A knowledge graph reasoning-based metal roof health early warning method

Abstract

This invention provides a knowledge graph-based method for early warning of health conditions in metal roofs, comprising the following steps: in response to a request to initiate health monitoring of a metal roof, collecting environmental load parameters and structural response parameters corresponding to a large-span metal roof; obtaining a parameter collection sequence within a preset time period and performing anomaly filtering and missing data completion processing on the parameter collection sequence to obtain a standardized monitoring parameter sequence; constructing a knowledge graph for early warning of health conditions in metal roofs, wherein the knowledge graph includes at least multiple types of entities and triplet relationships between entities, and defines contextual relationships and risk propagation logic; this invention expresses the relationship between environmental loads, roof components, structural response, failure modes, and health status through a knowledge graph, and can explain the reasons for triggering early warnings and the risk propagation path.

Description

Technical Field

[0001] This invention relates to the field of building structure monitoring technology, and in particular to a knowledge graph reasoning-based method for early warning of health conditions in metal roofs. Background Technology

[0002] Metal roofs of large-span public buildings are exposed to complex outdoor service environments for extended periods, continuously subjected to varying external forces such as wind loads, temperature loads, and rainfall loads. This makes them prone to structural defects such as deformation, stress concentration, component loosening, and even localized damage. If structural anomalies cannot be detected in a timely manner, the causes of defects cannot be traced, and the development trend cannot be predicted, it can easily lead to localized roof leaks, structural failures, or even overall safety hazards.

[0003] Current health monitoring solutions for metal roofs mostly rely on single-point sensor threshold comparison or simple data statistical analysis, which can only achieve passive alarms for data exceeding the standard, and have obvious technical shortcomings.

[0004] First, traditional monitoring only analyzes the time series data of each measuring point in isolation, without establishing the correlation between the environment, measuring points, roof components, structural response and failure modes. It cannot depict the propagation path of risk along the component level and physical connection, and the ability to trace the source of the disease is weak.

[0005] Second, conventional threshold alarms can only output abnormal results, lacking interpretable risk causes and transmission logic. The warning results are fragmented and cannot support maintenance personnel in accurately locating risk locations and judging the root cause of faults.

[0006] Third, existing data-driven models rely on a large number of labeled samples for training, have poor adaptability to small sample engineering test scenarios, and have limited generalization ability, making them unsuitable for complex service scenarios with multiple working conditions and multiple load couplings in metal roofing. Summary of the Invention

[0007] In view of this, the present invention aims to provide a knowledge graph reasoning-based method for early warning of health conditions of metal roofs, in order to solve or alleviate the technical problems existing in the prior art, and at least provide a beneficial option.

[0008] To address the aforementioned technical problems, this application adopts the following technical solution: A knowledge graph-based method for early warning of health conditions in metal roofs is provided, characterized by the following steps: In response to a metal roof health monitoring initiation request, environmental load parameters and structural response parameters corresponding to a large-span metal roof are collected; a parameter collection sequence within a preset time period is obtained, and anomaly filtering and missing data completion processing are performed on the parameter collection sequence to obtain a standardized monitoring parameter sequence; a metal roof health early warning knowledge graph is constructed, wherein the knowledge graph includes at least multiple types of entities and triplet relationships between entities, and defines contextual relationships and risk transmission logic; the standardized monitoring parameter sequence is state-identified and matched with entity nodes in the knowledge graph according to preset state criteria to generate a risk-triggered entity set; the knowledge graph association links are traversed using the risk-triggered entity set as the starting context to generate multiple candidate risk propagation paths and calculate the confidence level of each path based on the link relationship weight; target propagation paths are selected based on path confidence levels, and a comprehensive risk score is completed and health warning levels are divided by combining preset warning factors; early warning feedback information carrying the risk location, propagation path, and operation and maintenance strategy is output.

