A construction project quality detection supervision management method and system

Abstract

The application discloses a kind of construction engineering quality detection supervision management method and system, belong to engineering quality detection field, method includes: the relevant data of construction site is collected, data is converted into standardization semantic label;Establish initial engineering quality knowledge graph, when new construction material or new technology appears, extend knowledge graph by feature matching, realize dynamic adaptation;Adopt graph neural network to learn knowledge graph, form the potential dependence between index, while propose a kind of self-evolution reasoning system;Based on the reasoning result and historical evolution mode of knowledge graph, predict potential risk event, classify exception according to influence range, and automatically trigger corresponding supervision management measure;Rectification result and new detection data are written into knowledge graph, as subsequent self-evolution training data.The application uses the above method and system, can significantly improve the forward-looking of construction engineering quality detection and supervision management, guarantee the controllability and reliability of engineering quality.

Description

Technical Field

[0001] This invention relates to the field of engineering quality testing technology, and in particular to a method and system for supervising and managing the quality testing of construction projects. Background Technology

[0002] In the existing process of construction project quality inspection and supervision, although various data collection methods such as material testing, process testing, image testing, and environmental monitoring have been introduced, most of these data are stored in scattered numerical or textual forms, lacking unified semantic processing and logical connections. This results in supervision and management measures often lagging behind the occurrence of problems, lacking forward-looking prediction and dynamic optimization capabilities, and making it difficult to effectively ensure the overall quality and construction safety of construction projects. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for supervising and managing the quality inspection of construction projects, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides a method for supervising and managing the quality inspection of construction projects, comprising the following steps: S1. Collect relevant data from the construction site, convert the data into standardized semantic tags, and form a structured semantic vector; the relevant data includes material testing data, process testing data, image testing data, and environmental monitoring data; S2. Using engineering sub-sections, components, testing indicators, and testing results as basic nodes, an initial engineering quality knowledge graph is established. When new construction materials or new processes appear, the knowledge graph is expanded through feature matching to achieve dynamic adaptation. S3. A graph neural network is used to learn the knowledge graph to form potential dependencies between indicators. At the same time, a self-evolutionary reasoning system is proposed, which is as follows: when the detection result is abnormal, reasoning is carried out along the knowledge graph path to generate an abnormal causal chain, and the reasoning rules are continuously corrected so that the model is optimized over time. S4. Based on the reasoning results and historical evolution patterns of knowledge graphs, predict potential risk events, classify anomalies according to their scope of impact, and automatically trigger corresponding supervision and management measures. S5. Write the rectification results and new testing data into the knowledge graph as training data for subsequent self-evolution.

[0005] Preferably, S1 includes: S11. Preprocess all raw data, including noise reduction, missing data completion, and timestamp alignment. S12. Extract and label semantic features; S13. Convert each semantically encoded data into a vector form.

[0006] Preferably, S12 includes: S121. Threshold rule labeling: Map numerical values ​​to semantic labels according to national or industry standards; S122. Extract pattern features from images using clustering and classification algorithms; S123. Semanticization of process status: Standardized description of progress and process information; S124. Semanticization of environmental conditions: Matching environmental monitoring data with construction requirements.

[0007] Preferably, S2 includes: S21. In the scenario of construction project quality inspection, based on the characteristics of the project structure and the needs of quality supervision and management, the core components of the graph are set, including nodes and edges; and through the setting of nodes and edges, an initial quality inspection knowledge graph is constructed; the nodes include project nodes, component nodes, inspection index nodes, inspection result nodes, and environmental nodes; the edges include dependency relationships, causal relationships, constraint relationships, and influence relationships. S22. When the detected data cannot match the existing semantic labels, a new node generation mechanism is triggered; the unidentified data is grouped by a clustering algorithm to discover potential new detection indicators or new abnormal patterns. S23. Once a new node is confirmed, candidate edges are automatically generated based on the correlation of historical data; initial relationships are added to the new node based on statistical correlation and causal inference models; each new edge is assigned a confidence weight, which is continuously adjusted as data accumulates. S24. Periodically perform evolutionary calculations on the knowledge graph, including: deleting low-value nodes that have not been activated for a long time; adjusting relation weights to conform to the latest detection data distribution; and merging semantically similar nodes.

