Digital twinning dynamic modeling system for engineering quality monitoring

Abstract

The invention relates to the technical field of digital twinning and engineering monitoring, particularly discloses a digital twinning dynamic modeling system for engineering quality monitoring, and aims at solving the problems that a digital twinning model is poor in universality, multi-source heterogeneous data fusion is difficult, and monitoring is discontinuous and inaccurate due to static solidification of a model topological structure. The system comprises a multi-source heterogeneous data sensing module, a data fusion and semantic alignment engine, a dynamic topology evolution module and a twin model driving and visualization service layer which are connected in sequence. According to the system, by fusing multi-source data and constructing a fusion data field with unified semantics, model topology is driven to synchronously and adaptively evolve along with engineering entities, and quality analysis, simulation and visualization are performed based on an updated model, so that full-life-cycle continuous and accurate monitoring and prediction of engineering quality are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital twinning and engineering monitoring, in particular to a digital twinning dynamic modeling system for engineering quality monitoring. BACKGROUND

[0002] As a key enabling technology for deep integration of the physical world and the information space, digital twinning is profoundly transforming multiple fields such as industrial manufacturing, urban management, and infrastructure operation. Its core lies in data-driven construction of a virtual model that interacts with physical entities in real time and evolves synchronously, thereby achieving state perception, process simulation, and decision optimization of complex systems. In the field of engineering construction, digital twinning provides a new technical paradigm for improving the transparency, real-time performance, and intelligent level of engineering quality monitoring.

[0003] Among them, the digital twinning application for engineering quality monitoring aims to map the physical entities, construction processes, and multi-source monitoring data on the construction site into a virtual twin that is computable, analyzable, and predictable, in order to dynamically track and evaluate multi-dimensional quality indicators such as structural deformation, material performance, and construction process compliance. However, existing technical solutions face significant challenges in achieving this goal. Existing digital twinning-based engineering monitoring systems mostly focus on specific disaster types or equipment failure prediction, and their model construction heavily relies on prior knowledge bases and pre-defined physical or degradation paths, resulting in poor model versatility and difficulty in adapting to complex and variable quality monitoring scenarios in civil engineering. At the same time, these systems lack sufficient processing capabilities for multi-source heterogeneous monitoring data, and lack effective mechanisms for deep fusion and semantic alignment of geometric, physical, environmental, and process data from different sensors, different stages, and different formats to drive dynamic updates of the digital twinning model.

[0004] More critically, the topology of existing models is usually static or pre-defined, and cannot adaptively evolve and reconstruct with the construction progress, structural assembly, or modification activities of the engineering entity, resulting in a break in the synchronization between the digital twin and the physical entity, and the inability to truly reflect the continuous state changes of engineering quality throughout the life cycle. Therefore, how to construct a dynamic digital twinning model that can deeply integrate multi-source heterogeneous data and adaptively evolve with engineering progress has become a technical problem to be solved for realizing high-precision intelligent monitoring of engineering quality throughout the entire process. SUMMARY

[0005] The present application aims to provide a digital twinning dynamic modeling system for engineering quality monitoring to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solution: a digital twinning dynamic modeling system for engineering quality monitoring, comprising:

[0007] The multi-source heterogeneous data perception module is used to collect and access geometric morphology data, physical state data, environmental parameter data, and construction process data from the engineering quality monitoring site in real time. The data fusion and semantic alignment engine performs spatiotemporal benchmark unification and multimodal feature extraction on the input multi-source heterogeneous data, and performs cross-modal semantic association based on a predefined engineering quality monitoring ontology library to generate a fused data field with a unified spatiotemporal benchmark and engineering semantic annotation. The dynamic topology evolution module is used to receive the fused data field, and based on the construction activity semantic tags and associated geometric feature changes contained in the field, it uses a topology relationship inferencer to infer the topological relationships between components in the engineering entity in real time, and drives the model incremental updater to synchronously evolve and reconstruct the topology of the digital twin model. The twin model driving and visualization service layer performs numerical simulation analysis of engineering quality, quantitative evaluation of key performance indicators, and multi-dimensional visualization interaction based on the updated digital twin model.

