A Bridge Construction Quality Perception and Diagnosis Method Based on Knowledge Graph and BIM

Abstract

This invention proposes a method and system for bridge construction quality perception and diagnosis based on knowledge graphs and BIM, belonging to the field of intelligent bridge construction quality diagnosis technology. Addressing the problems of fragmented multi-source data, inability to predict dust pollution in advance, missing causal reasoning chains, and difficulties in root cause diagnosis, this invention integrates BIM static data, environmental monitoring data, and prior causal rules to construct a knowledge graph and establish a PDCA quality information loop network. During the inspection phase, the LightGBM prediction model and physical settlement model are used to predict the dust adhesion coverage on the surface of the closure components. Two-level reasoning is performed using causal rules and Bayesian networks to output a closure quality index. When the index is below a threshold, the SHAP interpreter and counterfactual reasoning are used to diagnose the root causes of environmental factors and generate treatment suggestions. This invention improves the timeliness of dust risk early warning and the accuracy of quality diagnosis, enhancing the level of intelligent construction quality management.

Description

Technical Field

[0001] This invention relates to the field of bridge construction technology, and in particular to a method for bridge construction quality perception and diagnosis based on knowledge graphs and BIM. Background Technology

[0002] With the continuous advancement of transportation infrastructure construction, long-span bridges have become crucial nodes in modern transportation networks. Bridge closure, as the core process of force system transformation in the entire bridge construction, directly determines the smoothness of the bridge's alignment, structural internal forces, and long-term durability after completion. Closure construction is highly sensitive to on-site environmental conditions. Environmental factors such as wind, precipitation, sunshine, air particulate matter concentration, and the construction location can easily cause dust pollution at the closure joint, thus affecting welding quality, the bond strength between new and old concrete surfaces, and the sliding performance of the bearings. Therefore, dynamically sensing environmental risks, accurately predicting the degree of dust pollution, and promptly diagnosing potential quality hazards during closure construction has become an important development direction for improving the intelligence level of bridge construction.

[0003] Existing methods for addressing the environmental dust impact during closure construction have significant shortcomings: First, BIM (Building Information Modeling) static data and dynamic environmental monitoring data are fragmented, lacking a unified knowledge representation and correlation analysis framework, making it difficult to integrate and utilize multi-source heterogeneous information; Second, existing methods cannot predict the degree of dust pollution on the surface of sensitive closure components based on real-time meteorological and location data, resulting in quality risks only being passively discovered after pollution occurs; Third, there is a lack of a causal reasoning chain from dust pollution to joint quality deterioration and then to the impact on the overall structural performance, making it impossible to output a quantitative closure quality index; Fourth, once quality risks are identified, existing technologies cannot automatically diagnose the main environmental factors causing the risks and their degree of contribution, making it difficult to provide accurate decision-making basis for on-site control.

[0004] Therefore, there is an urgent need for an integrated intelligent method that can combine BIM, dynamic environmental monitoring, and prior knowledge to achieve dynamic perception, prediction, diagnosis, and root cause analysis of closure quality. Summary of the Invention

[0005] To address these issues, this invention provides a bridge construction quality perception and diagnosis method based on knowledge graphs and BIM, which overcomes the problems of fragmented multi-source heterogeneous data, inability to predict dust pollution levels in advance, lack of causal reasoning chain from pollution to structural performance, and inability to automatically diagnose the root causes of environmental factors in existing technologies.

[0006] To achieve the above objectives, this invention provides a method for bridge construction quality perception and diagnosis based on knowledge graphs and BIM, comprising: S1: Obtain BIM static data, environmental dynamic monitoring data and prior knowledge of the bridge closure section, and extract attribute information, time-series monitoring characteristics and causal rules of environmental impact of the closure sensitive components. S2, using the closure-sensitive components as nodes and the attribute information and time-series monitoring features as attributes, construct a knowledge graph; S3. Based on the knowledge graph, a PDCA quality information loop network is established. During the inspection phase of the PDCA quality information loop network, a pre-trained prediction model is used to predict the degree of dust pollution on the surface of the closure sensitive components, and the closure quality index is output according to the prediction results and the preset quality diagnosis rules. S4. When the closure quality index is lower than the preset quality threshold, root cause diagnosis of environmental factors is performed, treatment suggestions are generated, and the treatment suggestions are pushed to the processing stage of the PDCA network.

[0007] Furthermore, the closure-sensitive components include closure segment components, supports, stiffening frames, prestressed ducts, and construction joints; the environmental dynamic monitoring data includes wind monitoring data, precipitation monitoring data, solar radiation intensity detection data, and the construction location; the attribute information of the closure-sensitive components includes geometric attributes, spatial location, and permissible construction environment parameters, which are obtained by analyzing the IFC model and combining it with the construction specification database.

[0008] Furthermore, using the closure-sensitive components as nodes and the aforementioned attribute information and time-series monitoring features as attributes, a knowledge graph is constructed, including: S21, instantiate the closure segment components, supports, stiffening frames, prestressed ducts and construction joints as solid nodes respectively, and use the geometric properties, spatial positions and allowable construction environment parameters as the properties of the corresponding solid nodes; S22 instantiates wind monitoring data, precipitation monitoring data, solar radiation intensity detection data, and the construction location as environmental factor nodes, and uses time-series monitoring characteristics as the dynamic attributes of the corresponding environmental factor nodes. S23, calculate the Euclidean distance between the monitoring point location and the working area on the surface of the closure sensitive component. When the Euclidean distance is less than a preset spatial threshold, establish a spatial binding relationship edge between the entity node and the environmental factor node. By aligning and matching the timestamp of the monitoring data with the time window of the closure construction process, establish a time alignment relationship edge between the entity node and the environmental factor node. Based on the causal rules extracted from prior knowledge, establish a causal relationship edge from the environmental factor node to the quality risk event node. S24 is a knowledge graph for diagnosing the quality of closure construction, consisting of entity nodes, environmental factor nodes, quality risk event nodes, as well as spatial binding relationship edges, time alignment relationship edges, and causal relationship edges.

[0009] Furthermore, in S3, the PDCA quality information cycle network includes four stages: planning, execution, inspection, and processing. The planning stage extracts initial benchmark thresholds from the knowledge graph, which include the allowable wind speed threshold for welding, the surface dust coverage threshold, and the design closure temperature window. The execution stage continuously updates the knowledge graph with real-time collected environmental dynamic monitoring data. The processing stage generates control instructions including suspending welding or pouring operations, installing dustproof sheds, watering or covering exposed ground upwind with dustproof nets, and adjusting at least one of the following within the closure locking time window:

[0010] Furthermore, in the inspection phase of the PDCA quality information loop network, a pre-trained prediction model is used to predict the degree of dust pollution on the surface of the closure-sensitive components, including: S31. Extract the dynamic attributes of environmental factor nodes at the current moment and within the historical window from the knowledge graph to form an input feature vector; the input feature vector includes 10-minute average wind speed, gust wind speed, the angle between wind direction and the location of the exposed sand source, relative humidity, cumulative rainfall in the past N hours, solar radiation intensity, near-surface vertical temperature gradient, closure height, surrounding exposed construction area, and underlying surface type code. S32, input the input feature vector into a pre-trained multi-factor spatiotemporal prediction model, wherein the multi-factor spatiotemporal prediction model is a LightGBM model; S33, the total suspended particulate matter concentration on the surface of the closure sensitive component working area within a preset time period is output by the multi-factor spatiotemporal prediction model, and is used as a quantitative indicator of the degree of dust pollution.

[0011] Furthermore, based on the prediction results and preset quality diagnosis rules, the closure quality index is output, including: S34, input the total suspended particulate matter concentration, real-time wind speed and real-time relative humidity into the physical settling model to obtain the dust adhesion coverage on the surface of the closure sensitive component; wherein, the physical settling model calculates the dust adhesion mass per unit area based on the relationship between particulate matter settling flux and concentration, wind speed and relative humidity in the air. S35. Based on the dust adhesion coverage, and combined with the causal rules stored in the knowledge graph, a two-level quality inference is performed: First, the dust adhesion coverage is compared with the surface dust coverage threshold, and the real-time wind speed is compared with the allowable wind speed threshold for welding to determine the joint quality risk level; Second, with the joint quality risk level as input, the probability of risk propagating to insufficient structural locking stiffness and abnormal bridge alignment is evaluated based on a Bayesian network to obtain the probability of closure quality deterioration. S36. Based on the comprehensive joint quality risk level and the probability of closure quality deterioration, a quantified closure quality index is generated and output.

[0012] Furthermore, the training process of the pre-trained multi-factor spatiotemporal prediction model includes: S321, Obtain historical training dataset, which includes total suspended particulate matter concentration measured at the same height as the closure face during multiple historical periods as labels, and historical feature data corresponding to the input feature vector collected synchronously. S322, perform data preprocessing on historical feature data and labels, including missing value imputation, outlier removal and feature normalization; S323, Build a LightGBM regression model and set the model hyperparameters, including learning rate, tree depth, number of leaf nodes and number of iterations; S324, The preprocessed historical feature data is used as the input feature of the training sample, and the measured total suspended particulate matter concentration at the corresponding time is used as the output target to train the LightGBM regression model. The model parameters are iteratively updated by minimizing the loss function between the predicted value and the measured value. S325. During the training process, cross-validation is used to evaluate the model performance, and the hyperparameters of the model are adjusted according to the validation results until the model converges, thus obtaining a pre-trained multi-factor spatiotemporal prediction model.

