A building engineering quality supervision method and system based on BIM

Abstract

The application discloses a kind of based on BIM's construction engineering quality supervision method and system, specifically related to building engineering quality management technical field, by collecting multiple project historical data, construct with design element and change operation as node, the cross-project knowledge graph of professional logic and time sequence relationship as edge, form risk mode basic data set;Risk propagation model of uncertain factor fusion is trained using graph neural network, Monte Carlo simulation is carried out for current design change, the propagation probability of risk on multi-professional path is calculated, and the output quantitative risk distribution map;Application reinforcement learning algorithm, generate adaptive optimization path and specific correction suggestion of minimizing high-risk node influence;After updating the suggestion to model, re-verify optimization effect and record feedback to knowledge graph, realize closed-loop learning and continuous evolution;Solve the problem that traditional method cannot model and prospective control risk propagation in complex professional logic network.

Description

Technical Field

[0001] This invention relates to the field of building engineering quality management technology, and more specifically, to a BIM-based building engineering quality supervision method and system. Background Technology

[0002] In building construction projects, design changes are an unavoidable norm. These changes may initially involve only a single discipline or a local component, but due to the complex logic and spatiotemporal relationships between various disciplines such as structure, MEP, and water supply and drainage in the Building Information Model (BIM), even minor changes can trigger a chain reaction through this network of relationships, causing deviations to be amplified at each level, ultimately leading to significant quality risks and cost overruns during the construction or operation and maintenance phases.

[0003] Currently, BIM-based quality supervision methods mainly focus on explicit, static conflict detection, such as pipeline collision checks. While these technologies can detect direct physical interference, they are powerless against implicit, dynamically evolving risks. Implicit risks are rooted in the logical dependencies of design information transmission and iteration across multiple disciplines, and their propagation paths are complex and probabilistic. Traditional methods primarily rely on engineers' personal experience for manual review of BIM models, which has significant limitations: First, it cannot formally construct and represent the risk propagation logic network across disciplines and projects, making it difficult to systematically learn risk patterns from massive amounts of historical data; second, it lacks the ability to quantitatively simulate and predict the risk chain caused by changes, and cannot calculate the probability distribution of risk propagation along different paths and its potential impact in the early stages; third, it lacks the ability to make adaptive, automated optimization suggestions based on prediction results, remaining at a passive response stage from problem discovery to manual correction.

[0004] Therefore, existing technologies essentially lack a closed-loop monitoring method capable of modeling, dynamically quantifying, predicting, and proactively generating optimization strategies for the implicit risks arising from design changes within the multidisciplinary logical network upon which they propagate. This results in many potential risks failing to be effectively identified during the design and review phases, only to surface during the construction phase, leading to severe rework and losses. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of existing technologies, this invention provides a BIM-based method and system for supervising the quality of building construction projects. By constructing a cross-project knowledge graph to encode historical risk propagation patterns, and by using graph neural networks and Monte Carlo simulations to achieve dynamic quantitative prediction of implicit risks caused by design changes, this invention addresses the problems mentioned in the background.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a BIM-based method for supervising the quality of building construction projects, comprising: Step S1: Collect building information model data, design change logs and actual fault records from multiple historical building projects, and construct a cross-project knowledge graph through a graph database, where nodes represent design elements and change operations, and edges represent logical connections and temporal dependencies between disciplines, forming a shared risk pattern basic dataset; Step S2: Based on the cross-project knowledge graph, a risk propagation model is trained using a graph neural network. By learning the propagation process of historical design changes in a multi-professional network, the law of gradual amplification of deviation is characterized. During the model training process, the parameters of uncertain factors are dynamically introduced into the hidden layer calculation of the graph neural network through an attention mechanism, and the output is a prediction function that can handle dynamic design change inputs. Step S3: Input the design changes in the current building information model into the risk propagation model for Monte Carlo simulation, calculate the propagation probability and potential deviation paths of the design changes in multiple professional paths, and output a quantitative risk distribution map; Step S4: Based on the quantitative risk distribution results, risk-driven parameter optimization is performed on the design changes. Design element nodes with a risk occurrence probability exceeding a preset risk judgment threshold are used as optimization target nodes. Implementable design change parameters in the building information model are iteratively adjusted to reduce the risk occurrence probability corresponding to the target node. During the iteration process, the adjustment effect is evaluated based on the change in the risk occurrence probability of the target node before and after the design change parameter adjustment, and design correction suggestions for revising the building information model are generated based on the evaluation results. Step S5: Apply the design modification suggestions to the current building information model, update and re-enter the risk propagation model to verify the optimization effect, insert new nodes in the verification process, use transaction management to roll back failures, and record the verification results to supplement the cross-project knowledge graph.

