A Smart Construction Intelligent Control Method and System Based on Multi-Source Sensing and Data Fusion
The intelligent management and control method for smart building construction, which integrates multi-source sensing and data fusion, solves the problem of insufficient logical constraints in multimodal data processing in existing technologies, realizes reliable generation and execution of management and control instructions, and improves the intelligent management and control capabilities of the construction process.
Patent Information
- Application Number
- CN202611628103.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-21
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies lack formal modeling of deep logical constraints and physical rules between processes in multimodal data processing, leading to frequent false alarms or missed alarms caused by abnormal data. This makes it impossible to reliably support the automatic generation and issuance of control instructions, especially in scenarios with strong process constraints such as concrete curing and acceptance of concealed works.
By collecting multi-source sensing data from the construction site, preprocessing and spatiotemporal alignment are performed. Semantic decoupling and filtering are carried out using a neural-symbolic dual processing module. Combined with the construction potential field engine, deviation tensors are generated, and control instructions are automatically fused to generate control commands. Based on the execution results, the multi-source sensing data is updated to form a closed-loop intelligent control system encompassing perception, decision-making, and execution.
It achieves logical filtering and semantic decoupling of multimodal data, eliminates the misleading influence of abnormal data on management decisions, improves the purity and semantic consistency of feature data, quantifies progress delays and quality risks, and ensures the reliable generation and execution of management instructions.
Smart Images

Figure CN123116186A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent construction management and control technology, and in particular to an intelligent construction management and control method and system for smart buildings based on multi-source sensing and data fusion. Background Technology
[0002] With the deep integration of industrialized construction and information technology, multi-source sensing data collaborative analysis based on the Internet of Things, computer vision, and BIM has become the mainstream technical approach for smart building construction management. Existing technologies simultaneously acquire personnel work status, environmental parameters, and material information by deploying video surveillance, environmental sensors, and RFID readers. After classification and identification using machine learning models at edge computing nodes, this information is compared with the preset construction schedule. When the actual completion time of a process exceeds the planned window, an early warning signal or scheduling suggestion is automatically generated, realizing the digital perception and automated comparison of construction elements.
[0003] Existing technologies typically base multimodal data processing and plan comparison on shallow feature extraction and statistical matching, lacking formal modeling of deep logical constraints and physical rules between processes. When a modality of data becomes abnormal due to site obstruction, equipment failure, or environmental interference, machine learning models struggle to accurately identify the semantic rationality of that modality in the current process stage, leading to frequent false alarms or missed alarms in warning signals. This makes it unreliable to support the automatic generation and issuance of control instructions, which is particularly prominent in scenarios with strong process constraints such as concrete curing and concealed works acceptance. It has become a major bottleneck restricting the leap from perceptual visibility to cognitive control in smart building construction. Summary of the Invention
[0004] In view of the aforementioned existing problems, the present invention is proposed.
[0005] Therefore, this invention provides a smart building construction intelligent management and control method based on multi-source perception and data fusion to solve the problem that existing technologies, in multimodal data processing, lack formal modeling of deep logical constraints and physical rules between processes, resulting in frequent false alarms or missed alarms caused by abnormal data, thus failing to reliably support the automatic generation and issuance of management and control instructions.
[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a smart building construction intelligent management and control method based on multi-source sensing and data fusion, which includes collecting multi-source sensing data at the construction site and preprocessing it to form a standardized multi-source sensing data stream; The standardized multi-source sensing data stream is spatiotemporally aligned to generate a spatiotemporally aligned multi-source data set. Feature extraction is performed on the spatiotemporally aligned multi-source dataset to obtain a conventional feature dataset; The regular feature dataset is input into the neural-symbolic dual processing module for semantic decoupling and filtering to obtain a clean feature dataset. The clean feature dataset is input into the construction potential field engine for reverse comparison, and the deviation tensor is calculated and generated automatically based on the deviation tensor to generate control instructions. Control commands are sent to the execution terminal, and the multi-source sensing data is updated based on the execution results.
[0007] As a preferred embodiment of the intelligent construction management and control method for smart buildings based on multi-source sensing and data fusion described in this invention, wherein: the formation of a standardized multi-source sensing data stream includes, The system collects video images, point cloud data, temperature and humidity data of the concrete solidification environment, stress time series data, and RFID material information of incoming steel bars and formwork at the construction site. After noise reduction, target detection, and protocol conversion preprocessing, it forms a standardized multi-source sensing data stream.
[0008] As a preferred embodiment of the intelligent construction management and control method for smart buildings based on multi-source sensing and data fusion described in this invention, the generated spatiotemporally aligned multi-source data set includes, By using a unified time reference and spatial coordinate system, the positions of personnel and machinery in video images, the outlines of components in point cloud data, the temperature and humidity of the concrete solidification environment, the physical quantity values of stress time series data, and the material codes of RFID material information of incoming steel bars and formwork are mapped to the corresponding floor and axis positions of the BIM construction model built inside the edge computing gateway, generating a spatiotemporally aligned multi-source data set.
[0009] As a preferred embodiment of the intelligent construction management and control method for smart buildings based on multi-source sensing and data fusion described in this invention, the obtained conventional feature dataset includes: The computer vision algorithm is used to identify the type of work of workers in video images in a spatiotemporally aligned multi-source dataset. Based on point cloud clustering and segmentation, the geometric dimensions and installation progress of components in the point cloud data of the spatiotemporally aligned multi-source dataset are extracted. The concrete strength development curve is calculated based on the temperature, humidity, and stress time series data of the concrete solidification environment in the spatiotemporally aligned multi-source dataset using sensor readings. The material type and quantity are obtained by matching the RFID material information of incoming steel bars and formwork in a multi-source dataset after spatiotemporal alignment using RFID. The types of work, the geometric dimensions and installation progress of components, the concrete strength development curve, and the material type and quantity are summarized into a set of conventional features including process semantics, geometric shape, physical indicators and material properties.
[0010] As a preferred embodiment of the intelligent construction management and control method for smart buildings based on multi-source sensing and data fusion described in this invention, the obtained pure feature dataset includes: A neural-symbolic dual processing module is deployed within the edge computing gateway. The neural-symbolic dual processing module has a built-in construction process ontology library, which stores the logical relationships and mutual exclusion rules of the construction process. The conventional feature dataset is input into the neural-symbolic dual processing module to extract the process semantic labels of each feature in the conventional feature dataset; The semantic tags of the work process are matched and verified with the logical relationship of the work process corresponding to the current planned work process in the construction process ontology library. The semantic tags of the work process are matched and verified with the mutual exclusion rules corresponding to the current planned work process in the construction process ontology library. When any of the following features are missing in the regular feature dataset: worker operation type feature of video image mode, component geometric size and installation progress feature of point cloud data mode, temperature and humidity of concrete solidification environment, concrete strength development curve feature of stress time series data mode, and material type and quantity feature of RFID material information mode of incoming steel bars and formwork, it is determined to be a cross-modal semantic conflict. When a cross-modal semantic conflict is determined, the dual verification between the neural layer and the symbolic layer is initiated. The neural layer outputs the confidence score of each modal feature, and the symbolic layer outputs the modal confidence determination based on the priority of the construction process ontology library. The confidence score is cross-compared with the modal credibility determination. When the confidence score of any modal output in the neural layer reaches the credibility threshold, a clean feature dataset after semantic decoupling and logical filtering is obtained.
