Airport runway maintenance reasoning enhancement system fusing large model and graph neural network

Abstract

This application relates to the field of data processing technology, specifically to an enhanced airport pavement maintenance reasoning system integrating a large model and graph neural networks. The system includes: a knowledge graph construction module for acquiring an airport pavement maintenance knowledge graph; a dormant edge generation module for predicting disease spread and generating dormant edges based on spatial topology and deterioration conditions; a connection activation module for activating dormant edges that meet environmental load conditions; a failure disconnection module for disconnecting failure state edges based on aging and cumulative fatigue, achieving closed-loop updates of the knowledge graph; and a visualization module for displaying changes in connections between nodes at different times. This application upgrades the static knowledge graph to a dynamic evolution model driven by physical laws, achieving closed-loop self-evolution of the knowledge graph. The system significantly reduces the false alarm rate, promoting a shift from reactive post-maintenance to intelligent pre-prediction for pavement maintenance, and providing highly timely and reliable decision support for airport pavement maintenance.

Description

Technical Field

[0001] This application belongs to the field of data processing technology, specifically relating to an enhanced airport pavement maintenance inference system that integrates large models and graph neural networks. Background Technology

[0002] With the rapid development of the civil aviation industry, knowledge graph technology has been widely introduced into the field of intelligent airport pavement maintenance to integrate heterogeneous data from multiple sources, including inspection, meteorology, and maintenance. Existing pavement knowledge graphs primarily focus on static information association and business retrieval, achieving structured storage of existing defect records and historical maintenance plans. This static graph, to a certain extent, breaks down data silos between various business systems, providing basic information query and traceability support for operation and maintenance personnel.

[0003] However, traditional knowledge graphs struggle to accurately map the dynamic evolution of pavement in the complex physical world. First, existing graphs lack the ability to predict the spread of damage based on spatial topology and relative deterioration levels, and they cannot dynamically stimulate and assess potential damage outbreaks by incorporating multi-dimensional external environmental conditions such as real-time weather and high-frequency flight loads. Second, existing systems typically set the "repaired" state as static and permanently valid, severely detaching from the objective reality of fatigue degradation of maintenance materials over service time and accumulated loads. This prevents the graph from automatically identifying failure states and disconnecting invalid links. This static model, lacking spatiotemporal extrapolation and fatigue closed-loop update mechanisms, severely restricts the intelligent leap of airport pavement maintenance from "reactive post-event repair" to "precise pre-event prediction." Summary of the Invention

[0004] To address the aforementioned problems, embodiments of this application provide an enhanced airport pavement maintenance inference system that integrates large models and graph neural networks. The system includes: The knowledge graph construction module is configured to acquire the airport pavement maintenance knowledge graph; The dormant edge generation module is configured to predict disease spread based on the node spatial topology and degradation status in the knowledge graph, so as to generate dormant edges in the knowledge graph. The contact activation module is configured to introduce multi-dimensional environmental load conditions and activate the dormant edge that meets the conditions based on the multi-dimensional environmental load conditions. The failure disconnection module is configured to disconnect the failure state edge based on the node's timeliness and cumulative fatigue decay, so as to achieve closed-loop update of the knowledge graph. The visualization module is configured to visualize the relationship between different nodes in the knowledge graph at different times. The knowledge graph construction module includes: The graph neural network submodule is configured to output the disease feature parameters of each node through graph convolution operations based on the pavement condition index score and topological distance. The large model processing submodule is configured to use a pre-trained Transformer architecture to extract the co-occurrence frequency of weather and disease types from historical inspection logs and maintenance plan texts, and fit an exponentially adjusted power. The feature parameter assignment submodule is configured to convert the damage measurement values ​​obtained by the laser pavement tester and ground penetrating radar into a dimensionless deterioration scalar based on the pavement condition index evaluation system, and assign the deterioration scalar as a defect feature parameter to the map node.

[0005] In one embodiment, the dormant edge generation module includes: The evolutionary transmission index calculation submodule is configured to obtain the evolutionary transmission index of the target pavement block node; The dormant edge generation submodule is configured to dynamically generate potential evolutionary association edges pointing from the target pavement block node to the expected disease type node when the evolutionary transmission index is greater than the evolutionary critical threshold, and to use the potential evolutionary association edges as the dormant edges.