[0013] Furthermore, the training process of the physical settlement model includes: S341, Obtain historical settlement measurement dataset. The historical settlement measurement dataset includes the dust adhesion mass per unit area measured in the working area on the surface of the closure sensitive component during multiple historical periods as labels, and the total suspended particulate matter concentration, wind speed and relative humidity recorded at the corresponding time as settlement input features. S342, construct a parameterized physical settling model framework, set the dust settling flux as a function of the total suspended particulate matter concentration, wind speed and relative humidity in the air, and introduce empirical parameters to be calibrated into the function; S343, input the sedimentation input features into the parameterized physical sedimentation model, with the goal of minimizing the error between the predicted dust adhesion mass per unit area and the label, and fit the empirical parameters to be calibrated until the preset convergence condition is met. S344, Substitute the empirical parameters determined after fitting into the physical sedimentation model framework to obtain the trained physical sedimentation model, which is used to calculate the dust adhesion coverage based on the real-time predicted total suspended particulate matter concentration, real-time wind speed and real-time relative humidity.

[0014] Furthermore, when the closure quality index is lower than a preset quality threshold, root cause diagnosis of environmental factors is performed, processing suggestions are generated, and the processing suggestions are pushed to the processing stage of the PDCA network, including: S41, when the closure quality index is lower than the preset quality threshold, the root cause diagnosis process is triggered; S42, invoke the interpreter of the multi-factor spatiotemporal prediction model, calculate the contribution value of each environmental factor feature in the input feature vector to the prediction result of total suspended particulate matter concentration, and identify the top K main contributing factors with the largest contribution values; wherein, the interpreter is a SHAP interpreter, and the contribution value is the SHAP value; S43, in the knowledge graph, the dynamic attribute values ​​of the environmental factor nodes corresponding to the main contributing factors are replaced with preset ideal values ​​in turn, and the prediction and diagnosis process of S31 to S36 is re-executed after each replacement to calculate the improvement of the closure quality index after replacement compared with that before replacement. S44, the environmental factor with the greatest improvement is identified as the key constraint bottleneck. Combining the contribution value ranking of the main contributing factors and the key constraint bottleneck, a diagnostic report containing the dominant environmental factor and its treatment recommendations is generated, and the diagnostic report is pushed to the processing stage of the PDCA quality information cycle network.

[0015] Furthermore, the interpreter of the multi-factor spatiotemporal prediction model is invoked to calculate the contribution value of each environmental factor feature in the input feature vector to the prediction result of total suspended particulate matter concentration, including: S421, extract the target environmental feature subset corresponding to each environmental factor node from the constructed input feature vector; the target environmental feature subset includes 10-minute average wind speed, gust wind speed, the angle between wind direction and the orientation of the exposed sand source, relative humidity, cumulative rainfall in the past N hours, solar radiation intensity, near-surface vertical temperature gradient, surrounding exposed construction site area and underlying surface type encoding. S422, the input feature vector and the multi-factor spatiotemporal prediction model are loaded into the SHAP interpreter. The prediction function of the multi-factor spatiotemporal prediction model is used as a black box function. Kernel SHAP is selected as the interpretation method to generate the SHAP value of each feature in the input feature vector. The SHAP value represents the marginal contribution of the feature to the predicted total suspended particulate matter concentration relative to the baseline value. S423, group and aggregate the SHAP values ​​of each feature according to their corresponding environmental factor nodes, and aggregate the SHAP values ​​of multiple features belonging to the same environmental factor node to obtain the comprehensive contribution value of the environmental factor node. S424: Sort the environmental factor nodes according to their comprehensive contribution value from largest to smallest, and select the top K environmental factor nodes as the main contributing factors.

[0016] Furthermore, in S1, the environmental dynamic monitoring data comes from a sensor group that includes at least an ultrasonic anemometer, a tipping bucket rain gauge, a solar radiation meter, a laser particulate sensor, and a GPS positioning module; the prior knowledge is extracted from the closure construction operation procedures, historical closure accident analysis reports, and expert experience texts in the field.

[0017] Furthermore, in S23, the causal rules extracted from prior knowledge are obtained by structurally extracting the prior knowledge through a preset rule template; the rule template defines at least environmental condition elements, relational predicate elements, and quality consequence elements, which are used to transform the unstructured prior knowledge text into causal relationship edges pointing from environmental factor nodes to quality risk event nodes.

[0018] Furthermore, in S44, generating a diagnostic report containing the dominant environmental factors and their treatment recommendations includes: Based on the type of key bottleneck, matching is performed in the preset control instruction library to extract corresponding processing suggestions; among them, when the key bottleneck is a wind speed-related factor, priority is given to matching instructions to suspend welding or install dustproof canopies; when the key bottleneck is a surface dryness-related factor, priority is given to matching instructions to spray water on exposed ground upwind or cover it with dustproof nets; when the key bottleneck is a sunshine-related factor, priority is given to matching instructions to adjust the closure lock time window. The matched processing suggestions, along with the contribution values ​​of the main contributing factors and key bottlenecks, are written into the diagnostic report.

[0019] This invention also provides a bridge construction quality perception and diagnosis system based on knowledge graphs and BIM. The system is used to implement any of the bridge construction quality perception and diagnosis methods based on knowledge graphs and BIM described above. The system includes: The data acquisition and knowledge extraction module is used to acquire BIM static data, environmental dynamic monitoring data and prior knowledge of the bridge closure section, and extract attribute information, time-series monitoring characteristics and causal rules of environmental impact of the sensitive components of the closure. The knowledge graph construction module, connected to the data acquisition and knowledge extraction module, is used to construct a knowledge graph with the closure sensitive components as nodes and the attribute information and time-series monitoring features as attributes. The quality prediction and diagnosis module, connected to the knowledge graph construction module, is used to establish a PDCA quality information loop network based on the knowledge graph. During the inspection phase of the PDCA quality information loop network, a pre-trained prediction model is used to predict the degree of dust pollution on the surface of the closure sensitive components, and the closure quality index is output based on the prediction results and preset quality diagnosis rules. The root cause analysis module, connected to the quality prediction and diagnosis module, is used to perform root cause diagnosis of environmental factors when the closure quality index is lower than a preset quality threshold, generate treatment suggestions, and push the treatment suggestions to the processing stage of the PDCA network.

[0020] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the bridge construction quality perception and diagnosis methods based on knowledge graphs and BIM as described in the present invention.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows: Firstly, this invention constructs a knowledge graph with sensitive closure components as entity nodes, dynamic environmental monitoring data as environmental factor nodes, and causal rules extracted from prior knowledge as relational edges. This graph integrates BIM static data, multi-source sensor time-series monitoring data, closure construction operation procedures, historical accident analysis reports, and expert experience texts into a graph-structured quality diagnosis knowledge base. This solves the problem in existing technologies where BIM data, environmental monitoring data, and prior knowledge are fragmented and lack a unified knowledge expression and correlation analysis framework. This improves the integration and reasoning ability of multi-source heterogeneous construction quality information. Secondly, this invention, through the inspection phase of the PDCA quality information loop network, sequentially executes the prediction of dust pollution level based on the LightGBM multi-factor spatiotemporal prediction model, the calculation of dust adhesion coverage based on the physical settlement model, and the two-level quality reasoning combining knowledge graph causal rules and Bayesian networks, forming a complete causal reasoning chain from environmental factors to structural performance degradation and outputting a quantitative closure quality index. This solves the problems of existing technologies lacking an end-to-end quality diagnosis link from pre-prediction to structural performance impact and being unable to output unified quantitative quality criteria, thus improving the foresight and accuracy of closure construction quality assessment. Thirdly, this invention calculates the contribution of each environmental factor to the prediction of dust concentration using the SHAP interpreter and identifies key bottlenecks by combining counterfactual reasoning. When the closure quality index is lower than a preset threshold, it automatically generates a diagnostic report containing the dominant environmental factors and their treatment suggestions and pushes it to the processing stage of the PDCA network. Overall, it realizes the intelligentization of the entire process of bridge closure construction quality data perception, knowledge fusion, prediction diagnosis, root cause tracing, and closed-loop control. It effectively overcomes the multiple defects of traditional methods, such as fragmented multi-source data, lack of pre-prediction, insufficient causal reasoning, and blind reliance on human experience in root cause diagnosis. It significantly improves the timeliness of closure construction quality risk warning, the scientific nature of diagnostic conclusions, and the accuracy of control decisions. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating the bridge construction quality perception and diagnosis method based on knowledge graphs and BIM provided in this embodiment of the invention. Figure 2 The structural block diagram of the bridge construction quality perception and diagnosis system based on knowledge graph and BIM provided in the embodiments of the present invention. Detailed Implementation

[0024] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0025] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0026] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0027] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0028] Example 1 like Figure 1 As shown, this invention provides a method for bridge construction quality perception and diagnosis based on knowledge graphs and BIM, including: S1: Obtain BIM static data, environmental dynamic monitoring data and prior knowledge of the bridge closure section, and extract attribute information, time-series monitoring characteristics and causal rules of environmental impact of the closure sensitive components. The closure-sensitive components include the closure segment components, supports, stiffening frames, prestressed ducts, and construction joints; the environmental dynamic monitoring data includes wind monitoring data, precipitation monitoring data, solar radiation intensity detection data, and the construction location; the attribute information of the closure-sensitive components includes geometric attributes, spatial location, and permissible construction environment parameters, which are obtained by analyzing the IFC model and combining it with the construction specification database.

