Bridge degradation identification and maintenance decision support method and system based on knowledge graph

Abstract

The present application relates to a bridge degradation identification and maintenance decision support method and system based on a knowledge graph, belonging to the technical field of intelligent operation and maintenance of transportation infrastructure and engineering information processing, and solving the defects of weak data fusion capability, opaque causal modeling and non-traceable decision support of the prior art. The method comprises: acquiring and preprocessing multi-source heterogeneous data to generate a standardized data set; performing semantic alignment to extract candidate entities and relationships; constructing a knowledge graph ontology and loading it as a knowledge graph; using an adaptive quantile threshold to binarize the factor intensity and disease severity sequence, and fusing a one-point smoothing mutual information and a Spearman correlation coefficient to learn the causal weight; completing the bridge factor influence relationship and calculating the weight through graph reasoning; and finally calculating the bridge degradation risk score and outputting the high-confidence influence path based on the shortest path algorithm. The present application realizes multi-source data fusion and explicit modeling of the degradation causal chain, and provides a scientific basis for bridge maintenance decision-making.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance and engineering information processing technology of transportation infrastructure, specifically involving a bridge degradation identification and maintenance decision support method and system based on knowledge graph. Background Technology

[0002] As a key node in the highway network, bridges are subject to the coupling effects of environmental factors (such as freeze-thaw cycles, chloride corrosion, and humidity changes) and load factors (such as the proportion of heavy vehicles, traffic volume, and oversized vehicles) during long-term service, making them prone to various types of defects such as cracks, ruts, potholes, and spalling. Existing technologies mostly rely on statistical regression models or black-box machine learning methods for state prediction. Although they have certain predictive capabilities, they have significant shortcomings in the following aspects: (1) It is difficult to integrate multi-source heterogeneous information such as structured inspection data and unstructured report texts; (2) It is impossible to explicitly express the causal relationship chain of "external factors - defect occurrence - structural response - temporal evolution"; (3) It lacks traceable and interpretable decision support paths, making it difficult to provide engineering basis for maintenance plans.

[0003] Knowledge graph technology, through its entity-relationship-attribute triple structure, can effectively organize domain knowledge and support graph reasoning, and has already achieved complex relationship mining in multiple domains. However, for bridge engineering scenarios, there is still a lack of systematic and engineering-applicable technical solutions for how to uniformly extract and semantically align multi-source data such as road / bridge section record tables, periodic inspection reports, and PDF documents, and combine time decay, causal weight learning, and adaptive quantile threshold strategies to build a degradation impact identification model with dynamic reasoning capabilities.

[0004] Therefore, there is an urgent need for a decision support method for bridge degradation identification and maintenance that integrates multi-source data and has causal interpretability. Summary of the Invention

[0005] This invention aims to overcome the shortcomings of existing technologies, such as weak data fusion capabilities, opaque causal modeling, and lack of traceability in decision support. It proposes a knowledge graph-based method and system for identifying the impact of bridge degradation and providing decision support for maintenance. This invention achieves multi-source data fusion and explicit modeling of degradation causal chains, improving the accuracy of risk assessment and the traceability of decision support, thus providing a scientific basis for bridge maintenance decisions.

[0006] This invention is achieved through the following technical solution:

[0007] A knowledge graph-based method for bridge degradation identification and maintenance decision support includes the following steps:

[0008] Step 1: Obtain multi-source heterogeneous raw data and preprocess the data to generate a standardized dataset;

[0009] Step 2: Based on the bridge domain vocabulary and mapping rules, perform semantic alignment and extract candidate entities and candidate relations from the standardized dataset. The candidate entities include bridges, routes, administrative regions, maintenance units, inspection events, disease types, environmental factors, load factors, and maintenance measures.

[0010] Step 3: Based on the candidate entities and the candidate relationships, define semantic relationships and constraints, and construct a knowledge graph ontology. The semantic relationships include: located, managed, inspected, disease detected, exposed to, contributing to / causing, and remedial.