[0009] Preferably, the environmental load parameters are monitoring parameters acting on the external service environment of the large-span metal roof, and the environmental load parameters include at least wind load parameters, ambient temperature load parameters, and rainfall load parameters; the structural response parameters are response parameters collected by the metal roof structure itself and the attached sensing devices, and the structural response parameters include at least roof deformation parameters, stress-strain parameters, and parameters collected by distributed sensor measuring points.

[0010] Preferably, the knowledge graph is divided into multiple entities according to the dimensions of metal roof health monitoring, including at least environmental entities corresponding to the external environment, component entities corresponding to roof components, measuring point entities corresponding to monitoring points, response entities corresponding to structural deformation responses, and state entities corresponding to fault types and health levels.

[0011] Preferably, the parameter acquisition sequence is subjected to anomaly filtering and missing data completion processing, including: removing discrete and abrupt abnormal data in the parameter acquisition sequence by threshold comparison; and completing the missing monitoring parameters with continuous temporal loss by neighborhood temporal interpolation combined with fitting of parameters associated with the same working condition, thereby generating a standardized monitoring parameter sequence with continuous temporal loss and uniform dimensions.

[0012] Preferably, the knowledge graph is constructed in the form of triples of head entity, relation, and tail entity, and the triple association between entities includes at least influence relationship, trigger relationship, transmission relationship and subordinate relationship; the context association is used to characterize the semantic association of entities under the same monitoring dimension, and the risk transmission logic is used to characterize the chain propagation law from abnormal measurement point to component failure.

[0013] Preferably, the standardized monitoring parameter sequence is identified according to a preset state criterion, including: pre-configuring the normal range, slightly abnormal range, moderately abnormal range and severely abnormal range corresponding to each monitoring parameter; matching the standardized monitoring parameter sequence with the corresponding range, and classifying the operating state level of the metal roof monitoring parameters step by step.

[0014] Preferably, the risk trigger entity set is generated by matching entity nodes of the knowledge graph, including: establishing a one-to-one mapping relationship between each monitoring parameter, operating status level and entity node of the knowledge graph; marking the entity node corresponding to the abnormal interval as the trigger node, and aggregating all trigger nodes to form the risk trigger entity set under the current operating condition.

[0015] Preferably, traversing the knowledge graph association links with the risk-triggered entity set as the starting context includes: using a multi-hop semantic traversal method, starting with the risk-triggered entity as the starting node, expanding outward along the preset risk transmission relationship to generate multiple candidate risk propagation paths with different transmission logics; wherein each candidate risk propagation path is composed of consecutive triple links spliced ​​together.

[0016] Preferably, the confidence of each path is calculated based on the link relationship weight, including: configuring fixed weight values ​​for different types of associations within the knowledge graph; and performing fusion calculation on all relationship weights in a single candidate risk propagation path to obtain the comprehensive inference confidence of the corresponding path.

[0017] Preferably, the method of selecting target propagation paths based on path confidence includes: setting a pre-set confidence threshold and eliminating invalid candidate paths below the threshold; sorting the remaining valid paths from high to low confidence and selecting one or more of the top-ranked paths as target propagation paths.

[0018] The embodiments of the present invention have the following advantages due to the adoption of the above technical solutions:

[0019] 1. Improve the interpretability of early warning results: This invention uses a knowledge graph to express the relationship between environmental loads, roof components, structural response, failure modes and health status, which can explain the cause of the early warning and the risk propagation path.

[0020] 2. Improve the engineering relevance of risk identification: This invention can locate risky parts based on the sensor measurement point location and structural response status, identify possible fault types, and provide a basis for on-site inspection and maintenance.

[0021] 3. Reduced reliance on complex data models: This invention does not rely on large-scale training data and complex generative models. Instead, it achieves risk warning through domain knowledge graphs and monitoring status mapping, making it more suitable for scenarios in actual engineering where data is incomplete and samples are limited.