[0029] In S1, the environmental dynamic monitoring data comes from a sensor group that includes at least an ultrasonic anemometer, a tipping bucket rain gauge, a solar radiation meter, a laser particulate sensor, and a GPS positioning module; the prior knowledge is extracted from the closure construction operation procedures, historical closure accident analysis reports, and expert experience texts in the field.

[0030] In one possible implementation, BIM static data of the bridge closure segment is acquired. This BIM model, stored in IFC format, includes the closure segment components, supports, stiffening frame, prestressed ducts, and construction joints—a three-dimensional geometric model of the interface between new and old concrete. By parsing the IFC model, attribute information of closure-sensitive components is extracted, including: cross-sectional dimensions, concrete grade, spatial coordinates, and design elevation of the closure segment components; type, sliding direction, design friction coefficient, and installation location of the supports; steel type, cross-sectional characteristics, and embedded location of the stiffening frame; diameter, coordinate trajectory, and joint location of the prestressed ducts; and location, area, and design bond strength requirements of the roughened surface of the construction joints. Simultaneously, by combining with the construction specification database, permissible construction environment parameters are extracted for each component, including allowable welding wind speed thresholds (e.g., ≤2 m / s), surface dust coverage thresholds (e.g., ≤0.5 g / m²), and design closure temperature window (e.g., 15℃ ± 3℃). This attribute information serves as the static attributes of entity nodes in the subsequent knowledge graph.

[0031] An environmental dynamic monitoring sensor array was deployed at the closure construction site, including: an ultrasonic anemometer for collecting wind speed and direction; a tipping bucket rain gauge for collecting minute-level rainfall; a solar radiation meter for collecting solar radiation intensity; a laser particulate matter sensor for collecting PM10 and TSP concentrations; and a GPS positioning module for recording the latitude, longitude, and elevation of the sensor installation locations. The sensor data was uploaded to a data processing server in real time via a wireless network. Time-series monitoring features were extracted from the uploaded data, resulting in data such as: June 15, 2026, 14:00:00, 10-minute average wind speed 5.3 m / s, gust wind speed 7.1 m / s, wind direction 315°, rainfall 0 mm, and solar radiation intensity 680 W / m². 2The TSP concentration was 120 μg / m³, and the monitoring point was located 50 m northwest of the closure joint, at the same height as the working surface, forming a structured record. These time-series monitoring characteristics serve as dynamic attributes of environmental factor nodes and are correlated with closure-sensitive components through timestamps and spatial coordinates.

[0032] Prior knowledge is acquired from sources including bridge closure construction operation procedures, historical analysis reports of similar bridge closure accidents, and experience summaries from domain experts. These unstructured texts are then structured using pre-defined rule templates. The rule templates define environmental conditions such as wind speed, humidity, and surface dryness, as well as relational predicates such as causing, aggravating, triggering, and quality consequences such as weld porosity, bonding failure at the joint surface, and obstructed bearing sliding. For example, a rule is extracted from the operation procedures: when the wind speed exceeds the permissible welding wind speed threshold and welding operations are underway, the risk of weld porosity increases; another rule is extracted from the accident reports: when the dust coverage on the roughened surface exceeds the surface dust coverage threshold and concrete pouring is being carried out, insufficient bonding strength between the old and new concrete joints is achieved.

[0033] This invention utilizes a multi-source sensor array consisting of an ultrasonic anemometer, a tipping bucket rain gauge, a solar radiometer, a laser particulate sensor, and a GPS positioning module to collect real-time temporal monitoring data on wind speed, precipitation, solar radiation, air particulate matter concentration, and the construction location. This addresses the problem in traditional methods where environmental data lacks spatial and temporal correlation with the work area on the component surface, failing to provide effective input for subsequent accurate predictions. It improves the spatiotemporal matching accuracy between monitoring data and construction procedures. Furthermore, by using pre-set rule templates to structurally extract closure construction operation procedures, historical accident analysis reports, and expert experience texts, it automatically generates causal rules pointing from environmental factors to quality risk events. This solves the problem that prior knowledge relies on manual summarization and is difficult for computer systems to directly utilize, improving the automation level of knowledge extraction and the reusability of rules.

[0034] S2, using the closure-sensitive components as nodes and the aforementioned attribute information and time-series monitoring features as attributes, construct a knowledge graph, including: S21, instantiate the closure segment components, supports, stiffening frames, prestressed ducts and construction joints as solid nodes respectively, and use the geometric properties, spatial positions and allowable construction environment parameters as the properties of the corresponding solid nodes; S22 instantiates wind monitoring data, precipitation monitoring data, solar radiation intensity detection data, and the construction location as environmental factor nodes, and uses time-series monitoring characteristics as the dynamic attributes of the corresponding environmental factor nodes. S23, calculate the Euclidean distance between the monitoring point location and the working area on the surface of the closure sensitive component. When the Euclidean distance is less than a preset spatial threshold, establish a spatial binding relationship edge between the entity node and the environmental factor node. By aligning and matching the timestamp of the monitoring data with the time window of the closure construction process, establish a time alignment relationship edge between the entity node and the environmental factor node. Based on the causal rules extracted from prior knowledge, establish a causal relationship edge from the environmental factor node to the quality risk event node. S24 is a knowledge graph for diagnosing the quality of closure construction, consisting of entity nodes, environmental factor nodes, quality risk event nodes, as well as spatial binding relationship edges, time alignment relationship edges, and causal relationship edges.

[0035] The causal rules extracted from prior knowledge are obtained by structurally extracting the prior knowledge through a preset rule template; the rule template defines at least environmental condition elements, relational predicate elements, and quality consequence elements, which are used to transform unstructured prior knowledge text into causal relationship edges pointing from environmental factor nodes to quality risk event nodes.

[0036] In one possible implementation, Neo4j graph database is used as the knowledge graph storage and query platform. The extracted closure-sensitive components are instantiated one by one as entity nodes, and each node is assigned a type label and attributes. Specifically, entity node E1, type closure segment component, is created, with attributes including: cross-sectional dimensions (box girder top plate width 12.0m, bottom plate width 6.0m, beam height 3.5m); concrete grade C55; design elevation; and spatial coordinate range. Entity node E2, type support, is created, with attributes including: support type, sliding direction, design friction coefficient, and installation location. Entity node E3, type stiffener frame, is created, with attributes including: steel type Q345B, cross-sectional type, and embedded location. Entity node E4, type prestressed duct, is created, with attributes including: duct diameter, material, coordinate trajectory, and joint location. Create a solid node E5 of type construction joint, with attributes including: location of the roughened surface, area of ​​the roughened surface, top plate bonding surface 6.0m², bottom plate bonding surface 4.8m², web plate bonding surface 3.2m² each, and design bond strength requirements. The permissible construction environment parameters for each solid node are stored as attributes. For example, the allowable wind speed threshold for welding is stored in the attributes of the stiffening frame node E3, while the surface dust coverage threshold and the design closure temperature window are stored in the attributes of the closure segment member node E1.

[0037] Similarly, the acquired environmental dynamic monitoring data and the construction location are instantiated as environmental factor nodes, and node type labels and dynamic attributes are assigned.

[0038] Construct spatial binding edges. Calculate the Euclidean distance between the monitoring point W4 for each environmental factor and the working area on the surface of each closure-sensitive component. Taking the working area on the surface of the stiffening frame as an example, this working area is the welding operation surface of the upper and lower edge steel components of the closure joint, and its center point coordinates are taken from the spatial position attribute of entity node E3. The calculated Euclidean distance between the monitoring point W4 and the center point of the working area E3 is 35.4m, which is less than the preset spatial threshold, set to 50m in this embodiment. Therefore, a spatial binding edge is established between W4 and E3, and the edge attribute records the actual distance value and the threshold.

[0039] Based on spatial binding relationships, wind node W1, precipitation node W2, and solar radiation node W3 are indirectly associated with each entity node through location node W4. That is, "located" relationship edges are established between W1, W2, W3, and W4, thus forming spatial association paths from environmental factor nodes to entity nodes. "Location" is a type of spatial dependency relationship edge in a knowledge graph, originating from an environmental factor node and ending at a location node, used to describe the physical deployment of a certain environmental monitoring sensor at a specific construction location.

[0040] Construct time alignment edges. Align and match the timestamps of the monitoring data with the time windows of the closure construction process. For example, the current construction process is the welding of the stiffening frame, and its time window is from 13:30 to 15:30 on June 15, 2026. The monitoring data timestamp 2026-06-15T14:00:00 falls within this window. Therefore, time alignment edges are established between wind node W1, precipitation node W2, solar radiation node W3, and entity node E3, respectively. The edge attributes record the matched process name and time window.