[0007] Preferably, in step S1, when collecting data from historical building engineering projects, the design elements are first pre-classified according to architectural, structural, and mechanical and electrical engineering disciplines, and attribute labels reflecting uncertainty are marked on the design elements; when constructing the cross-project knowledge graph, the weights of the professional association edges are calculated based on the causal relationship between the amplified deviations in historical design changes and actual fault records, so as to form a basic dataset of risk patterns that can characterize the intensity of risk propagation.

[0008] Preferably, when training the risk propagation model in step S2, historical deviation information from different upstream nodes is weighted and aggregated based on the professional relationships between design elements to simulate the uneven impact of different professions in the risk propagation process, so that the prediction function can reflect the characteristics of cross-professional risk transmission.

[0009] Preferably, during the training of the risk propagation model, the uncertain factor parameters marked in the design elements are mapped to external features, and these features participate in the calculation together with the original features of the nodes during the node representation update process, so that the risk propagation model can reflect the risk propagation trend under the participation of uncertain factors; the uncertain factor parameters are used to characterize the impact of material properties or construction condition fluctuations on the risk propagation path.

[0010] Preferably, in step S3, the multi-disciplinary propagation paths related to design changes are evaluated based on the prediction results of the risk propagation model, the cumulative risk impact intensity of each propagation path is calculated, and random sampling calculation is performed only on propagation paths whose impact intensity exceeds a preset risk screening threshold. The random sampling calculation is performed under the constraints of the prediction results of the risk propagation model to reduce the complexity of risk propagation probability calculation and focus on key risk paths.

[0011] Preferably, in step S3, the multi-disciplinary path refers to the propagation path formed by starting from the initial node where the design change occurs and following a sequence of directed edges that spans two or more different professional nodes.

[0012] Preferably, during the random sampling calculation process, a risk event is defined as a situation where the risk response value of a design element node or propagation path exceeds a preset risk judgment threshold in the sampling results. The probability of occurrence of the corresponding risk is obtained by statistically analyzing the frequency of occurrence of the risk event in all sampling times. The design change parameters are design parameters in the building information model that can be directly modified and take effect in the engineering implementation. The design correction suggestion is generated by reducing the risk level of the design element node whose probability of occurrence exceeds the preset threshold.

[0013] Preferably, in step S5, when the verification results show that the design modification can reduce the probability of the risk occurring at the target risk node, the design modification path and its corresponding risk change results are written into the cross-project knowledge graph as new risk propagation samples; when the verification results do not meet the preset risk reduction conditions, the corresponding knowledge graph update operation is canceled to maintain the validity and consistency of the risk pattern data.

[0014] Preferably, in step S5, after the verified design correction path and its corresponding risk change results are written into the cross-project knowledge graph, the prediction results of the risk propagation model for the same or similar design changes will change, thereby realizing the adaptive update of the risk propagation model input structure based on historical verification results.

[0015] To achieve the above objectives, the present invention provides the following technical solution: a BIM-based method and system for supervising the quality of building construction projects, comprising: Knowledge graph construction module: Collect building information model data, design change logs and actual failure records from multiple historical building projects, and construct a cross-project knowledge graph through a graph database to form a shared risk pattern basic dataset; Risk propagation modeling module: Based on cross-project knowledge graph, a risk propagation model is trained using graph neural network. By learning the propagation process of historical design changes in multi-professional related networks, the model can depict the pattern of gradual amplification of deviations. During the model training process, the attention mechanism is used to dynamically introduce the parameters of uncertain factors into the hidden layer calculation of the graph neural network, and the output is a prediction function that can handle dynamic design change inputs. Risk Quantification Simulation Module: Inputs design changes from the current Building Information Model into the risk propagation model for Monte Carlo simulation, calculates the propagation probability and potential deviation paths of design changes across multiple professional paths, and outputs a quantitative risk distribution map; The optimization suggestion module optimizes design changes based on the quantified risk distribution results, using risk-driven parameter optimization. Design element nodes with a risk occurrence probability exceeding a preset risk threshold are selected as optimization target nodes. The module iteratively adjusts the implementable design change parameters in the building information model to reduce the risk occurrence probability corresponding to the target node. During the iteration process, the adjustment effect is evaluated based on the change in the risk occurrence probability of the target node before and after the design change parameter adjustment, and design correction suggestions for revising the building information model are generated based on the evaluation results. Verification and update feedback module: Apply design correction suggestions to the current building information model, re-enter the risk propagation model after the update to verify the optimization effect, insert new nodes in the verification process, use transaction management to roll back failures, and record the verification results to supplement the cross-project knowledge graph.