[0022] 4. Improve the efficiency of metal roof operation and maintenance decision-making: The output of this invention includes early warning level, risk type, risk path and handling suggestions, which can help operation and maintenance personnel quickly identify the source of risk and take corresponding measures.

[0023] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the invention will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0026] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0027] It should be understood that the following specific examples illustrate the implementation of this disclosure, and those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0028] Figure 1 This is a flowchart illustrating a knowledge graph reasoning-based health early warning method for metal roofs according to an embodiment of the present invention. It should be noted that if substantially the same result is obtained, the method of this application is not necessarily identical. Figure 1 The illustrated process sequence is limited. For example... Figure 1 As shown: A knowledge graph reasoning-based method for early warning of health conditions in metal roofs includes the following steps:

[0029] S100: In response to the request to start the health monitoring of the metal roof, collect the environmental load parameters and structural response parameters corresponding to the large-span metal roof, obtain the parameter collection sequence within a preset time period, and perform anomaly filtering and missing completion processing on the parameter collection sequence to obtain a standardized monitoring parameter sequence.

[0030] Among them, the environmental load parameters are the monitoring parameters acting on the external service environment of the large-span metal roof, and the environmental load parameters include at least wind load parameters, ambient temperature load parameters, and rainfall load parameters; the structural response parameters are the response parameters collected by the metal roof structure itself and the attached sensing devices, and the structural response parameters include at least roof deformation parameters, stress and strain parameters, and parameters collected by distributed sensor measuring points.

[0031] The anomaly filtering and missing data completion for the parameter acquisition sequence includes: removing discrete and abrupt abnormal data in the parameter acquisition sequence by threshold comparison; and completing the missing monitoring parameters with continuous time-series loss by using neighborhood time-series interpolation combined with fitting of parameters associated with the same working conditions, thereby generating a standardized monitoring parameter sequence with continuous time-series loss and uniform dimensions.

[0032] For large-span metal roofs exposed to the elements for extended periods and subjected to complex service conditions involving alternating wind, temperature, and rain, distributed sensor arrays need to be deployed at key components and locations to collect multi-source monitoring data on environmental loads and structural responses in real time. However, on-site sensor signals are highly susceptible to electromagnetic interference, equipment malfunctions, and signal jitter, often resulting in abrupt changes and missing data in the original acquisition sequences. To address this, this invention first performs anomaly identification and missing data completion processing on the original parameter sequences. On one hand, it eliminates random, abrupt noise data; on the other hand, it repairs missing segments through temporal interpolation and fitting of related parameters, thereby forming a standardized monitoring parameter sequence with regular temporal order and unified dimensions. This lays a reliable data foundation for subsequent state recognition and knowledge graph matching.

[0033] More specifically, the present invention also provides specific ways of applying the method steps according to the present invention, as shown in Table 1, which is a table of monitored data types:

[0034] Table 1. Data Types for Monitoring

[0035] Meteorological load data Wind speed, wind direction, temperature, rainfall Judging the effects of the external environment Structural response data Stress, strain, displacement Determine the stress and deformation state of the roof Measurement point status data Measurement point number, measurement point location, sensor status Locating abnormal areas Historical data Historical number of warnings, historical risk types Assist in risk assessment

[0036] Let the first Class monitoring variables at time The observed value is

[0037] All monitored variables are combined into a multi-source monitoring vector:

[0038]

[0039] in, To monitor the number of variables, The multi-source monitoring state vector is defined at any given time. For missing or abnormal data, common engineering methods can be used for processing, such as filling in data with data from nearby time points, replacing data from similar measurement points, or filling in data with historical averages.

[0040] Let the first The normal range of values ​​for each variable Then the exception detection function is:

[0041]

[0042] in, This indicates that the variable is in a normal state. This indicates that the variable is in an abnormal state. To facilitate subsequent knowledge graph mapping, the structural response can be further divided into 5 levels:

[0043] in, For the first The state level of the class monitoring variable, to This is the threshold for grading.