[0008] Preferably, the overall framework of the self-evolutionary reasoning mechanism includes a knowledge graph input layer, a reasoning engine layer, a causal chain generation layer, and an evolutionary optimization layer; the knowledge graph input layer maps the semantically processed detection data to the attributes of knowledge graph nodes and edges; the reasoning engine layer uses graph neural networks to embed the knowledge graph and mine potential dependencies between nodes; the causal chain generation layer performs causal reasoning in the knowledge graph and generates "abnormal causal chains"; the evolutionary optimization layer evaluates the error of the causal reasoning results and dynamically corrects the rules and edge weights based on historical feedback.

[0009] Preferably, the process of generating abnormal causal chains includes: When the detection result of a certain node is abnormal, the inference engine automatically triggers a causal chain search; Backtrack along the causal relationships in the knowledge graph to find possible upstream influencing factors; Combining multiple related causal relationships into an abnormal causal chain; The causal chain results are displayed graphically.

[0010] Preferably, the self-evolving reasoning base in S3 includes: When the inferred causal chain is inconsistent with subsequent manual detection or actual results, error feedback is recorded, and the weights of the relation edges are adjusted through reinforcement learning methods to gradually optimize the inference path. Periodically perform correlation analysis on historical data; if two nodes are found to co-occur frequently in historical anomalies, it is recommended to establish a new causal edge; the initial confidence of newly added relationship edges is low, and it is gradually corrected as subsequent data accumulates; Multiple inference sub-models are used at different engineering stages; the optimal sub-model is adaptively selected to participate in inference through a model selection algorithm; the inference mechanism is evolved in stages to ensure that it can adapt to different construction conditions.

[0011] Preferably, S4 includes: S41. Based on the causal chains in the knowledge graph, deduce the potential transmission path of anomalies; S42. Quantify the probability of an anomaly occurring through Monte Carlo simulation or Bayesian inference. The output of the quantification includes two dimensions: probability of occurrence and scope of impact. S43. Abnormal risk events are divided into four levels and linked to different supervision and management measures. The abnormal risk events include Level 1, Level 2, Level 3 and Level 4. S44. After each implementation of the graded supervision and management measures, the rectification results and new testing data shall be written into the database.

[0012] Preferably, S5 includes: S51. When an anomaly is discovered and tiered supervision and management measures are taken, the construction party will implement rectification and return the rectification results to the system in the form of new test data. The rectification results include: post-rectification test index values, rectification time and process adjustments, and actual risk changes. S52. Map the rectification results to node attribute updates in the knowledge graph, and automatically extract patterns by statistically analyzing the results of multiple similar rectifications. S53. For causal edges related to the knowledge graph, recalculate the confidence level based on the rectification results; if the harm of a certain type of anomaly is lower than expected after multiple verifications, automatically lower its risk level. S54. Generate an effectiveness assessment report for rectification based on feedback data to evaluate the effects of different measures; recommend the best rectification measures for similar future problems; and at the supervision and management level, incorporate the feedback results into the project quality evaluation indicators to form a quantitative assessment of the construction party.

[0013] This invention also proposes a construction project quality inspection and supervision management system, comprising: The data acquisition unit is used to acquire multi-source detection data; The semantic processing unit is used to convert raw detection data into standardized semantic labels; The knowledge graph engine unit is used to construct an engineering quality knowledge graph and dynamically expand nodes and edges; The reasoning and evolution unit is used to perform graph neural network reasoning and self-evolutionary rule learning. Anomaly prediction and supervision unit, used for risk classification and generation of supervision and management strategies; The user interaction terminal is used to visually display abnormal causal chains, risk prediction, and supervision and management instructions.

[0014] Therefore, the present invention, employing the above-described method and system for supervising and managing construction project quality inspection, has the following beneficial effects: (1) Convert numerical, textual and image detection data into standardized semantic labels and semantic vectors to eliminate data heterogeneity and enable information from different sources to be calculated and compared in the same semantic space.

[0015] (2) By establishing a causal chain between engineering components, testing indicators and environmental conditions through knowledge graphs, it is possible not only to identify an abnormality of a certain indicator, but also to trace its cause and transmission path, thereby improving the interpretability and traceability of quality problems.

[0016] (3) Based on the reasoning results and historical evolution patterns, make probabilistic predictions of potential anomalies and match differentiated supervision and management measures according to risk level to achieve the transformation from "post-event discovery" to "pre-event prevention".