[0011] As a preferred embodiment of the intelligent construction management and control method for smart buildings based on multi-source sensing and data fusion described in this invention, the generation of management and control instructions includes: Based on the logical dependencies and durations of various construction tasks in the 4D BIM schedule, calculate the total float time for the construction activities corresponding to each structural component. The path formed by connecting construction activities with a total floating time less than or equal to a preset time threshold is defined as a high-priority construction path. The potential field value corresponding to each structural component on the high-priority construction path is assigned a value greater than the potential field value corresponding to the structural component on the non-high-priority construction path. The potential field value corresponding to the monitoring area specified in the quality acceptance specification is assigned in a way that decreases layer by layer according to the spatial distance from the high-priority construction path, thus forming a progress-quality potential field. The clean feature dataset is input into the construction potential field engine, which then performs a reverse comparison between the actual construction status in the clean feature dataset and the corresponding planned status in the progress-quality potential field. Based on the inverse comparison results between the actual construction status in the pure feature dataset and the corresponding planned status in the schedule-quality potential field, the difference between the actual construction status in the pure feature dataset and the planned status in the schedule-quality potential field is calculated, and a deviation tensor reflecting the degree of schedule lag and the quality risk level is generated. The expression for the difference between the actual construction state and the planned state in the schedule-quality potential field is: When the deviation tensor indicates that the actual formwork has been completed but the concrete temperature and humidity have not reached the conditions for demolding and the time has exceeded the planned node, it will automatically adjust to postpone the subsequent pouring plan and send a work order for additional curing agent to the material management system. The deviation tensor will be fed back to the progress-quality potential field to update the monitoring threshold for the next period.
[0012] As a preferred embodiment of the intelligent control method for smart building construction based on multi-source perception and data fusion described in this invention, the control command is sent to the execution terminal, and the execution terminal adjusts the operating parameters of the on-site construction equipment or the material delivery plan according to the control command. Based on the execution result feedback generated by the execution terminal's execution control instructions, update the video images, point cloud data, temperature, humidity and stress time series data of the concrete solidification environment, and RFID material information of incoming steel bars and formwork in the spatiotemporally aligned multi-source data set.
[0013] Secondly, the present invention provides a smart building construction intelligent management and control system based on multi-source sensing and data fusion, including a preprocessing module, which collects multi-source sensing data from the construction site and preprocesses it to form a standardized multi-source sensing data stream; The alignment module performs spatiotemporal alignment processing on standardized multi-source sensing data streams to generate a spatiotemporally aligned multi-source data set. The feature extraction module extracts features from the spatiotemporally aligned multi-source dataset to obtain a conventional feature dataset. The decoupling filtering module inputs the regular feature dataset into the neural-symbolic dual processing module for semantic decoupling filtering to obtain a clean feature dataset. The comparison module inputs the clean feature dataset into the construction potential field engine for reverse comparison, calculates and generates a deviation tensor, and automatically fuses and generates control instructions based on the deviation tensor. The execution module sends control commands to the execution terminal and updates the multi-source sensing data based on the execution results.
[0014] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the intelligent management and control method for smart building construction based on multi-source sensing and data fusion as described in the first aspect of the present invention.
[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent control method for smart building construction based on multi-source sensing and data fusion as described in the first aspect of the present invention.
[0016] The beneficial effects of this invention are as follows: A neural-symbolic dual processing module performs semantic decoupling filtering on conventional feature datasets. It utilizes the process logic relationships and mutual exclusion rules of the construction technology ontology library for verification. In the event of cross-modal semantic conflicts, the neural layer and symbolic layer initiate dual verification. Based on the cross-comparison of confidence scores and modal credibility adjudication, the weight of untrustworthy modalities is automatically downgraded. This achieves logical filtering and semantic decoupling of multimodal data, eliminating the misleading influence of abnormal data on control decisions and improving the purity and semantic consistency of feature data. Furthermore, a construction potential field engine performs a reverse comparison between the pure feature dataset and the progress-quality potential field constructed based on the total floating time calculated using a graph attention network. It calculates the difference between the actual construction state and the planned state to generate a deviation tensor, achieving a quantitative assessment of progress delays and quality risks. Then, based on the deviation tensor, control instructions are automatically generated and issued to the execution terminal. Based on the execution results, multi-source sensing data is updated, forming a closed-loop intelligent control system encompassing perception, decision-making, and execution. This solves the problems of existing technologies lacking deep logical constraints and physical rule modeling, leading to frequent false alarms and missed alarms, and the inability to reliably support the automatic generation and issuance of control instructions. Attached Figure Description
[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a flowchart of a smart building construction intelligent management and control method based on multi-source sensing and data fusion.
[0019] Figure 2 This is a schematic diagram of a smart building construction intelligent management and control system based on multi-source sensing and data fusion. Detailed Implementation
[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.
[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.
[0023] Reference Figures 1-2 This is one embodiment of the present invention, which provides a smart building construction intelligent management and control method based on multi-source sensing and data fusion, including the following steps: S1. Collect multi-source sensing data from the construction site and preprocess it to form a standardized multi-source sensing data stream.
[0024] S1.1 Collect video images, point cloud data, temperature and humidity data of the concrete solidification environment, stress time series data, and RFID material information of incoming steel bars and formwork at the construction site. After noise reduction, target detection, and protocol conversion preprocessing, a standardized multi-source sensing data stream is formed.
[0025] Furthermore, video images from the construction site are simultaneously captured as continuous frame sequences by fixed and mobile camera equipment to obtain information on personnel behavior, machinery operation, and changes in the work interface; point cloud data is obtained by 3D laser scanning equipment to acquire spatial reflection echoes and form structured point sets; temperature and humidity of the concrete solidification environment are acquired by distributed environmental sensing units, and stress time series data are continuously recorded by embedded strain sensing units to form a time series; RFID material information of incoming steel bars and formwork is automatically captured by material channel radio frequency identification reading devices to obtain identity codes and flow records; video images are separated from effective work areas and background interference through noise suppression and target detection algorithms constrained by time domain continuity; point cloud data is filtered to remove false point sets through spatial outlier removal and surface continuity screening; temperature, humidity, and stress time series data are eliminated by time synchronization interpolation to eliminate time offset caused by inconsistent sampling frequencies; RFID material information is unified by communication protocol parsing and encoding standard conversion to unify different supply chain coding systems, and the processed information is bound according to a unified timestamp and spatial identifier, so that visual information, spatial structure information, physical environment information, and material information complete the correspondence within the same construction space unit and time unit, forming a standardized multi-source sensing data stream.
[0026] S2. Perform spatiotemporal alignment processing on the standardized multi-source sensing data stream to generate a spatiotemporally aligned multi-source data set.
[0027] S2.1. Using a unified time reference and spatial coordinate system, the positions of personnel and machinery in video images, the outlines of components in point cloud data, the temperature and humidity of the concrete solidification environment, the physical quantity values of stress time series data, and the material codes of RFID material information of incoming steel bars and formwork are mapped to the corresponding floor and axis positions of the BIM construction model built inside the edge computing gateway, generating a spatiotemporally aligned multi-source data set.
[0028] Furthermore, the positions of personnel and machinery in the video images are geometrically constrained and transformed using pre-set spatial calibration points at the construction site, mapping the two-dimensional image coordinates to the construction space coordinate system; the component outlines in the point cloud data are morphologically matched with the geometric benchmarks of the construction structure using a point cloud registration algorithm, mapping the set of spatial points to the corresponding component axes and floor units; the temperature, humidity, and stress time series data of the concrete solidification environment are aligned with the structural units using timestamps, attributing the changes in physical quantities to the environmental action area of the corresponding construction components; the material codes of the RFID material information of the incoming steel bars and formwork are associated and matched through the construction component coding system, mapping the material flow information to the corresponding component installation unit.