[0016] The data transfer relationship between modules is as follows: The risk pattern dataset output by the knowledge graph construction module is directly input into the risk propagation modeling module to train the prediction model. When a new design change occurs, the risk quantification simulation module calls the model to perform risk simulation and passes the generated quantified risk distribution map to the optimization suggestion module. The optimization suggestion module drives the parameter optimization process based on this, and the generated design correction suggestions are sent to the verification update feedback module for application and verification. This module uses the verification results (whether successful or not) as new learning cases and structurally feeds them back to the knowledge graph construction module, thereby realizing the continuous evolution of the knowledge base and the iterative improvement of model performance, forming a self-learning and self-optimizing intelligent quality supervision closed loop.

[0017] The technical effects and advantages of this invention are as follows: This invention constructs a cross-project knowledge graph that integrates historical project data and professional correlation logic, and uses graph neural networks to train a predictive model that can learn the dynamic propagation law of risks and integrate uncertain factors such as material variation. This enables accurate quantitative modeling of the implicit propagation risks caused by design changes in building information models. It effectively solves the fundamental problem that traditional BIM quality review methods rely on static rules and human experience, and cannot formally represent and probabilistically predict the dynamic accumulation and evolution of risks in complex multi-professional networks.

[0018] This invention further outputs a visualized quantitative risk distribution map through Monte Carlo simulation and drives a reinforcement learning algorithm to generate adaptive optimization paths, forming an intelligent closed loop of prediction, optimization, and verification. It not only provides specific and actionable design modification suggestions for current changes to proactively suppress high risks, but also continuously feeds verification results back to the knowledge graph, enabling the risk pattern database and prediction model to evolve and reinforce each other. This effectively solves the shortcomings of existing technologies, which are limited to passive problem discovery, lack proactive optimization capabilities, and cannot accumulate and iterate knowledge. Attached Figure Description

[0019] Figure 1 This is a simplified flowchart of the BIM-based building engineering quality supervision method of the present invention. Detailed Implementation

[0020] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0021] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0022] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the scope of this application and its application or use.

[0023] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0024] See Figure 1 A simplified flowchart of a BIM-based building construction quality supervision method is provided in this invention. Figure 1 The BIM-based building construction quality supervision method shown includes the following steps: Step S1: Collect building information model data, design change logs and actual fault records from multiple historical building projects, and construct a cross-project knowledge graph through a graph database, where nodes represent design elements and change operations, and edges represent logical connections and temporal dependencies between disciplines, forming a shared risk pattern basic dataset; The Building Information Model (BIM) is a structured information model that takes the entire life cycle of a building project as its object and uses digital methods to uniformly express and manage buildings and their constituent elements. It uses a three-dimensional geometric model as its carrier and integrates spatial form information, physical attribute information, professional attribute information, logical associations and temporal relationships of design elements from multiple disciplines such as architecture, structure, and mechanical and electrical engineering. It is used to support information sharing, collaborative analysis and decision management in the design, construction and operation and maintenance stages of building projects.

[0025] Specifically, the specific implementation of step S1 is as follows: Step 101: Collect historical data from multiple completed building projects as input, including Building Information Modeling (BIM) data, design change logs, and actual fault records. BIM data provides geometric information, attribute information, and professional affiliation identifiers for design elements. Design change logs record the time points when design changes occurred and the professional scope involved. Actual fault records characterize the final results of design deviations during the construction or operation and maintenance phases. Step 102: Standardize and parse the collected historical data, extracting design elements from the Building Information Modeling (BIM) into independent data objects and classifying them according to their respective disciplines; parse the design change log, abstracting each design change into corresponding change operation data; annotate actual fault records with causal relationships to establish the correspondence between design changes and final fault results; simultaneously, label design elements with attribute tags reflecting uncertainty to describe the impact of fluctuations in materials or construction conditions; Step 103: Based on the preprocessed data, construct a cross-project knowledge graph using a graph database, storing design elements and change operations as different types of nodes. Establish logical connections based on the professional dependencies between design elements to represent the influence relationships between different disciplines; establish temporal dependency edges between change operation nodes based on the chronological order of design changes. Through unified node identification rules, achieve the associated expression of similar design elements across different projects. Step 104: Based on historical fault records, weight modeling is performed on the logically related edges in the knowledge graph to characterize the amplification effect of design deviations in professional related paths. The edge weights are calculated based on the relationship between the initial deviation and the final fault result, and are limited within a preset range to ensure numerical stability. By writing the weight information into the graph database, the knowledge graph can reflect the strength differences in risk propagation. The constructed knowledge graph serves as a shared risk pattern dataset for subsequent steps.