[0011] Step 4: Load the standardized dataset into a knowledge graph based on the knowledge graph ontology;

[0012] Step 5: For any factor and disease, calculate the adaptive quantile thresholds for the factor's intensity sequence and the disease's severity sequence, respectively, and define factor occurrence events and disease occurrence events based on the adaptive quantile thresholds; calculate the point mutual information with a smoothing effect and obtain the first weight through Sigmoid function mapping; calculate the Spearman correlation coefficient between the two sequences and linearly normalize it to [0,1] as the second weight; fuse the first weight and the second weight using a linear fusion method to generate causal weights. ;

[0013] Step 6: Perform graph reasoning and relation completion, specifically: for bridges With factors If diseases exist and Then the reasoning generates a bridge. Factors Determine the relationship of influence and calculate the corresponding weights. ,in, The inference threshold;

[0014] Step 7: Calculate the degradation risk score of the bridge and perform shortest path search based on the shortest path algorithm with negative logarithm of edge weight as the cost function, and output several high-confidence influence paths from the factors through the defects and inspection events to the bridge.

[0015] Accordingly, this invention also proposes a knowledge graph-based bridge degradation identification and maintenance decision support system, which includes:

[0016] The data acquisition and standardization module is used to acquire multi-source heterogeneous raw data, preprocess the data, and generate standardized datasets.

[0017] The semantic alignment and entity mapping module is used to perform semantic alignment based on the bridge domain vocabulary and mapping rules, and to extract candidate entities and candidate relations from the standardized dataset. The candidate entities include bridges, routes, administrative regions, maintenance units, inspection events, disease types, environmental factors, load factors, and maintenance measures.

[0018] The ontology construction module is used to define semantic relationships and constraints based on the candidate entities and the candidate relationships, and to construct a knowledge graph ontology. The semantic relationships include being located, being managed, having inspection, discovering diseases, being exposed to, causing / leading to, and being able to alleviate.

[0019] A knowledge graph loading module is used to load the standardized dataset into a knowledge graph based on the knowledge graph ontology;

[0020] The causal knowledge and weight learning module is used to calculate adaptive quantile thresholds for the intensity sequence of a factor and the severity sequence of a disease for any given factor and disease, respectively, and to define factor occurrence events and disease occurrence events based on the adaptive quantile thresholds; calculate the point mutual information with a smoothing factor and map it using the Sigmoid function to obtain the first weight; calculate the Spearman correlation coefficient between the two sequences and linearly normalize it to [0,1] as the second weight; and fuse the first weight and the second weight using a linear fusion method to generate causal weights. ;

[0021] The graph reasoning and relation completion module is used to perform graph reasoning and relation completion, specifically for bridges. With factors If diseases exist and Then the reasoning generates a bridge. Factors Determine the relationship of influence and calculate the corresponding weights. ,in, The inference threshold;

[0022] The risk assessment and path interpretation module is used to calculate the degradation risk score of the bridge and perform shortest path search based on the shortest path algorithm with negative logarithm of edge weight as the cost function, outputting several high-confidence impact paths from factors through defects and inspection events to the bridge.

[0023] This invention constructs a structured knowledge graph in the bridge field, integrates multi-source heterogeneous data, learns the causal weights between factors and defects, and performs forward chain reasoning and time decay modeling to generate interpretable degradation risk scores and impact paths, achieving closed-loop support of "data → knowledge → decision-making" and improving the scientific, accurate and engineering practicality of bridge maintenance. Attached Figure Description

[0024] Figure 1 This is a flowchart of the knowledge graph-based bridge degradation identification and maintenance decision support method described in an embodiment of the present invention;

[0025] Figure 2 A heatmap showing the causal weights of factors on the disease;

[0026] Figure 3 A schematic diagram of the three-layer knowledge graph structure for the target year;

[0027] Figure 4 A bar chart for bridge degradation risk scoring;

[0028] Figure 5 A network diagram showing the impact of factors on bridges. Detailed Implementation

[0029] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings and preferred embodiments.

[0030] This invention provides a knowledge graph-based method for bridge degradation identification and maintenance decision-making. Its core process mainly includes data standardization, semantic alignment, ontology construction, graph loading, causal weight learning, graph reasoning, and risk assessment.

[0031] Step 1 (S101): Obtain multi-source heterogeneous raw data and perform standardization processing to generate a standardized dataset.

[0032] The multi-source heterogeneous raw data obtained in this step includes:

[0033] The road section or bridge section record form includes route code, chainage interval, technical grade, pavement type and width, MQI (Pavement Performance Index), PCI (Pavement Condition Index), quantitative indicators and classification information of defects such as cracks, ruts, potholes, and subsidence.

[0034] The bridge periodic inspection checklist includes the year, bridge name or bridge code, center pile number, classification by span, rating grade and score, administrative code, bridge length and total span length, and maintenance unit.

[0035] The PDF inspection report includes text and images, containing unstructured descriptions, disease cause analysis, and on-site photos.