[0041] Constructing causal edges. Based on causal rules extracted from prior knowledge using rule templates, causal edges are established from environmental factor nodes to quality risk event nodes. For example, according to a pre-defined causal rule, when wind speed exceeds the permissible welding wind speed threshold and welding operations are underway, the risk of weld porosity increases. A node R1 of type quality risk event is created in the knowledge graph, with attributes including risk name weld porosity, associated component stiffness skeleton, and confidence level of 0.9. Then, a causal edge is established from wind node W1 to quality risk event node R1, with the edge attributes recording the triggering condition and rule source. Similarly, according to the rule, when the dust coverage on the roughened surface exceeds the surface dust coverage threshold and concrete pouring is carried out, insufficient bonding strength at the construction joint is caused, creating a quality risk event node R2. A causal edge is then established from wind node W1 (the driving factor of dust) to R2 to exacerbate the problem.

[0042] Causal rules refer to structured knowledge extracted from prior knowledge such as closure construction operation procedures, historical accident analysis reports, and expert experience texts, using pre-defined rule templates to describe the causal relationship between environmental factors and quality risks. Each causal rule consists of three elements: environmental condition elements, which define the specific environmental state that triggers the rule, such as wind speed exceeding the allowable wind speed threshold for welding; relational predicate elements, which indicate the type of causal relationship, such as causing or aggravating; and quality consequence elements, which describe the quality risk events that may be triggered, such as increased risk of weld porosity. In the knowledge graph, causal rules are stored in the form of causal relationship edges pointing from environmental factor nodes to quality risk event nodes. The attributes of these edges record the triggering conditions, confidence levels, and rule sources, thus forming a complete reasoning link from environmental factors to component quality deterioration.

[0043] Quality risk event nodes are a type of node in a knowledge graph specifically used to characterize potential risk events triggered by environmental factors that may lead to quality defects in closure components, such as weld porosity and insufficient bonding strength of construction joints. They serve as intermediate nodes in causal reasoning from environmental factors to component quality deterioration.

[0044] A knowledge graph was formed, comprising 5 entity nodes (E1 to E5), 4 environmental factor nodes (W1 to W4), 2 quality risk event nodes (R1 and R2), as well as spatial binding edges, temporal alignment edges, and causal relationship edges. This knowledge graph, in a graph structure, uniformly expresses the static correlation and dynamic evolution relationship between component attributes, environmental conditions, and quality risks during the closure construction, forming a complete knowledge foundation for quality diagnosis of the closure construction.

[0045] Meanwhile, this invention instantiates wind force, precipitation, solar radiation, and construction location as environmental factor nodes, and continuously updates time-series monitoring features as dynamic attributes. This solves the problem of disconnect between environmental monitoring data and construction objects, and the difficulty in systematically recording and utilizing dynamic changes in traditional methods, thus improving the real-time characterization capability of environmental state changes. By calculating the Euclidean distance between monitoring points and the work area on the component surface and establishing spatial binding relationships, and by matching timestamps with process time windows to establish time alignment relationships, this invention solves the problem of not being able to accurately determine which sensor data truly affects the construction quality of which component at which moment in traditional methods, thus improving the spatiotemporal matching accuracy between environmental data and construction objects. By using preset rule templates to structurally extract prior knowledge text and automatically generate causal relationship edges from environmental factor nodes to quality risk event nodes, this invention solves the problem that causal knowledge in expert experience, accident reports, and operating procedures is difficult for computer systems to understand and reason about, thus improving the completeness and automated construction degree of causal reasoning chains in knowledge graphs.

[0046] S3. Based on the knowledge graph, a PDCA quality information loop network is established. During the inspection phase of the PDCA quality information loop network, a pre-trained prediction model is used to predict the degree of dust pollution on the surface of the closure sensitive components, and the closure quality index is output according to the prediction results and the preset quality diagnosis rules. In S3, the PDCA quality information cycle network includes four stages: planning, execution, inspection, and processing. The planning stage extracts initial benchmark thresholds from the knowledge graph, which include the allowable wind speed threshold for welding, the surface dust coverage threshold, and the design closure temperature window. The execution stage continuously updates the knowledge graph with real-time environmental dynamic monitoring data. The processing stage generates control instructions including suspending welding or pouring operations, installing dustproof sheds, watering or covering exposed ground upwind with dustproof netting, and adjusting at least one of the following within the closure locking time window:

[0047] During the inspection phase of the PDCA quality information cycle network, a pre-trained prediction model is used to predict the degree of dust pollution on the surface of the closure-sensitive components, including: S31. Extract the dynamic attributes of environmental factor nodes at the current moment and within the historical window from the knowledge graph to form an input feature vector; the input feature vector includes 10-minute average wind speed, gust wind speed, the angle between wind direction and the location of the exposed sand source, relative humidity, cumulative rainfall in the past N hours, solar radiation intensity, near-surface vertical temperature gradient, closure height, surrounding exposed construction area, and underlying surface type code. S32, input the input feature vector into a pre-trained multi-factor spatiotemporal prediction model, wherein the multi-factor spatiotemporal prediction model is a LightGBM model; S33, the total suspended particulate matter concentration on the surface of the closure sensitive component working area within a preset time period is output by the multi-factor spatiotemporal prediction model, and is used as a quantitative indicator of the degree of dust pollution.

[0048] Based on the prediction results and the preset quality diagnosis rules, the closure quality index is output, including: S34, input the total suspended particulate matter concentration, real-time wind speed and real-time relative humidity into the physical settling model to obtain the dust adhesion coverage on the surface of the closure sensitive component; wherein, the physical settling model calculates the dust adhesion mass per unit area based on the relationship between particulate matter settling flux and concentration, wind speed and relative humidity in the air. S35. Based on the dust adhesion coverage, and combined with the causal rules stored in the knowledge graph, a two-level quality inference is performed: First, the dust adhesion coverage is compared with the surface dust coverage threshold, and the real-time wind speed is compared with the allowable wind speed threshold for welding to determine the joint quality risk level; Second, with the joint quality risk level as input, the probability of risk propagating to insufficient structural locking stiffness and abnormal bridge alignment is evaluated based on a Bayesian network to obtain the probability of closure quality deterioration. S36. Based on the comprehensive joint quality risk level and the probability of closure quality deterioration, a quantified closure quality index is generated and output.

[0049] The training process of a pre-trained multi-factor spatiotemporal prediction model includes: S321, Obtain historical training dataset, which includes total suspended particulate matter concentration measured at the same height as the closure face during multiple historical periods as labels, and historical feature data corresponding to the input feature vector collected synchronously. S322, perform data preprocessing on historical feature data and labels, including missing value imputation, outlier removal and feature normalization; S323, Build a LightGBM regression model and set the model hyperparameters, including learning rate, tree depth, number of leaf nodes and number of iterations; S324, The preprocessed historical feature data is used as the input feature of the training sample, and the measured total suspended particulate matter concentration at the corresponding time is used as the output target to train the LightGBM regression model. The model parameters are iteratively updated by minimizing the loss function between the predicted value and the measured value. S325. During the training process, cross-validation is used to evaluate the model performance, and the hyperparameters of the model are adjusted according to the validation results until the model converges, thus obtaining a pre-trained multi-factor spatiotemporal prediction model.

[0050] The training process of the physical settlement model includes: S341, Obtain historical settlement measurement dataset. The historical settlement measurement dataset includes the dust adhesion mass per unit area measured in the working area on the surface of the closure sensitive component during multiple historical periods as labels, and the total suspended particulate matter concentration, wind speed and relative humidity recorded at the corresponding time as settlement input features. S342, construct a parameterized physical settling model framework, set the dust settling flux as a function of the total suspended particulate matter concentration, wind speed and relative humidity in the air, and introduce empirical parameters to be calibrated into the function; S343, input the sedimentation input features into the parameterized physical sedimentation model, with the goal of minimizing the error between the predicted dust adhesion mass per unit area and the label, and fit the empirical parameters to be calibrated until the preset convergence condition is met. S344, Substitute the empirical parameters determined after fitting into the physical sedimentation model framework to obtain the trained physical sedimentation model, which is used to calculate the dust adhesion coverage based on the real-time predicted total suspended particulate matter concentration, real-time wind speed and real-time relative humidity.

[0051] In one possible implementation, a PDCA quality information cycle network is established based on the knowledge graph constructed by S2, and its four stages operate as follows: In the planning phase (P), initial baseline thresholds are extracted from the entity node attributes of the knowledge graph. Specifically, the allowable welding wind speed threshold of 2.0 m / s is extracted from the attributes of the stiffening frame node E3; the surface dust coverage threshold of 0.5 g / m² is extracted from the attributes of the closure segment component node E1; and the design closure temperature window of 15℃±3℃ (i.e., 12℃ to 18℃) is extracted from the attributes of the closure segment component node E1. These thresholds are stored in the baseline configuration table of the PDCA cycle as the basis for subsequent checks and judgments. In the execution phase (D), real-time environmental monitoring data is continuously updated to the knowledge graph. For example, at 14:00:00 on June 15, 2026, the ultrasonic anemometer recorded a 10-minute average wind speed of 5.3 m / s, a gust wind speed of 7.1 m / s, and a wind direction of 315°; the tipping bucket rain gauge recorded a current hourly rainfall of 0 mm; the solar radiation meter recorded a solar radiation intensity of 680 W / m²; and the laser particulate matter sensor recorded a TSP concentration of 120 μg / m³. These data, through timestamps and spatial binding relationships, are automatically updated to the dynamic attributes of the corresponding wind node W1, precipitation node W2, and solar radiation node W3 in the knowledge graph. During the inspection phase (C), dust pollution levels are predicted and quality is diagnosed, and the closure quality index is output. In the processing phase (A), when the closure quality index output in the inspection phase is lower than the preset quality threshold, a control instruction is generated and pushed to the on-site construction management system. The control instructions include: suspending welding or pouring operations, installing a dustproof canopy at the closure joint, spraying water or covering the exposed ground upwind with dustproof netting, and adjusting the closure locking time window to the low-temperature period at night.