[0026] Step S2: Based on the cross-project knowledge graph, a risk propagation model is trained using a graph neural network. By learning the propagation process of historical design changes in a multi-professional network, the law of gradual amplification of deviation is characterized. During the model training process, the parameters of uncertain factors are dynamically introduced into the hidden layer calculation of the graph neural network through an attention mechanism, and the output is a prediction function that can handle dynamic design change inputs. Specifically, the risk propagation model training method in step S2 is as follows: Knowledge graph data loading and subgraph extraction: The shared risk pattern basic dataset constructed in step S1 is loaded into the model training environment, and the corresponding local subgraphs are extracted from the knowledge graph according to the target design elements; the local subgraphs include design element nodes, change operation nodes, and logical association edges and temporal dependency edges between nodes; the attribute information and edge weights of the subgraph nodes are encoded to form input features that the model can recognize; Risk propagation feature modeling: A risk propagation model is constructed based on subgraphs. Node features and edge weights are iteratively updated to learn the propagation patterns of risk in professionally related networks. During the node representation update process, messages from different upstream nodes are weighted and aggregated according to the historical deviation amplification characteristics corresponding to the upstream related edges of the nodes, so that the influence from different professional nodes presents an unbalanced distribution in the aggregation result. The edge weights are determined at least in part based on the professional type of the node and its association frequency in historical fault records. Meanwhile, for the uncertain factor parameters marked in the design elements, the parameters are mapped to external feature vectors and participate in the calculation together with the original features of the nodes during the node representation update process, so that the node representation can reflect the risk propagation trend under the participation of uncertain factors after multiple iterations; the node representation obtained by training is used to characterize the potential risk status of design changes in the multi-disciplinary network and serves as the input basis for subsequent risk propagation simulation and optimization steps. Model training and convergence: The risk propagation model is trained using historical design changes and actual failure results as supervision signals; the loss value is calculated by comparing the difference between the risk deviation results predicted by the model and the historical records, and the model parameters are updated accordingly; the above training process is repeated until the model output results are stable; after training, a risk propagation model for predicting the propagation of design change risks is obtained.

[0027] Furthermore, during the training of the risk propagation model, a custom neighborhood aggregation mechanism is introduced to summarize the historical deviation information of upstream edges of nodes and simulate the unevenness of the influence between different professions. The aggregation mechanism weights the historical deviation amplification features of upstream edges, enabling the model to prioritize high-risk propagation paths when updating node representations. The weight allocation is dynamically adjusted based on the causal chain reflected in historical fault records, thereby avoiding the weakening of key risk paths caused by uniform aggregation. Through the hierarchical aggregation of multi-layer neighborhoods, the model forms a risk propagation view from local to global, which is more in line with the actual logic of cross-professional risk transmission in construction engineering.

[0028] Furthermore, to enhance the model's adaptability to uncertainties, material variation parameters labeled in the design elements are introduced as external embedding vectors into the hidden layer calculation process of the risk propagation model. These embedding vectors are dynamically injected during training iterations and fused with node feature vectors to simulate the impact of material performance fluctuations on the risk propagation path. During training, model parameters are progressively optimized through multiple iterations, and the convergence of the model is determined by analyzing the loss trend, thus avoiding overfitting or ineffective training. This approach enables the model to consistently output reasonable risk propagation predictions even when facing design change scenarios involving uncertainties.