[0036] The above-mentioned multi-source heterogeneous raw data are preprocessed. The preprocessing process includes:

[0037] Outlier handling, missing value completion, unit conversion and dimension unification, primary key alignment and disambiguation are performed on road or bridge section record tables and bridge periodic inspection tables to form structured and standardized data. Among them, Z-score or quantile method is used to identify and remove outliers; missing values ​​are completed using a combination of linear interpolation, mean imputation, and forward chronological imputation strategies; unit conversion and dimension unification are performed, such as converting area, length, and quantity to "per kilometer" standard; primary key alignment is performed with bridge code as primary key and route code and station interval as secondary key, and disambiguation is performed for homonymous or heteronymous codes.

[0038] Unstructured information is extracted from the text and figures of the PDF inspection report to generate candidate triples that have been screened with confidence. The process of extracting unstructured information from the text and figures of the PDF inspection report includes: layout parsing and table extraction, optical character recognition, terminology standardization and synonym merging, and extraction of "disease-location-measurement-cause" quadruples based on regular expressions and named entity recognition. After confidence screening and manual verification, high-quality candidate triples are generated.

[0039] Finally, the structured and standardized data are combined with the candidate triples to form a unified standardized dataset.

[0040] Step 2 (S102): Semantic alignment is performed based on the bridge domain vocabulary and mapping rules, and candidate entities and candidate relationships are extracted from the standardized dataset.

[0041] This step extracts candidate entities and candidate relationships from a standardized dataset based on a pre-built or configurable bridge domain lexicon and mapping rules. The extracted candidate entities include, but are not limited to, bridges, routes, administrative regions, maintenance units, inspection events, disease types, environmental factors, load factors, and maintenance measures.

[0042] Step 3 (S103): Based on the extracted candidate entities and candidate relations, define semantic relations and constraints, and construct the knowledge graph ontology.

[0043] Using candidate entities as core entities, a knowledge graph ontology for the bridge domain is constructed by defining semantic relationships and constraints. The semantic relationships include:

[0044] Location: Spatial affiliation between bridges and routes or between bridges and administrative regions;

[0045] Maintenance and upkeep: The relationship of responsibility between the bridge and the maintenance and upkeep unit;

[0046] It includes an inspection of the relationship between the bridge and the inspection event;

[0047] Disease detection: Examine the observed relationship between events and disease types or between bridges and disease types, including severity or quantity attributes;

[0048] Exposure to (or having factors): the exposure relationship between bridges, routes (or administrative regions) and environmental and load factors, along with intensity attributes;

[0049] Contributing / Causing: The causal relationship between factors and disease types;

[0050] It can alleviate the relationship between the type of disease and maintenance measures.

[0051] Step 4 (S104): Load the standardized dataset into a knowledge graph based on the constructed knowledge graph ontology.

[0052] Based on the knowledge graph ontology constructed in step 3, the standardized dataset is loaded into the knowledge graph. This process includes: creating nodes to represent entities and establishing edges to represent relationships. Specifically, inspection events are modeled using a combination of "bridge identifier + year," linked to indicators such as the MQI, PCI, and rating level of the current year; the severity or quantity of diseases is stored as edge attributes or instance attributes. All nodes and edges carry timestamps and data source metadata.

[0053] Step 5 (S105): Learn causal knowledge and assign weights based on standardized data. Specifically, for any factor and disease, calculate the adaptive quantile thresholds for the factor's intensity sequence and the disease's severity sequence, and define factor occurrence events and disease occurrence events according to the adaptive quantile thresholds; calculate the point mutual information with one-smoothness. The first weights are obtained by mapping using the Sigmoid function; the Spearman correlation coefficient between the two sequences is then calculated. The first and second weights are linearly normalized to [0,1] and used as the second weight; a linear fusion method is used to fuse the first and second weights to generate causal weights. .

[0054] This step involves learning the causal weights between factors and diseases based on standardized data to quantify their impact strength. The specific process is as follows:

[0055] For any factor Diseases Using each factor (environmental factor and load factor) and each type of disease as the statistical object, the intensity sequence of each factor on different bridges and in different years was obtained. And to obtain the severity sequence of each disease on different bridges and in different years. ,in, For bridge signage, A collection of bridges; Indicates the year. A set of years; For a set of factors; This is a collection of diseases.

[0056] (1) Calculate the adaptive quantile threshold and perform binarization:

[0057] For each factor Calculate its intensity sequence quantiles For each disease Calculate its severity sequence quantiles .

