A physical constraint driven method and system for intelligent diagnosis of existing building beam columns

Abstract

The application provides a kind of physical constraint driven existing building beam column intelligent diagnosis method and system in the cross technical field of intelligent construction and structure health monitoring, method includes: step S1, construct multimodal basic data, obtain functional target parameter;Step S2, based on multimodal basic data, identify the key stress area of existing building and collect field monitoring data;Step S3, the feature vector of pre-processing field monitoring data is uploaded to cloud;Step S4, cloud is fused with feature vector by structure state cognitive agent, generates structure real mechanics image;Step S5, based on functional target parameter and structure real mechanics image, the continuous evaluation and intelligent diagnosis of beam column demolition effect is carried out, and the function target compliance judgment result is output.The application has the advantages that: it greatly improves the transparent perception of beam column demolition whole process, the dynamic correction of calculation model and the closed-loop intelligent evaluation ability of structure performance.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of intelligent construction and structural health monitoring, and in particular to a method and system for intelligent diagnosis of existing building beams and columns driven by physical constraints. Background Technology

[0002] In the functional upgrading, spatial renovation, or safety reinforcement projects of existing buildings, the demolition, alteration, and reinforcement of critical load-bearing components such as beams and columns is a common and complex technical approach. These methods include, but are not limited to, increasing cross-sections, steel plate reinforcement, carbon fiber reinforcement, and the highly complex replacement of columns with supporting beams. Such construction significantly alters the force transmission path and mechanical properties of the original structure; therefore, accurate assessment of structural safety during the demolition and alteration process and long-term service performance after the alteration is crucial.

[0003] Currently, the evaluation of the effects of beam and column alteration in engineering practice mainly relies on two types of technologies: one is discretized, phased on-site testing, and the other is theoretical calculation and simulation before construction. On-site testing typically uses methods such as rebound hammer testing, ultrasonic testing, and core drilling to obtain material strength, supplemented by total stations and levels for settlement and displacement observation. Theoretical calculation mainly relies on finite element analysis software to perform stress verification based on design drawings and idealized material constitutive and boundary conditions.

[0004] However, existing technologies have significant limitations and are insufficient to meet the needs for comprehensive, detailed, and intelligent assessment of the entire demolition and renovation project process. Specifically, these limitations manifest in the following aspects: 1. The "black box" nature of the process and the unpredictable redistribution of internal forces: During the dynamic construction process of beam and column alteration, the structure undergoes complex redistribution of internal forces due to unloading, reloading, and the collaborative work of old and new materials. Existing discretized and static detection methods cannot continuously capture this time-varying process, resulting in a "black box effect" in the evolution of the stress state at key interfaces (such as the interface between old and new concrete and the area where rebar is installed). The evaluation of the reinforcement effect is often lagging and remains at the theoretical level.

[0005] 2. Limited and Coupled Monitoring Data: Existing monitoring technologies present limited data types (e.g., measuring only strain or displacement) and are discontinuous in time, making it difficult to comprehensively reflect the coupling relationship between structural strain, vibration characteristics, crack evolution, and environmental temperature and humidity. Therefore, they cannot effectively provide early warnings and long-term monitoring of potential risks such as cumulative damage and stiffness degradation that may occur early after demolition or alteration.

[0006] 3. Rigid computational models, disconnected from actual working conditions: Finite element analyses conducted before construction are mostly one-time "static dead models." These models are based on ideal assumptions and cannot be adaptively corrected according to real-time monitoring data during construction and service. This leads to systematic deviations between model parameters (such as material properties and boundary constraints) and the actual state of the structure, greatly reducing the reliability and guiding significance of the calculation results.

[0007] 4. Weak correlation between assessment results and functional objectives: Traditional assessment methods often focus on immediate safety judgments and lack a continuous, quantitative assessment system guided by the pre-set "functional objectives" of the demolition and alteration project (such as the proportion of bearing capacity improvement, deflection control indicators, and seismic performance requirements). This makes it difficult to directly use the assessment results to guide subsequent maintenance, performance optimization, or secondary reinforcement decisions.

[0008] Therefore, how to provide a physical constraint-driven intelligent diagnostic method and system for existing building beams and columns, so as to achieve transparent perception of the entire process of beam and column demolition and modification, dynamic correction of the calculation model, and closed-loop intelligent evaluation of structural performance, has become an urgent technical problem to be solved. Summary of the Invention

[0009] The technical problem to be solved by this invention is to provide a physical constraint-driven intelligent diagnostic method and system for existing building beams and columns, which can improve the transparency of the entire process of beam and column demolition and modification, the dynamic correction of the calculation model, and the closed-loop intelligent evaluation capability of structural performance.