[0006] In one implementation, the evolutionary transmission index calculation submodule is configured to calculate the evolutionary transmission index based on the baseline evolution probability of the disease extracted from historical maintenance text, the disease characteristic parameters of the target pavement block node, the standard permissible state parameters of the target pavement block node, the disease characteristic parameters of the physical adjacent nodes of the target pavement block node, the spatial distance between the target pavement block node and the physical adjacent nodes, and the inherent resistance constant of the pavement structure.

[0007] In one implementation, the contact activation module includes: The stress amplification factor calculation submodule is configured to construct the stress amplification factor of the airport pavement; The trigger risk value calculation submodule is configured to obtain the trigger risk value based on the evolution transmission index and the stress amplification factor; The activation submodule is configured to activate and solidify dormant potential evolutionary association edges into deterministic predicted edges when the trigger risk value is greater than the trigger confirmation threshold.

[0008] In one embodiment, the stress amplification factor calculation submodule is configured to calculate the stress amplification factor based on the cumulative flight equivalent load during the monitoring period, the pavement design reference bearing capacity, the maximum diurnal temperature difference during the monitoring period, the reference temperature difference that causes thermal expansion and contraction, the cumulative precipitation during the monitoring period, the reference discharge capacity of the airport drainage system, and the exponential adjustment power.

[0009] In one embodiment, the failure disconnect module includes: The maintenance effectiveness weight calculation submodule is configured to obtain the maintenance effectiveness weight of the repair association edge connecting the target pavement block node and the maintenance plan node; The graph reconstruction submodule is configured to disconnect the repair association edge and establish a new association edge pointing to the node to be re-inspected or the node at risk of secondary deterioration when the maintenance effectiveness weight is less than the maintenance failure critical ratio.

[0010] In one implementation, the maintenance effectiveness weight calculation submodule is configured to calculate the maintenance effectiveness weight for the repair-related edge based on the initial mechanical property constant of the maintenance material, the actual service time of the maintenance material, the nominal design life of the maintenance material, the cumulative dynamic environmental load since the repair of the target pavement block node, and the fatigue limit load of the maintenance material, wherein the cumulative dynamic environmental load is obtained based on the stress amplification factor.

[0011] In one embodiment, the system further includes: The early warning linkage module is configured to link with an external early warning system when the connection activation module activates a dormant edge as a deterministic prediction edge.

[0012] This application provides an enhanced airport pavement maintenance reasoning system that integrates a large model and graph neural networks. The system includes: a knowledge graph construction module for acquiring an airport pavement maintenance knowledge graph; a dormant edge generation module for predicting disease spread and generating dormant edges based on spatial topology and deterioration; a connection activation module for activating dormant edges that meet environmental load conditions; a failure disconnection module for disconnecting failure state edges based on aging and cumulative fatigue, achieving closed-loop graph updates; and a visualization module for displaying changes in connections between nodes at different times. This application upgrades the static knowledge graph to a dynamic evolution model driven by physical laws. It predicts disease spread through spatial deterioration transmission, filters invalid triggers using environmental load factors, and automatically disconnects failed maintenance links based on fatigue attenuation, achieving closed-loop self-evolution of the knowledge graph. The system significantly reduces the false alarm rate, accurately reflects the lifecycle of maintenance materials, and promotes the intelligent leap from reactive post-maintenance to pre-precise prediction in pavement maintenance, providing highly timely and reliable decision support for airport pavement maintenance.

[0013] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description

[0014] To more clearly illustrate the implementation schemes of this application, the accompanying drawings used in the implementation schemes will be briefly introduced below. It should be understood that the accompanying drawings only show some implementation schemes of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from the accompanying drawings without creative effort.

[0015] Figure 1 This is a block diagram illustrating an airport pavement maintenance inference enhancement system that integrates a large model and a graph neural network, according to an exemplary embodiment. Figure 2 This is a block diagram illustrating a knowledge graph construction module according to an exemplary embodiment; Figure 3 This is a block diagram illustrating yet another knowledge graph construction module according to an exemplary embodiment; Figure 4 This is a block diagram illustrating a dormant edge generation module according to an exemplary embodiment; Figure 5 This is a block diagram illustrating a contact activation module according to an exemplary embodiment; Figure 6 This is a block diagram illustrating a failure disconnect module according to an exemplary embodiment; Figure 7 This is a block diagram illustrating yet another airport pavement maintenance inference enhancement system that integrates a large model and a graph neural network, according to an exemplary embodiment. Detailed Implementation

[0016] The specific embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this application.

[0017] It should be understood that the term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the following description.