[0058] Using quantiles as adaptive quantile thresholds, respectively based on the adaptive quantile thresholds Define the event as: ;

[0059] Based on adaptive quantile threshold The disease occurrence event is defined as: .

[0060] Adaptive quantile thresholds can better adapt to the differences in data distribution across different bridges and years, reducing binarization bias caused by different distributions.

[0061] (2) Calculate the point mutual information (PMI) with one-smoothness added. :

[0062] Let the total number of samples be ,calculate:

[0063] (1);

[0064] Introducing plus-one smoothing:

[0065] (2);

[0066] Add a smooth point mutual information The definition of is:

[0067] (3);

[0068] in, This represents the number of samples in the bridge set; This represents the number of samples in the year set; This represents the number of samples in the factor set, which is the complete set including environmental factors and load factors. This represents the number of samples in the disease set; Indicator Factors The number of times the intensity of a bridge exceeds its adaptive quantile threshold across all bridges and years; Indicates disease The severity of the sample number that exceeds its adaptive quantile threshold across all bridges and years; This indicates that within the same bridge and the same year, factors... Diseases The number of samples that simultaneously exceed their respective adaptive quantile thresholds; Indicator Factors The marginal probability of occurrence; Indicates disease The marginal probability of occurrence; Indicator Factors Diseases Joint probability of co-occurrence.

[0069] (3) Calculate the Spearman correlation coefficient and combine the weights:

[0070] Calculation factors intensity sequence Diseases Severity sequence Spearman correlation coefficient and linearly normalized to the interval : Thus, the second weight is obtained.

[0071] Let the fusion coefficient be set The preferred value is 0.4, and the causal weights are generated using a linear fusion method. Causal weight The calculation formula is:

[0072] (4);

[0073] in, The Sigmoid function is used to map point mutual information with a smoothing factor to... The first weight is obtained from the interval.

[0074] Since point mutual information captures co-occurrence patterns and reflects co-occurrence strength, while the Spearman correlation coefficient reflects monotonic trends, fusing point mutual information and the Spearman correlation coefficient using the fusion strategy shown in formula (4) can enhance the causal weight. The robustness and stability of the causal weights obtained after fusion More stable and more evenly distributed.

[0075] Step 6 (S106): Perform graph reasoning and relation completion.

[0076] This step performs forward chain reasoning to complete the relationship of "factors affecting the bridge". Specifically, for the bridge... With factors When the bridge Diseases exist When recording observations, i.e., the bridge Diseases exist and satisfy Then the reasoning generates a bridge. Factors Determine the relationship of influence and calculate the corresponding influence weights. .in The inference threshold is preferably 0.2 to 0.4.

[0077] For each bridge Diseases Obtain the severity of its most recent observation. Time interval from the current evaluation time The severity is then normalized to a minimum-maximum value based on the disease type, using the following formula:

[0078] (5);

[0079] in, Indicates the severity of normalization. Bridge Diseases The severity of the most recent observation, and These represent the maximum and minimum values ​​for the same type of bridge defects, respectively.

[0080] If the denominator in formula (5) is zero (i.e., all bridges have the same severity, which is a constant series), then the uniform order is... To avoid all values ​​being zero.

[0081] when At that time, the disease accumulated. Calculate the weights of the contributions of "factor effects" and consider time decay. .

[0082] Weight The calculation formula is as follows:

[0083] (6);

[0084] in, For bridges A collection of diseases; This represents the time decay coefficient, used to control the rate at which the effects of historical diseases decay over time. Time decay coefficient The value can be set between 0 and 1, depending on the inspection cycle and corrosion development rate, and is preferably between 0.05 and 0.15.

[0085] If all If all values ​​are zero, the inference threshold will be automatically lowered. For example, the inference threshold Reduce the value to half of its original value, then recalculate the weights. If it is still zero, then cancel the inference threshold. We directly sum the results (as a fallback) to ensure that subsequent scoring and visualization are not blank, thus ensuring stable output.

[0086] Step 7 (S107): Conduct a bridge degradation risk assessment and generate interpretable impact paths.

[0087] (1) Degradation risk assessment:

[0088] bridge Degradation risk score The calculation formula is:

[0089] (7);

[0090] in, To affect the bridge A set of factors; Indicator Factors The global weights are obtained by minimizing and maximizing the average strength of each bridge corresponding to the factor in the target year; when the normalization results are equal, i.e., all factor weights are equal (constant), the weights revert to uniform weights. .