[0008] Furthermore, the data fusion and semantic alignment engine includes: a spatiotemporal reference unification module, which uses the global satellite navigation system timing signal as a unified time reference and the global coordinate system of the engineering design as a unified spatial reference to convert and align the time and space information of the input data; a multimodal feature extraction module, which extracts surface curvature, edge feature points, and volume change rate from the geometric morphology data to form a geometric feature vector, extracts spectral features, statistical features, and temporal correlation from the physical state data to form a physical feature vector, extracts temperature gradient and wind speed and direction statistics from the environmental parameter data to form an environmental feature vector, and extracts semantic feature vectors from the text information in the construction process data through a pre-trained natural language processing model; and a semantic association module, which maps the multimodal feature vectors to the corresponding engineering concept nodes in the ontology based on the engineering quality monitoring ontology and establishes semantic association relationships between the feature vectors.

[0009] Furthermore, the dynamic topology evolution module includes: a model primitive library, which pre-stores parameterized three-dimensional geometric primitives with associated physical attribute templates; a topology relationship inferencer, which infers the topology relationships between components based on the semantic tags of construction activities and changes in geometric features in the fused data field; and a model incremental updater, which performs addition, deletion, or modification operations on components and their topology relationships in the digital twin model based on the output of the topology relationship inferencer.

[0010] Furthermore, the topology relationship inference engine operates based on a hybrid rule- and data-driven decision framework, which includes: an explicit engineering rule knowledge base, which encodes the standard component connection structure and process requirements using generative rules; and an implicit graph neural network model, which takes the current scene feature map constructed from the fused data field as input and performs partially supervised training based on the output of the engineering rule knowledge base. In the partially supervised training, the determined topological relationships (such as beam A and column B being rigidly connected) output by the engineering rule knowledge base for the sample scene are encoded as one-hot vectors, serving as soft labels for the output layer of the graph neural network. The loss function consists of a weighted cross-entropy loss and a self-supervised contrastive loss, with a weight ratio of 7:3. The final output of the topology relationship inference engine is a weighted fusion of the inference result from the rule knowledge base and the prediction result from the graph neural network model, with the weights dynamically adjusted according to the confidence level of the input data.

[0011] Furthermore, the semantic association module achieves cross-modal semantic alignment through a two-layer graph neural network architecture, including: an intra-modal feature graph network, used to perform graph structure modeling on geometric, physical, environmental and text semantic feature vectors respectively, and learn the feature relationships within the modality; and a cross-modal alignment graph network, which, through a learnable attention mechanism, performs multiple rounds of message passing and feature aggregation between nodes in each modal feature graph and concept nodes in the engineering quality monitoring ontology library, so that different modal feature nodes describing the same engineering entity or phenomenon have similar embedding vector representations in the ontology concept space.

[0012] Furthermore, the operations performed by the model incremental updater include: when a new component is inferred, instantiating the corresponding parameterized geometric primitive from the model primitive library, assigning geometric and physical parameters according to the fused data field, and connecting it to the existing model network according to the inferred topology; when a component is inferred to be removed, deleting the corresponding primitive instance and all its associated relationships from the model network; and when a change in component attributes is detected, updating the parameter values ​​in the physical attribute template associated with the corresponding primitive instance.

[0013] Furthermore, the twin model-driven and visualization service layer includes: a numerical simulation kernel, which performs simulation calculations of structural mechanics, temperature field, or durability based on the current geometric topology and physical properties of the digital twin model; a quality index evaluation module, which predefines the calculation logic of key performance indicators, extracts parameters from the digital twin model and calls the calculation logic to output quantitative evaluation results; and a time series prediction submodule, which takes historical fused data field and quality evaluation results as input, learns the evolution law of indicators through a time series prediction model, predicts the quality index values ​​under future operating conditions, and generates early warning information.

[0014] Furthermore, the time-series prediction submodule uses a long short-term memory network as its time-series prediction model. The long short-term memory network regulates cell states through gating mechanisms such as input gates, forget gates, and output gates to learn the long-term dependencies of time-series data on quality indicators.

[0015] Furthermore, the multi-source heterogeneous data perception module, data fusion and semantic alignment engine, dynamic topology evolution module, and twin model-driven and visualization service layer are encapsulated as independently deployable microservices; the microservices communicate with each other through message queues and event buses, and data streams are transmitted in an event-driven manner.