[0044] Table 2 Structural Response Status Levels

[0045] Level 1 normal The data is within the normal range. Normal monitoring Level 2 Minor abnormality Slight fluctuations exist Continue to observe Level 3 moderate abnormality Local response is too high Encrypted monitoring Level 4 Significant abnormality There are obvious signs of risk. Triggering risk reasoning Level 5 Serious abnormality The response clearly exceeded the limit. Triggering a high-level warning

[0046] S200. Construct a health early warning knowledge graph for metal roofs. The knowledge graph should include at least multiple types of entities and triple relationships between entities, and should define contextual relationships and risk transmission logic.

[0047] Among them, the various entities in the knowledge graph are divided according to the dimension of metal roof health monitoring, including at least environmental entities corresponding to the external environment, component entities corresponding to the roof components, measuring point entities corresponding to the monitoring points, structural deformation response entities, and entities corresponding to fault and health status.

[0048] Specifically, the knowledge graph is constructed in the form of triples consisting of head entity, relation, and tail entity. The triple relationships between entities include at least influence, trigger, transmission, and subordinate relationships. Contextual association is used to characterize the semantic association of entities under the same monitoring dimension, and risk transmission logic is used to characterize the chain propagation pattern from abnormal measurement points to component failures.

[0049] Most existing monitoring solutions only perform data-level analysis and do not semantically model elements such as the roof environment, components, measuring points, and faults. This results in the monitoring elements being isolated from each other and failing to reflect their inherent connections. To overcome this deficiency, this invention is designed for the specific application scenario of metal roofs and builds a domain knowledge graph adapted to the characteristics of the industry. First, it divides entities into multiple categories according to monitoring dimensions. Second, it defines multiple semantic relationships based on a head-relationship-tail triplet architecture and configures contextual association and risk transmission logic. This transforms the originally discrete monitoring elements into a structured semantic system, thereby providing professional knowledge support for subsequent risk reasoning and path tracing.

[0050] More specifically, the present invention also provides specific ways of applying the method steps according to the present invention, as shown in Table 3. Table 3 is a table of entity types for the knowledge graph of health early warning for metal roofs:

[0051] Table 3. Entity Types of the Knowledge Graph for Health Warning of Metal Roofing

[0052] Environmental load entity Strong winds, rain, high temperatures, sudden temperature changes Intensity, duration, and time of occurrence Environmental load entity Roofing component solid Roof panels, supports, connectors, edge clips Materials, Location, Region Roofing component solid Sensor measuring point entity Strain gauges, displacement sensors, weather stations Number, Location, Working Status Sensor measuring point entity Structural Response Entity Increased stress, increased displacement, abnormal strain Value, threshold, measurement point number Structural Response Entity Failure Mode Entities Panel deformation, loose connections, localized cracks, and risk of wind damage. Risk level and scope of impact Failure Mode Entities Health Status Entity Healthy, relatively healthy, weakly healthy, in danger, in peril Grade number, handling requirements Health Status Entity

[0053] A knowledge graph for health early warning of metal roofs is constructed to express the semantic relationships between environmental loads, roof components, sensor measurement points, structural response, failure modes, and health states. The basic form of the knowledge graph is a triple:

[0054]

[0055] in, For the head entity, For entity relationships, It is a tail entity.

[0056] As shown in Table 4, Table 4 is a table of relationship types in the knowledge graph of health early warning for metal roofs:

[0057] Table 4. Relationship Types of the Knowledge Graph for Health Early Warning of Metal Roofing

[0058] Relationship Name Relationship meaning Example of a triplet Acting on Environmental loads acting on roof components Strong winds – acting on – roof panels lead to An event triggers a response Strong winds lead to increased stress. reflect Response data reflects structural status Increased displacement — reaction — loose connection monitor Sensors monitor a certain type of response Strain gauge — Monitoring — Strain anomaly lie in The measuring point is located in a certain area Measuring point S1 is located in the windward area. Related Response and Failure Mode Related Stress rise—correlation—fatigue crack trigger Risk conditions trigger warning status Displacement Exceeds Limits - Trigger - Emergency Warning

[0059] Based on the above entities and relationships, the following knowledge representation can be formed:

[0060]

[0061] This knowledge graph is used for subsequent risk transmission path reasoning and health warning interpretation.