[0091] (2) Explainable influence path:

[0092] Construct a three-layer directed graph consisting of "factors → diseases → inspection events → bridges".

[0093] To ensure consistency in the weights of different semantic relations during graph reasoning, edge weights are defined for all edges in the knowledge graph according to the rules. The rules include: the boundary weights between factors and diseases are determined by causal weights. Normalized severity of the boundary weights between disease and inspection events. The edge weight between the event and the bridge is set to a unit weight of 1 to represent the structural membership relationship without introducing additional influence bias.

[0094] negative logarithm of edge weights As the cost function of the shortest path algorithm, for bridges Using each factor as the source and the bridge as the sink, shortest path algorithms such as Dijkstra's are employed to search for the path from each factor through defects and inspection events to the bridge. The minimum cost path is identified, and the top-ranked high-confidence influence paths are output to form a traceable chain of evidence.

[0095] The method of this invention constructs a structured knowledge graph in the bridge domain, integrates multi-source heterogeneous data, learns the causal weights between factors and defects, and performs forward chain reasoning and time decay modeling to generate interpretable degradation risk scores and impact paths. This achieves closed-loop support from "data → knowledge → decision-making," improving the scientific rigor, accuracy, and engineering practicality of bridge maintenance. This method also has the following advantages:

[0096] (1) Robust causal learning: The adaptive quantile threshold combined with the fusion of PMI and Spearman correlation coefficient effectively alleviates the bias caused by the differences in data distribution between different bridges and different years;

[0097] (2) Engineering Explainable: The degradation risk score consists of factors-disease-time decay, and is combined with the shortest path output evidence chain, which facilitates scheme demonstration and on-site verification;

[0098] (3) Adaptive and fallback strategies: threshold self-degradation, constant column rollback, and uniform weight rollback ensure that stable results can still be produced in the case of sparse or "homogeneous" data.

[0099] To more intuitively demonstrate the implementation process and technical effects of the above method, the present invention will be described below with reference to a specific example.

[0100] In this example, the data is organized as follows:

[0101] Factor intensity dictionary: `exposure[(bridge ID, year, factor name)]∈[0,1]`;

[0102] Disease severity dictionary: `severity[(bridge ID, year, disease name)]∈[0,1]`;

[0103] Example entity collection: Bridge collection Year set Factor set Disease collection .

[0104] Next, the preprocessing method from step 1 is used to perform outlier removal (Z-score or quantile method), missing value completion (linear interpolation, mean of similar values, forward time-series filling), unit conversion and standardization of dimensions (area, length, and quantity are standardized to "per kilometer"), and primary key alignment and disambiguation (bridge code as primary key, route code and station number as secondary keys). After processing, numerical sequences of two standardized dimensions, `exposure` and `severity`, are obtained.

[0105] Then, steps 2 through 7 are performed to obtain the bridge's degradation risk score and explainable impact paths.

[0106] Figure 2 This is a heatmap showing the causal weights of factors on diseases. The horizontal axis represents disease type, and the vertical axis represents environmental or load factors. The cells display the merged results. And the corresponding PMI value;

[0107] Each cell displays the merged causal weights. And the corresponding PMI value, with colors ranging from light to dark representing causal weights. From small to large. From Figure 2 It is evident that combinations such as "heavy vehicle - ruts" and "chloride exposure - flaking" have high causal weights. This aligns with engineering principles.

[0108] Figure 3 The diagram illustrates the structure of a three-layer knowledge graph for the target year (2025). The leftmost column represents factor nodes, the middle column represents disease nodes and inspection event nodes (displayed by multiple short labels in the format "Insp + bridge identifier + year", where "Insp" is an abbreviation for Inspection), and the rightmost column represents bridge nodes. Figure 3 It visually demonstrates the multi-hop semantic link "from factor to bridge".

[0109] Figure 4 This is a bar chart showing the degradation risk score of each bridge. Numerical values ​​are used to visually compare the level of degradation risk. Figure 4 The horizontal axis represents the bridge, and the vertical axis represents the degradation risk score. If the data is constant or sparse, a strategy of automatically decreasing the inference threshold, backtracking the constant column, and uniformly backtracking the weights is employed to ensure that the histogram is readable and not blank. High-risk bridges correspond to... Figure 5 It has more high-weight connections.