[0016] Furthermore, the multi-source heterogeneous data sensing module has a built-in unified data interface protocol stack, which includes: a point cloud data protocol for point cloud data transmission, an industrial IoT protocol for sensor time-series data transmission, a real-time streaming protocol for video streaming, and an application programming interface for business system data interaction. The protocol stack encapsulates heterogeneous data sources into standardized data packets containing metadata headers and payload data bodies for access. The metadata header at least includes a data source identifier, a collection timestamp, a spatial reference frame identifier, and a data type encoding.

[0017] Compared with the prior art, the beneficial effects of the present invention are:

[0018] This invention fundamentally solves the problems of data silos and static models in engineering quality monitoring by constructing a complete system architecture that includes a multi-source heterogeneous data perception module, a data fusion and semantic alignment engine, a dynamic topology evolution module, and a twin model-driven and visualization service layer. The data fusion and semantic alignment engine utilizes spatiotemporal registration, multimodal feature extraction, and ontology-based semantic alignment techniques to deeply fuse multi-source heterogeneous data from geometric, physical, environmental, and process sources into a unified semantic data field. This provides a high-quality, highly consistent data foundation for model construction, overcoming the shortcomings of existing technologies such as superficial data utilization and poor correlation.

[0019] The dynamic topology evolution module proposed in this invention, through the collaborative work of the model primitive library, topology relationship inferencer, and model incremental updater, enables the adaptive following and reconstruction of the digital twin model's topology structure to the construction, assembly, and modification activities of the physical engineering entity. This mechanism ensures that the digital twin maintains geometric and logical synchronization with the physical entity throughout its entire lifecycle, overcoming the limitations of existing models with fixed topologies that cannot reflect the continuous changes in the engineering process, and providing a realistic and reliable virtual carrier for full-process, high-precision quality monitoring.

[0020] This invention deeply integrates physical simulation, quality indicator evaluation, and time-series prediction capabilities at the twin model-driven layer. It not only achieves accurate assessment of the current engineering status but also enables future performance prediction and risk warning based on dynamically updated models. This closed-loop "perception-modeling-analysis-prediction" elevates engineering quality monitoring from passive response to a new level of proactive prediction and decision support, significantly enhancing the predictability and scientific rigor of quality management.

[0021] This invention is implemented using a microservice architecture, with each core functional module decoupled and independent. Through standardized interface communication, the system has high scalability, maintainability, and elastic deployment capabilities, and can flexibly adapt to engineering quality monitoring scenarios of different scales and complexities, from single buildings to large infrastructures. It has broad engineering applicability and promotional value. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the core principle framework of the data fusion and semantic alignment engine in this invention; Figure 2 This is a logical flow diagram of the dynamic topology evolution module in this invention. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0024] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” and “described” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the described features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0025] This invention provides a digital twin dynamic modeling system for engineering quality monitoring, comprising:

[0026] The multi-source heterogeneous data sensing module is used to collect and access multi-dimensional heterogeneous data streams from the engineering quality monitoring site in real time and perform normalization processing. The data streams specifically include geometric morphology data, physical state data, environmental parameter data, and construction process data. Geometric morphology data originates from laser scanners, photogrammetric equipment, and deformation monitoring stations based on the Global Navigation Satellite System deployed at key parts of the engineering structure. Physical state data originates from strain sensors, acceleration sensors, tilt sensors, temperature sensors, and humidity sensors embedded or attached inside the engineering structure. Environmental parameter data originates from weather stations, anemometers, noise monitors, and video surveillance equipment deployed at the construction site. Construction process data originates from the Building Information Modeling (BIM) schedule module, material arrival record database, and construction log text records.

[0027] The multi-source heterogeneous data sensing module also includes a data access and normalization processing unit. This unit accesses the original multi-heterogeneous data streams through adapted point cloud data protocols, industrial IoT protocols, real-time streaming protocols, and application programming interfaces. It decodes the data, appends spatiotemporal stamps, unifies units, and restructures the data to complete normalization. Finally, it encapsulates the data into standardized data packets containing a unified metadata header and a payload data body. Each data packet includes a mandatory metadata header and payload data body. The metadata header is required to include a globally unique identifier for the data source, a data acquisition timestamp, a data spatial reference frame identifier, data type encoding, and a data packet sequence number. The payload data body carries the actual monitoring data or business data that has undergone preliminary verification and format conversion.