[0052] The dynamic attributes of environmental factor nodes within the current moment and the previous 6-hour historical window are extracted from the knowledge graph to form an input feature vector. Specifically, these include the 10-minute average wind speed of 5.3 m / s, gust wind speed of 7.1 m / s, the angle between the wind direction and the location of the exposed sand source of 45°, relative humidity of 28%, cumulative rainfall of 0 mm in the past 6 hours, solar radiation intensity of 680 W / m², near-surface vertical temperature gradient of 0.12 ℃ / m, closure height of 187.632 m, surrounding exposed construction site area of ​​3500 m², and underlying surface type code 2, where code 2 represents dry exposed sand. The coding system is as follows: 1-concrete pavement, 2-dry exposed sand, 3-moist exposed soil, 4-vegetation cover, 5-water area.

[0053] Historical training dataset was obtained. In this embodiment, temporary monitoring points were set up at the same height as the closure face six months before the closure construction, and TSP concentration was continuously collected as a label, while various feature data were recorded synchronously. A total of 120 days of valid data were obtained, with approximately 2880 training samples at the hourly granularity. For missing values, forward imputation was used, such as filling in the previous time value if the sensor occasionally lost packets; outliers were removed using the 3σ principle, such as values ​​where the sensor reading jumps more than the mean ± 3 times the standard deviation were considered outliers; and Min-Max normalization was used for numerical features, mapping each feature value to the [0,1] interval. A LightGBM regression model was constructed. The model hyperparameters were set as follows: learning rate 0.05, tree depth 8, number of leaf nodes 31, number of iterations 500, and mean squared error as the loss function. The 2880 samples were randomly divided into a training set of 2304 samples and a validation set of 576 samples in an 8:2 ratio. The LightGBM model was trained using preprocessed historical feature data as input features and the measured TSP concentration at the corresponding time point as the output target. In each iteration, the model updated its parameters based on the gradient of the loss function between the current predicted and measured values. A 5-fold cross-validation method was used to evaluate model performance, calculating the root mean square error (RMSE) and coefficient of determination (R²) on the validation set. Hyperparameters were adjusted based on the validation results: iterations were terminated early to avoid overfitting when the RMSE on the validation set began to decrease gradually. After parameter tuning, the final model achieved an RMSE of 18.7 μg / m³ and an R² of 0.91 on the validation set, satisfying the preset convergence condition, thus obtaining a well-trained multi-factor spatiotemporal prediction model.

[0054] The constructed feature vector for the current moment is input into the trained LightGBM model. The model outputs a predicted TSP concentration of 485 μg / m³ for the work area on the surface of the closure-sensitive component within a preset time period, i.e., the next hour. This value serves as a quantitative indicator of dust pollution levels. In the model application (prediction) phase, the input is real-time environmental data from the current moment and within a historical window; the model output is the predicted TSP concentration for the future period. In the model training phase, historical feature data is used as training samples, and corresponding historical measured concentrations are used to teach the model how to predict, ultimately resulting in a well-trained model.

[0055] Historical settlement measurement datasets were obtained. Before the closure construction, under similar environmental conditions (wind speed, humidity, and TSP concentration range consistent with the expected conditions during the closure period), dust collection plates were placed on the surface of a temporary test platform. The dust adhesion mass per unit area was collected every 2 hours as a label, and the corresponding TSP concentration, wind speed, and relative humidity were recorded simultaneously. A total of 360 sets of valid data were obtained. A parameterized physical settlement model framework was constructed. According to particulate matter settling theory, the dust settling flux F can be expressed as… Where C is the TSP concentration (μg / m³), RH is the relative humidity (%), U is the wind speed (m / s), and α, β, and γ are empirical parameters to be calibrated. The 360 ​​sets of data were divided into training and validation sets in an 8:2 ratio. The goal was to minimize the mean square error between the predicted dust adhesion mass per unit area and the measured label. The least squares method was used to fit the parameters α, β, and γ until the sum of squared residuals decreased less than the preset convergence condition, i.e., 10... -6 The parameters were determined through fitting: α=0.0012, β=0.018, γ=0.35. Substituting these values ​​into the model yielded a trained physical sedimentation model. The predicted TSP concentration of 485 μg / m³, the current relative humidity of 28%, and the current wind speed of 5.3 m / s were input into the trained physical sedimentation model. The model first calculates the dust sedimentation flux based on the functional relationship between dust sedimentation flux and the total suspended particulate matter concentration, wind speed, and relative humidity. The dust sedimentation flux refers to the mass of dust settling per unit area per unit time. Integrating this flux based on the predicted time period length yields a dust attachment mass of 0.62 g / m² per unit area. Dividing this dust attachment mass per unit area by the allowable attachment mass per unit area corresponding to the surface dust coverage threshold gives the dust attachment coverage. In this embodiment, the surface dust coverage threshold is set to 0.5 g / m², and the dust attachment coverage is 1.24, exceeding the threshold by 24%.

[0056] The real-time average wind speed of 5.3 m / s was compared with the allowable wind speed threshold of 2.0 m / s for welding. 5.3 > 2.0, exceeding the threshold. Since both exceed the limits, based on the causal rules in the knowledge graph, the joint quality risk level is determined to be high risk.

[0057] Using the high-risk level of joint quality as input, inference is performed based on a Bayesian network. The conditional probability table of this Bayesian network is determined by historical accident data and expert experience, and the key parameters are as follows: When the joint quality risk is high, the probability of insufficient structural locking stiffness is 0.72. When the structural locking stiffness is insufficient, the probability of abnormal bridge alignment is 0.65.

[0058] Insufficient structural locking stiffness refers to a situation where, after the completion of the stiffening frame welding, prestressed duct connection, and concrete pouring, defects in the joint quality cause the actual bending and shear stiffness of the closure section to fall below the critical design values. This results in the closure section experiencing relative displacement or rotation exceeding the allowable range under subsequent construction loads and temperature effects. The determination of this condition is based on establishing a stiffness calculation model for the closure section according to the geometric and material properties of the sensitive closure components. When the joint quality risk level is high, based on historical accident statistics and expert experience, welding defects (such as weld porosity) can reduce the stiffness of the stiffening frame connection to 60% to 75% of the design value, and insufficient bonding strength of the construction joint can reduce the shear stiffness of the new and old concrete interface to 50% to 70% of the design value. The probability that the equivalent stiffness of the closure section, calculated by considering these reduction factors, is lower than the design stiffness threshold is 0.72. The stiffness reduction factor and probability value are derived from statistical analysis of similar bridge closure accidents and evaluation by experts in the field, and are stored in the conditional probability table of the Bayesian network. They can be calibrated and adjusted according to the actual bridge design parameters and construction conditions.

[0059] Therefore, the probability of quality degradation after closure is calculated to be 0.72 × 0.65 ≈ 0.47, or 47%.

[0060] The overall joint quality risk level is high, and the probability of closure quality deterioration is 47%. A weighted scoring method is used to generate a quantified closure quality index. In this embodiment, the risk level is mapped to a score, i.e., high risk = 30 points, medium risk = 60 points, and low risk = 90 points. The probability of deterioration is deducted as a percentage, i.e., 47 points are deducted for a 47% probability. The calculated closure quality index is 30 - 47 = -17 points, which is judged as a quality failure. This score is lower than the preset quality threshold of 60 points, triggering the root cause diagnosis process in S4.

[0061] This invention addresses the problems of disconnect between planning benchmarks and dynamic monitoring, and the inability to provide closed-loop feedback of inspection results in traditional construction quality management by establishing a PDCA quality information loop network on top of a knowledge graph, encompassing four stages: planning, execution, inspection, and processing. This improves the continuous improvement capability and automation level of quality management. Furthermore, this invention uses a pre-trained LightGBM multi-factor spatiotemporal prediction model to predict the total suspended particulate matter concentration on the surface of closure-sensitive components. This solves the problem that traditional methods cannot quantitatively predict the degree of dust pollution before construction and can only passively detect it after pollution occurs, improving the lead time and timeliness of dust risk warnings. Finally, this invention further inputs the predicted total suspended particulate matter concentration into... Based on the principle of particulate matter settling flux, a physical settling model is used to calculate the dust adhesion coverage on the surface of components. This solves the problem that relying solely on air concentration indicators cannot directly determine the actual pollution state of the component surface, and improves the direct correlation between pollution assessment results and the impact on construction quality. This invention uses a two-level quality inference by combining causal rules in a knowledge graph. The first level compares the dust adhesion coverage and real-time wind speed with corresponding thresholds to determine the joint quality risk level. The second level uses a Bayesian network to assess the probability of this risk propagating to insufficient structural locking stiffness and abnormal bridge alignment. This solves the problem that existing methods lack a complete causal inference chain from surface pollution to structural performance degradation, and improves the accuracy of quality diagnosis conclusions.