[0110] Figure 5 This is a network diagram showing the impact of factors on bridges. Figure 5In the diagram, the left column represents factors, and the right column represents bridges. The directed edges from factors to bridges represent the "affected" relationship, and the width of the directed edges is determined by weight. Normalized display, weights The values ​​are labeled as edge labels (not shown in the figure). Figure 5 Used to identify the main influencing factors of each bridge and to provide a basis for maintenance strategies.

[0111] Finally, the console synchronously prints several shortest interpretation paths for each bridge, for example: "Freeze-thaw → Crack → → B1 (no arrow symbol is used in the text output, only conjunctions are used). These paths correspond to Figure 3 The connections within form a complete chain of evidence.

[0112] In another embodiment, the present invention also provides a knowledge graph-based bridge degradation identification and maintenance decision support system, including a data acquisition and standardization module, a semantic alignment and entity mapping module, an ontology construction module, a knowledge graph loading module, a causal knowledge and weight learning module, a graph reasoning and relation completion module, and a risk assessment and path interpretation module.

[0113] The data acquisition and standardization module is used to acquire multi-source heterogeneous raw data and preprocess the data to generate standardized datasets.

[0114] The multi-source heterogeneous raw data acquired by the data acquisition and standardization module includes:

[0115] The road section or bridge section record form includes route code, chainage interval, technical grade, pavement type and width, MQI (Pavement Performance Index), PCI (Pavement Condition Index), quantitative indicators and classification information of defects such as cracks, ruts, potholes, and subsidence.

[0116] The bridge periodic inspection checklist includes the year, bridge name or bridge code, center pile number, classification by span, rating grade and score, administrative code, bridge length and total span length, and maintenance unit.

[0117] The PDF inspection report includes text and images, containing unstructured descriptions, disease cause analysis, and on-site photos.

[0118] The data acquisition and standardization module preprocesses the aforementioned multi-source heterogeneous raw data. The preprocessing process includes:

[0119] Outlier handling, missing value completion, unit conversion and dimension unification, primary key alignment and disambiguation are performed on road or bridge section record tables and bridge periodic inspection tables to form structured and standardized data. Among them, Z-score or quantile method is used to identify and remove outliers; missing values ​​are completed using a combination of linear interpolation, mean imputation, and forward chronological imputation strategies; unit conversion and dimension unification are performed, such as converting area, length, and quantity to "per kilometer" standard; primary key alignment is performed with bridge code as primary key and route code and station interval as secondary key, and disambiguation is performed for homonymous or heteronymous codes.

[0120] Unstructured information is extracted from the text and figures of the PDF inspection report to generate candidate triples that have been screened with confidence. The process of extracting unstructured information from the text and figures of the PDF inspection report includes: layout parsing and table extraction, optical character recognition, terminology standardization and synonym merging, and extraction of "disease-location-measurement-cause" quadruples based on regular expressions and named entity recognition. After confidence screening and manual verification, high-quality candidate triples are generated.

[0121] Finally, the data acquisition and standardization module combines the structured and standardized data with the candidate triples to form a unified standardized dataset.

[0122] The semantic alignment and entity mapping module is used to perform semantic alignment based on a pre-built or configurable bridge domain lexicon and mapping rules, and to extract candidate entities and candidate relationships from a standardized dataset. Candidate entities include, but are not limited to, bridges, routes, administrative regions, management units, inspection events, disease types, environmental factors, load factors, and maintenance measures.

[0123] The ontology construction module is used to define semantic relationships and constraints based on candidate entities and candidate relationships extracted by the semantic alignment and entity mapping module, and to construct a knowledge graph ontology. The semantic relationships include being located, being maintained, having inspection, discovering diseases, being exposed to, contributing to / causing, and being remediable.

[0124] The knowledge graph loading module loads standardized datasets into a knowledge graph based on the knowledge graph ontology. The module uses the knowledge graph ontology constructed by the ontology building module to load the standardized datasets into a knowledge graph. This process includes creating nodes to represent entities and establishing edges to represent relationships. Specifically, inspection events are modeled using a combination of "bridge identifier + year," and linked to indicators such as MQI, PCI, and rating level for that year. The severity or quantity of diseases are stored as edge attributes or instance attributes. All nodes and edges carry timestamps and data source metadata.

[0125] The causal knowledge and weight learning module is used to perform the operation in step 5 of the aforementioned method, namely, for any factor and disease, it calculates the adaptive quantile thresholds for the intensity sequence of the factor and the severity sequence of the disease, and defines the factor occurrence event and the disease occurrence event according to the adaptive quantile thresholds; it calculates the point mutual information with a smoothing effect and obtains the first weight by mapping it through the Sigmoid function; it calculates the Spearman correlation coefficient of the two sequences and linearly normalizes it to [0,1] as the second weight; and it fuses the first weight and the second weight using a linear fusion method to generate the causal weight. .