[0028] The data fusion and semantic alignment engine receives standardized, multi-dimensional heterogeneous data streams and performs spatiotemporal benchmark unification and multimodal feature extraction on the input multi-source heterogeneous data. Based on a predefined engineering quality monitoring ontology library, it performs cross-modal semantic association to generate a fused data field with a unified spatiotemporal benchmark and engineering semantic annotations. The engineering quality monitoring ontology library is constructed using the OWL language and includes engineering construction classes, construction activity classes, attribute classes, and object attributes.

[0029] The unified spatiotemporal reference module uses the global navigation satellite system (GNSS) timing signal as the unified time reference and the engineering-designed global coordinate system as the unified spatial reference to transform and align the time and spatial information of the input data. The internal timestamp of any incoming data packet is converted to this unified Coordinated Universal Time (UTC) reference. The spatial reference adopts the engineering-designed global coordinate system, which was defined in the early stages of system design; specifically, it is a geodetic coordinate system. All data containing spatial information, such as laser point cloud coordinates, GNSS monitoring point coordinates, and sensor installation locations, must have their coordinates transformed to this global coordinate system using preset transformation parameters.

[0030] The multimodal feature extraction module extracts digital feature vectors that can characterize the essence of the engineering state from the spatiotemporally unified original data.

[0031] Geometric feature vector extraction: For geometric data such as point clouds and triangular meshes, the corresponding geometric feature vectors are generated by calculating the surface curvature distribution of the point cloud or mesh model, identifying feature point sets such as structural edges and corners, and calculating the volume change rate or displacement field of a specific region in a continuous time interval.

[0032] Physical feature vector extraction: First, the time-series signal is converted to the frequency domain, and its dominant frequency, spectral energy distribution and other spectral features are extracted. At the same time, statistical features such as mean, variance, peak factor and kurtosis are calculated in the time domain, and the time-domain correlation coefficient between multi-channel signals is analyzed to form a physical feature vector.

[0033] Environmental feature vector extraction: Calculate the temperature gradient field, wind speed and direction statistical characteristics such as average wind speed and wind rose diagram, and the diurnal variation curve characteristics of light intensity within the construction site space to form an environmental feature vector;

[0034] Text semantic feature vector extraction: By calling a pre-trained natural language processing model (a model based on the Transformer architecture) to understand the text, semantic feature vectors related to construction techniques, quality inspection items, and abnormal events are extracted.

[0035] The semantic association module establishes feature vectors for different modalities based on a pre-defined engineering quality monitoring ontology library and a dual-layer neural network architecture. The dual-layer neural network architecture includes an intra-modal feature graph network and a cross-modal alignment graph network. The intra-modal feature graph network constructs independent graph structures from the extracted geometric, physical, environmental, and textual semantic feature vector sets, with nodes representing feature vectors and edges defined by spatial proximity, temporal continuity, or semantic similarity between features, aiming to learn and enhance the relationships within each modality. The cross-modal alignment graph network constructs a unified heterogeneous knowledge graph by combining nodes in each modal feature graph with concept nodes in the engineering quality monitoring ontology library. Through a learnable attention mechanism, the network performs multiple rounds of message passing between nodes in different modalities and ontology concept nodes. Feature nodes from the same physical entity (e.g., a specific beam) but belonging to different modalities (e.g., the beam's "deformation" geometric feature and "strain" physical feature) will aggregate their features towards the ontology concept node (e.g., the "concrete beam_01" instance) during message passing. After training, these heterogeneous feature nodes describing the same entity will tend to be highly similar in their embedding vector representations in the ontology concept space, thereby completing semantic alignment and association, and outputting a structured fused data field. The fused data field not only includes feature data, timestamps and spatial locations, but also carries semantic tags from the ontology library obtained during the alignment process, as well as semantic links between entries.

[0036] The dynamic topology evolution module is connected to the data fusion and semantic alignment engine, and is used to receive and process the fused data field, thereby driving the topology of the digital twin model to adaptively evolve in real time as the physical engineering entity is built, assembled, modified, or demolished. The dynamic topology evolution module also includes a model primitive library, a topology relationship inferencer, and a model incremental updater.