[0010] In a first aspect, the present invention provides a physically constrained intelligent diagnostic method for existing building beams and columns, comprising the following steps: Step S1: Construct multimodal basic data including the original structural design data of existing buildings and the beam and column demolition and modification design data, and obtain the input functional target parameters encoded as target constraint vectors; Step S2: Identify the key stress areas of the existing building based on the multimodal basic data, and deploy a multimodal sensing network based on the key stress areas to collect on-site monitoring data. The on-site monitoring data includes strain parameters, vibration parameters, crack parameters, and environmental parameters. Step S3: Perform real-time preprocessing on the field monitoring data through edge computing nodes, including at least data noise reduction, timestamp alignment, and feature extraction, to obtain standardized feature vectors and upload them to the cloud; Step S4: Construct and run a physical constraint-driven structural state cognitive agent in the cloud. The structural state cognitive agent dynamically infers and self-evolves to update the structural state of the beams and columns after dismantling and modification by integrating prior physical knowledge with the feature vector, thereby generating a true mechanical profile of the structure. Step S5: Based on the functional target parameters and the actual mechanical profile of the structure, continuously evaluate and intelligently diagnose the beam and column demolition and modification effects, and output the functional target compliance judgment results. Step S6: Based on the functional objective compliance determination results, output visualized diagnostic results and optimization suggestions, and form a full lifecycle data archive covering the entire transformation process.

[0011] Furthermore, in step S1, the multimodal basic data includes the structural topology, geometric parameters, and initial mechanical parameters obtained by parsing the structural drawings of the CAD or BIM model. The functional target parameters include at least one of the following: bearing capacity improvement ratio, maximum allowable displacement, and seismic performance requirements.

[0012] Furthermore, in step S2, the multimodal sensing network includes at least three of the following sensors: fiber optic strain sensor, vibrating wire stress gauge, triaxial MEMS accelerometer, biaxial tilt sensor, industrial-grade vision sensor, and temperature and humidity sensor. The multimodal sensing network adopts a dynamic trigger sampling mechanism, using a high sampling frequency during key construction processes and a low sampling frequency or threshold-triggered sampling during long-term service monitoring.

[0013] Furthermore, in step S4, constructing and running the physically constrained structure state cognitive agent includes: The cognitive kernel of the structure state cognitive agent is constructed. The cognitive kernel has a built-in physical prior knowledge base, which stores the physical prior knowledge of the basic constraint rules of structural mechanics and the nonlinear mechanical constraint operators for the connection interface of new and old components. A neural operator inference engine is used as the core of the structural response calculation of the structural state cognitive agent. It learns the operator mapping relationship from structural parameters to structural response to replace the traditional explicit solution of finite element method. In the reasoning process of the structure state cognitive agent, the node force balance residual and energy conservation deviation are embedded as constraints in the loss function to ensure the physical consistency of the reasoning results. The feature vector is dynamically compared with the predicted state of the structural state cognitive agent. When the deviation exceeds a preset threshold, an adaptive update mechanism based on physical constraints is triggered to correct the key state variables that reflect the evolution of structural performance, and a true mechanical profile of the structure is generated based on the corrected key state variables. The adaptive update mechanism employs reinforcement learning or adversarial optimization strategies to learn the evolution of structural internal force redistribution, damage accumulation, and cooperative force relationships during the dismantling and modification process, while satisfying physical constraints.

[0014] Furthermore, in step S5, the continuous assessment and intelligent diagnosis specifically include: Calculate the load-bearing capacity reserve, stiffness variation, and structural dynamic performance indicators of beams and columns; Based on the above indicators, long-term damage accumulation and structural stability analysis are conducted to identify accumulated damage, stiffness degradation or instability risks, and thus obtain the assessment results. The evaluation results are compared with the functional target parameters to complete the functional target compliance determination; In step S6, the optimization suggestions include secondary grouting, adding carbon fiber reinforcement, or adjusting the construction plan.

[0015] Secondly, the present invention provides a physically constrained intelligent diagnostic system for existing building beams and columns, comprising the following modules: The initialization module is used to construct multimodal basic data, including the original structural design data and beam and column demolition and modification design data of existing buildings, and to obtain the input functional target parameters encoded as target constraint vectors. The on-site monitoring data acquisition module is used to identify key stress areas of existing buildings based on the multimodal basic data, and to deploy a multimodal sensing network based on the key stress areas to collect on-site monitoring data, including strain parameters, vibration parameters, crack parameters and environmental parameters. The on-site monitoring data preprocessing module is used to perform real-time preprocessing on the on-site monitoring data through edge computing nodes, including at least data noise reduction, timestamp alignment and feature extraction, to obtain standardized feature vectors and upload them to the cloud. The structural real mechanics profile generation module is used to build and run a physical constraint-driven structural state cognitive intelligent agent in the cloud. The structural state cognitive intelligent agent dynamically infers and self-evolves to update the structural state of beams and columns after dismantling and modification by integrating physical prior knowledge with the feature vector, thereby generating a structural real mechanics profile. The intelligent diagnostic module is used to continuously evaluate and intelligently diagnose the effects of beam and column demolition and modification based on the functional target parameters and the actual mechanical profile of the structure, and output the functional target compliance judgment result. The results output module is used to output visualized diagnostic results and optimization suggestions based on the functional target compliance judgment results, and to form a full life cycle data archive covering the entire transformation process.