[0018] It should be noted that the concepts of "first" and "second" mentioned in this application are only used to distinguish different devices, modules, or units, and are not used to limit the order of functions performed by these devices, modules, or units or their interdependencies. The modifications of "one" and "multiple" mentioned in this application are illustrative rather than restrictive, and those skilled in the art should understand that, unless explicitly stated in the context, they should be understood as "one or more". In the description of this application, unless otherwise stated, "multiple" means two or more, and other quantifiers are similar; "at least one item", "one item or multiple items", or similar expressions refer to any combination of these items, including any combination of single or multiple items.

[0019] Although operations or steps are described in a specific order in the accompanying drawings in the embodiments of this application, this should not be construed as requiring these operations or steps to be performed in the specific order or serial order shown, or requiring all of the shown operations or steps to be performed to obtain the desired result. In the embodiments of this application, these operations or steps may be performed serially; they may be performed in parallel; or a portion of these operations or steps may be performed.

[0020] The names of the messages or information exchanged between multiple devices in the embodiments of this application are for illustrative purposes only and are not intended to limit the scope of these messages or information. It is understood that before using the technical solutions of the various embodiments of this application, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this application in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained. The application will now be described in conjunction with specific embodiments.

[0021] Figure 1 This is a block diagram illustrating an enhanced airport pavement maintenance inference system that integrates a large model and a graph neural network, according to an exemplary embodiment. Figure 1 As shown in the embodiments of this application, an enhanced airport pavement maintenance inference system integrating large models and graph neural networks is provided. The system may include the following modules: The knowledge graph construction module is configured to acquire the airport pavement maintenance knowledge graph; The dormant edge generation module is configured to predict disease spread based on the node spatial topology and degradation status in the knowledge graph, so as to generate dormant edges in the knowledge graph. The contact activation module is configured to introduce multi-dimensional environmental load conditions and activate the dormant edge that meets the conditions based on the multi-dimensional environmental load conditions. The failure disconnection module is configured to disconnect the failure state edge based on the node's timeliness and cumulative fatigue decay, so as to achieve closed-loop update of the knowledge graph. The visualization module is configured to visualize the relationship between different nodes in the knowledge graph at different times.

[0022] For example, during system operation, the knowledge graph construction module first obtains an airport pavement maintenance knowledge graph. This graph is constructed by integrating multi-source heterogeneous data and includes entity nodes such as pavement blocks, defect types, maintenance plans, and environmental factors, as well as their initial relationships. Subsequently, the dormant edge generation module predicts the defect spread trend based on the spatial topology and deterioration of nodes in the knowledge graph. Specifically, it calculates the evolutionary transmission index of the central node, combines its own deterioration degree with the defect penetration pressure of neighboring nodes, and dynamically generates a potential evolutionary association edge from the target pavement block node to the expected defect type node when the index exceeds the evolutionary critical threshold. This edge is temporarily stored in the graph as a dormant edge, representing a potential hidden danger connection that has not yet erupted but has the possibility of physical transmission.

[0023] Next, the activation module introduces multi-dimensional environmental load conditions, including real-time meteorological data, cumulative flight load, and other external contextual information, to construct a stress amplification factor. This factor is then multiplied by the aforementioned evolutionary transmission index to obtain a trigger risk value. When this value exceeds the trigger confirmation threshold, the system activates dormant edges and solidifies them into deterministic predictive edges, triggering the early warning logic. Simultaneously, the failure disconnection module periodically calculates the maintenance effectiveness weight of the repair association edges connecting pavement nodes and maintenance plan nodes based on node aging and cumulative fatigue decay. This weight decreases non-linearly with increasing service time and cumulative load. When the weight falls below the maintenance failure critical ratio, it is determined that the maintenance material is nearing its fatigue limit. The original repaired relationship edges are then disconnected, and a new connection edge pointing to the node awaiting re-inspection or the node at risk of secondary deterioration is forcibly established, thereby achieving a closed-loop update of the knowledge graph. Finally, the visualization module uses the global timeline as its driving engine to intuitively display the node status and edge connections of different nodes in the knowledge graph at different times. It uses visual elements such as color gradients, line solidity and dashedness, and high brightness frequency to map the degree of disease deterioration, connection strength, and dynamic evolution trajectory in real time.

[0024] This application's embodiments break through the limitations of traditional static maps, elevating airport pavement maintenance from a reactive, post-event response to a precise, pre-event prediction, significantly reducing the false alarm rate, and realizing the self-evolution of knowledge graphs based on physical laws, providing highly timely and reliable airport pavement maintenance decision support.