[0037] The model primitive library is a pre-defined digital component repository that stores parametric three-dimensional geometric primitives corresponding to various standard engineering components. Each geometric primitive is defined by a set of parameters and associated with a physical property template. The physical property template defines the relevant material and mechanical property parameters involved in this type of component, such as elastic modulus, Poisson's ratio, density, compressive strength, coefficient of thermal expansion, etc. These parameters can be assigned values ​​based on measured or design information in the fused data field when the component is instantiated.

[0038] The topology inference engine monitors the changing trends of semantic labels and geometric features of construction activities parsed from the fused data field, and infers the topological relationships (such as connections, assembly, support, and embedding) between components in the engineering entity in real time. The inference process is based on a hybrid rule- and data-driven decision framework. The decision framework includes an explicit engineering rule knowledge base and an implicit graph neural network model. The knowledge base encodes standard engineering experience and specification requirements in the form of production rules. For example, the rule is stated as follows: if the activity label is "steel beam hoisting and positioning", and the geometry detects that a new elongated entity is vertically aligned with the top of an existing column, and the process semantics extracted from the construction process data include "high-strength bolt connection", then the inferred topological relationship is "rigid beam-column joint connection". The implicit graph neural network model takes a scene feature map constructed from the current fused data field as input. The nodes of the feature map are the detected entities or features, and the edges initially represent their spatial adjacency or semantic association. The graph neural network is trained with a large amount of historical engineering data and uses the output of the rule knowledge base as a partial supervision signal, thereby enabling it to handle complex inference scenarios where rules are not covered, non-standard, or the data contains noise. During inference, the inference engine executes rule-based inference and graph neural network prediction in parallel. The outputs of both are weighted and fused to form the final decision. The weights are dynamically adjusted based on the confidence level of the current input data (e.g., based on factors such as the signal-to-noise ratio of sensor data and image sharpness). When the rule match is clear and the data confidence is high, the rule result has a dominant weight; conversely, the weight of the graph neural network prediction result is increased. The final output is a clear set of topology change instructions.

[0039] The model incremental updater receives change instructions output by the topology relationship inferencer and performs real-time, incremental evolution operations on the current digital twin model.

[0040] Evolutionary operations are divided into the following three categories:

[0041] Topology addition: When an instruction indicates the addition of a new component, the updater instantiates the corresponding parametric geometric primitive from the model primitive library and assigns values ​​to the primitive parameters and associated physical attribute templates based on the measured geometric and physical data of the component in the fused data field. Subsequently, the new primitive instance is connected to the existing model network strictly according to the inferred topological relationships;

[0042] Topology deletion: When an instruction indicates the removal of a component, the updater deletes the corresponding primitive instance from the model network and simultaneously removes the connection between that instance and all other components to maintain the topological integrity of the model network.

[0043] Attribute Modification: When the fused data field indicates that the attribute (such as material strength, cross-sectional dimensions) of a component has changed, the updater locates the corresponding primitive instance and updates the relevant parameter values ​​in its physical attribute template.

[0044] By continuously performing the above incremental operations, the topology and physical properties of the digital twin model are dynamically updated, thereby achieving synchronous evolution with the state of the physical engineering entity.

[0045] The twin model-driven and visualization service layer is connected to the dynamic topology evolution module. Based on a dynamically updated digital twin model synchronized with the physical entity, it provides engineering quality analysis, prediction, and immersive visualization interaction services. The twin model-driven and visualization service layer integrates a physics engine and a numerical simulation kernel. The physics engine handles rigid body dynamics and collision detection to simulate dynamic scenarios such as hoisting and assembly during construction. The numerical simulation kernel utilizes the finite element method and finite volume method to perform simulation analysis of various physical fields based on the current accurate geometric topology and physical properties of the digital twin model. This includes structural mechanical response simulation, calculating the internal forces, deformation, and stress distribution of the structure under self-weight, construction loads, and wind loads; temperature field simulation, analyzing the structural temperature gradient and thermal stress caused by the heat of hydration of large-volume concrete or changes in ambient temperature; and durability analysis, predicting time-varying loss processes such as chloride ion erosion and carbonation depth development.

[0046] The twin model-driven and visualization service layer also includes a built-in quality indicator evaluation module, which contains a predefined library of calculation logic for key performance indicators of engineering quality. Each indicator is associated with specific engineering specifications and calculation formulas. The quality indicator evaluation module automatically extracts the required geometric and physical parameters from the driven digital twin model, calls the corresponding indicator calculation logic, and performs quantitative evaluation. The indicator calculation logic is as follows:

[0047] Maximum deflection at mid-span: Extract the vertical displacement value of the mid-span node of the specified beam member by calling the displacement field obtained from the structural simulation.