[0029] During the process of establishing a BIM construction model within the edge computing gateway, a basic building information model is constructed based on construction drawings and structural design documents. Floor divisions, axis divisions, and component numbering systems serve as the spatial skeleton, and the process nodes in the construction schedule are embedded into the corresponding component's time attribute dimension, forming a construction expression framework with both spatial structure and time attributes. The model coordinate system is uniformly calibrated using pre-set spatial reference points at the construction site, ensuring geometric alignment between video image information and point cloud data under a unified spatial reference. The intersection points of component axes serve as spatial anchor points, mapping spatial expressions from different data sources to the same structural semantic unit.
[0030] Specifically, the positions of personnel and machinery in video images, the outlines of components in point cloud data, the temperature and humidity of the concrete solidification environment, the physical quantities of stress time series data, and the material codes of RFID material information are projected onto the floor and axis units of the BIM construction model. The observation information is assigned to the corresponding component entities through spatial attribution rules, and the observation information of the same component under different time states is continuously expressed through a time base unification mechanism, forming a spatiotemporally aligned multi-source data set with spatial and temporal consistency within the edge computing gateway.
[0031] S3. Perform feature extraction on the spatiotemporally aligned multi-source dataset to obtain a conventional feature dataset.
[0032] S3.1. Identify the work type of workers in video images in the spatiotemporally aligned multi-source dataset using computer vision algorithms, and extract the geometric dimensions and installation progress of components in the point cloud data in the spatiotemporally aligned multi-source dataset based on point cloud clustering and segmentation.
[0033] Furthermore, the human body contours and key joints within the construction area are spatially located. Then, based on the human body posture change trajectory and the interaction with construction tools, behavioral segments are divided. On the basis of behavioral segments, rebar tying actions, formwork support actions, and concrete pouring actions are categorized to form the worker's work type output result. When a worker performs multiple construction actions simultaneously, the action with the highest proportion and the strongest spatial consistency is used as the basis for work type output by judging the proportion of action duration and spatial position stability.
[0034] Specifically, the point cloud is divided into candidate regions for components based on the continuity of spatial density. Then, the candidate regions are reconstructed by geometric boundary consistency. The external dimensions of the components are matched with the construction axis reference to obtain the geometric dimension information of the components. By comparing the spatial consistency between the completeness of the component point cloud coverage and the design geometric model, when the coverage of the component point cloud gradually approaches the complete structural form, the corresponding installation progress is judged as the gradual completion state. When there are local missing areas, the construction stage status is judged according to the distribution position of the missing areas in the spatial structure.
[0035] It should be noted that when there are multiple components being constructed at the construction site and there is obstruction, the obstructed area is separated from the unobstructed area by identifying the spatial continuity breaks in the point cloud data. The missing part is then assigned and corrected based on the spatial topology of the adjacent components, so that the geometric dimensions of the components are restored to a consistent expression, thus avoiding structural misjudgment caused by construction interference.
[0036] S3.2 Calculate the concrete strength development curve corresponding to the temperature, humidity and stress time series data of the concrete solidification environment in the spatiotemporally aligned multi-source dataset by using sensor readings.
[0037] Furthermore, the temperature information of the concrete solidification environment is collected by temperature sensors placed on the concrete surface and temperature sensors embedded in the concrete component to form a dual-source temperature observation sequence. By performing consistent fusion processing on the surface temperature change and the internal core temperature change, the temperature information can simultaneously reflect the effects of the external environment and the internal hydration heat release process. When the surface temperature and the internal temperature change differently, the contributions of the two types of temperature are weighted by the depth coefficient that varies with the thickness of the concrete component, so that the temperature gradient at different depths maintains a continuous mapping relationship in a unified expression.
[0038] The humidity information of the concrete curing environment is collected by humidity sensors placed on the concrete surface to collect changes in relative humidity. Combined with the self-drying process of the concrete, a humidity attenuation constraint expression is formed. During the gradual consumption of moisture inside the concrete, when the ambient humidity is higher than the self-drying humidity of the concrete, the self-drying humidity is used as the constraint boundary. When the ambient humidity is lower than the self-drying humidity, the ambient humidity is used as the control boundary, thus forming a constraint expression relationship of the humidity of the concrete curing environment.
[0039] Specifically, stress time series data, through the conversion results of stress sensors and strain gauges, constitute a sequence of stress changes inside concrete. The strain rate change trend continuously expresses the microcrack propagation state inside the concrete structure. When stress abrupt changes occur, the consistency constraint judgment of strain increments in adjacent time series is applied to attribute abnormal stress changes to construction load disturbances or changes in hydration reactions inside the structure.
[0040] It should be noted that the concrete strength development curve is obtained by jointly calculating the temperature information, humidity information and stress time series data by inputting them into the strength mapping layer. Among them, the gated time-varying memory unit performs state modeling on the evolution process of concrete strength over time, and the long short-term memory network continuously updates the strength development trend through the recursive relationship between the gated hidden state at the previous moment and the input features at the current moment. This allows the concrete compressive strength value to reflect the coupled influence of environmental temperature and humidity changes and structural stress changes, thus forming the concrete strength development curve.
[0041] The expression for the temperature of the concrete solidification environment is: ; in, For at any time The temperature of the concrete setting environment. This represents the depth coefficient that varies with the thickness of the concrete member. For at any time The ambient temperature was measured by temperature and humidity sensors placed on the concrete surface. For at any time The core temperature was measured by a temperature sensor embedded inside the concrete component. For a moment.
[0042] The expression for the humidity of the concrete solidification environment is: ; in, For at any time The humidity of the environment in which the concrete solidifies. For at any time The ambient relative humidity was measured by humidity sensors placed on the concrete surface. It is the smaller of the ambient humidity and the self-drying humidity.
[0043] The expression for the compressive strength value on the concrete strength development curve is: ; in, For at any time The compressive strength value on the concrete strength development curve This refers to the standard value of the compressive strength of a concrete cube. The weight matrix of the intensity mapping layer. For a moment The gated hidden state. This is the bias vector for the intensity mapping layer.
[0044] ; in, It is a long short-term memory network with a gating mechanism. For the previous moment The gated hidden state. For at any time The feature vector input to the gated time-varying memory unit. This is the bias vector for calculating the hidden state layer.
[0045] ; in, For at any time The ambient temperature was measured by temperature and humidity sensors placed on the concrete surface. For at any time The ambient relative humidity was measured by humidity sensors placed on the concrete surface. For at any time The actual internal stress of concrete obtained by stress sensors For at any time The measured strain rate or the effective strain increment after pretreatment.
[0046] S3.3. The material type and quantity of the incoming steel bars and formwork are obtained from the RFID material information of the multi-source data set after spatiotemporal alignment by matching with RFID.
[0047] Furthermore, RFID material information is first continuously collected by radio frequency identification (RFID) readers deployed at the material entry channels and stacking areas of the construction site. During the material flow process, the unique coding information of steel bars and formwork is captured in real time, and the coding information is initially bound to the axial unit of the construction component. Subsequently, in the spatiotemporally aligned multi-source data set, the consistency comparison between the RFID material code and the component numbering system is performed. When there is a one-to-one correspondence between the material code and the component code, the material is assigned to the corresponding component installation unit. When there is a one-to-many or many-to-one mapping relationship, the constraint judgment is made by the process dependency relationship in the construction schedule, and the material is preferentially allocated to the corresponding component unit of the current construction stage.