[0025] Figure 2 This is a block diagram illustrating a knowledge graph construction module according to an exemplary embodiment. For example... Figure 2 As shown, the knowledge graph construction module may include the following modules: The graph neural network submodule is configured to output the disease feature parameters of each node through graph convolution operations based on the pavement condition index score and topological distance. The large model processing submodule is configured to use a pre-trained Transformer architecture to extract the co-occurrence frequency of weather and disease types from historical inspection logs and maintenance plan texts, and fit an exponentially adjusted power.

[0026] For example, in the knowledge graph construction module, the system further integrates a graph neural network submodule and a large model processing submodule to improve the accuracy and dimensionality of feature extraction. The graph neural network submodule receives pavement condition index scores from automated collection equipment and topological distance matrices calculated based on physical coordinates as input. It uses graph convolution operations to perform multi-layer spatial feature aggregation, passing and aggregating the feature information of adjacent nodes layer by layer, and finally outputs the disease feature parameters of each node i after unifying the dimensionality. This parameter accurately reflects the severity of damage to each pavement block at the current moment.

[0027] Meanwhile, the large model processing submodule employs a pre-trained Transformer architecture, receiving historical inspection logs and maintenance plan text as input. It guides the model through pre-defined domain cue words to perform named entity recognition and relation extraction, extracting the co-occurrence frequency and association patterns of meteorological conditions and disease types from unstructured text. The hidden layer representation vector, containing the aforementioned frequency and association pattern information, output by the Transformer architecture is used as the co-occurrence frequency feature vector and input to a fully connected regression layer. Parameter optimization is performed by calculating the mean squared error loss using historical actual disease outbreak samples, ultimately mapping and outputting an exponentially adjusted power in continuous numerical form. And output it to the downstream module.

[0028] In this embodiment, the two sub-modules work together. The graph neural network provides structured features based on physical measurements and spatial topology, while the large model mines implicit empirical knowledge from historical operation and maintenance texts. Together, they provide high-quality input features for subsequent calculation of evolutionary transmission index and construction of stress amplification factor, which significantly enhances the system's predictive reasoning ability.

[0029] Figure 3 This is a block diagram illustrating yet another knowledge graph construction module according to an exemplary embodiment. For example... Figure 3 As shown, the knowledge graph construction module may further include the following modules: The feature parameter assignment submodule is configured to convert the damage measurement values ​​obtained by the laser pavement tester and ground penetrating radar into a dimensionless deterioration scalar based on the pavement condition index evaluation system, and assign the deterioration scalar as a defect feature parameter to the map node.

[0030] For example, the knowledge graph construction module may also include a feature parameter assignment submodule, which converts the original physical measurement values ​​into unified dimensionless scalars to adapt to the processing requirements of graph neural networks. In specific operation, this submodule first obtains the absolute values ​​of the underlying physical measurements by periodically importing the apparent crack length and width data collected by the laser pavement tester and the void volume data detected by ground penetrating radar.

[0031] Subsequently, based on the pavement condition index (PCI) evaluation system in the "Technical Standards for Civil Airport Flight Areas," damage measurement values ​​of different dimensions can be substituted into the corresponding single-item defect deduction model. After standardization and conversion, these values ​​are uniformly converted into a dimensionless relative scalar of 0 to 100 points, representing the severity of damage in that area, namely the pavement condition index (PCI standard score). 100 points represent good condition, and 0 points represent severe damage. On this basis, the system performs a reverse mapping, calculating the difference between 100 and the PCI standard score as a dimensionless deterioration scalar, which serves as a defect characteristic parameter. The values ​​are directly assigned to each graph node; the larger the value, the more severe the disease deterioration in the corresponding block.

[0032] This embodiment of the application uses this processing to uniformly map heterogeneous measurement data from different devices with different physical dimensions onto the same numerical scale. This eliminates the impact of dimensional differences on graph neural network training while retaining the relative severity information in the original measurement data, ensuring the accuracy and consistency of subsequent spatial feature aggregation and evolutionary transmission index calculation.

[0033] Figure 4 This is a block diagram illustrating a dormant edge generation module according to an exemplary embodiment. For example... Figure 4 As shown, the dormant edge generation module may include the following modules: The evolutionary transmission index calculation submodule is configured to obtain the evolutionary transmission index of the target pavement block node; The dormant edge generation submodule is configured to dynamically generate potential evolutionary association edges pointing from the target pavement block node to the expected disease type node when the evolutionary transmission index is greater than the evolutionary critical threshold, and to use the potential evolutionary association edges as the dormant edges.