[0016] Furthermore, in the initialization module, the multimodal basic data includes the structural topology, geometric parameters, and initial mechanical parameters obtained by parsing the structural drawings of the CAD or BIM model. The functional target parameters include at least one of the following: bearing capacity improvement ratio, maximum allowable displacement, and seismic performance requirements.

[0017] Furthermore, in the field monitoring data acquisition module, the multimodal sensing network includes at least three of the following sensors: fiber optic strain sensor, vibrating wire stress gauge, triaxial MEMS accelerometer, biaxial tilt sensor, industrial-grade vision sensor, and temperature and humidity sensor. The multimodal sensing network adopts a dynamic trigger sampling mechanism, using a high sampling frequency during key construction processes and a low sampling frequency or threshold-triggered sampling during long-term service monitoring.

[0018] Furthermore, in the structural real-mechanical profile generation module, the construction and operation of the physically constraint-driven structural state cognitive agent includes: The cognitive kernel of the structure state cognitive agent is constructed. The cognitive kernel has a built-in physical prior knowledge base, which stores the physical prior knowledge of the basic constraint rules of structural mechanics and the nonlinear mechanical constraint operators for the connection interface of new and old components. A neural operator inference engine is used as the core of the structural response calculation of the structural state cognitive agent. It learns the operator mapping relationship from structural parameters to structural response to replace the traditional explicit solution of finite element method. In the reasoning process of the structure state cognitive agent, the node force balance residual and energy conservation deviation are embedded as constraints in the loss function to ensure the physical consistency of the reasoning results. The feature vector is dynamically compared with the predicted state of the structural state cognitive agent. When the deviation exceeds a preset threshold, an adaptive update mechanism based on physical constraints is triggered to correct the key state variables that reflect the evolution of structural performance, and a true mechanical profile of the structure is generated based on the corrected key state variables. The adaptive update mechanism employs reinforcement learning or adversarial optimization strategies to learn the evolution of structural internal force redistribution, damage accumulation, and cooperative force relationships during the dismantling and modification process, while satisfying physical constraints.

[0019] Furthermore, in the intelligent diagnostic module, the continuous assessment and intelligent diagnosis specifically include: Calculate the load-bearing capacity reserve, stiffness variation, and structural dynamic performance indicators of beams and columns; Based on the above indicators, long-term damage accumulation and structural stability analysis are conducted to identify accumulated damage, stiffness degradation or instability risks, and thus obtain the assessment results. The evaluation results are compared with the functional target parameters to complete the functional target compliance determination; In the result output module, the optimization suggestions include secondary grouting, adding carbon fiber reinforcement, or adjusting the construction plan.

[0020] The advantages of this invention are: 1. By constructing multimodal foundational data including the original structural design data and beam-column alteration design data of existing buildings, functional target parameters encoded as target constraint vectors are obtained as input. Then, based on the multimodal foundational data, key stress areas of the existing buildings are identified, and a multimodal sensing network is deployed based on these key stress areas to collect on-site monitoring data. Next, the on-site monitoring data is preprocessed in real time using edge computing nodes to obtain standardized feature vectors, which are then uploaded to the cloud. The cloud uses a structural state cognitive agent to fuse prior physical knowledge with feature vectors, dynamically inferring and self-evolving the structural state of the altered beams and columns, generating a realistic structural mechanical profile. Based on the functional target parameters and the realistic structural mechanical profile, the alteration effect of beams and columns is continuously evaluated and intelligently diagnosed, outputting a functional target compliance judgment result. Finally, based on the functional target compliance judgment result, a visualized diagnostic result and optimization suggestions are output, forming a full lifecycle data archive covering the entire renovation process. This involves deploying a multimodal sensing network and leveraging edge computing... The system performs real-time data processing, enabling continuous and high-frequency acquisition of key parameters such as beam and column strain, vibration, and cracks during the demolition and alteration process, thereby enhancing the transparency of the entire process. Based on this, a cloud-based, physically constrained, structure-state cognitive intelligent agent integrates real-time data with built-in physical prior knowledge and uses neural operators for reasoning. When the deviation between prediction and measurement exceeds limits, it triggers a self-evolutionary update mechanism, dynamically correcting key state variables of the model. This allows the computational model to continuously evolve with the actual structural state, achieving a fundamental shift from static rigidity to dynamic correction. Finally, the system continuously assesses the conformity between the real-time updated structural mechanical profile and pre-set functional target parameters, outputting quantitative evaluation results and optimization suggestions. This forms a closed-loop intelligent evaluation system from perception and cognition to decision-making, overcoming the limitations of traditional methods that are lagging and disconnected. Ultimately, this significantly improves the transparency of the entire beam and column demolition and alteration process, the dynamic correction of the computational model, and the closed-loop intelligent evaluation capabilities of structural performance.