[0048] Specifically, in the material model identification process, the model information of incoming steel bars and formwork is structured and parsed using RFID coding parsing rules. This maps different supplier coding systems to a unified construction material standard coding system, ensuring that the same type of material maintains consistent representation under different source conditions. When multiple batches of steel bars or formwork correspond to the same component on the construction site, the material batch flow sequence is matched with the component construction time sequence. Materials that arrive first are preferentially bound to components that are constructed first, forming a temporal correspondence between material model and component installation status.
[0049] During the material quantity calculation process, the entry and exit records in the RFID material information are accumulated and statistically analyzed. The changes in the quantity of incoming steel bars and formwork are compared with the component installation progress in sync. When there is a difference between the actual consumption of materials and the design requirements of the construction components, the component coding system is used for reverse positioning to map the difference to the corresponding construction component unit, thereby realizing the consistency verification between the material quantity and the construction status.
[0050] S3.4. Summarize the job type, component geometry and installation progress, concrete strength development curve, material type and quantity into a routine feature dataset of process semantics, geometric shape, physical indicators and material properties.
[0051] Furthermore, the work type is transformed into process semantics through construction behavior semantic mapping. Construction behaviors such as rebar tying, formwork installation, and concrete pouring are structurally categorized according to the logical relationship of construction processes, so that different work behaviors form continuous semantics within the same construction process. The geometric dimensions of components and the installation progress are jointly represented by spatial morphological features and the construction completion ratio. The axial dimensions, spatial contour changes, and installation completion degree of components are uniformly mapped to the geometric shape, so that the spatial state of components is expressed in a structurally consistent form.
[0052] Specifically, the concrete strength development curve is derived by using time-series data of temperature, humidity, and stress to summarize the strength change trajectory. This transforms the continuous process of concrete from initial setting to strength growth into a physical index, ensuring that changes in material properties correspond consistently with the construction stage. Material types and quantities are structurally categorized using RFID material coding matching results. This maps the specifications and quantities of reinforcing bars and formwork to material attributes, ensuring a consistent correspondence between material supply status and construction component demand status.
[0053] During the induction process, when there is an inconsistency between the operation type and the component installation progress, consistency verification is performed through the logical relationship of construction procedures, and abnormal operation behaviors are classified into the corresponding semantic categories of procedures; when there is a deviation between the component geometric dimensions and the installation progress, correction and classification are performed through spatial structural integrity judgment; when there is a deviation between the concrete strength development curve and the trend of physical index changes, induction and adjustment are performed through time series consistency constraints; when there are multiple batches of duplicate material models and quantities, unique classification is performed through the component coding system, forming a stable four-category feature expression system.
[0054] S4. Input the regular feature dataset into the neural-symbolic dual processing module for semantic decoupling and filtering to obtain a clean feature dataset.
[0055] S4.1 Deploy a neural-symbolic dual processing module within the edge computing gateway. The neural-symbolic dual processing module has a built-in construction process ontology library, which stores the logical relationships and mutual exclusion rules of the construction process.
[0056] Furthermore, during the deployment of the neural-symbolic dual processing module within the edge computing gateway, multi-source construction features such as video image recognition results, point cloud structure information, concrete strength development curves, and RFID material attributes are input into a unified processing entry point. Within this processing environment, a construction process ontology library is constructed. The construction process ontology library stores the logical dependencies and mutual exclusion rules between construction procedures. The logical relationships between procedures are used to describe the sequence and dependencies between construction steps, while the mutual exclusion rules are used to describe the constraints that prevent different construction procedures from being simultaneously established.
[0057] Specifically, in the actual implementation process, when there is a situation where rebar tying and concrete pouring are identified simultaneously at the construction site, the sequence consistency of the two types of processes is checked through the pre-set process logic relationship in the construction process body library. When there is a conflict, the process status is constrained and judged according to the logic relationship. When there is a situation where formwork removal and concrete not reaching the curing conditions occur simultaneously at the construction site, the conflict of construction status is identified through mutual exclusion rules, and the process status that does not conform to the construction specifications is marked as a conflict status, forming a process logic consistency constraint.
[0058] In the neural-symbolic dual processing, the neural computation path is used to probabilistically correlate and express the multi-source features of the construction site, and output the confidence distribution of different construction procedures. The symbolic computation path performs deterministic reasoning on the logical relationship and mutual exclusion rules of the procedures based on the construction process ontology. The results of the two types of paths are compared for consistency under the same processing framework. When there is a deviation between the output of the neural path and the judgment of the symbolic path, the procedure state is reconstructed and adjusted through priority rules to keep the construction procedure state consistent under logical constraints.
[0059] S4.2 Input the regular feature dataset into the neural-symbolic dual processing module to extract the process semantic labels of each feature in the regular feature dataset.
[0060] Furthermore, the semantic features of the construction process, geometric features, physical index features, and material attribute features are uniformly structured and encoded to maintain computable relationships between features from different sources within the same semantic expression space. The semantic features of the construction process are used to describe the categories of construction behaviors and the attributes of construction stages; the geometric features are used to describe the spatial shape and installation status of components; the physical index features are used to describe the concrete strength development curve and changes due to environmental effects; and the material attribute features are used to describe the type and quantity information of steel bars and formwork.
[0061] Specifically, during the extraction of process semantic labels, a neural computation path is used to perform probability mapping on the correlation strength between various features. The spatial dependency between process semantic features and geometric shape features is matched and analyzed. When the geometric shape features show that the component installation completion rate is high and the spatial shape is relatively complete, the corresponding process semantic label is determined to be the formwork installation completion stage. When the physical index features show that the concrete strength development curve is in a continuous growth state and the ambient temperature and humidity are stable, the corresponding process semantic label is determined to be the concrete curing stage. When the material attribute features show that the steel bars and formwork are continuously arriving on site and the component geometry is in an unclosed state, the corresponding process semantic label is determined to be the structural construction preparation stage.
[0062] In the symbolic computation path, the process logic relationships in the construction process ontology are used to perform consistency constraint verification on the process semantic labels. For example, when the process semantic labels show both formwork removal and concrete not meeting strength requirements, conflict identification is performed on the process semantic labels through mutual exclusion rules, and labels that do not conform to the construction logic relationships are corrected and constrained to ensure that the process semantic labels are consistent with the construction process logic. Under the synergistic effect of the neural path and the symbolic path, through the dual-path alignment between process semantic features and structured construction knowledge, various features in the conventional feature dataset are mapped to process semantic labels with construction logic constraints, forming a stable and consistent process semantics.
[0063] S4.3 Match and verify the semantic tags of the work process with the logical relationships of the work processes corresponding to the current planned work processes in the construction technology ontology library, and match and verify the semantic tags of the work process with the mutual exclusion rules corresponding to the current planned work processes in the construction technology ontology library.
[0064] Furthermore, the semantic tags of the work processes are transformed into structured semantic expression units and aligned with the logical nodes of the currently planned work processes in the construction process ontology library. The logical relationship of the work processes corresponding to the currently planned work processes is used to describe the dependency and constraint relationships between the preceding and subsequent work processes in the construction flow. When the construction behavior corresponding to the work process semantic tag has missing preconditions, such as the concrete pouring semantic tag appearing before the formwork installation is completed, the sequential constraint judgment is performed based on the logical relationship of the work processes, and the work process semantic tag is marked as logically inconsistent. When the work process semantic tag and the currently planned work process are in the same construction stage but have a sequence conflict, the dependency path in the construction process ontology library is used for verification, and the semantic tags that do not conform to the dependency path are reconstructed and constrained so that they are assigned to the correct logical node of the corresponding construction stage.