[0017] (4) By reporting back the rectification results and evaluating the effectiveness, the knowledge graph and reasoning rules are dynamically updated, and the best rectification measures are automatically recommended in subsequent similar problems, forming a closed loop of detection-prediction-supervision and management-feedback-optimization, thereby improving the pertinence and effectiveness of supervision and management.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] Figure 1 This is a flowchart of a construction project quality inspection and supervision method according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0021] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0022] Example: like Figure 1 As shown, the present invention provides a method for supervising and managing the quality inspection of construction projects, comprising the following steps: S1. Collect material testing data, process testing data, image testing data, and environmental monitoring data from the construction site, and convert the data into standardized semantic tags, such as "concrete strength ≥ 35MPa" and "steel corrosion = slight", to form a structured semantic vector.

[0023] Material testing data, such as concrete strength, rebar diameter, and mortar density, are originally numerical values; process testing data, such as pouring completion time, curing period, and welding qualification rate, are originally numerical values ​​plus text; image testing data, such as crack width and hole morphology, are converted into values ​​such as crack width = 0.2mm and hole area = 5cm² using computer vision algorithms. 2 Indicators such as temperature, humidity, and wind speed; environmental monitoring data, such as temperature, humidity, and wind speed, which are originally real-time sensor data.

[0024] In this embodiment, S1 includes: S11. Preprocess all raw data, including noise reduction, missing data completion, and timestamp alignment.

[0025] S12. Extract and label semantic features, specifically including: S121. Threshold rule labeling: Based on national or industry standards, numerical values ​​are mapped to semantic labels, such as: when the concrete strength is ≥35MPa, it is marked as qualified; when the steel reinforcement corrosion depth is >2mm, it is marked as severely corroded.

[0026] S122. Extract pattern features from the image using clustering and classification algorithms, such as: crack orientation including longitudinal cracks, transverse cracks and diagonal cracks; crack width levels including minor (<0.3mm), moderate (0.3-0.6mm) and severe (>0.6mm).

[0027] S123. Semanticization of process status: Standardize the description of progress and process information, such as marking insufficient curing if the curing completion time is less than 7 days; marking unqualified process if the welding qualification rate is less than 90%.

[0028] S124. Semanticization of environmental conditions: Match environmental monitoring data with construction requirements. For example, when the temperature is <5℃, mark it as having a risk of low-temperature construction; when the humidity is >85%, mark it as having a risk of humid environment.

[0029] S13. Convert each semantically encoded data point into a vector form for easier computation in knowledge graphs and graph neural networks. The semantic vector contains three elements: the first element represents the data category (material / structure / process / environment); the second element represents the detection index (strength / cracks / curing / humidity); and the third element represents the semantic evaluation result (-1 = unqualified, 0 = medium risk, +1 = qualified). The semantic vectors are shown in Table 1. Table 1 provides examples of semantic vectors.

[0030] S14. Use semantic data as attributes of nodes and edges in the knowledge graph, for example: Node: Concrete component; Attributes: {Strength = Insufficient, Crack Status = Moderate}; Edge: Influence relationship → from excessive humidity to insufficient intensity.

[0031] Therefore, in subsequent reasoning, we can not only identify "a certain indicator is not up to standard", but also identify the causal chain of "due to excessive humidity → insufficient strength → risk of cracking".

[0032] S2. Using engineering sub-sections, components, testing indicators, and testing results as basic nodes, an initial engineering quality knowledge graph is established. When new construction materials or new processes appear, the knowledge graph is expanded through feature matching to achieve dynamic adaptation.

[0033] In this embodiment, S2 includes: S21. In the scenario of construction project quality inspection, based on the characteristics of the project structure and the needs of quality supervision and management, the core components of the graph are defined. These core components include nodes and edges. An initial quality inspection knowledge graph is then constructed by defining the nodes and edges. This graph can be stored and queried using a graph database (such as Neo4j).

[0034] The nodes include engineering nodes, component nodes, testing index nodes, testing result nodes, and environmental nodes. Engineering nodes include the main structure, secondary structure, and decoration works; component nodes include beams, columns, walls, reinforcing steel, and concrete; testing index nodes include strength, crack width, moisture content, and welding quality; testing result nodes include qualified, unqualified, and medium risk; environmental nodes include temperature, humidity, wind speed, and rainwater infiltration; and process nodes include pouring, curing, welding, and waterproofing construction.