[0029] Step S3: Input the design changes in the current building information model into the risk propagation model for Monte Carlo simulation, calculate the propagation probability and potential deviation path of the design changes on the multi-professional path (multi-professional path refers to the propagation path formed by the directed edge sequence that crosses two or more different professional nodes starting from the starting node where the design change occurs, and output a quantitative risk distribution map; Among them, the quantitative risk distribution map is used to characterize the risk state space of design changes in the multi-disciplinary network and serves as the input basis for state definition in subsequent optimization steps; Specifically, the risk propagation prediction and quantification process in step S3 includes: Step 301: Design Change Mapping and Related Subgraph Construction For design changes in the current Building Information Model (BIM), the design element nodes and change operation nodes corresponding to the changes are identified and mapped to node representations in the risk propagation model. Starting from the change node, design element nodes with logical or temporal dependencies are extracted from the cross-project knowledge graph to construct a relevant subgraph for risk propagation calculation. The relevant subgraph contains multi-disciplinary design elements that may be affected by the design changes and their relationships, thus limiting the scope of risk propagation calculation. Step 302: Initial Risk State Construction and Basic Prediction Based on the specific content of the design change, an initial risk state is constructed to describe the initial deviation conditions introduced by the design change to the corresponding nodes when it occurs; this initial risk state serves as the model input and is correlated with relevant sub-nodes. Figure 1 The trained risk propagation model is input, and forward propagation calculation is performed. The model outputs the risk response value of each node, which is used to characterize the potential propagation trend of design changes in a multi-disciplinary network. This output result serves as the basis for prediction in subsequent stochastic simulations. Step 303: Pre-screening of high-impact paths Based on the basic prediction results, potential risk propagation paths in the relevant subgraphs are evaluated, and the cumulative impact intensity of each path is calculated. Paths are sorted according to their impact intensity, and paths with an impact intensity exceeding a preset threshold are selected as high-impact paths. This pre-screening step narrows down the path range for subsequent simulations, allowing risk assessment to focus on propagation paths that have a significant impact on project quality. The preset threshold is determined by statistically analyzing the cumulative risk response values ​​of each potential propagation path in the relevant subgraphs and combining the verified risk levels in historical design changes and actual failure records. A critical value that can distinguish between high-risk and low-risk paths is selected as the threshold, so that paths exceeding the threshold are identified as high-impact paths that have a significant impact on the quality of the project.

[0030] Step 304: Monte Carlo Risk Propagation Simulation Within the high-impact path range, the initial risk state is used as input conditions to perform multiple rounds of random sampling simulation of the risk propagation process. In each round of simulation, the deviation impact is propagated step by step according to the logical correlation and edge weights on the path, and random perturbations are introduced in the propagation process to reflect the impact of uncertainty factors. Through multiple random sampling, the risk response of each node under different propagation results is obtained. The simulation process is used to characterize the uncertainty features of risk propagation. Step 305: Statistical and Quantitative Output of Risk Probability Distribution Statistical analysis is performed on the risk propagation results obtained from multiple rounds of random simulation. When the risk response value of each design element node or propagation path exceeds the preset risk judgment threshold during the simulation, it is regarded as a risk event. The occurrence frequency of the risk event in all simulation rounds is counted. By calculating the ratio of the occurrence frequency to the total number of simulations, the risk occurrence probability of the corresponding node or path is obtained, which is used to quantify the risk propagation characteristics of design changes on multi-disciplinary paths.

[0031] Furthermore, after completing the pre-screening of high-impact paths, real-time design change data is injected as initial state variables, and multiple rounds of random sampling simulation are performed on the paths. During the simulation, the impact of deviations is accumulated step by step according to the logical relationships on the paths to characterize the propagation process of design changes in the multi-disciplinary network. By repeatedly sampling and statistically analyzing different propagation results, a probability distribution information reflecting the degree of deviation accumulation is formed. The simulation is used to quantify the uncertainty of risk propagation and its potential evolution trend.