[0065] Specifically, during the mutual exclusion rule matching and verification process, the semantic tags of the work processes are compared one by one with the mutual exclusion constraints corresponding to the currently planned work processes in the construction technology ontology library. The mutual exclusion rules describe construction states that cannot be simultaneously established within the same construction space unit or the same time unit. When both formwork removal and concrete not yet reaching curing conditions are identified simultaneously at the construction site, conflict detection is performed using the mutual exclusion rules, determining this combination as an incompatible construction state, and suppressing the semantic tags of the work processes based on priority constraints in the construction technology ontology library. When rebar tying and concrete pouring occur simultaneously within the same component unit at the construction site, conflict identification of the semantic tags of the work processes is performed using spatial consistency constraints to prevent logical overlap of construction states within the same component unit.
[0066] During the comprehensive matching and verification process, the semantic tags of the process are dually verified by the constraints of the logical relationship of the process and the mutual exclusion rule, so that the semantic tags of the process simultaneously meet the constraints of the construction process ontology library in both the time and space dimensions. Furthermore, by correcting and filtering conflicting tags, the semantic expression of the process in the final output is made to maintain a consistent mapping relationship with the current planned process.
[0067] S4.4 When any of the following features are missing from the regular feature dataset: worker operation type feature of video image mode, component geometric size and installation progress feature of point cloud data mode, temperature and humidity of concrete solidification environment, concrete strength development curve feature of stress time series data mode, and material type and quantity feature of RFID material information mode of incoming steel bars and formwork, it is determined to be a cross-modal semantic conflict.
[0068] Furthermore, integrity and consistency verification is performed based on the multimodal constraint relationships corresponding to the current planned procedures in the construction process ontology library. This process is achieved by verifying the existence of each necessary feature set required in the logical relationship of the procedure. When a construction procedure explicitly requires both operational behavior information and component spatial state information in its logical structure, but operational behavior information is missing in the actual feature set, it is determined to be in an incomplete feature state.
[0069] Specifically, during the judgment process, for example, when the logical relationship of the construction process requires the formwork installation process to include both worker operation type features and component geometric dimensions and installation progress features, but the conventional feature dataset only contains component geometric dimension information but lacks operation type information, it is judged as a cross-modal semantic conflict based on the logical relationship of the process; when the concrete pouring process requires both concrete strength development curve features and material type and quantity features, but the conventional feature dataset lacks the concrete strength development curve features corresponding to stress time series data, it is judged as a cross-modal semantic conflict; when the rebar tying process requires both worker operation type features and RFID material information features in the video image to be true, but only RFID material information features actually exist, it is judged as a cross-modal semantic conflict.
[0070] In the process of cross-modal semantic conflict determination, the dependency structure of the feature set is constrained by the logical relationship of the construction process. The necessary features corresponding to each construction process are regarded as components of the complete semantic expression unit. When any component is missing, the semantic incompleteness determination is triggered, and the construction process is transformed from a complete expression state to a conflict state.
[0071] S4.5 When a cross-modal semantic conflict is determined, the dual verification between the neural layer and the symbolic layer is initiated. The neural layer outputs the confidence score of each modal feature, and the symbolic layer outputs the modal confidence determination according to the priority of the construction process ontology library.
[0072] Furthermore, the neural layer performs probabilistic correlation calculations on worker operation type features from video image modalities, component geometric dimensions and installation progress features from point cloud data modalities, concrete strength development curve features from temperature, humidity, and stress time series data modalities of the concrete curing environment, and material type and quantity features from RFID material information modalities of incoming steel bars and formwork. It outputs the confidence score of each modal feature under the current construction state, which characterizes the strength of support for the semantic validity of the current process by each modal feature. During the neural layer calculation process, when a modality exhibits a stable and consistent trend within a continuous time window, the corresponding confidence score remains high. Conversely, when a modality experiences a sudden change or loss, the corresponding confidence score decreases, forming a multimodal confidence distribution expression.
[0073] Specifically, the symbol layer performs deterministic reasoning on each modal feature based on the pre-defined process logic relationships and mutual exclusion rules in the construction process ontology library, and determines the credibility of each modality in combination with the process dependency priority. When the construction process ontology library explicitly stipulates that the semantics of the process must preferentially depend on the component geometric dimensions and installation progress features of the point cloud data modality, then a higher credibility determination weight is given to that modality in the process of determining that process. When the construction process ontology library stipulates that the concrete strength development curve feature is a key constraint, then a priority determination level is given to that physical index feature.
[0074] During the dual verification process, for example, when the neural layer outputs a high confidence level for the video image modality while the symbolic layer determines, based on the construction process ontology, that process must rely on the point cloud data modality, the confidence level of the video image modality is corrected through the priority determination of the symbolic layer; when the neural layer outputs a low confidence level for the RFID material information modality while the symbolic layer determines, based on the construction process ontology, that material information is a critical control condition, the confidence level of that modality is increased through the determination of the symbolic layer; when the neural layer and the symbolic layer output consistent judgment results for the same modality, the original confidence level of that modality remains unchanged.
[0075] The confidence score expression for the video image modality is: ; in, The confidence score for the video image modality. For normalized exponential functions, A convolutional neural network for video image modalities. This refers to the raw input data for the video image modality.
[0076] The confidence score expression for point cloud data modes is: ; in, The confidence score for the point cloud data mode. For intersection, union, and comparison, For point cloud segmentation function, The raw input data for the point cloud data modality. This is a standard construction plan template.
[0077] The confidence score expressions for the temperature, humidity, and stress time series data modes of the concrete curing environment are as follows: ; in, The confidence scores are for the temperature, humidity, and stress time series data modes of the concrete curing environment. It is a natural exponential function. It is a temporal convolutional network. Input temperature time series data. Input humidity time series data. Input data for stress timing. For physical law models or ontology libraries, For the current moment, This refers to the current construction plan or maintenance scheme.
[0078] The confidence score expression for the RFID material information modal of incoming steel bars and formwork is as follows: ; in, The confidence scores are for the RFID material information modes of incoming steel bars and formwork. For the matching degree calculation function, This is an RFID data decoding and parsing module. This is the raw input data for RFID material information modality. This is the bill of materials for the currently planned process.
[0079] S4.6 Cross-compare the confidence score with the modal credibility determination. When the confidence score of any modal output in the neural layer determines that the video image modal, point cloud data modal, temperature, humidity and stress time series data modal of concrete solidification environment, or RFID material information modal of incoming steel bars and formwork reaches the credibility threshold, a clean feature dataset after semantic decoupling and logical filtering is obtained.
[0080] Furthermore, in the process of cross-comparing the confidence scores with the modal credibility determination, the confidence scores of each modality output by the neural layer are consistent with the modal credibility determination results generated by the symbol layer based on the construction process ontology library. Among them, the video image modality, point cloud data modality, temperature, humidity and stress time series data modality of concrete solidification environment, and RFID material information modality of incoming steel bars and formwork are used as independent comparison units to participate in the cross-comparison. When a certain modality is in a high confidence state in the output of the neural layer but has a low priority in the corresponding construction process ontology library in the symbol layer determination, the modal weight adjustment mechanism is triggered.
[0081] Specifically, in the priority determination process, the construction process ontology library performs a structured sorting of the dependence of each modality on the current planned process according to the logical relationship of the process. The dependence is expressed by the necessity of the modality information during the execution of the process. For example, in the rebar tying process, the RFID material information modality and the point cloud data modality are identified as key constraint modalities. In the formwork installation process, the point cloud data modality and the video image modality are identified as key constraint modalities. In the concrete curing process, the temperature, humidity and stress time series data modality of the concrete curing environment are identified as key constraint modalities, forming a modal priority sequence under different process conditions.