[0062] S4, when the closure quality index is lower than the preset quality threshold, root cause diagnosis of environmental factors is performed, treatment suggestions are generated, and the treatment suggestions are pushed to the processing stage of the PDCA network, including: S41, when the closure quality index is lower than the preset quality threshold, the root cause diagnosis process is triggered; S42, invoke the interpreter of the multi-factor spatiotemporal prediction model, calculate the contribution value of each environmental factor feature in the input feature vector to the prediction result of total suspended particulate matter concentration, and identify the top K main contributing factors with the largest contribution values; wherein, the interpreter is a SHAP interpreter, and the contribution value is the SHAP value; S43, in the knowledge graph, the dynamic attribute values ​​of the environmental factor nodes corresponding to the main contributing factors are replaced with preset ideal values ​​in turn, and the prediction and diagnosis process of S31 to S36 is re-executed after each replacement to calculate the improvement of the closure quality index after replacement compared with that before replacement. S44, the environmental factor with the greatest improvement is identified as the key constraint bottleneck. Combining the contribution value ranking of the main contributing factors and the key constraint bottleneck, a diagnostic report containing the dominant environmental factor and its treatment recommendations is generated, and the diagnostic report is pushed to the processing stage of the PDCA quality information cycle network.

[0063] The interpreter of the multi-factor spatiotemporal prediction model is invoked to calculate the contribution of each environmental factor feature in the input feature vector to the prediction result of total suspended particulate matter concentration, including: S421, extract the target environmental feature subset corresponding to each environmental factor node from the constructed input feature vector; the target environmental feature subset includes 10-minute average wind speed, gust wind speed, the angle between wind direction and the orientation of the exposed sand source, relative humidity, cumulative rainfall in the past N hours, solar radiation intensity, near-surface vertical temperature gradient, surrounding exposed construction site area and underlying surface type encoding. S422, the input feature vector and the multi-factor spatiotemporal prediction model are loaded into the SHAP interpreter. The prediction function of the multi-factor spatiotemporal prediction model is used as a black-box function. The prediction function refers to the callable function interface after encapsulating the trained LightGBM multi-factor spatiotemporal prediction model. Its input is a feature vector containing features such as 10-minute average wind speed, gust wind speed, the angle between wind direction and the location of the exposed sand source, relative humidity, cumulative rainfall in the past N hours, solar radiation intensity, near-surface vertical temperature gradient, closure height, surrounding exposed construction area, and underlying surface type encoding. The output is the predicted value of the total suspended particulate matter concentration on the surface of the closure sensitive component within the corresponding future preset time period. This prediction function encapsulates the integrated calculation results of all decision trees in the LightGBM model. For the SHAP interpreter, its internal operation logic is regarded as an opaque black box, and the marginal contribution of each feature is calculated only through the mapping relationship between the input feature vector and the output prediction value. Kernel SHAP is selected as the interpretation method to generate SHAP values ​​for each feature in the input feature vector; wherein, the SHAP value represents the marginal contribution of the feature to the predicted total suspended particulate matter concentration relative to the baseline value. S423, group and aggregate the SHAP values ​​of each feature according to their corresponding environmental factor nodes, and aggregate the SHAP values ​​of multiple features belonging to the same environmental factor node to obtain the comprehensive contribution value of the environmental factor node. S424: Sort the environmental factor nodes according to their comprehensive contribution value from largest to smallest, and select the top K environmental factor nodes as the main contributing factors.

[0064] The generation of the diagnostic report, which includes the dominant environmental factors and their treatment recommendations, includes: Based on the type of key bottleneck, matching is performed in the preset control instruction library to extract corresponding processing suggestions; among them, when the key bottleneck is a wind speed-related factor, priority is given to matching instructions to suspend welding or install dustproof canopies; when the key bottleneck is a surface dryness-related factor, priority is given to matching instructions to spray water on exposed ground upwind or cover it with dustproof nets; when the key bottleneck is a sunshine-related factor, priority is given to matching instructions to adjust the closure lock time window. The matched processing suggestions, along with the contribution values ​​of the main contributing factors and key bottlenecks, are written into the diagnostic report.

[0065] In one possible implementation, if the closure quality index of -17 points is lower than the preset quality threshold of 60 points, the inspection phase of the PDCA quality information cycle network will automatically mark the diagnostic result as non-conforming and send a trigger signal to the root cause analysis module to start the root cause diagnosis process.

[0066] From the input feature vector constructed in step S31, extract the target environmental feature subsets corresponding to each environmental factor node. The 10-minute average wind speed is 5.3 m / s, gust wind speed is 7.1 m / s, wind direction is 315°, rainfall is 0 mm, solar radiation intensity is 680 W / m², and TSP concentration is 120 μg / m². 3 The monitoring point is located 50m away from the work surface on the northwest side of the closure point, at the same height as the work surface.

[0067] The complete input feature vector and the trained LightGBM multi-factor spatiotemporal prediction model are loaded into the SHAP interpreter. Using the prediction function of the LightGBM model as a black-box function, Kernel SHAP is selected as the interpretation method. Kernel SHAP calculates the marginal contribution of each feature under different feature combinations by performing multiple perturbation samplings on the features (1000 sampling times in this embodiment), ultimately generating the SHAP value for each feature. The baseline value refers to the baseline predicted value in the SHAP interpretation method, specifically the average of the predicted total suspended particulate matter concentrations of all samples in the training dataset. Within the SHAP interpretation framework, the SHAP value of each feature represents the marginal contribution of the feature's value at the current time relative to the baseline value to the prediction result; that is, the degree to which the feature value causes the prediction result to deviate from the average level. For example, the SHAP value of the 10-minute average wind speed is +42.3, indicating that this feature increases the predicted TSP concentration by 42.3 μg / m³ compared to the average predicted concentration of all training samples; the SHAP value of the relative humidity is -18.5, indicating that this feature decreases the predicted TSP concentration by 18.5 μg / m³ compared to the average predicted concentration.

[0068] The SHAP values ​​for each feature are calculated as follows: The 10-minute average wind speed, with a SHAP value of +42.3, caused the predicted TSP concentration to increase by 42.3 μg / m³ compared to the baseline value. The gust wind speed, with a SHAP value of +28.7, caused the predicted TSP concentration to increase by 28.7 μg / m³ compared to the baseline value; The angle between the wind direction and the azimuth of the exposed sand source has a SHAP value of +15.2, which is conducive to transporting dust from the sand source to the closure point. The relative humidity, SHAP value is -18.5, the low humidity weakens the dust suppression effect; The cumulative rainfall over the past 6 hours, with a SHAP value of -35.8, indicates that the lack of rainfall has dried the ground and significantly increased the predicted concentration. Solar radiation intensity, SHAP value +12.3, radiative heating of the Earth's surface enhances turbulence and promotes dust; The near-surface vertical temperature gradient has a SHAP value of +8.9, and the unstable temperature stratification is conducive to the vertical diffusion of particulate matter. The closure height has a SHAP value of +5.6, which matches the dust conveying path. The surrounding exposed construction site area has a SHAP value of +22.4, indicating that the large exposed area provides ample dust sources. The underlying surface type is coded with a SHAP value of +16.1, indicating that the dry, exposed sandy soil type is highly prone to dust generation.

[0069] The SHAP values ​​of the above features are grouped and aggregated according to their corresponding environmental factor nodes. The grouping results are as follows: The wind factor node includes three characteristics: 10-minute average wind speed (SHAP value +42.3), gust wind speed (SHAP value +28.7), and the angle between the wind direction and the azimuth of the exposed sand source (SHAP value +15.2). The precipitation factor node includes two characteristics: relative humidity (SHAP value -18.5) and cumulative rainfall in the past 6 hours (SHAP value -35.8). The surface and location factor node includes two features: the area of ​​the surrounding exposed construction site (SHAP value +22.4) and the underlying surface type code (SHAP value +16.1). The thermal factor node contains two features: solar radiation intensity (SHAP value +12.3) and near-surface vertical temperature gradient (SHAP value +8.9). Other factor nodes include one feature: closure height (SHAP value + 5.6).

[0070] Aggregating STAP values ​​for multiple features belonging to the same environmental factor node involves summing the absolute values ​​of the STAP values ​​for all features belonging to the same environmental factor node to reflect the overall influence of that environmental factor node on the total suspended particulate matter concentration prediction results. The STAP values ​​of each feature are grouped and aggregated according to their corresponding environmental factor nodes. The absolute values ​​of the STAP values ​​for multiple features belonging to the same environmental factor node are then summed to obtain the comprehensive contribution value of that environmental factor node. For wind, the contribution is |+42.3| + |+28.7| + |+15.2| = 86.2; for precipitation, it is |-18.5| + |-35.8| = 54.3; for surface and location factors, it is |+22.4| + |+16.1| = 38.5; for thermal factors, it is |+12.3| + |+8.9| = 21.2; and for other factors, it is |+5.6| = 5.6.

[0071] With K=3 pre-set, the top 3 environmental factor nodes were selected as the main contributing factors according to their comprehensive contribution values ​​from largest to smallest: wind factor (comprehensive contribution value 86.2, ranked 1st), precipitation factor (comprehensive contribution value 54.3, ranked 2nd), and surface and location factor (comprehensive contribution value 38.5, ranked 3rd).