[0048] Maximum stress ratio: Extract the maximum stress value of the component obtained from the simulation and divide it by the material design strength value in its physical property template.

[0049] Crack width development rate: calculated by inversion based on long-term strain monitoring data and concrete constitutive model.

[0050] Concrete carbonation depth: predicted using an empirical model based on environmental temperature and humidity data, concrete mix proportions, and time.

[0051] The quality indicator evaluation module is periodically or triggered by model update events, performing batch calculations and outputting a structured evaluation report. The report includes the current value, design allowable value, specification limit, and out-of-limit status of each indicator.

[0052] To further enhance the system's predictive and early warning capabilities, the quality indicator evaluation module also integrates a time-series prediction submodule. This submodule predicts the future evolution trend of quality indicators based on historical data and generates early warnings. It employs a Long Short-Term Memory (LSTM) network as its core prediction model. LSM networks regulate internal cell states through a collaborative gating mechanism of input, forget, and output gates, thereby enabling them to learn long-term dependencies in time series data. The time-series prediction submodule uses a fused historical time-series data field and corresponding historical quality evaluation results as its training dataset.

[0053] During the training phase, the Long Short-Term Memory (LSTM) network learns the evolution of quality indicators under the coupled effects of multiple factors such as load, environment, and material aging. In the prediction phase, for the current digital twin model, given a future time point or assumed load conditions, this submodule can output predicted values ​​for various quality indicators. These predicted values ​​are compared with preset multi-level safety thresholds. If the predicted value exceeds the warning threshold, a warning message is automatically generated, including the risky component, risk indicator, predicted time of exceedance, and possible causes, thus achieving proactive risk warning.

[0054] The twin model-driven and visualization service layer uses a graphical user interface to integrate and display the dynamically evolving 3D digital twin model, real-time monitoring data curves, simulation analysis result cloud maps, and quality assessment reports and early warning information in a unified and fusion manner. Specifically, this includes:

[0055] It can render dynamically evolving digital twin models, supporting arbitrary angle rotation, scaling, and translation, as well as cross-sectional viewing of internal structure and transparent display.

[0056] Real-time monitoring data is overlaid on the corresponding components in the 3D scene in the form of curves, dashboards, etc.; simulation analysis results (such as stress and deformation cloud maps) are mapped onto the model surface, and the physical field distribution is displayed intuitively with color gradients.

[0057] Users are provided with a variety of interactive operations, including but not limited to: switching to view the state snapshot of the model at any point in history, and realizing historical backtracking;

[0058] Set different future loads or environmental conditions to drive the prediction submodule to simulate future performance.

[0059] Dynamically adjust the early warning thresholds for various quality indicators;

[0060] Generate quality monitoring and analysis reports for a specified time period with a single click.

[0061] Example 1:

[0062] In the digital twin monitoring system of a certain type of bridge project, the construction activity of "hoisting and positioning the mid-span steel beam E01 and connecting it with the existing bridge piers P01 and P02" was carried out on site. The following data streams were collected in real time through the multi-source heterogeneous data sensing module:

[0063] Geometric data: Point cloud generated by a laser scanner deployed on the bridge deck shows a new elongated entity appearing between the tops of P01 and P02;

[0064] Physical state data: Tilt sensors installed on top of P01 and P02 detected minute attitude changes; strain gauges installed on the new steel beam detected initial strain;

[0065] Environmental parameter data: Wind speed and temperature data reported by the on-site weather station;

[0066] Construction process data: A text log was entered from the construction management system: "Today, the E01 steel beam was hoisted, and high-strength bolts were used to rigidly connect the beam end to the corbel."

[0067] Subsequently, the data fusion and semantic alignment engine processes the above data, aligning all timestamps to UTC time and unifying spatial coordinates to the global coordinate system. Geometric feature vectors of new entities (30 meters long, I-shaped cross-section, spatial location) are extracted from the point cloud; semantic feature vectors of construction activities are extracted from text logs using a trained NIL model, containing key concepts such as "steel beam," "hoisting," "high-strength bolts," and "rigid connection." Based on the bridge engineering monitoring ontology library, the data fusion and semantic alignment engine maps these feature vectors to concept nodes in the ontology: steel beam E01, pier (P01, P02), hoisting activity, and rigid connection. The system thus understands that "a steel beam component named E01 is undergoing a hoisting activity, forming a rigid connection with piers P01 and P02," and this understanding is structurally stored in the fused data field.