[0062] S300: Based on preset state criteria, identify the state of the standardized monitoring parameter sequence and match it with entity nodes of the knowledge graph to generate a risk-triggered entity set;

[0063] The standardized monitoring parameter sequence is identified according to preset status criteria, including: pre-configuring normal intervals, slightly abnormal intervals, moderately abnormal intervals and severely abnormal intervals for each monitoring parameter; matching the standardized monitoring parameter sequence with the corresponding intervals, and classifying the operating status level of the metal roof monitoring parameters step by step.

[0064] Specifically, the process of generating a risk-triggered entity set by matching entity nodes in the knowledge graph includes: establishing a one-to-one mapping relationship between each monitoring parameter, operating status level and entity nodes in the knowledge graph; marking entity nodes in the abnormal interval as trigger nodes; and aggregating all trigger nodes to form a risk-triggered entity set under the current operating condition.

[0065] After completing the data standardization process, it is necessary to further classify the status of the monitoring sequence. Given the significant differences in the service thresholds of various parameters of the metal roof, a multi-level interval criterion is pre-configured for this invention to divide the operating status into multiple levels. At the same time, a mapping rule is established between monitoring parameters, status levels and graph entities. Once a monitoring parameter falls into an abnormal interval, the corresponding entity node is immediately associated and marked as a trigger node. Finally, all abnormal entities are aggregated to form a risk trigger entity set, which serves as the starting context for subsequent link traversal.

[0066] When monitoring data shows abnormalities or reaches a set level, it is mapped to a structural response entity or environmental load entity in the knowledge graph.

[0067] Set monitoring variables The status level is Then the mapping rule can be expressed as:

[0068]

[0069] in, For the mapped knowledge graph entities, For entity mapping functions, This corresponds to the measurement point number or location.

[0070] As shown in Table 5, Table 5 is a mapping table between monitoring status and knowledge graph entities:

[0071] Table 5 Mapping Table of Monitoring Status and Knowledge Graph Entities

[0072] Monitoring status Status Level Mapping Entities Warning meaning Wind speed increased significantly Level 4 Strong wind event May cause wind-induced response Stress continues to rise Level 3 or 4 Stress Anomaly There may be localized stress concentration. The displacement increased significantly. Level 4 Significant displacement anomalies There may be deformation of the sheet metal. Displacement severely exceeded limits Level 5 Severe displacement anomaly Potential risk of being blown away by wind Strain abnormal mutation Level 4 or 5 strain anomaly Local deformation or cracks may exist.

[0073] When multiple entities trigger simultaneously, an entity set can be formed:

[0074]

[0075] in, For a moment The set of risky entities that are triggered To trigger the number of entities.

[0076] S400: Starting with the set of risk-triggered entities, traverse the knowledge graph's related links to generate multiple candidate risk propagation paths and calculate the confidence of each path based on the link relationship weights.

[0077] The knowledge graph association links traversal with the risk-triggered entity set as the starting context includes: using a multi-hop semantic traversal method, starting with the risk-triggered entity as the starting node, expanding outward along the preset risk transmission relationship to generate multiple candidate risk propagation paths with different transmission logics; each candidate risk propagation path is composed of consecutive triple links spliced ​​together.

[0078] The confidence score of each path is calculated based on the link relationship weights, including: configuring fixed weight values ​​for different types of relationships within the knowledge graph; and merging the weights of all relationships in a single candidate risk propagation path to obtain the comprehensive inference confidence score of the corresponding path.