[0126] The graph reasoning and relation completion module is used to perform the graph reasoning and relation completion in step 6 of the aforementioned method, specifically: for bridges With factors If diseases exist and Then the reasoning generates a bridge. Factors Determine the relationship of influence and calculate the corresponding weights. ,in, This is the inference threshold.

[0127] The risk assessment and path interpretation module is used to perform the operation in step 7 of the aforementioned method, namely, to calculate the degradation risk score of the bridge, and to perform shortest path search based on the shortest path algorithm with negative logarithm of edge weight as the cost function, and output several high-confidence impact paths from the factors through the defects and inspection events to the bridge.

[0128] This invention can be directly deployed in highway bridge management information systems, supporting the automatic generation of knowledge graphs from historical inspection data and annual inspection reports. It outputs root cause identification, risk ranking, and evidence pathways to assist in the formulation of annual maintenance plans and specific treatment solutions. This invention features adaptive and fallback mechanisms for data quality, demonstrating good engineering robustness and widespread applicability.

[0129] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0130] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A knowledge graph-based method for bridge degradation identification and maintenance decision support, characterized in that, Includes the following steps: Step 1: Obtain multi-source heterogeneous raw data and preprocess the data to generate a standardized dataset; Step 2: Based on the bridge domain vocabulary and mapping rules, perform semantic alignment and extract candidate entities and candidate relations from the standardized dataset. The candidate entities include bridges, routes, administrative regions, maintenance units, inspection events, disease types, environmental factors, load factors, and maintenance measures. Step 3: Based on the candidate entities and the candidate relationships, define semantic relationships and constraints, and construct a knowledge graph ontology. The semantic relationships include: located, managed, inspected, disease detected, exposed to, contributing to / causing, and remedial. Step 4: Load the standardized dataset into a knowledge graph based on the knowledge graph ontology; Step 5: For any factor and disease, calculate the adaptive quantile thresholds for the intensity sequence of the factor and the severity sequence of the disease, and define the occurrence events of the factor and the occurrence events of the disease according to the adaptive quantile thresholds; calculate the point mutual information with a smoothing factor and obtain the first weight by mapping it through the Sigmoid function; calculate the Spearman correlation coefficient of the two sequences and linearly normalize it to [0,1] as the second weight. The first weight and the second weight are fused using a linear fusion method to generate causal weights. ; Step 6: Perform graph reasoning and relation completion, specifically: for bridges With factors If diseases exist and Then the reasoning generates a bridge. Factors Determine the relationship of influence and calculate the corresponding weights. ,in, The inference threshold; Step 7: Calculate the degradation risk score of the bridge and perform shortest path search based on the shortest path algorithm with negative logarithm of edge weight as the cost function, and output several high-confidence influence paths from the factors through the defects and inspection events to the bridge.

2. The bridge degradation identification and maintenance decision support method based on knowledge graphs according to claim 1, characterized in that, In step 1, the multi-source heterogeneous raw data includes road or bridge section record sheets, bridge periodic inspection forms, and PDF periodic inspection reports with attached figures; The road section or bridge section record form includes quantitative indicators and classification information for route code, chainage interval, technical grade, pavement type and width, MQI, PCI, cracking disease, rutting disease, pothole disease, and subsidence disease. The bridge periodic inspection checklist includes the year, bridge name or bridge code, center pile number, classification by span, rating grade and score, administrative code, bridge length and total span length, and maintenance unit. The PDF routine inspection report text and accompanying images include unstructured descriptions, disease cause analysis, and on-site photos.

3. The bridge degradation identification and maintenance decision support method based on knowledge graphs according to claim 2, characterized in that, Data preprocessing includes outlier handling, missing value completion, unit conversion and dimension unification, primary key alignment and disambiguation of road or bridge section record tables and bridge periodic inspection tables, forming structured and standardized data. Outlier handling uses Z-score or quantile method; missing value completion uses a combination of linear interpolation, mean imputation, and forward time-series imputation strategies; unit conversion and dimension unification convert area, length, and quantity to "per kilometer" standard; primary key alignment is performed with bridge code as primary key and route code and station interval as secondary key, and disambiguation is performed for homonymous or heteronymous codes. Data preprocessing also includes unstructured information extraction from the PDF inspection report text and figures, including: layout parsing and table extraction, optical character recognition, terminology standardization and synonym merging, and extraction of "disease-location-measurement-cause" quadruples based on regular expressions and named entity recognition; after confidence screening and manual verification, candidate triples are generated. The structured and standardized data is combined with the candidate triples to obtain the standardized dataset.