[0082] During the weight adjustment process, when the neural layer determines that the confidence level of a certain modality has reached a credible state while the symbolic layer determines that the modality is in a low priority state, the modality is downgraded according to the modality dependency level of the corresponding process in the construction process ontology. When both the neural layer and the symbolic layer determine that a certain modality is in a high confidence state and has a high priority, the weight of that modality remains unchanged. When the judgment results of the neural layer and the symbolic layer conflict, the low priority modality is suppressed according to the priority rules of the construction process ontology. During the semantic decoupling process, the expression ratio of each modality in the regular feature dataset is redistributed through the modality contribution after weight adjustment, so that the influence of low confidence or low priority modalities on the final feature expression is gradually weakened, forming a pure feature dataset after semantic decoupling and logical filtering.
[0083] S5. Input the clean feature dataset into the construction potential field engine for reverse comparison, calculate and generate the deviation tensor, and automatically fuse and generate control instructions based on the deviation tensor.
[0084] S5.1. Based on the logical dependencies and durations of various construction tasks in the 4D BIM schedule, calculate the total float time for the construction activities corresponding to each structural component.
[0085] Furthermore, the construction task nodes in the 4D BIM schedule are mapped to the structural component coding system, so that each construction task is bound to the corresponding structural component in the time dimension. The logical dependency relationship of construction tasks is used to describe the constraint path between the preceding and subsequent tasks, and the duration is used to describe the time range occupied by a single construction task in the planned execution process. For example, when the construction task corresponding to a certain structural component has multiple preceding dependent tasks, the paths of the preceding tasks are analyzed in series, all possible construction paths are logically combined, and the path with the highest degree of influence on the path constraint is selected as the benchmark calculation path. Based on this path, the maximum time offset range allowed for the structural component without affecting the start of subsequent key construction tasks is calculated, and the total floating time of the construction activities corresponding to the structural component is obtained.
[0086] The total float time expression is: ; in, structural components The total floating time corresponding to the construction activities, For parameter-based Graph attention network, For the construction task attribute matrix, This is an adjacency matrix representing logical dependencies. Let the duration of the construction activity be the vector. For indexes of structural components.
[0087] S5.2. The path formed by connecting construction activities with a total floating time less than or equal to a preset time threshold is determined as a high-priority construction path. The potential field value corresponding to each structural component on the high-priority construction path is assigned a value greater than the potential field value corresponding to the structural component on the non-high-priority construction path. The potential field value corresponding to the monitoring area specified in the quality acceptance specification is assigned in a way that decreases layer by layer according to the spatial distance from the high-priority construction path to form a progress-quality potential field.
[0088] Furthermore, based on the total floating time of the construction activities corresponding to the structural components, the construction activities with the total floating time in the low elasticity range are regarded as progress-sensitive nodes, and the paths are connected according to the logical dependencies between construction tasks, so that the construction activities that satisfy the continuous dependency relationship form a complete construction path sequence, where the path sequence represents the key construction link with the highest impact on the overall construction progress.
[0089] Specifically, in the process of setting the preset threshold, the preset threshold is determined comprehensively based on the constraints of the construction schedule plan, the strength of the critical path constraints, and the importance level of the structural components. The importance level of the structural components is expressed by assigning weights to the classification constraints of load-bearing components, critical node components, and ordinary components in the construction process ontology library. When the total floating time of the construction activity is less than or equal to the comprehensive constraint threshold, the construction activity is included in the set of high-priority construction paths and forms a continuous dependency path structure.
[0090] During the potential field assignment process, the potential field values corresponding to each structural component on the high-priority construction path are set to a high potential energy state, so that the components on this path have higher constraint weight in progress control and quality control. The potential field values corresponding to the structural components on the non-high-priority construction path are set to a relatively low potential energy state, forming an expression of the potential energy gradient difference between different construction paths.
[0091] In the process of processing the monitoring area of the quality acceptance specification, the monitoring area specified in the quality acceptance specification is mapped to the spatial distribution location of the structural components, and the potential field attenuation value is assigned according to the spatial distance between the monitoring area and the high-priority construction path. When the monitoring area is close to the high-priority construction path, a higher potential field value is assigned. As the distance gradually increases, the potential field value decreases layer by layer, forming a quality risk gradient expression based on the propagation of spatial influence.
[0092] It should be noted that, for example, when a high-priority construction path passes through the core load-bearing component area, the potential field value of the surrounding monitoring area remains at a high level; when a high-priority construction path is located in a non-critical auxiliary component area, the potential field value of the surrounding monitoring area shows a gradual decrease; and when the monitoring area is located in an independent area far away from the construction path, the potential field value is at its lowest level.
[0093] S5.3 Input the clean feature dataset into the construction potential field engine, and use the construction potential field engine to perform a reverse comparison between the actual construction status in the clean feature dataset and the corresponding planned status in the progress-quality potential field.
[0094] Furthermore, during the input of the clean feature dataset into the construction potential field engine, the semantic features of the work process, geometric shape features, physical index features, and material attribute features contained in the clean feature dataset are uniformly vectorized, enabling different types of construction information to have a comparable structure in the same expression space. This expression result is then mapped and matched with the corresponding planned state feature vector in the schedule-quality potential field. In the reverse comparison process, using the schedule-quality potential field as the planned state constraint benchmark, the actual construction state feature vector is traced backward along the construction work process dependency path, allowing the planned state corresponding to each structural component at the current construction stage to be analyzed item by item.
[0095] Specifically, for example, if the geometric shape of the component in the actual construction state shows that the formwork has been closed, but the corresponding planned state in the progress-quality potential field is still in the formwork installation stage, it is judged as a geometric state ahead of schedule deviation; if the concrete strength development curve in the actual construction state is lower than the strength development trajectory corresponding to the planned state in the progress-quality potential field, it is judged as a physical index lagging deviation; if the material property characteristics in the actual construction state show that the supply of steel bars and formwork has been completed, but the corresponding planned state in the progress-quality potential field has not yet entered the material arrival stage, it is judged as a material state ahead of schedule deviation.
[0096] S5.4 Based on the inverse comparison results between the actual construction status in the pure feature dataset and the corresponding planned status in the schedule-quality potential field, calculate the difference between the actual construction status in the pure feature dataset and the planned status in the schedule-quality potential field, and generate a deviation tensor that reflects the degree of schedule lag and the quality risk level.
[0097] Furthermore, the feature vectors of the actual construction state and the feature vectors of the planned state are aligned in the same dimension, and different types of construction features are differentiated and weighted by a feature weight matrix, so that the semantic features of the process, the geometric shape features, the physical index features, and the material attribute features have different contribution strengths in the difference calculation. Among them, the semantic features of the process are used to reflect the degree of deviation of the construction stage, the geometric shape features are used to reflect the degree of deviation of the component shape, the physical index features are used to reflect the degree of deviation of the concrete strength development, and the material attribute features are used to reflect the degree of deviation of the resource supply.
[0098] Specifically, in the process of expressing differences, when the actual construction state and the planned state are inconsistent in the semantic dimension of the process, the difference is manifested as a construction stage shift; when there is an inconsistency in the geometric shape dimension, the difference is manifested as a structural form shift; when there is an inconsistency in the physical index dimension, the difference is manifested as a material performance shift; and when there is an inconsistency in the material attribute dimension, the difference is manifested as a resource supply shift.
[0099] In the expression for the difference between the actual construction state and the planned state in the schedule-quality potential field, the feature vector of the actual construction state extracted from the pure feature dataset is used to represent the comprehensive feature set of the current construction state, the feature vector of the planned state mapped in the schedule-quality potential field is used to represent the target state expression under the constraints of the construction plan, and the feature weight matrix is used to structurally adjust the importance of different feature dimensions to form a unified difference calculation result.