[0079] Traditional early warning methods can only achieve single-point alarms and cannot deduce how risks are transmitted from measurement points to components. In view of this, the present invention takes the risk-triggered entity as the starting point and carries out multi-hop semantic traversal based on the pre-set transmission relationship of the graph. On the one hand, it extends along the associated links to generate multiple candidate propagation paths, and on the other hand, it configures weights differently for different semantic relationships. Then, it quantifies the confidence of reasoning by fusion of weights within the path, so that each candidate path has a quantifiable evaluation basis, overcoming the shortcomings of traditional early warning methods that are unexplainable and lack path traceability.

[0080] After completing entity mapping, risk propagation reasoning is performed based on the relationship paths in the knowledge graph. Let the triggering entity be... The target risk entity is The risk propagation path is then represented as:

[0081]

[0082] in, Represents the relationships between entities. This indicates an intermediate or risky entity in the path.

[0083] When multiple risk propagation paths exist, path confidence can be calculated based on relationship weights:

[0084]

[0085] in, For the first The weight of each relation Path confidence

[0086] The path with the highest confidence level was selected as the primary risk propagation path:

[0087]

[0088] Table 6 Examples of Risk Transmission Paths

[0089] Triggering Entity Reasoning path Risk Conclusion Strong wind event Strong winds → Increased stress on roof panels → Increased stress → Risk of loosening of connectors Loose connector warning Severe displacement anomaly Increased displacement → Localized warping of roof panels → Reduced wind uplift resistance → Risk of wind rupture Wind-induced risk warning strain anomaly Sudden temperature change → thermal expansion and contraction → abnormal strain → risk of sheet metal deformation Plate deformation warning Rainfall events Rainfall → Joint area affected → Abnormal displacement → Local connection abnormalities Local inspection and early warning

[0090] S500 filters target propagation paths based on path confidence, combines preset early warning factors to complete a comprehensive risk score and classify health early warning levels, and outputs early warning feedback information carrying risk location, propagation path and operation and maintenance strategy.

[0091] After obtaining the confidence levels of each candidate path, invalid paths are eliminated through threshold screening, and high-confidence target paths are retained. On this basis, multi-dimensional early warning factors are introduced to carry out weighted fusion scoring, and then multiple early warning levels are divided according to the score range. At the same time, the risk location is accurately located, the complete propagation path is restored, and a hierarchical operation and maintenance strategy is matched. Finally, a closed loop of the entire process from data collection, knowledge modeling, risk reasoning to hierarchical early warning and operation and maintenance guidance is realized, which greatly improves the practicality and implementation of engineering applications.

[0092] Based on the structural response prediction level and the reasoning results of the risk propagation path in the knowledge graph, a health warning result for the metal roof is output.

[0093] The comprehensive risk score is defined as follows:

[0094] in, This is the normalized value for the structural response risk level. Confidence level of the main risk transmission path For abnormal data or historical early warning factors, Let be the weighting coefficients, and satisfy:

[0095]

[0096] The health alert level is determined as follows:

[0097]

[0098] Table 7 Health Warning Level Table

[0099] Warning Level health status Risk Statement Recommended measures Level 1 healthy Structural response is normal Periodic monitoring Level 2 healthier Slight fluctuations exist Continue to observe Level 3 Weak health Local response anomaly Encrypted monitoring Level 4 In danger There is a clear risk transmission path. Arrange on-site inspection Level 5 Danger Structural failure may occur. Immediate inspection or safety measures

[0100] The final output includes:

[0101]

[0102] in, The health alert level is set at [level]. As a risk type, As the main route of risk transmission, As a result of risk explanation, Recommendations for handling this matter.

[0103] The output of the early warning results is shown in Table 8:

[0104] Table 8. Early Warning Result Output Table

[0105] Output items Output content Health alert level Healthy, relatively healthy, weakly healthy, in danger, in peril Risk type Risks include wind damage, loose connectors, panel deformation, and localized cracks. Risk areas Corresponding measurement point area, windward area, edge area, connecting part area, etc. Risk transmission path Environmental load → Structural response → Failure mode → Healthy state Risk Explanation Results Explain the main reasons for triggering the warning. Recommendations for handling Continuous monitoring, key inspections, time-limited maintenance, and emergency handling.