[0072] In the knowledge graph, the dynamic attribute values ​​of the environmental factor nodes corresponding to the above three main contributing factors are replaced with preset ideal values ​​in turn, and the prediction and diagnosis process of S31 to S36 is re-executed after each replacement.

[0073] In this embodiment, the ideal values ​​are set as follows: the ideal state of wind factor is that the average wind speed drops to the welding allowable threshold of 2.0 m / s and the gust wind speed is 3.0 m / s; the ideal state of precipitation factor is that the cumulative rainfall in the past 6 hours is 5 mm and the relative humidity is 60%; the ideal state of the surrounding exposed construction site area in the location factor is that the exposed area is reduced by 90%, that is, 350 m², which is achieved by covering with dustproof netting.

[0074] After replacing and calculating the wind nodes in sequence, the closure quality index improved the most (+89 points), so wind was identified as the key bottleneck.

[0075] Based on the contribution value ranking of wind force (1st), precipitation (2nd), location (3rd), and the critical bottleneck wind force, a matching command is performed from a pre-set control command library. Since the critical bottleneck is a wind speed-related factor, the priority control command to match is: suspend welding operations and immediately install a dustproof canopy at the closure joint.

[0076] The generated diagnostic report contains the following: Diagnosis time: June 15, 2026, 14:05:00; Current closure quality index: -17 points; Predicted TSP concentration: 485 μg / m³; Ranking of main contributing factors: Wind force factor (comprehensive contribution value 86.2), 10-minute average wind speed 5.3 m / s, gust 7.1 m / s, wind direction is conducive to transporting dust from exposed sand sources to the closure point; Precipitation factor (comprehensive contribution value 54.3): There has been no rainfall in the past 6 hours, the relative humidity is only 28%, and the surface is extremely dry; Location factors (overall contribution value 44.1): There is 3500m² of dry, exposed sandy soil within 500m of the closure point; The key bottleneck is wind force. Through counterfactual reasoning, controlling the wind speed can improve the closure quality index by 89 points to the qualified level.

[0077] Recommendations: Immediately suspend the welding of the stiffening frame at the closure joint, install a mobile dustproof shed above the closure joint and upwind, reduce the wind speed on the work surface to below 2.0 m / s, spray water on the exposed ground upwind to fix the dust, and lay dustproof netting. After the wind speed decreases and the dustproof measures are in place, clean the surface of the components again, and carry out the closure and locking during the stable low temperature period at night after 23:00.

[0078] This invention addresses the problems of reliance on manual identification of quality anomalies and delayed diagnosis in traditional methods by setting preset quality thresholds and automatically triggering root cause diagnosis processes, thereby improving the speed of quality risk response and the degree of automation in diagnosis. By calling the SHAP interpreter to calculate the contribution of each environmental factor characteristic to the dust concentration prediction results and aggregating them into a comprehensive contribution value for environmental factor nodes, it solves the problem of existing technologies being unable to quantify the impact of different environmental factors on dust pollution, improving the accuracy of root cause localization. By performing counterfactual reasoning in the knowledge graph, replacing the dynamic attribute values ​​of the main contributing factors one by one with ideal values ​​and recalculating the closure quality index, it solves the decision-making blind spot problem of not being able to determine which environmental factor has the greatest improvement benefit, improving the scientific nature and pertinence of the selection of control measures. By combining the ranking of contribution values ​​with the improvement magnitude of counterfactual reasoning, it identifies key bottlenecks and matches corresponding treatment suggestions in a preset control instruction library based on the bottleneck type, solving the problem of control measures being disconnected from risk causes and relying on subjective experience in traditional methods, thus improving the accuracy of treatment suggestions.

[0079] Example 2 like Figure 2 As shown, this invention provides a bridge construction quality perception and diagnosis system based on knowledge graphs and BIM. The system is used to implement the bridge construction quality perception and diagnosis method based on knowledge graphs and BIM described in any one of Embodiment 1. The system includes: The data acquisition and knowledge extraction module is used to acquire BIM static data, environmental dynamic monitoring data and prior knowledge of the bridge closure section, and extract attribute information, time-series monitoring characteristics and causal rules of environmental impact of the sensitive components of the closure. The knowledge graph construction module, connected to the data acquisition and knowledge extraction module, is used to construct a knowledge graph with the closure sensitive components as nodes and the attribute information and time-series monitoring features as attributes. The quality prediction and diagnosis module, connected to the knowledge graph construction module, is used to establish a PDCA quality information loop network based on the knowledge graph. During the inspection phase of the PDCA quality information loop network, a pre-trained prediction model is used to predict the degree of dust pollution on the surface of the closure sensitive components, and the closure quality index is output based on the prediction results and preset quality diagnosis rules. The root cause analysis module, connected to the quality prediction and diagnosis module, is used to perform root cause diagnosis of environmental factors when the closure quality index is lower than a preset quality threshold, generate treatment suggestions, and push the treatment suggestions to the processing stage of the PDCA network.

[0080] Example 3 The present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the bridge construction quality perception and diagnosis method based on knowledge graph and BIM as described in any one of Embodiment 1.

[0081] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for bridge construction quality perception and diagnosis based on knowledge graphs and BIM, characterized in that, include: S1: Obtain BIM static data, environmental dynamic monitoring data and prior knowledge of the bridge closure section, and extract attribute information, time-series monitoring characteristics and causal rules of environmental impact of the closure sensitive components. S2, using the closure-sensitive components as nodes and the attribute information and time-series monitoring features as attributes, construct a knowledge graph; S3, based on knowledge graphs, establishes a PDCA quality information circulation network; During the inspection phase of the PDCA quality information loop network, a pre-trained prediction model is used to predict the degree of dust pollution on the surface of the closure sensitive components, and the closure quality index is output based on the prediction results and the preset quality diagnosis rules. S4. When the closure quality index is lower than the preset quality threshold, root cause diagnosis of environmental factors is performed, treatment suggestions are generated, and the treatment suggestions are pushed to the processing stage of the PDCA network.

2. The bridge construction quality perception and diagnosis method based on knowledge graph and BIM according to claim 1, characterized in that, The closure-sensitive components include closure segment components, supports, stiffening frames, prestressed ducts, and construction joints; the environmental dynamic monitoring data includes wind monitoring data, precipitation monitoring data, solar radiation intensity detection data, and the construction location; the attribute information of the closure-sensitive components includes geometric attributes, spatial location, and permissible construction environment parameters, which are obtained by analyzing the IFC model and combining it with the construction specification database.

3. The bridge construction quality perception and diagnosis method based on knowledge graph and BIM according to claim 2, characterized in that, Using the closure-sensitive components as nodes and the aforementioned attribute information and time-series monitoring features as attributes, a knowledge graph is constructed, including: S21, instantiate the closure segment components, supports, stiffening frames, prestressed ducts and construction joints as solid nodes respectively, and use the geometric properties, spatial positions and allowable construction environment parameters as the properties of the corresponding solid nodes; S22 instantiates wind monitoring data, precipitation monitoring data, solar radiation intensity detection data, and the construction location as environmental factor nodes, and uses time-series monitoring characteristics as the dynamic attributes of the corresponding environmental factor nodes. S23, calculate the Euclidean distance between the monitoring point location and the working area on the surface of the closure sensitive component. When the Euclidean distance is less than a preset spatial threshold, establish a spatial binding relationship edge between the entity node and the environmental factor node. By aligning and matching the timestamp of the monitoring data with the time window of the closure construction process, establish a time alignment relationship edge between the entity node and the environmental factor node. Based on the causal rules extracted from prior knowledge, establish a causal relationship edge from the environmental factor node to the quality risk event node. S24 is a knowledge graph for diagnosing the quality of closure construction, consisting of entity nodes, environmental factor nodes, quality risk event nodes, as well as spatial binding relationship edges, time alignment relationship edges, and causal relationship edges.

4. The bridge construction quality perception and diagnosis method based on knowledge graph and BIM according to claim 1, characterized in that, In S3, the PDCA quality information cycle network includes four stages: planning, execution, inspection, and processing. The planning stage extracts initial benchmark thresholds from the knowledge graph, which include the allowable wind speed threshold for welding, the surface dust coverage threshold, and the design closure temperature window. The execution stage continuously updates the knowledge graph with real-time environmental dynamic monitoring data. The processing stage generates control instructions including suspending welding or pouring operations, installing dustproof sheds, watering or covering exposed ground upwind with dustproof netting, and adjusting at least one of the following within the closure locking time window:

5. The bridge construction quality perception and diagnosis method based on knowledge graph and BIM according to claim 1, characterized in that, During the inspection phase of the PDCA quality information loop network, a pre-trained prediction model is used to predict the degree of dust pollution on the surface of the closure-sensitive components, including: S31. Extract the dynamic attributes of environmental factor nodes at the current moment and within the historical window from the knowledge graph to form an input feature vector; the input feature vector includes 10-minute average wind speed, gust wind speed, the angle between wind direction and the location of the exposed sand source, relative humidity, cumulative rainfall in the past N hours, solar radiation intensity, near-surface vertical temperature gradient, closure height, surrounding exposed construction area, and underlying surface type code. S32, input the input feature vector into a pre-trained multi-factor spatiotemporal prediction model, wherein the multi-factor spatiotemporal prediction model is a LightGBM model; S33, the total suspended particulate matter concentration on the surface of the closure sensitive component working area within a preset time period is output by the multi-factor spatiotemporal prediction model, and is used as a quantitative indicator of the degree of dust pollution.