[0032] Step S4: Based on the quantitative risk distribution results, risk-driven parameter optimization is performed on the design changes. Design element nodes with a risk occurrence probability exceeding a preset risk judgment threshold are used as optimization target nodes. Implementable design change parameters in the building information model are iteratively adjusted to reduce the risk occurrence probability corresponding to the target node. During the iteration process, the adjustment effect is evaluated based on the change in the risk occurrence probability of the target node before and after the design change parameter adjustment, and design correction suggestions for revising the building information model are generated based on the evaluation results. The high-probability risk node is a design element node in the risk distribution result obtained in step S3 whose risk occurrence probability exceeds a preset risk judgment threshold. It is used to characterize the key risk location that has a significant impact on the quality of the project. The reward value is determined based on the change in the risk occurrence probability of the high-probability risk node before and after the adjustment of the design parameters. It is used to characterize the contribution of the adjustment to the risk reduction effect. A smoothing constant is introduced in the reward calculation process to perform numerical processing on the case where the risk probability is close to zero or close to one, so as to ensure the stability of the optimization process. Specifically, step S4: the design change optimization path generation process includes: Step 401: Optimize state and action definitions Based on the risk distribution results obtained in step S3, a state representation for design change optimization is constructed. The state includes at least the current design change parameters and the probability information of the risk occurrence corresponding to the parameters. The ways to adjust the design change parameters are defined as a set of actions. Each action corresponds to one or more adjustment operations of the design parameters. Through the combination of the state and actions, the optimization process can describe the relationship between the adjustment of design change parameters and risk changes. Step 402: Optimize Path Search and Evaluation During the optimization process, different design parameter adjustment paths are explored through multiple rounds of state updates. After each adjustment, the effect of the adjustment is evaluated based on the changes in risk distribution. The evaluation results are used to determine whether the current adjustment effectively reduces the impact of high-risk nodes. By continuously iterating and screening, adjustment paths with better risk reduction effects are selected. The risk state space constructed based on the risk distribution results is searched iteratively, so that the adjustment of design parameters is always constrained by the risk propagation characteristics. The adjustment effect is evaluated by comparing the changes in the probability of high-probability risk nodes before and after the adjustment; if the adjustment reduces the probability of risk occurrence corresponding to high-probability risk nodes, it is considered an effective adjustment; through multiple rounds of iterative search, a design parameter adjustment path that can reduce the overall risk level is gradually formed. Step 403: Generation of Design Revision Suggestions The effective adjustment paths obtained during the optimization process are compiled into design correction suggestions, which include the direction of adjustment of design change parameters and the corresponding risk reduction effect. The design correction suggestions are output in a structured form to guide the correction of design changes in the building information model.

[0033] Step S5: Apply the design modification suggestions to the current building information model, update and re-enter the risk propagation model to verify the optimization effect, insert new nodes in the verification process, use transaction management to roll back failures, and record the verification results to supplement the cross-project knowledge graph and realize the continuous evolution of risk patterns; Specifically, the optimization verification and knowledge update process in step S5 includes: Step 501: Design Revision Application and Model Update The design correction suggestions generated in step S4 are applied to the current Building Information Model (BIM) to update the corresponding design elements. The updated model is then remapped to the input space of the risk propagation model. New initial risk conditions are constructed based on the updated design state, providing a consistent input basis for subsequent validation. Step 502: Optimization effect verification The updated design model is input into the risk propagation model, and the risk propagation simulation calculation is repeated. The risk distribution results before and after optimization are compared to assess the changes in risk. The effectiveness of the design modification is determined by comparing the changes in the number of high-risk nodes or the risk level. The verification results are used to confirm the feasibility of the optimization path. Step 503: Risk Model Knowledge Update Record the validated design modification paths and their corresponding risk changes. Write these records into a shared risk pattern dataset as new nodes or relationships. Continuously add new risk propagation and optimization cases to achieve dynamic evolution of the risk pattern. The updated dataset is used to support risk prediction and optimization in subsequent projects. Furthermore, the design correction suggestions generated in step S4 are applied to the current building information model, and the risk propagation calculation is re-executed under the same risk propagation model conditions as in step S3 to verify the risk reduction effect of the design correction on high-probability risk nodes. When the verification results show that the design correction can reduce the probability of risk occurrence corresponding to the target risk node, the design correction path and its corresponding risk change results are written as new risk propagation samples into the cross-project knowledge graph to update the input structure of the subsequent risk propagation model. When the verification results do not meet the preset risk reduction conditions, the corresponding knowledge graph update operation is canceled to ensure the effectiveness and consistency of the risk patterns in the cross-project knowledge graph.

[0034] Furthermore, in step S1, when collecting data from historical building engineering projects, the design elements are first pre-classified according to architectural, structural, and mechanical and electrical engineering disciplines, and attribute labels reflecting uncertainty are marked on the design elements. When constructing the cross-project knowledge graph, the weights of the professional association edges are calculated based on the causal relationship between the amplified deviations in historical design changes and actual failure records, so as to form a basic dataset of risk patterns that can characterize the intensity of risk propagation.