[0068] The dynamic topology evolution module receives the fused data field and triggers the following process:

[0069] Rule knowledge base trigger: The rule "If the activity is steel beam hoisting, and the new component has a long strip geometry and is aligned with the tops of the two columns, and the process text contains high-strength bolts and rigid connections, then the topological relationship is inferred to be a beam-column rigid connection" is triggered, and a high-confidence inference result is output.

[0070] Graph Neural Network Validation: Simultaneously, the system constructs a feature map of the current scene (including entity nodes such as E01, P01, and P02, and their spatial and semantic relationships), and inputs it into the trained graph neural network. Based on patterns learned from historical data, the graph neural network also outputs a prediction of "beam-column rigid connection."

[0071] Weighted fusion decision: Due to the high quality of sensor data and clear rule matching, the system assigns higher weights to the rule outputs and ultimately determines that a "rigid connection" topology relationship is established between E01 and P01, and between E01 and P02.

[0072] Subsequently, the model incremental updater instantiates an "I-beam" from the model primitive library, parameterizes several primitives corresponding to E01, assigns it precise geometric parameters (length, width, height, and spatial coordinates) based on point cloud data, and creates a new E01 component node in the digital twin model's topology network based on the inference results, establishing rigid connections between it and nodes P01 and P02. The model topology is then dynamically reconstructed.

[0073] Finally, the numerical simulation kernel, based on the updated model incorporating new topological relationships, automatically calculates the stress and deflection of steel beam E01 under its own weight and current wind load in real time. The quality index evaluation module calculates its "mid-span deflection," "stress ratio," and other indicators, with results all showing within safe limits. The time-series prediction submodule, based on historical structural response data after this hoisting, predicts the vibration amplitude of E01 under future extreme wind loads, and no risk of exceeding limits is found. Furthermore, in the 3D visualization interface, users can clearly see that the E01 component has "grown" in the model, with its connection area with the pier highlighted in a high-key color. Stress cloud maps and monitoring curves are updated synchronously.

[0074] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A digital twin dynamic modeling system for engineering quality monitoring, characterized in that, include: The multi-source heterogeneous data sensing module is used to collect and access geometric morphology data, physical state data, environmental parameter data, and construction process data from the engineering quality monitoring site in real time. The data fusion and semantic alignment engine unifies the spatiotemporal benchmark and extracts multimodal features from the input multi-source heterogeneous data. Based on a predefined engineering quality monitoring ontology library, it performs cross-modal semantic association to generate a fused data field with a unified spatiotemporal benchmark and engineering semantic annotations. The dynamic topology evolution module receives the fused data field and, based on the semantic tags of construction activities and associated geometric feature changes contained in the field, infers the topological relationships between components in the engineering entity in real time through a topology relationship inferencer. It also drives the model incremental updater to synchronously evolve and reconstruct the topology of the digital twin model. The twin model driving and visualization service layer performs numerical simulation analysis of engineering quality, quantitative evaluation of key performance indicators, and multi-dimensional visualization interaction based on the updated digital twin model.

2. The digital twin dynamic modeling system for engineering quality monitoring according to claim 1, characterized in that: The data fusion and semantic alignment engine includes: a spatiotemporal reference unification module, which uses the global satellite navigation system timing signal as a unified time reference and the global coordinate system of the engineering design as a unified spatial reference to convert and align the time and space information of the input data; a multimodal feature extraction module, which extracts surface curvature, edge feature points, and volume change rate from the geometric morphology data to form a geometric feature vector, extracts spectral features, statistical features, and temporal correlation from the physical state data to form a physical feature vector, extracts temperature gradient and wind speed and direction statistics from the environmental parameter data to form an environmental feature vector, and extracts semantic feature vectors from the text information in the construction process data through a pre-trained natural language processing model; and a semantic association module, which maps the multimodal feature vectors to the corresponding engineering concept nodes in the ontology library based on the engineering quality monitoring ontology library, and establishes semantic association relationships between the feature vectors.