6. The bridge construction quality perception and diagnosis method based on knowledge graph and BIM according to claim 5, characterized in that, Based on the prediction results and preset quality diagnosis rules, the closure quality index is output, including: S34, input the total suspended particulate matter concentration, real-time wind speed and real-time relative humidity into the physical settling model to obtain the dust adhesion coverage on the surface of the closure sensitive component; wherein, the physical settling model calculates the dust adhesion mass per unit area based on the relationship between particulate matter settling flux and concentration, wind speed and relative humidity in the air. S35. Based on the dust adhesion coverage, and combined with the causal rules stored in the knowledge graph, a two-level quality inference is performed: First, the dust adhesion coverage is compared with the surface dust coverage threshold, and the real-time wind speed is compared with the allowable wind speed threshold for welding to determine the joint quality risk level; Second, with the joint quality risk level as input, the probability of risk propagating to insufficient structural locking stiffness and abnormal bridge alignment is evaluated based on a Bayesian network to obtain the probability of closure quality deterioration. S36. Based on the comprehensive joint quality risk level and the probability of closure quality deterioration, a quantified closure quality index is generated and output.

7. The bridge construction quality perception and diagnosis method based on knowledge graph and BIM according to claim 5, characterized in that, The training process of the pre-trained multi-factor spatiotemporal prediction model includes: S321, Obtain historical training dataset, which includes total suspended particulate matter concentration measured at the same height as the closure face during multiple historical periods as labels, and historical feature data corresponding to the input feature vector collected synchronously. S322, perform data preprocessing on historical feature data and labels, including missing value imputation, outlier removal and feature normalization; S323, Build a LightGBM regression model and set the model hyperparameters, including learning rate, tree depth, number of leaf nodes and number of iterations; S324, The preprocessed historical feature data is used as the input feature of the training sample, and the measured total suspended particulate matter concentration at the corresponding time is used as the output target to train the LightGBM regression model. The model parameters are iteratively updated by minimizing the loss function between the predicted value and the measured value. S325. During the training process, cross-validation is used to evaluate the model performance, and the hyperparameters of the model are adjusted according to the validation results until the model converges, thus obtaining a pre-trained multi-factor spatiotemporal prediction model.

8. The bridge construction quality perception and diagnosis method based on knowledge graph and BIM according to claim 6, characterized in that, The training process of the physical settlement model includes: S341, Obtain historical settlement measurement dataset. The historical settlement measurement dataset includes the dust adhesion mass per unit area measured in the working area on the surface of the closure sensitive component during multiple historical periods as labels, and the total suspended particulate matter concentration, wind speed and relative humidity recorded at the corresponding time as settlement input features. S342, construct a parameterized physical settling model framework, set the dust settling flux as a function of the total suspended particulate matter concentration, wind speed and relative humidity in the air, and introduce empirical parameters to be calibrated into the function; S343, input the sedimentation input features into the parameterized physical sedimentation model, with the goal of minimizing the error between the predicted dust adhesion mass per unit area and the label, and fit the empirical parameters to be calibrated until the preset convergence condition is met. S344, Substitute the empirical parameters determined after fitting into the physical sedimentation model framework to obtain the trained physical sedimentation model, which is used to calculate the dust adhesion coverage based on the real-time predicted total suspended particulate matter concentration, real-time wind speed and real-time relative humidity.

9. The bridge construction quality perception and diagnosis method based on knowledge graph and BIM according to claim 5, characterized in that, When the closure quality index is lower than a preset quality threshold, root cause diagnosis of environmental factors is performed, treatment suggestions are generated, and the treatment suggestions are pushed to the processing stage of the PDCA network, including: S41, when the closure quality index is lower than the preset quality threshold, the root cause diagnosis process is triggered; S42, invoke the interpreter of the multi-factor spatiotemporal prediction model, calculate the contribution value of each environmental factor feature in the input feature vector to the prediction result of total suspended particulate matter concentration, and identify the top K main contributing factors with the largest contribution values; wherein, the interpreter is a SHAP interpreter, and the contribution value is the SHAP value; S43, in the knowledge graph, the dynamic attribute values ​​of the environmental factor nodes corresponding to the main contributing factors are replaced with preset ideal values ​​in turn, and the prediction and diagnosis process of S31 to S36 is re-executed after each replacement to calculate the improvement of the closure quality index after replacement compared with that before replacement. S44, the environmental factor with the greatest improvement is identified as the key constraint bottleneck. Combining the contribution value ranking of the main contributing factors and the key constraint bottleneck, a diagnostic report containing the dominant environmental factor and its treatment recommendations is generated, and the diagnostic report is pushed to the processing stage of the PDCA quality information cycle network.

10. The bridge construction quality perception and diagnosis method based on knowledge graph and BIM according to claim 9, characterized in that, The interpreter of the multi-factor spatiotemporal prediction model calculates the contribution of each environmental factor feature in the input feature vector to the prediction result of total suspended particulate matter concentration, including: S421, extract the target environmental feature subset corresponding to each environmental factor node from the constructed input feature vector; the target environmental feature subset includes 10-minute average wind speed, gust wind speed, the angle between wind direction and the orientation of the exposed sand source, relative humidity, cumulative rainfall in the past N hours, solar radiation intensity, near-surface vertical temperature gradient, surrounding exposed construction site area and underlying surface type encoding. S422, the input feature vector and the multi-factor spatiotemporal prediction model are loaded into the SHAP interpreter. The prediction function of the multi-factor spatiotemporal prediction model is used as a black box function. Kernel SHAP is selected as the interpretation method to generate the SHAP value of each feature in the input feature vector. The SHAP value represents the marginal contribution of the feature to the predicted total suspended particulate matter concentration relative to the baseline value. S423, group and aggregate the SHAP values ​​of each feature according to their corresponding environmental factor nodes, and aggregate the SHAP values ​​of multiple features belonging to the same environmental factor node to obtain the comprehensive contribution value of the environmental factor node. S424: Sort the environmental factor nodes according to their comprehensive contribution value from largest to smallest, and select the top K environmental factor nodes as the main contributing factors.

11. The bridge construction quality perception and diagnosis method based on knowledge graph and BIM according to claim 1, characterized in that, In S1, the environmental dynamic monitoring data comes from a sensor group that includes at least an ultrasonic anemometer, a tipping bucket rain gauge, a solar radiation meter, a laser particulate sensor, and a GPS positioning module; the prior knowledge is extracted from the closure construction operation procedures, historical closure accident analysis reports, and expert experience texts in the field.

12. The bridge construction quality perception and diagnosis method based on knowledge graph and BIM according to claim 3, characterized in that, In S23, the causal rules extracted from prior knowledge are obtained by structurally extracting the prior knowledge through a preset rule template; the rule template defines at least environmental condition elements, relational predicate elements, and quality consequence elements, which are used to transform the unstructured prior knowledge text into causal relationship edges pointing from environmental factor nodes to quality risk event nodes.

13. The bridge construction quality perception and diagnosis method based on knowledge graph and BIM according to claim 9, characterized in that, In S44, generating a diagnostic report containing the dominant environmental factors and their treatment recommendations includes: Based on the type of key bottleneck, matching is performed in the preset control instruction library to extract corresponding processing suggestions; among them, when the key bottleneck is a wind speed-related factor, priority is given to matching instructions to suspend welding or install dustproof canopies; when the key bottleneck is a surface dryness-related factor, priority is given to matching instructions to spray water on exposed ground upwind or cover it with dustproof nets; when the key bottleneck is a sunshine-related factor, priority is given to matching instructions to adjust the closure lock time window. The matched processing suggestions, along with the contribution values ​​of the main contributing factors and key bottlenecks, are written into the diagnostic report.

14. A bridge construction quality perception and diagnosis system based on knowledge graphs and BIM, characterized in that, The system is used to implement the bridge construction quality perception and diagnosis method based on knowledge graph and BIM as described in any one of claims 1 to 13, and the system includes: The data acquisition and knowledge extraction module is used to acquire BIM static data, environmental dynamic monitoring data and prior knowledge of the bridge closure section, and extract attribute information, time-series monitoring characteristics and causal rules of environmental impact of the sensitive components of the closure. The knowledge graph construction module, connected to the data acquisition and knowledge extraction module, is used to construct a knowledge graph with the closure sensitive components as nodes and the attribute information and time-series monitoring features as attributes. The quality prediction and diagnosis module, connected to the knowledge graph construction module, is used to establish a PDCA quality information loop network based on the knowledge graph. During the inspection phase of the PDCA quality information loop network, a pre-trained prediction model is used to predict the degree of dust pollution on the surface of the closure sensitive components, and the closure quality index is output based on the prediction results and preset quality diagnosis rules. The root cause analysis module, connected to the quality prediction and diagnosis module, is used to perform root cause diagnosis of environmental factors when the closure quality index is lower than a preset quality threshold, generate treatment suggestions, and push the treatment suggestions to the processing stage of the PDCA network.

15. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the bridge construction quality perception and diagnosis method based on knowledge graph and BIM as described in any one of claims 1 to 13.

Images