[0035] Furthermore, when training the risk propagation model in step S2, historical deviation information from different upstream nodes is weighted and aggregated based on the professional relationships between design elements to simulate the uneven impact of different professions in the risk propagation process, so that the prediction function can reflect the characteristics of cross-professional risk transmission.

[0036] Furthermore, during the training process of the risk propagation model, the uncertain factor parameters marked in the design elements are mapped to external features, and these features participate in the calculation together with the original features of the nodes during the node representation update process, so that the risk propagation model can reflect the risk propagation trend under the participation of uncertain factors.

[0037] Furthermore, in step S3, the multi-disciplinary propagation paths related to design changes are evaluated based on the prediction results of the risk propagation model, the cumulative risk impact intensity of each propagation path is calculated, and only the propagation paths whose impact intensity exceeds the preset risk screening threshold are randomly sampled and calculated to reduce the complexity of risk propagation probability calculation and focus on key risk paths.

[0038] Furthermore, during the random sampling calculation process, a risk event is defined as a situation where the risk response value of a design element node or propagation path exceeds a preset risk judgment threshold in the sampling results. The probability of the risk event is obtained by statistically analyzing the frequency of occurrence of the risk event in all sampling times.

[0039] Furthermore, in step S4, the design change parameters are design parameters in the building information model that can be directly modified and take effect in the project implementation. The design correction suggestions are generated by reducing the risk level corresponding to the design element node whose probability of risk occurrence exceeds a preset threshold.

[0040] Furthermore, when generating design modification suggestions, the adjustment effect is evaluated based on the change in the probability of occurrence of high-probability risk nodes before and after the design change parameter adjustment, and the design parameter adjustment path with better risk reduction effect is selected based on the evaluation results.

[0041] Furthermore, in step S5, when the verification results show that the design modification can reduce the probability of the risk occurrence corresponding to the target risk node, the design modification path and its corresponding risk change results are written into the cross-project knowledge graph as new risk propagation samples; when the verification results do not meet the preset risk reduction conditions, the corresponding knowledge graph update operation is canceled to maintain the validity and consistency of the risk pattern data.

[0042] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A BIM-based method for supervising the quality of building construction projects, characterized in that, include: Step S1: Collect building information model data, design change logs and actual fault records from multiple historical building projects, and construct a cross-project knowledge graph through a graph database, where nodes represent design elements and change operations, and edges represent logical connections and temporal dependencies between disciplines, forming a shared risk pattern basic dataset; Step S2: Based on the cross-project knowledge graph, a risk propagation model is trained using a graph neural network. By learning the propagation process of historical design changes in a multi-professional network, the law of gradual amplification of deviation is characterized. During the model training process, the parameters of uncertain factors are dynamically introduced into the hidden layer calculation of the graph neural network through an attention mechanism, and the output is a prediction function that can handle dynamic design change inputs. Step S3: Input the design changes in the current building information model into the risk propagation model for Monte Carlo simulation, calculate the propagation probability and potential deviation paths of the design changes in multiple professional paths, and output a quantitative risk distribution map; Step S4: Based on the quantitative risk distribution results, risk-driven parameter optimization is performed on the design changes. Design element nodes with a risk occurrence probability exceeding a preset risk judgment threshold are used as optimization target nodes. Implementable design change parameters in the building information model are iteratively adjusted to reduce the risk occurrence probability corresponding to the target node. During the iteration process, the adjustment effect is evaluated based on the change in the risk occurrence probability of the target node before and after the design change parameter adjustment, and design correction suggestions for revising the building information model are generated based on the evaluation results. Step S5: Apply the design modification suggestions to the current building information model, update and re-enter the risk propagation model to verify the optimization effect, insert new nodes in the verification process, use transaction management to roll back failures, and record the verification results to supplement the cross-project knowledge graph.

2. The BIM-based building construction quality supervision method according to claim 1, characterized in that, In step S1, when collecting data from historical building engineering projects, the design elements are first pre-classified according to architectural, structural and mechanical and electrical engineering disciplines, and attribute labels reflecting uncertainty are marked on the design elements. When constructing the cross-project knowledge graph, the weights of professional association edges are calculated based on the causal relationship of amplified deviations in historical design changes and actual failure records, so as to form a basic dataset of risk patterns that can characterize the intensity of risk propagation.

