, abnormal time periods, and abnormality levels generated by the intelligent agent; The causal discovery agent constructs a causal graph of equipment anomaly propagation based on component relationships, process sequence, control variables, and observation variables constraints in the equipment health knowledge graph; the root cause analysis agent generates a ranking of candidate root causes.

7. The industrial equipment full lifecycle health management system based on digital twins and AI knowledge graphs according to claim 6, characterized in that: The health prediction and handling decision module includes a remaining life prediction unit, a maintenance priority ranking unit, a handling strategy generation unit, a downtime impact assessment unit, and a decision interpretation output unit. The remaining life prediction unit generates remaining life prediction results based on equipment health score, component degradation status, historical failure interval, operating load, and operating condition fluctuations. The maintenance priority ranking unit generates maintenance priorities based on equipment importance, health level, downtime impact, spare parts inventory, and maintenance window. The disposal strategy generation unit generates one or more disposal strategies from among: continue operation, reduced load operation, planned maintenance, immediate shutdown, spare parts replacement, process parameter adjustment, and equipment modification.

8. The industrial equipment full lifecycle health management system based on digital twins and AI knowledge graphs according to claim 1, characterized in that: The lifecycle closed-loop update module includes a maintenance result write-back unit, a retest data access unit, a graph relationship correction unit, a twin model correction unit, a knowledge accumulation unit, and a lifecycle report generation unit. The graph relationship correction unit corrects the relationship weights between fault nodes, cause nodes, and handling nodes based on abnormal changes before and after maintenance. The twin model correction unit corrects the digital twin parameters based on retest data and equipment operation recovery status. The knowledge accumulation unit writes successful handling cases, failed handling cases, manual review opinions, and newly added fault modes into the equipment health knowledge graph; the life cycle report generation unit generates life cycle health reports.

Images