[0100] The expression for the difference between the actual construction state and the planned state in the schedule-quality potential field is: ; in, This represents the difference between the actual construction status and the planned status in the schedule-quality potential field. This is the actual construction state feature vector extracted from the pure feature dataset. This represents the planned state feature vector mapped in the schedule-mass potential field. This is the feature weight matrix.
[0101] S5.5 When the deviation tensor shows that the actual formwork has been completed but the concrete temperature and humidity have not reached the demolding conditions and the time has exceeded the planned node, it will automatically adjust to postpone the subsequent pouring plan and send a work order for adding curing agent to the material management system, and feed the deviation tensor back to the progress-quality potential field to update the monitoring threshold for the next period.
[0102] Furthermore, when the deviation tensor shows that the actual formwork has been completed but the concrete temperature and humidity have not reached the demolding conditions and the time has exceeded the planned node, the process semantic offset information and physical index offset information in the deviation tensor are jointly judged to identify this state as a conflict state where the progress is ahead of schedule and the quality conditions are insufficient. In this case, the subsequent pouring plan is automatically adjusted, the pouring process in the original plan is adjusted to the delayed execution state, and a work order instruction for adding curing agent is sent to the material management system to compensate for the quality risk caused by insufficient concrete curing conditions.
[0103] During the deviation tensor feedback process, the deviation tensor is fed back to the schedule-quality potential field update unit, so that the monitoring threshold in the original schedule-quality potential field is dynamically corrected according to the current construction deviation status. When the construction process frequently results in the schedule being ahead of schedule but the quality indicators lagging behind, the constraint weight of the quality indicators in the potential field is increased. When the construction process frequently results in the resource lagging behind, the influence range of the material properties in the potential field is adjusted to form a dynamically updated schedule-quality potential field.
[0104] S6. Send control instructions to the execution terminal and update the multi-source sensing data based on the execution results.
[0105] S6.1. Issue control instructions to the execution terminal, and the execution terminal adjusts the operating parameters of the on-site construction equipment or the material delivery plan according to the control instructions.
[0106] Furthermore, during the process of issuing control instructions to the execution terminal, the control instructions generated by the construction potential field engine are structurally broken down according to the construction equipment category and material delivery category, and mapped to the corresponding construction equipment control unit and material scheduling execution unit respectively. The construction equipment control unit is used to adjust the operating parameters of tower cranes, pumping equipment and vibrating equipment, and the material scheduling execution unit is used to adjust the delivery sequence and supply rhythm of steel bars and formwork, so that different types of control instructions form a hierarchical execution path within the execution terminal.
[0107] Specifically, during the adjustment of operating parameters, for example, when the control command requires a delay in the concrete pouring pace, the construction equipment control unit adjusts the output pace of the pumping equipment and simultaneously coordinates and corrects the working rhythm of the vibrating equipment; when the control command requires an acceleration of the formwork installation progress, the operation priority of the hoisting equipment is adjusted so that the formwork components can enter the construction area first; when the control command involves material supply optimization, the material scheduling execution unit rearranges the order of steel bars and formwork arrival to ensure that the material supply matches the construction pace.
[0108] During the adjustment of the material distribution plan, when multiple construction components request the same type of material at the same time, the material allocation order is determined by the construction priority constraint, so that the components corresponding to the high-priority construction path can obtain material supply first, forming an execution mapping relationship consistent with the schedule-quality potential field.
[0109] S6.2. Based on the execution result feedback generated by the execution terminal executing the control command, update the video images, point cloud data, temperature and humidity and stress time series data of the concrete solidification environment, and RFID material information of the incoming steel bars and formwork in the spatiotemporally aligned multi-source data set.
[0110] Furthermore, during the feedback process of execution results generated by the execution terminal's execution control commands, information on changes in the operating status of construction equipment, changes in material delivery, and changes in the status of the construction site are uniformly collected and remapped to a spatiotemporally aligned multi-source data set. Among them, video image information is used to reflect changes in the movement of construction equipment and changes in personnel's work behavior, point cloud data is used to reflect changes in the spatial shape of components, temperature, humidity, and stress time series data of the concrete solidification environment are used to reflect changes in the material's environmental response, and RFID material information of incoming steel bars and formwork is used to reflect changes in the material flow status.
[0111] Specifically, during video image updates, when the operating parameters of construction equipment are adjusted, the equipment's movement trajectory is updated synchronously through visual capture of the construction area, ensuring that the adjusted equipment operating status is continuously expressed in the image sequence. During point cloud data updates, when the component installation progress changes, the component morphology is updated through spatial point cloud reconstruction, allowing the completed construction status to be remapped in the spatial structure. During concrete curing environment updates, when temperature, humidity, and stress states change, environmental changes are continuously updated through time-series synchronization, ensuring that the concrete strength development process is consistently expressed in the time dimension. During RFID material information updates, when the material delivery plan is adjusted, the component correspondence is updated through material coding rebinding, ensuring that the material flow status and the construction component status maintain a consistent correspondence.
[0112] When the execution result feedback deviates from the original spatiotemporally aligned multi-source data set, the updated data status is corrected for consistency through priority rules, so that the construction site status maintains a synchronous update relationship among the multi-source information.
[0113] This embodiment also provides a smart building construction intelligent management and control system based on multi-source sensing and data fusion, including: a preprocessing module, which collects multi-source sensing data from the construction site and preprocesses it to form a standardized multi-source sensing data stream; The alignment module performs spatiotemporal alignment processing on standardized multi-source sensing data streams to generate a spatiotemporally aligned multi-source data set. The feature extraction module extracts features from the spatiotemporally aligned multi-source dataset to obtain a conventional feature dataset. The decoupling filtering module inputs the regular feature dataset into the neural-symbolic dual processing module for semantic decoupling filtering to obtain a clean feature dataset. The comparison module inputs the clean feature dataset into the construction potential field engine for reverse comparison, calculates and generates a deviation tensor, and automatically fuses and generates control instructions based on the deviation tensor. The execution module sends control commands to the execution terminal and updates the multi-source sensing data based on the execution results.
[0114] This embodiment also provides a computer device applicable to the intelligent management and control method for smart building construction based on multi-source sensing and data fusion, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent management and control method for smart building construction based on multi-source sensing and data fusion as proposed in the above embodiment.
[0115] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.
[0116] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the intelligent control method for smart building construction based on multi-source sensing and data fusion, as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.
[0117] In summary, this invention performs semantic decoupling and filtering on conventional feature datasets through a neural-symbolic dual processing module. It utilizes the process logic relationships and mutual exclusion rules of the construction technology ontology library for verification. When cross-modal semantic conflicts occur, the neural layer and symbolic layer initiate dual verification. Based on the cross-comparison of confidence scores and modal credibility determination, it automatically downgrades the weight of untrustworthy modalities, achieving logical filtering and semantic decoupling of multimodal data. This eliminates the misleading influence of abnormal data on control decisions and improves the purity and semantic consistency of feature data. Furthermore, through a construction potential field engine, it performs a reverse comparison between the pure feature dataset and a progress-quality potential field constructed based on the total floating time calculated using a graph attention network. It calculates the difference between the actual construction state and the planned state to generate a deviation tensor, enabling quantitative assessment of progress delays and quality risks. Then, based on the deviation tensor, it automatically fuses and generates control instructions and issues them to the execution terminal. Based on the execution results, it updates multi-source sensing data, forming a closed-loop intelligent control system of perception, decision-making, and execution. This solves the problems of existing technologies lacking deep logical constraints and physical rule modeling, leading to frequent false alarms and missed alarms, and the inability to reliably support the automatic generation and issuance of control instructions.