[0106] The present invention also provides embodiments in which the above method is applied:

[0107] The following description uses the metal roof of a large public building as an example to illustrate the specific implementation of this invention. A weather station, strain gauges, stress sensors, and laser displacement sensors are installed on a large-span metal roof to collect data such as wind speed, wind direction, temperature, rainfall, stress, strain, and displacement. The sampling period is set to 1 minute.

[0108] As shown in Table 9:

[0109] Table 9 Example of Monitoring Data Collection

[0110] Data categories Collect variables Sensor type effect Meteorological load data Wind speed, wind direction, temperature, rainfall Weather station Describe the role of the external environment Structural response data Stress, strain stress sensor, strain gauge Describe the local stress state Structural response data Displacement Laser displacement sensor Describe the roof deformation state Measurement point status data Measurement point number, sensor status Monitoring System Locating abnormal areas

[0111] During a certain monitoring period, the system identified the following as shown in Table 10:

[0112] Table 10 Examples of Monitoring Status Identification

[0113] Monitoring objects Monitoring results Status Level Mapping Entities wind speed Significantly increased Level 4 Strong wind event Displacement measuring point D1 Continue to increase Level 5 Severe displacement anomaly Stress measuring point S1 Continue to rise Level 4 Stress Anomaly Sensor status normal Level 1 Data valid

[0114] Mapping the aforementioned "strong wind events," "severe displacement anomalies," and "stress anomalies" onto a knowledge graph of metal roof health early warning systems, the system performs path reasoning based on entity relationships to derive the main risk propagation paths:

[0115]

[0116] If the confidence level of the second path is high, the system will take "wind-induced risk" as the main risk type and the roof panel and connecting parts area corresponding to the displacement anomaly measurement point as the key risk area.

[0117] Table 11 Example of Health Warning Results

[0118] Output items Output Health alert level Level 5 health status Danger Risk type Wind-induced risks Risk areas Area of ​​roof panels and connectors near displacement anomaly measuring points Risk transmission path Severe abnormal displacement → Localized warping of roof panels → Abnormal stress on connectors → Risk of wind rupture Risk Explanation The predicted continuous increase in displacement may lead to local warping and abnormal stress on the connectors. Recommendations for handling Arrange an on-site inspection immediately, focusing on the connectors and roof panels near the abnormal measuring points.

[0119] This embodiment demonstrates that the present invention can combine metal roof monitoring data with engineering knowledge, identify potential risk propagation paths through knowledge graph reasoning, and output interpretable health warning results, thereby providing a basis for decision-making in the operation and maintenance management of metal roofs.

[0120] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A knowledge graph reasoning-based method for early warning of health conditions in metal roofs, characterized in that, Includes the following steps: In response to the request to start health monitoring of metal roofs, environmental load parameters and structural response parameters corresponding to large-span metal roofs are collected, parameter collection sequences within a preset time period are obtained, and anomaly filtering and missing completion processing are performed on the parameter collection sequences to obtain standardized monitoring parameter sequences. Construct a health early warning knowledge graph for metal roofs. The knowledge graph includes at least multiple types of entities and triplet relationships between entities, and defines contextual relationships and risk transmission logic. Based on preset state criteria, the standardized monitoring parameter sequence is identified and matched with entity nodes of the knowledge graph to generate a risk-triggered entity set. Starting with the set of risk-triggered entities, the knowledge graph is traversed to generate multiple candidate risk propagation paths, and the confidence of each path is calculated based on the link relationship weights. Based on path confidence, target propagation paths are selected, and a comprehensive risk score is completed by combining preset early warning factors. Health warning levels are then classified, and early warning feedback information carrying risk location, propagation path, and operation and maintenance strategies is output.