[0035] Edges include dependencies, causal relationships, constraint relationships, and influence relationships. For example, dependency relationship: concrete strength → affected by temperature; causal relationship: excessive humidity → insufficient strength; constraint relationship: curing time < 7 days → strength decrease; influence relationship: insufficient welding quality → decreased overall reliability of the component.

[0036] S22. When the detected data cannot match the existing semantic labels, a new node generation mechanism is triggered. The unidentified data is grouped by a clustering algorithm to discover potential new detection indicators or new abnormal patterns.

[0037] S23. Once a new node is confirmed, candidate edges are automatically generated based on the correlation of historical data. For example, new material X → strength fluctuation → crack risk. Initial relationships are added to the new node based on statistical correlation and causal inference models. Each new edge is assigned a confidence weight, which is continuously corrected as data accumulates.

[0038] S24. Periodically perform evolutionary calculations on the knowledge graph, including: deleting low-value nodes that have not been activated for a long time; adjusting relation weights to conform to the latest detection data distribution; and merging semantically similar nodes (e.g., nodes with large crack width and severe crack can be merged into one node).

[0039] Steps S22-S24 all involve expanding the knowledge graph. The following are examples illustrating S22-S24: A self-compacting concrete material was introduced into the engineering construction: A new metric, slump extension value, was detected, which is not present in the existing knowledge graph. Feature clustering revealed that this index is highly correlated with concrete workability, and a candidate node, namely the slump extension value, was automatically generated. After experts confirmed the validity of the indicator, it was written into the knowledge graph and an edge relationship with the node of concrete strength bleeding rate was automatically established. In subsequent detection data, the new node continuously accumulates data, and its edge weights are automatically adjusted according to the statistical results, thereby completing the expansion and evolution of the knowledge graph.

[0040] S3. A graph neural network (GNN) is used to learn the knowledge graph to form potential dependencies between indicators. At the same time, a self-evolutionary reasoning system is proposed. Specifically, when an anomaly occurs in the detection result, reasoning is performed along the knowledge graph path to generate an abnormal causal chain, and the reasoning rules are continuously corrected so that the model is optimized over time.

[0041] The overall framework of the self-evolving reasoning mechanism includes a knowledge graph input layer, a reasoning engine layer, a causal chain generation layer, and an evolutionary optimization layer. The knowledge graph input layer maps semantically processed detection data to the attributes of knowledge graph nodes and edges. The reasoning engine layer uses graph neural networks (GNNs) to embed the knowledge graph and mine potential dependencies between nodes. The causal chain generation layer performs causal reasoning in the knowledge graph, generating "abnormal causal chains." The evolutionary optimization layer evaluates the errors in the causal reasoning results and dynamically adjusts the rules and edge weights based on historical feedback.

[0042] The generation process of anomaly causal chains includes anomaly triggering, path tracing, chain combination, and interpretable output. Anomaly triggering: When the detection result of a node is abnormal (e.g., insufficient concrete strength), the inference engine automatically triggers a causal chain search. Path tracing: Backtracking along the causal relationship edges in the knowledge graph to find possible upstream influencing factors (e.g., insufficient curing → decreased strength). Chain combination: Combining multiple related causal relationships into an anomaly causal chain, such as: excessively high environmental humidity → hindered concrete hydration reaction → insufficient strength → increased risk of cracking. Interpretable output: The causal chain results are displayed graphically to facilitate understanding of the problem's origin by relevant personnel.

[0043] In this embodiment, the self-evolving reasoning base in S3 includes: When the inferred causal chain is inconsistent with subsequent manual detection or actual results, error feedback is recorded, and the weights of the relation edges are adjusted through reinforcement learning methods to gradually optimize the inference path. Periodically perform correlation analysis on historical data; if two nodes are found to co-occur frequently in historical anomalies, it is recommended to establish a new causal edge; the initial confidence of newly added relationship edges is low, and it is gradually corrected as subsequent data accumulates; Multiple inference sub-models are used at different engineering stages (such as foundation construction, main structure construction, and decoration stage); the optimal sub-model is adaptively selected to participate in inference through a model selection algorithm; the inference mechanism is evolved in stages to ensure that it can adapt to different construction conditions.