[0034] For example, the dormant edge generation module can be composed of an evolutionary transmission index calculation submodule and a dormant edge generation submodule, which work together to dynamically identify potential disease spread relationships. During runtime, the evolutionary transmission index calculation submodule first obtains the necessary parameters of the target pavement block node, including the baseline evolution probability of this type of disease extracted from historical maintenance text. The current disease feature parameters output by the graph neural network. Standard permissible condition parameters for this type of pavement Disease characteristic parameters of physically adjacent node j Spatial distance between nodes i and j and the inherent hindering constant of pavement structure on the spread of disease Based on these parameters, the submodule uses a multiplicative model of the deterioration state of the central node and its physical neighborhood to calculate the evolutionary transmission index. .

[0035] The dormant edge generation submodule will use this index With the preset evolutionary critical threshold By comparison, when the evolutionary transmission index is greater than the threshold, it is determined that the physical evolutionary conditions of the neighborhood are mature. The system dynamically generates a potential evolutionary association edge in the graph from the target pavement block node to the expected disease type node, and marks the edge as a dormant edge.

[0036] This mechanism in the embodiments of this application enables the system to identify potential pathways with a risk of spreading in advance based on the multiplicative evaluation of spatial topology and deterioration rate before the actual outbreak of the disease, providing a set of candidate connections for subsequent environmental load verification.

[0037] Optionally, the evolutionary transmission index calculation submodule is configured to calculate the evolutionary transmission index based on the baseline evolution probability of the disease extracted from historical maintenance text, the disease characteristic parameters of the target pavement block node, the standard permissible state parameters of the target pavement block node, the disease characteristic parameters of the physical adjacent nodes of the target pavement block node, the spatial distance between the target pavement block node and the physical adjacent nodes, and the inherent resistance constant of the pavement structure.

[0038] For example, the evolutionary transmission index calculation submodule can use a fractional multiplication model containing multiple physically meaningful parameters to quantify the probability of disease spread. Specifically, the submodule first extracts the baseline evolution probability of this type of disease from historical maintenance data parsed by the larger model. As a fundamental term, this probability reflects the frequency of occurrence of a specific disease type in a historical statistical sense. Subsequently, the relative degradation degree of the central node itself is introduced, i.e., the ratio of the current disease feature parameter output by the graph neural network to the standard permissible state parameter. ratio The larger this ratio, the more severe the damage to the central node itself. Then multiply by the physical neighborhood influence factor, which is calculated as the average of the ratios of 1 plus the product of the adjacent node disease characteristic parameters, spatial distance, and inherent retardation constant, representing the cumulative boundary pressure of external diseases penetrating into the central node.

[0039] For example, this will be described using any one of the nodes as an example, and will not be repeated hereafter. Evolutionary transmission index It can be obtained from the following formula: Formula 1 in, For nodes The set of physical adjacency nodes, Not zero, Non-zero, N is The number of nodes in the set.

[0040] This application embodiment uses a multiplication model to systematically integrate information from four dimensions: historical statistical patterns, current state, neighborhood transmission effect, and physical resistance. The resulting evolutionary transmission index can truly reflect the spread trend of pavement defects in spatial physical topology. Compared with a simple linear weighted model, it has higher prediction accuracy and physical interpretability.

[0041] Figure 5 This is a block diagram illustrating a contact activation module according to an exemplary embodiment. For example... Figure 5 As shown, the contact activation module may include the following modules: The stress amplification factor calculation submodule is configured to construct the stress amplification factor of the airport pavement; The trigger risk value calculation submodule is configured to obtain the trigger risk value based on the evolution transmission index and the stress amplification factor; The activation submodule is configured to activate and solidify dormant potential evolutionary association edges into deterministic predicted edges when the trigger risk value is greater than the trigger confirmation threshold.

[0042] For example, the activation module can include a stress amplification factor calculation submodule, a trigger risk value calculation submodule, and an activation submodule, used to verify whether the dormant edge meets the induction conditions for the outbreak of physical disease. During runtime, the stress amplification factor calculation submodule first extracts key external parameters within the monitoring period (e.g., 30 days, the current monitoring period corresponding to the most recent 30 days), including cumulative flight equivalent load, maximum diurnal temperature range, and cumulative precipitation, and combines this with the exponential adjustment power of the large model fitting to construct a stress amplification factor based on the over-limit ratio. This factor reflects the nonlinear catalytic effect of external environment and load conditions on pavement defects.