2. The knowledge graph reasoning-based health early warning method for metal roofs according to claim 1, characterized in that, The environmental load parameters are monitoring parameters acting on the external service environment of the large-span metal roof. The environmental load parameters include at least wind load parameters, ambient temperature load parameters, and rainfall load parameters. The structural response parameters are response parameters collected by the metal roof structure itself and its associated sensing devices. The structural response parameters include at least roof deformation parameters, stress-strain parameters, and parameters collected by distributed sensor measuring points.

3. The knowledge graph reasoning-based health early warning method for metal roofs according to claim 1, characterized in that, The knowledge graph is divided into multiple entities according to the dimensions of metal roof health monitoring, including at least environmental entities corresponding to the external environment, component entities corresponding to roof components, measuring point entities corresponding to monitoring points, response entities corresponding to structural deformation responses, and state entities corresponding to fault types and health levels.

4. The knowledge graph reasoning-based health early warning method for metal roofs according to claim 1, characterized in that, The parameter acquisition sequence is subjected to anomaly filtering and missing data completion processing, including: Abnormal data with discrete mutations in the parameter acquisition sequence are removed by threshold comparison; for monitoring parameters with continuous time loss, the missing data are filled by neighborhood time interpolation combined with fitting of parameters associated with the same working condition, so as to generate a standardized monitoring parameter sequence with continuous time loss and uniform dimensions.

5. The knowledge graph reasoning-based health early warning method for metal roofs according to claim 1, characterized in that, The knowledge graph is constructed in the form of triples consisting of head entity, relation, and tail entity. The triple relationships between entities include at least influence, trigger, transmission, and dependency relationships. The context association is used to characterize the semantic association of entities under the same monitoring dimension, and the risk transmission logic is used to characterize the chain propagation law from abnormal measurement points to component failures.

6. The knowledge graph reasoning-based health early warning method for metal roofs according to claim 1, characterized in that, The standardized monitoring parameter sequence is identified based on preset state criteria, including: Pre-configure the normal range, mild abnormal range, moderate abnormal range and severe abnormal range corresponding to each monitoring parameter; The standardized monitoring parameter sequence is matched with the corresponding interval, and the operating status level of the monitoring parameters of the metal roof is classified step by step.

7. The knowledge graph reasoning-based health early warning method for metal roofs according to claim 6, characterized in that, Match entity nodes from the knowledge graph to generate a risk-triggered entity set, including: Establish a one-to-one mapping relationship between each monitoring parameter, operating status level and knowledge graph entity node; mark the entity node corresponding to the abnormal interval as the trigger node, and aggregate all trigger nodes to form a risk trigger entity set under the current operating condition.

8. The knowledge graph reasoning-based health early warning method for metal roofs according to claim 1, characterized in that, Starting with the set of risk-triggered entities, the knowledge graph association links are traversed, including: A multi-hop semantic traversal approach is adopted, starting with the risk-triggered entity as the starting node and expanding outward along the preset risk transmission relationship to generate multiple candidate risk propagation paths with different transmission logics; each candidate risk propagation path is composed of a continuous triple link spliced ​​together.

9. The knowledge graph reasoning-based health early warning method for metal roofs according to claim 1, characterized in that, The confidence score of each path is calculated based on the link relationship weights, including: Fixed weight values ​​are assigned to different types of relationships within the knowledge graph; the weights of all relationships in a single candidate risk propagation path are fused and calculated to obtain the comprehensive inference confidence of the corresponding path.

10. The knowledge graph reasoning-based health early warning method for metal roofs according to claim 9, characterized in that, Target propagation paths are selected based on path confidence, including: A pre-set confidence threshold is used to eliminate invalid candidate paths that are below the threshold; the remaining valid paths are sorted from high to low confidence, and one or more of the top-ranked paths are selected as the target propagation path.

Images