3. The digital twin dynamic modeling system for engineering quality monitoring according to claim 1, characterized in that: The dynamic topology evolution module includes: a model primitive library, which pre-stores parameterized three-dimensional geometric primitives with associated physical attribute templates; a topology relationship inferencer, which infers the topology relationships between components based on the semantic tags of construction activities and changes in geometric features in the fused data field; and a model incremental updater, which performs addition, deletion, or modification operations on components and their topology relationships in the digital twin model based on the output of the topology relationship inferencer.

4. The digital twin dynamic modeling system for engineering quality monitoring according to claim 3, characterized in that: The topology relation inference engine operates based on a hybrid rule- and data-driven decision framework, which includes: an explicit engineering rule knowledge base, which encodes standard component connection structures and process requirements using generative rules; and an implicit graph neural network model, which takes the current scene feature map constructed from the fused data field as input and performs partially supervised training based on the output of the engineering rule knowledge base. The final output of the topology relation inference engine is a weighted fusion of the inference results from the rule knowledge base and the prediction results from the graph neural network model, with the weights dynamically adjusted according to the confidence level of the input data.

5. The digital twin dynamic modeling system for engineering quality monitoring according to claim 2, characterized in that: The semantic association module achieves cross-modal semantic alignment through a two-layer graph neural network architecture, including: an intra-modal feature graph network, used to perform graph structure modeling on geometric, physical, environmental and text semantic feature vectors respectively, and learn the feature relationships within the modality; and a cross-modal alignment graph network, used to perform multi-round message passing and feature aggregation between nodes in each modal feature graph and concept nodes in the engineering quality monitoring ontology library through a learnable attention mechanism, so that different modal feature nodes describing the same engineering entity or phenomenon have similar embedding vector representations in the ontology concept space.

6. The digital twin dynamic modeling system for engineering quality monitoring according to claim 3, characterized in that: The operations performed by the model incremental updater include: when a new component is inferred, instantiating the corresponding parameterized geometric primitive from the model primitive library, assigning geometric and physical parameters according to the fused data field, and connecting it to the existing model network according to the inferred topology; when a component is inferred to be removed, deleting the corresponding primitive instance and all its associated relationships from the model network; and when a change in component attributes is detected, updating the parameter values ​​in the physical attribute template associated with the corresponding primitive instance.

7. The digital twin dynamic modeling system for engineering quality monitoring according to claim 1, characterized in that: The twin model-driven and visualization service layer includes: a numerical simulation kernel, which performs simulation calculations of structural mechanics, temperature field, or durability based on the current geometric topology and physical properties of the digital twin model; a quality index evaluation module, which predefines the calculation logic of key performance indicators, extracts parameters from the digital twin model and calls the calculation logic to output quantitative evaluation results; and a time series prediction submodule, which takes historical fused data fields and quality evaluation results as input, learns the evolution law of indicators through a time series prediction model, predicts the quality index values ​​under future operating conditions, and generates early warning information.

8. The digital twin dynamic modeling system for engineering quality monitoring according to claim 7, characterized in that: The time series prediction submodule uses a long short-term memory network as its time series prediction model. The long short-term memory network regulates cell states through gating mechanisms such as input gates, forget gates, and output gates to learn the long-term dependencies of time series data on quality indicators.

9. The digital twin dynamic modeling system for engineering quality monitoring according to claim 1, characterized in that: The multi-source heterogeneous data perception module, data fusion and semantic alignment engine, dynamic topology evolution module, and twin model-driven and visualization service layer are encapsulated as independently deployable microservices. The microservices communicate with each other through message queues and event buses, and data is transmitted in an event-driven manner.

10. A digital twin dynamic modeling system for engineering quality monitoring according to claim 1, characterized in that: The multi-source heterogeneous data sensing module has a built-in unified data interface protocol stack, which includes: a point cloud data protocol for point cloud data transmission, an industrial IoT protocol for sensor time-series data transmission, a real-time streaming protocol for video streaming, and an application programming interface for business system data interaction. The protocol stack encapsulates heterogeneous data sources into standardized data packets containing metadata headers and payload data bodies for access. The metadata header at least includes a data source identifier, acquisition timestamp, spatial reference frame identifier, and data type encoding.

Images