[0118] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A smart building construction intelligent management and control method based on multi-source sensing and data fusion, characterized in that: include, Collect multi-source sensing data from the construction site and preprocess it to form a standardized multi-source sensing data stream; The standardized multi-source sensing data stream is spatiotemporally aligned to generate a spatiotemporally aligned multi-source data set. Feature extraction is performed on the spatiotemporally aligned multi-source dataset to obtain a conventional feature dataset; The regular feature dataset is input into the neural-symbolic dual processing module for semantic decoupling and filtering to obtain a clean feature dataset. The clean feature dataset is input into the construction potential field engine for reverse comparison, and the deviation tensor is calculated and generated automatically based on the deviation tensor to generate control instructions. Control commands are sent to the execution terminal, and the multi-source sensing data is updated based on the execution results.
2. The intelligent control method for smart building construction based on multi-source sensing and data fusion as described in claim 1, characterized in that: The formation of a standardized multi-source sensing data stream includes... The system collects video images, point cloud data, temperature and humidity data of the concrete solidification environment, stress time series data, and RFID material information of incoming steel bars and formwork at the construction site. After noise reduction, target detection, and protocol conversion preprocessing, it forms a standardized multi-source sensing data stream.
3. The intelligent control method for smart building construction based on multi-source sensing and data fusion as described in claim 2, characterized in that: The generated spatiotemporally aligned multi-source data set includes, By using a unified time reference and spatial coordinate system, the positions of personnel and machinery in video images, the outlines of components in point cloud data, the temperature and humidity of the concrete solidification environment, the physical quantity values of stress time series data, and the material codes of RFID material information of incoming steel bars and formwork are mapped to the corresponding floor and axis positions of the BIM construction model built inside the edge computing gateway, generating a spatiotemporally aligned multi-source data set.
4. The intelligent control method for smart building construction based on multi-source sensing and data fusion as described in claim 3, characterized in that: The obtained conventional feature dataset includes, The computer vision algorithm is used to identify the type of work of workers in video images in a spatiotemporally aligned multi-source dataset. Based on point cloud clustering and segmentation, the geometric dimensions and installation progress of components in the point cloud data of the spatiotemporally aligned multi-source dataset are extracted. The concrete strength development curve is calculated based on the temperature, humidity, and stress time series data of the concrete solidification environment in the spatiotemporally aligned multi-source dataset using sensor readings. The material type and quantity are obtained by matching the RFID material information of incoming steel bars and formwork in a multi-source dataset after spatiotemporal alignment using RFID. The types of work, the geometric dimensions and installation progress of components, the concrete strength development curve, and the material type and quantity are summarized into a set of conventional features including process semantics, geometric shape, physical indicators and material properties.
5. The intelligent control method for smart building construction based on multi-source sensing and data fusion as described in claim 4, characterized in that: The resulting pure feature dataset A neural-symbolic dual processing module is deployed within the edge computing gateway. The neural-symbolic dual processing module has a built-in construction process ontology library, which stores the logical relationships and mutual exclusion rules of the construction process. The conventional feature dataset is input into the neural-symbolic dual processing module to extract the process semantic labels of each feature in the conventional feature dataset; The semantic tags of the work process are matched and verified with the logical relationship of the work process corresponding to the current planned work process in the construction process ontology library. The semantic tags of the work process are matched and verified with the mutual exclusion rules corresponding to the current planned work process in the construction process ontology library. When any of the following features are missing in the regular feature dataset: worker operation type feature of video image mode, component geometric size and installation progress feature of point cloud data mode, temperature and humidity of concrete solidification environment, concrete strength development curve feature of stress time series data mode, and material type and quantity feature of RFID material information mode of incoming steel bars and formwork, it is determined to be a cross-modal semantic conflict. When a cross-modal semantic conflict is determined, the dual verification between the neural layer and the symbolic layer is initiated. The neural layer outputs the confidence score of each modal feature, and the symbolic layer outputs the modal confidence determination based on the priority of the construction process ontology library. The confidence score is cross-compared with the modal credibility determination. When the confidence score of any modal output in the neural layer reaches the credibility threshold, a clean feature dataset after semantic decoupling and logical filtering is obtained.
6. The intelligent control method for smart building construction based on multi-source sensing and data fusion as described in claim 5, characterized in that: The generated control instructions include, Based on the logical dependencies and durations of various construction tasks in the 4D BIM schedule, calculate the total float time for the construction activities corresponding to each structural component. The path formed by connecting construction activities with a total floating time less than or equal to a preset time threshold is defined as a high-priority construction path. The potential field value corresponding to each structural component on the high-priority construction path is assigned a value greater than the potential field value corresponding to the structural component on the non-high-priority construction path. The potential field value corresponding to the monitoring area specified in the quality acceptance specification is assigned in a way that decreases layer by layer according to the spatial distance from the high-priority construction path, thus forming a progress-quality potential field. The clean feature dataset is input into the construction potential field engine, which then performs a reverse comparison between the actual construction status in the clean feature dataset and the corresponding planned status in the progress-quality potential field. Based on the inverse comparison results between the actual construction status in the pure feature dataset and the corresponding planned status in the schedule-quality potential field, the difference between the actual construction status in the pure feature dataset and the planned status in the schedule-quality potential field is calculated, and a deviation tensor reflecting the degree of schedule lag and the quality risk level is generated. When the deviation tensor indicates that the actual formwork has been completed but the concrete temperature and humidity have not reached the conditions for demolding and the time has exceeded the planned node, it will automatically adjust to postpone the subsequent pouring plan and send a work order for adding curing agent to the material management system. The deviation tensor will be fed back to the progress-quality potential field to update the monitoring threshold for the next period.
7. The intelligent control method for smart building construction based on multi-source sensing and data fusion as described in claim 6, characterized in that: The updating of the multi-source sensing data includes Control instructions are issued to the execution terminal, which then adjusts the operating parameters of on-site construction equipment or the material delivery plan based on the control instructions. Based on the execution result feedback generated by the execution terminal's execution control instructions, update the video images, point cloud data, temperature, humidity and stress time series data of the concrete solidification environment, and RFID material information of incoming steel bars and formwork in the spatiotemporally aligned multi-source data set.
8. A smart building construction intelligent management and control system based on multi-source sensing and data fusion, based on the smart building construction intelligent management and control method based on multi-source sensing and data fusion as described in any one of claims 1 to 7, characterized in that: This includes a preprocessing module, which collects and preprocesses multi-source sensing data from the construction site to form a standardized multi-source sensing data stream; The alignment module performs spatiotemporal alignment processing on standardized multi-source sensing data streams to generate a spatiotemporally aligned multi-source data set. The feature extraction module extracts features from the spatiotemporally aligned multi-source dataset to obtain a conventional feature dataset. The decoupling filtering module inputs the regular feature dataset into the neural-symbolic dual processing module for semantic decoupling filtering to obtain a clean feature dataset. The comparison module inputs the clean feature dataset into the construction potential field engine for reverse comparison, calculates and generates a deviation tensor, and automatically fuses and generates control instructions based on the deviation tensor. The execution module sends control commands to the execution terminal and updates the multi-source sensing data based on the execution results.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the intelligent management and control method for smart building construction based on multi-source perception and data fusion as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the intelligent management and control method for smart building construction based on multi-source perception and data fusion as described in any one of claims 1 to 7.