[0043] The trigger risk value calculation submodule multiplies the evolutionary transmission index obtained during the dormant edge generation stage with the stress amplification factor in the current cycle to obtain the final trigger risk value. This product model reflects the coupling effect between the physical propagation probability and the external triggering conditions.

[0044] Example, trigger risk value It can be obtained from the following formula: Formula 2 The activation submodule will trigger the risk value. With preset trigger confirmation threshold By comparison, if and only if the trigger risk value is greater than the threshold, the system determines that the external triggering conditions are sufficient, activates and solidifies the previously dormant potential evolutionary association edges into deterministic predictive edges, and simultaneously links the early warning system.

[0045] This mechanism in the embodiments of this application effectively avoids false alarms caused by relying solely on spatial spread prediction. The system will only issue an early warning when external conditions such as weather and load are indeed severe enough to trigger a disease outbreak, thus significantly improving the reliability of decision-making.

[0046] Optionally, the stress amplification factor calculation submodule is configured to calculate the stress amplification factor based on the cumulative flight equivalent load during the monitoring period, the pavement design reference bearing capacity, the maximum diurnal temperature difference during the monitoring period, the reference temperature difference that causes thermal expansion and contraction, the cumulative precipitation during the monitoring period, the reference discharge capacity of the airport drainage system, and the exponential adjustment power.

[0047] For example, the stress amplification factor calculation submodule can use a fractional multiplication model containing multiple physical ratios to accurately quantify the comprehensive stress of the external environment. In specific calculations, the submodule first obtains the monitoring period. Cumulative flight equivalent load within integral value Compare it with the pavement design reference load capacity (This amount is for a single monitoring period) The corresponding design baseline (which can be obtained from the design manual) is used to calculate the load exceedance rate. This ratio reflects the extent to which the actual traffic load exceeds the design capacity.

[0048] Next, the temperature stress rate is calculated, and the period is... Maximum diurnal temperature range The temperature difference that causes thermal expansion and contraction In comparison, as temperature stress rate This characterizes the level of thermal stress in the material caused by temperature changes. Simultaneously, water wetting permeability is calculated, and the periodicity is considered. Cumulative precipitation Compared with the airport drainage system's benchmark discharge capacity In comparison, as a water wetting permeability This reflects the deteriorating effect of moisture infiltration on the pavement base layer. The system uses the first exponential adjustment power of the model fitting with a large load over-limit rate. The product of temperature stress rate and water wetting permeability is adjusted to a second exponential power. Then multiply the two results to obtain the final stress amplification factor. These two exponential adjustment powers are derived by fitting the large model based on the co-occurrence frequency and correlation strength between weather, load, and disease types in historical data, enabling the calculation of the stress amplification factor to adapt to the climate characteristics and operating modes of different airports.

[0049] Example, stress amplification factor It can be obtained from the following formula: Formula 3 in, Not zero, Not zero, Not zero.

[0050] The embodiments of this application achieve a refined and quantitative characterization of the impact of external environmental factors on disease outbreaks.

[0051] Figure 6 This is a block diagram illustrating a failure disconnect module according to an exemplary embodiment. For example... Figure 6 As shown, the failure disconnect module may include the following modules: The maintenance effectiveness weight calculation submodule is configured to obtain the maintenance effectiveness weight of the repair association edge connecting the target pavement block node and the maintenance plan node; The graph reconstruction submodule is configured to disconnect the repair association edge and establish a new association edge pointing to the node to be re-inspected or the node at risk of secondary deterioration when the maintenance effectiveness weight is less than the maintenance failure critical ratio.

[0052] For example, the failure disconnection module may include a maintenance effectiveness weight calculation submodule and a map reconstruction submodule, used to simulate the objective fatigue degradation process of maintenance materials over service time. During runtime, the maintenance effectiveness weight calculation submodule, for the repair-related edges connecting pavement nodes and maintenance scheme nodes, obtains the initial mechanical property constants, actual service time, nominal design life, cumulative dynamic environmental load since repair, and fatigue limit load of the maintenance material. Based on these parameters, it uses a fractional decay model to calculate the maintenance effectiveness weight at the current moment. The weight decreases smoothly over time and load in the range of 0 to 1, which truly reflects the gradual degradation of the performance of maintenance materials.