3. The BIM-based building construction quality supervision method according to claim 1, characterized in that, In step S2, when training the risk propagation model, historical deviation information from different upstream nodes is weighted and aggregated based on the professional relationships between design elements to simulate the uneven impact of different professions in the risk propagation process, so that the prediction function can reflect the characteristics of cross-professional risk transmission.

4. The BIM-based building construction quality supervision method according to claim 3, characterized in that, During the training of the risk propagation model, the uncertain factor parameters marked in the design elements are mapped to external features and participate in the calculation together with the original features of the nodes during the node representation update process, so that the risk propagation model can reflect the risk propagation trend under the participation of uncertain factors; the uncertain factor parameters are used to characterize the impact of material properties or construction condition fluctuations on the risk propagation path.

5. A BIM-based method for supervising the quality of building construction projects according to claim 1, characterized in that, In step S3, the multi-disciplinary propagation paths related to design changes are evaluated based on the prediction results of the risk propagation model. The cumulative risk impact intensity of each propagation path is calculated, and random sampling calculation is performed only on propagation paths whose impact intensity exceeds the preset risk screening threshold. The random sampling calculation is performed under the constraints of the prediction results of the risk propagation model to reduce the complexity of risk propagation probability calculation and focus on key risk paths.

6. A BIM-based method for supervising the quality of building construction projects according to claim 5, characterized in that, In step S3, the multi-disciplinary path refers to the propagation path formed by starting from the initial node where the design change occurs and following a sequence of directed edges that spans two or more different discipline nodes.

7. A BIM-based method for supervising the quality of building construction projects according to claim 6, characterized in that, During the random sampling calculation process, a risk event is defined as a situation in which the risk response value of a design element node or propagation path exceeds a preset risk judgment threshold in the sampling results. The probability of the risk event is obtained by statistically analyzing the frequency of occurrence of the risk event in all sampling times. The design change parameters are design parameters in the building information model that can be directly modified and take effect in the project implementation. The design correction suggestions are generated by reducing the risk level of the design element node whose probability of risk occurrence exceeds a preset threshold.

8. A BIM-based method for supervising the quality of building construction projects according to claim 1, characterized in that, In step S5, when the verification results show that the design modification can reduce the probability of the risk occurrence corresponding to the target risk node, the design modification path and its corresponding risk change results are written into the cross-project knowledge graph as new risk propagation samples; when the verification results do not meet the preset risk reduction conditions, the corresponding knowledge graph update operation is canceled to maintain the validity and consistency of the risk pattern data.

9. A BIM-based method for supervising the quality of building construction projects according to claim 8, characterized in that, In step S5, after the verified design correction path and its corresponding risk change results are written into the cross-project knowledge graph, the prediction results of the risk propagation model for the same or similar design changes will change, thereby realizing the adaptive update of the risk propagation model input structure based on historical verification results.

10. A BIM-based building construction quality supervision system, used to implement the method described in any one of claims 1-9, characterized in that, include: Knowledge graph construction module: Collect building information model data, design change logs and actual failure records from multiple historical building projects, and construct a cross-project knowledge graph through a graph database to form a shared risk pattern basic dataset; Risk propagation modeling module: Based on cross-project knowledge graph, a risk propagation model is trained using graph neural network. By learning the propagation process of historical design changes in multi-professional related networks, the model can depict the pattern of gradual amplification of deviations. During the model training process, the attention mechanism is used to dynamically introduce the parameters of uncertain factors into the hidden layer calculation of the graph neural network, and the output is a prediction function that can handle dynamic design change inputs. Risk Quantification Simulation Module: Inputs design changes from the current Building Information Model into the risk propagation model for Monte Carlo simulation, calculates the propagation probability and potential deviation paths of design changes across multiple professional paths, and outputs a quantitative risk distribution map; The optimization suggestion module optimizes design changes based on the quantified risk distribution results, using risk-driven parameter optimization. Design element nodes with a risk occurrence probability exceeding a preset risk threshold are selected as optimization target nodes. The module iteratively adjusts the implementable design change parameters in the building information model to reduce the risk occurrence probability corresponding to the target node. During the iteration process, the adjustment effect is evaluated based on the change in the risk occurrence probability of the target node before and after the design change parameter adjustment, and design correction suggestions for revising the building information model are generated based on the evaluation results. Verification and update feedback module: Apply design correction suggestions to the current building information model, re-enter the risk propagation model after the update to verify the optimization effect, insert new nodes in the verification process, use transaction management to roll back failures, and record the verification results to supplement the cross-project knowledge graph.

Images