[0044] For example: In a certain project, a crack width of 0.6mm (severe) was detected: The initial causal chain reasoning result is: insufficient concrete strength → risk of cracking; However, subsequent manual re-inspection revealed that the crack was more related to the corrosion of the reinforcing steel. After receiving feedback, the weight of the edge related to "steel corrosion → crack risk" will be increased; At the same time, the weight of "insufficient strength → risk of cracks" is reduced to make the next reasoning more consistent with the actual situation.

[0045] S4. Based on the reasoning results and historical evolution patterns of knowledge graphs, predict potential abnormal risk events; classify abnormal risk events according to their scope of impact (local anomaly, system anomaly, global risk), and automatically trigger corresponding supervision and management measures, such as rectification, re-inspection, or work stoppage warning.

[0046] In this embodiment, S4 includes: S41. Based on the causal chains in the knowledge graph, infer the potential transmission path of anomalies. For example: excessively high environmental humidity → hindered concrete hydration → insufficient strength → increased risk of cracking. This can predict the structural risks that a certain environmental anomaly may cause in the next few days.

[0047] S42. Quantify the probability of anomalies occurring through Monte Carlo simulation or Bayesian inference. The output includes two dimensions: probability of occurrence (e.g., 70% chance of crack propagation within 10 days) and scope of impact (e.g., involving main beam and column components, accounting for 15%).

[0048] S43. Abnormal risk events are classified into four levels and linked to different supervisory and management measures. Abnormal risk events include Level 1 (low risk), Level 2 (medium risk), Level 3 (relatively high risk), and Level 4 (high risk).

[0049] When the detection index deviates slightly from the standard, such as a crack width of 0.25mm (slight), it is judged as a level one abnormal risk event. At this time, the supervising engineer is notified to pay attention, no mandatory rectification is required, and it is marked as a yellow warning on the supervision and management terminal.

[0050] When an indicator is clearly abnormal, but only in a few components or local areas, such as insufficient concrete cover thickness for a single column, it is judged as a level two abnormal risk event. In this case, local work stoppage, re-inspection, or local reinforcement is required, and the abnormal node is marked in the knowledge graph and a rectification work order is generated.

[0051] When an anomaly occurs in multiple components or multiple detection indicators and shows a propagation trend, it is judged as a level three anomaly risk event. At this time, systemic rectification measures are triggered, the relevant processes are suspended, and the information is automatically pushed to the superior supervision and management department through the supervision and management platform.

[0052] When the prediction results indicate that it may cause structural failure or serious safety hazards, such as the crack is expected to expand to 1.2mm in 7 days, affecting the load-bearing capacity, it is judged as a level four abnormal risk event. At this time, work should be stopped immediately, emergency reinforcement or rework should be implemented, and a red warning should be generated in the supervision and management system.

[0053] S44. After each implementation of the graded supervision and management measures, the rectification results and new testing data will be written into the database. These data will update the knowledge graph and prediction mechanism, making the subsequent graded supervision and management judgments more accurate.

[0054] For example: During the construction of a certain project, test data showed that the current concrete strength was 28 MPa (standard ≥30 MPa), while the ambient humidity remained at 88% for an extended period. Prediction: The strength may drop to 26 MPa within the next 5 days, which may trigger crack propagation; The risk level was determined to be Level 3, and a rectification notice was immediately issued, requiring the suspension of the relevant pouring process and reinforcement. After the construction team rectified the issue, the data was transmitted back, confirming that the strength had rebounded to 30MPa, the risk level had been reduced to Level 1 warning, and the closed-loop process was completed.

[0055] S5. The rectification results and new testing data will be written into the knowledge graph as training data for subsequent self-evolution, ensuring that the judgment ability gradually improves as the project progresses.

[0056] In this embodiment, S5 includes: S51. Once an anomaly is discovered and tiered supervision and management measures are implemented, the construction party will carry out rectification (such as reinforcement, re-inspection, and rework) and transmit the rectification results back to the system in the form of new test data. The data is uploaded to the supervision and management system via mobile terminal to ensure real-time updates. Rectification results include: post-rectification test index values ​​(e.g., crack width reduced from 0.6mm to 0.25mm), rectification time and process adjustments, and actual risk changes (e.g., load-bearing capacity recovery).

[0057] S52. Map the rectification results to the node attribute updates in the knowledge graph. By statistically analyzing the results of multiple similar rectifications (for example, the node "crack width = 0.6mm (severe)" is updated to "crack width = 0.25mm (minor)" after rectification, the causal relationship of "rectification measures → indicator improvement" is recorded to form a new causal edge), and the rules are automatically extracted (for example, increasing the maintenance time by 5 days → increasing the strength by 10%, which is solidified into a new reasoning rule).