[0053] The map reconstruction submodule periodically checks this weight value. When it is below the maintenance failure threshold ratio (like When the physical material is nearing its fatigue limit, the original repair effect is essentially lost. The system then triggers a graph reconstruction command, disconnects the currently repaired relationship edges, and forcibly establishes new relationship edges pointing to the node to be re-inspected or the node at risk of secondary degradation.

[0054] This application's embodiment breaks through the limitation of setting the repaired state as permanently valid in traditional static knowledge graphs, and restores the real physical life cycle of pavement maintenance in a closed loop, ensuring that the knowledge graph can automatically identify the failure state and update it in a timely manner.

[0055] Optionally, the maintenance effectiveness weight calculation submodule is configured to calculate the maintenance effectiveness weight for the repair-related edge based on the initial mechanical property constant of the maintenance material, the actual service time of the maintenance material, the nominal design life of the maintenance material, the cumulative dynamic environmental load since the repair of the target pavement block node, and the fatigue limit load of the maintenance material, wherein the cumulative dynamic environmental load is obtained based on the stress amplification factor.

[0056] For example, the maintenance effectiveness weight calculation submodule uses a fractional model that includes the ratio of both time aging and fatigue wear to accurately calculate the remaining effective weight of the repair-related edges. The numerator of this fraction is the initial mechanical property constant of the maintenance material. (This quantity is dimensionless; if it has dimensions, it can be eliminated by dividing by a unit dimensional constant.) It characterizes the reference strength of the material immediately after repair. The denominator is composed of this constant plus the sum of the time aging rate and the fatigue wear rate, where the time aging rate is the actual service time. ( For the current moment, (Initial service time) and material nominal design life ratio The fatigue wear rate is the cumulative dynamic environmental load since the repair. Material fatigue limit load ratio The cumulative dynamic environmental load is obtained by accumulating the stress amplification factors of each cycle calculated in the connection activation module. As the service time increases and the cumulative load increases, the product term in the denominator gradually increases, causing the value of the entire fraction to gradually decrease from around 1, forming a smooth nonlinear decay curve.

[0057] It should be noted that initial mechanical property constants, nominal design life of materials, and material fatigue limit loads can be obtained through design manuals, laboratory simulation experiments, and other means.

[0058] Example: Maintenance effectiveness weight It can be obtained from the following formula: Formula 4 in, Greater than zero, and All are not less than zero.

[0059] This model in the embodiments of this application realistically simulates the environmental erosion and load fatigue effects suffered by maintenance materials during long-term service. Compared with simple linear subtraction or fixed-duration failure determination, it can more accurately reflect the differentiated degradation rate of maintenance effect under different working conditions, and provides a scientific quantitative basis for the triggering time of failure disconnection.

[0060] Figure 7 This is a block diagram illustrating yet another airport pavement maintenance inference enhancement system that integrates a large model and a graph neural network, according to an exemplary embodiment. Figure 7 As shown, the system may further include the following modules: The early warning linkage module is configured to link with an external early warning system when the connection activation module activates a dormant edge as a deterministic prediction edge.

[0061] For example, the system may also include an early warning linkage module, which works closely with the connection activation module to achieve an automated closed loop from prediction to early warning. When the connection activation module determines that the trigger risk value is greater than the trigger confirmation threshold and activates a dormant potential evolutionary correlation edge as a deterministic prediction edge, the early warning linkage module is immediately triggered. This module first receives all the information carried by the activated prediction edge, including the location and current degradation state of the source pavement block node, the target expected disease type, the evolutionary transmission index, the trigger risk value, and the external environmental factors that dominate the trigger. Subsequently, the module encapsulates the above information into a standardized early warning message and sends it to an external early warning system according to a preset communication protocol.

[0062] Based on the information received, the external early warning system can issue tiered early warning notices on channels such as the terminal building operation and maintenance screen, mobile inspection terminals, or centralized alarm platforms, reminding operation and maintenance personnel to pay attention to the potential risk of disease outbreaks in specific pavement areas and recommending preventive maintenance measures.

[0063] This application embodiment uses this linkage mechanism to enable the system's high-confidence prediction results to be converted into executable operation and maintenance instructions with extremely low latency, effectively avoiding human delays in the information transmission chain and significantly improving the response speed and execution capability of proactive maintenance of airport pavements.

[0064] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application.