4. The bridge degradation identification and maintenance decision support method based on knowledge graphs according to claim 1, characterized in that, In step 4, loading the knowledge graph includes creating nodes and establishing edges. The inspection event is modeled based on the combination of "bridge identifier + year" and linked to the road surface performance index, road surface condition index and rating level of the current year. The severity or quantity of defects is stored as edge attributes or instance attributes. All nodes and edges in the knowledge graph carry timestamps and data source metadata.

5. The bridge degradation identification and maintenance decision support method based on knowledge graphs according to claim 1, characterized in that, In step 5, the event of factor occurrence is defined as: The disease occurrence event is defined as: ;in, and To achieve adaptive quantile thresholds, factors are calculated separately. intensity sequence quantiles and diseases Severity sequence The quantiles were obtained.

6. The bridge degradation identification and maintenance decision support method based on knowledge graphs according to claim 1, characterized in that, Causal weight The calculation formula is: （4）； in, For the Sigmoid function, To add a smooth point mutual information, The Spearman correlation coefficient is... This is the fusion coefficient.

7. The bridge degradation identification and maintenance decision support method based on knowledge graphs according to claim 1, characterized in that, Weight The calculation formula is: （6）； in, For bridges Diseases Normalized severity after minimum-maximum normalization of severity according to disease type; For bridges The most recent observation of the disease Severity The time interval from the current evaluation time; This is the time decay coefficient; For bridges A collection of diseases.

8. The bridge degradation identification and maintenance decision support method based on knowledge graphs according to claim 7, characterized in that, bridge Degradation risk score The calculation formula is: （7）； in, To affect the bridge A set of factors; Indicator Factors The global weight is obtained by minimizing and maximizing the average intensity of the corresponding factor for each bridge in the target year; when the normalization results are equal, it reverts to uniform weight. .

9. The bridge degradation identification and maintenance decision support method based on knowledge graphs according to claim 1, characterized in that, The edge weights of all edges in the knowledge graph are defined according to the following rules: the edge weights between factors and diseases are determined by causal weights. Normalized severity of the boundary weights between disease and inspection events. The edge weight between the inspection event and the bridge is set to a unit weight of 1.

10. A bridge degradation identification and maintenance decision support system based on knowledge graphs, characterized in that, include: The data acquisition and standardization module is used to acquire multi-source heterogeneous raw data, preprocess the data, and generate standardized datasets. The semantic alignment and entity mapping module is used to perform semantic alignment based on the bridge domain vocabulary and mapping rules, and to extract candidate entities and candidate relations from the standardized dataset. The candidate entities include bridges, routes, administrative regions, maintenance units, inspection events, disease types, environmental factors, load factors, and maintenance measures. The ontology construction module is used to define semantic relationships and constraints based on the candidate entities and the candidate relationships, and to construct a knowledge graph ontology. The semantic relationships include being located, being managed, having inspection, discovering diseases, being exposed to, causing / leading to, and being able to alleviate. A knowledge graph loading module is used to load the standardized dataset into a knowledge graph based on the knowledge graph ontology; The causal knowledge and weight learning module is used to calculate the adaptive quantile thresholds of the intensity sequence of the factor and the severity sequence of the disease for any factor and disease, respectively, and define the occurrence events of the factor and the occurrence events of the disease according to the adaptive quantile thresholds; calculate the point mutual information with a smoothing factor and obtain the first weight by mapping it through the Sigmoid function; calculate the Spearman correlation coefficient of the two sequences and linearly normalize it to [0,1] as the second weight. The first weight and the second weight are fused using a linear fusion method to generate causal weights. ; The graph reasoning and relation completion module is used to perform graph reasoning and relation completion, specifically for bridges. With factors If diseases exist and Then the reasoning generates a bridge. Factors Determine the relationship of influence and calculate the corresponding weights. ,in, The inference threshold; The risk assessment and path interpretation module is used to calculate the degradation risk score of the bridge and perform shortest path search based on the shortest path algorithm with negative logarithm of edge weight as the cost function, outputting several high-confidence impact paths from factors through defects and inspection events to the bridge.

Images