[0058] S53. For causal edges related to the knowledge graph, recalculate the confidence level based on the rectification results; if the harm of a certain type of anomaly is lower than expected after multiple verifications, automatically lower its risk level.

[0059] S54. Generate an effectiveness assessment report for rectification based on feedback data to evaluate the effects of different measures; for similar future problems, the best rectification measures can be automatically recommended. For example, when "excessive humidity → insufficient strength" occurs again, "extended curing + admixture" can be directly recommended as a rectification measure; at the supervision and management level, the feedback results are incorporated into the project quality evaluation indicators to form a quantitative assessment of the construction party.

[0060] This invention also proposes a construction project quality inspection and supervision management system, comprising: Data acquisition unit: used to acquire multi-source detection data (materials, processes, images, environment).

[0061] Semantic processing unit: used to convert raw detection data into standardized semantic labels.

[0062] Knowledge Graph Engine Unit: Used to build engineering quality knowledge graphs and dynamically expand nodes and edges.

[0063] Reasoning and Evolution Unit: Used to perform graph neural network reasoning and self-evolutionary rule learning.

[0064] Anomaly prediction and supervision unit: used for risk classification and generation of supervision and management strategies.

[0065] User interaction terminal: used to visually display abnormal causal chains, risk prediction, and supervision and management instructions.

[0066] Therefore, the above-mentioned construction project quality inspection and supervision management method and system can significantly improve the intelligence and foresight of construction project quality inspection and supervision management, reduce safety hazards, and ensure the controllability and reliability of project quality.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for supervising and managing the quality inspection of construction projects, characterized in that, Includes the following steps: S1. Collect relevant data from the construction site, convert the data into standardized semantic tags, and form a structured semantic vector; the relevant data includes material testing data, process testing data, image testing data, and environmental monitoring data; S2. Using engineering sub-sections, components, testing indicators, and testing results as basic nodes, an initial engineering quality knowledge graph is established. When new construction materials or new processes appear, the knowledge graph is expanded through feature matching to achieve dynamic adaptation. S3. A graph neural network is used to learn the knowledge graph to form potential dependencies between indicators. At the same time, a self-evolutionary reasoning system is proposed. Specifically, when the detection result is abnormal, reasoning is performed along the knowledge graph path to generate an abnormal causal chain, and the reasoning rules are continuously corrected so that the model is optimized over time. S4. Based on the reasoning results and historical evolution patterns of knowledge graphs, predict potential risk events, classify anomalies according to their scope of impact, and automatically trigger corresponding supervision and management measures. S5. Write the rectification results and new testing data into the knowledge graph as training data for subsequent self-evolution.

2. The construction project quality inspection and supervision supervision method according to claim 1, characterized in that, S1 includes: S11. Preprocess all raw data, including noise reduction, missing data completion, and timestamp alignment. S12. Extract and label semantic features; S13. Convert each semantically encoded data into a vector form.

3. The construction project quality inspection and supervision supervision method according to claim 2, characterized in that, S12 includes: S121. Threshold rule labeling: Map numerical values ​​to semantic labels according to national or industry standards; S122. Extract pattern features from images using clustering and classification algorithms; S123. Semanticization of process status: Standardized description of progress and process information; S124. Semanticization of environmental conditions: Matching environmental monitoring data with construction requirements.

4. The construction project quality inspection and supervision supervision method according to claim 1, characterized in that, S2 includes: S21. In the scenario of construction project quality inspection, based on the characteristics of the project structure and the needs of quality supervision and management, the core components of the graph are set, including nodes and edges; and through the setting of nodes and edges, an initial quality inspection knowledge graph is constructed; the nodes include project nodes, component nodes, inspection index nodes, inspection result nodes, and environmental nodes; the edges include dependency relationships, causal relationships, constraint relationships, and influence relationships. S22. When the detected data cannot match the existing semantic labels, a new node generation mechanism is triggered; the unidentified data is grouped by a clustering algorithm to discover potential new detection indicators or new abnormal patterns. S23. Once a new node is confirmed, candidate edges are automatically generated based on the correlation of historical data; initial relationships are added to the new node based on statistical correlation and causal inference models; each new edge is assigned a confidence weight, which is continuously adjusted as data accumulates. S24. Periodically perform evolutionary calculations on the knowledge graph, including: deleting low-value nodes that have not been activated for a long time; adjusting relation weights to conform to the latest detection data distribution; and merging semantically similar nodes.