Claims

1. An airport runway maintenance reasoning enhancement system fusing large models and graph neural networks, characterized in that, The system includes: The knowledge graph construction module is configured to acquire the airport pavement maintenance knowledge graph; The dormant edge generation module is configured to predict disease spread based on the node spatial topology and degradation status in the knowledge graph, so as to generate dormant edges in the knowledge graph. The contact activation module is configured to introduce multi-dimensional environmental load conditions and activate the dormant edge that meets the conditions based on the multi-dimensional environmental load conditions. The failure disconnection module is configured to disconnect the failure state edge based on the node's timeliness and cumulative fatigue decay, so as to achieve closed-loop update of the knowledge graph. The visualization module is configured to visualize the relationship between different nodes in the knowledge graph at different times. The knowledge graph construction module includes: The graph neural network submodule is configured to output the disease feature parameters of each node through graph convolution operations based on the pavement condition index score and topological distance. The large model processing submodule is configured to use a pre-trained Transformer architecture to extract the co-occurrence frequency of weather and disease types from historical inspection logs and maintenance plan texts, and fit an exponentially adjusted power. The feature parameter assignment submodule is configured to convert the damage measurement values ​​obtained by the laser pavement tester and ground penetrating radar into a dimensionless deterioration scalar based on the pavement condition index evaluation system, and assign the deterioration scalar as a defect feature parameter to the map node.

2. The airport runway maintenance reasoning enhancement system of claim 1, wherein, The dormant edge generation module includes: The evolutionary transmission index calculation submodule is configured to obtain the evolutionary transmission index of the target pavement block node; The dormant edge generation submodule is configured to dynamically generate potential evolutionary association edges pointing from the target pavement block node to the expected disease type node when the evolutionary transmission index is greater than the evolutionary critical threshold, and to use the potential evolutionary association edges as the dormant edges.

3. The airport runway maintenance reasoning enhancement system of claim 2, wherein, The evolutionary transmission index calculation submodule is configured to calculate the evolutionary transmission index based on the baseline evolution probability of the disease extracted from historical maintenance text, the disease characteristic parameters of the target pavement block node, the standard permissible state parameters of the target pavement block node, the disease characteristic parameters of the physical adjacent nodes of the target pavement block node, the spatial distance between the target pavement block node and the physical adjacent nodes, and the inherent resistance constant of the pavement structure.

4. The airport runway maintenance reasoning enhancement system of claim 2, wherein, The contact activation module includes: The stress amplification factor calculation submodule is configured to construct the stress amplification factor of the airport pavement; The trigger risk value calculation submodule is configured to obtain the trigger risk value based on the evolution transmission index and the stress amplification factor; The activation submodule is configured to activate and solidify dormant potential evolutionary association edges into deterministic predicted edges when the trigger risk value is greater than the trigger confirmation threshold.

5. The airport runway maintenance reasoning enhancement system of claim 4, wherein, The stress amplification factor calculation submodule is configured to calculate the stress amplification factor based on the cumulative flight equivalent load during the monitoring period, the pavement design reference bearing capacity, the maximum diurnal temperature difference during the monitoring period, the reference temperature difference that causes thermal expansion and contraction, the cumulative precipitation during the monitoring period, the reference discharge capacity of the airport drainage system, and the exponential adjustment power.

6. The airport runway maintenance reasoning enhancement system of claim 4, wherein, The failure disconnect module includes: The maintenance effectiveness weight calculation submodule is configured to obtain the maintenance effectiveness weight of the repair association edge connecting the target pavement block node and the maintenance plan node; The graph reconstruction submodule is configured to disconnect the repair association edge and establish a new association edge pointing to the node to be re-inspected or the node at risk of secondary deterioration when the maintenance effectiveness weight is less than the maintenance failure critical ratio.

7. The airport runway maintenance reasoning enhancement system of claim 6, wherein, The maintenance effectiveness weight calculation submodule is configured to calculate the maintenance effectiveness weight for the repair-related edge based on the initial mechanical property constant of the maintenance material, the actual service time of the maintenance material, the nominal design life of the maintenance material, the cumulative dynamic environmental load since the repair of the target pavement block node, and the fatigue limit load of the maintenance material, wherein the cumulative dynamic environmental load is obtained based on the stress amplification factor.

8. The airport runway maintenance reasoning enhancement system of claim 1, wherein, The system also includes: The early warning linkage module is configured to link with an external early warning system when the connection activation module activates a dormant edge as a deterministic prediction edge.

Images