5. A method for supervising and managing the quality inspection of construction projects according to claim 1, characterized in that: The overall framework of the self-evolving reasoning mechanism includes a knowledge graph input layer, a reasoning engine layer, a causal chain generation layer, and an evolutionary optimization layer. The knowledge graph input layer maps semantically processed detection data to the attributes of knowledge graph nodes and edges. The reasoning engine layer uses graph neural networks to embed the knowledge graph and mine potential dependencies between nodes. The causal chain generation layer performs causal reasoning in the knowledge graph, generating "abnormal causal chains." The evolutionary optimization layer evaluates the errors in the causal reasoning results and dynamically adjusts the rules and edge weights based on historical feedback.

6. A method for supervising and managing the quality inspection of construction projects according to claim 5, characterized in that, The generation process of abnormal causal chains includes: When the detection result of a certain node is abnormal, the inference engine automatically triggers a causal chain search; Backtrack along the causal relationships in the knowledge graph to find possible upstream influencing factors; Combining multiple related causal relationships into an abnormal causal chain; The causal chain results are displayed graphically.

7. A method for supervising and managing the quality inspection of construction projects according to claim 6, characterized in that, The self-evolutionary reasoning base in S3 includes: When the inferred causal chain is inconsistent with subsequent manual detection or actual results, error feedback is recorded, and the weights of the relation edges are adjusted through reinforcement learning methods to gradually optimize the inference path. Periodically perform correlation analysis on historical data; if two nodes are found to co-occur frequently in historical anomalies, it is recommended to establish a new causal edge; the initial confidence of newly added relationship edges is low, and it is gradually corrected as subsequent data accumulates; Multiple inference sub-models are used at different engineering stages; the optimal sub-model is adaptively selected to participate in inference through a model selection algorithm; the inference mechanism is evolved in stages to ensure that it can adapt to different construction conditions.

8. A method for supervising and managing the quality inspection of construction projects according to claim 1, characterized in that, S4 includes: S41. Based on the causal chains in the knowledge graph, deduce the potential transmission path of anomalies; S42. Quantify the probability of an anomaly occurrence through Monte Carlo simulation or Bayesian inference. The output of the quantification includes two dimensions: probability of occurrence and scope of impact. S43. Abnormal risk events are divided into four levels and linked to different supervision and management measures. The abnormal risk events include Level 1, Level 2, Level 3 and Level 4. S44. After each implementation of the graded supervision and management measures, the rectification results and new testing data shall be written into the database.

9. A method for supervising and managing the quality inspection of construction projects according to claim 1, characterized in that, S5 includes: S51. When an anomaly is discovered and tiered supervision and management measures are taken, the construction party will implement rectification and return the rectification results to the system in the form of new test data. The rectification results include: post-rectification test index values, rectification time and process adjustments, and actual risk changes. S52. Map the rectification results to node attribute updates in the knowledge graph, and automatically extract patterns by statistically analyzing the results of multiple similar rectifications. S53. For causal edges related to the knowledge graph, recalculate the confidence level based on the rectification results; if the harm of a certain type of anomaly is lower than expected after multiple verifications, automatically lower its risk level. S54. Generate an effectiveness assessment report for rectification based on feedback data to evaluate the effects of different measures; recommend the best rectification measures for similar future problems; and at the supervision and management level, incorporate the feedback results into the project quality evaluation indicators to form a quantitative assessment of the construction party.

10. A construction project quality inspection and supervision management system, employing the construction project quality inspection and supervision management method as described in any one of claims 1-9, characterized in that, include: The data acquisition unit is used to acquire multi-source detection data; The semantic processing unit is used to convert raw detection data into standardized semantic labels; The knowledge graph engine unit is used to construct an engineering quality knowledge graph and dynamically expand nodes and edges; The reasoning and evolution unit is used to perform graph neural network reasoning and self-evolutionary rule learning. Anomaly prediction and supervision unit, used for risk classification and generation of supervision and management strategies; The user interaction terminal is used to visually display abnormal causal chains, risk prediction, and supervision and management instructions.

Images