[0021] 2. By integrating the original structural design data of existing buildings, beam and column demolition and modification design data, and on-site monitoring data (such as strain, vibration, cracks, and environmental parameters), a multimodal basic data system was constructed. This multi-source data fusion method overcomes the limitations of traditional diagnostic methods that rely on a single data source, and can more comprehensively capture changes in structural state. In particular, during the demolition and modification process, by analyzing CAD or BIM models to obtain structural topological relationships, geometric parameters, etc., the accuracy and reliability of diagnosis are enhanced, thus providing a solid data foundation for subsequent intelligent diagnosis.

[0022] 3. Edge computing nodes are used to perform real-time preprocessing of on-site monitoring data, including data noise reduction, timestamp alignment, and feature extraction. After generating standardized feature vectors, the data is uploaded to the cloud. This design significantly reduces data transmission latency and cloud computing burden, achieving near real-time data processing capabilities. It is particularly suitable for dynamic monitoring scenarios in the renovation of existing buildings, avoiding the diagnostic lag caused by data transmission bottlenecks in traditional methods, and improving the overall system response speed and efficiency.

[0023] 4. By constructing a physical constraint-driven structural state cognitive agent in the cloud, physical prior knowledge (such as basic constraint rules of structural mechanics and constraint operators of nonlinear mechanics) is fused with feature vectors to perform dynamic reasoning and self-evolutionary updates. This physical constraint embedding mechanism ensures the physical rationality of the reasoning results and avoids non-physical errors that may be generated by pure data-driven methods. At the same time, the neural operator reasoning engine is used to replace the traditional finite element solution, which improves the computational efficiency and makes the diagnostic process more intelligent and adaptable.

[0024] 5. The structural state cognitive agent has an adaptive update mechanism. When the deviation between the monitored data and the predicted state exceeds a threshold, it triggers a correction strategy based on reinforcement learning or adversarial optimization to update key state variables. This feature enables the system to continuously learn the evolution laws of structural internal force redistribution and damage accumulation, enhancing the robustness and long-term stability of diagnosis. It is particularly suitable for monitoring the entire life cycle of existing building renovations and can promptly identify accumulated damage or instability risks.

[0025] 6. By covering the entire process from data collection to diagnostic output and forming a full lifecycle data archive, including visualized diagnostic results and optimization suggestions, this systematic management approach supports long-term tracking and retrospection of renovation projects, improves data traceability and management efficiency, provides decision-makers with a complete chain of evidence, helps optimize subsequent maintenance strategies, and is in line with the development trend of modern building information management.

[0026] 7. Based on functional target parameters and the actual mechanical profile of the structure, the system automatically performs continuous evaluation and intelligent diagnosis, outputting the functional target compliance judgment results and optimization suggestions (such as secondary grouting or carbon fiber reinforcement). This automated process reduces manual intervention, improves diagnostic efficiency and accuracy, and lowers the professional threshold through visual output, enabling non-professionals to quickly understand the results, thereby enhancing the practicality and promotional value of the technology.

[0027] 8. The multimodal sensing network adopts a dynamic trigger sampling mechanism, using a high sampling frequency during critical construction processes and a low sampling or threshold-triggered sampling during long-term service. This intelligent scheduling method optimizes the use of sensor resources and reduces energy consumption and data storage costs. This not only improves economic efficiency but also conforms to the concept of green building. By reducing unnecessary monitoring activities, it achieves energy-saving and environmental protection effects.

[0028] 9. By constructing an intelligent assessment system that integrates "end-cloud-model" collaboration, continuous monitoring and intelligent diagnosis of the entire process of demolition and renovation of existing building beams and columns, as well as their service life, have been achieved, significantly improving the accuracy and reliability of the renovation quality assessment. A multimodal sensing network is used to collect structural strain, vibration, cracks, and environmental parameters. Combined with design drawings and functional target data, real-time preprocessing and feature extraction are performed through edge computing. This forms a dual-drive closed loop with the cloud-based structural state cognitive intelligent agent, enabling dynamic adaptive correction of model parameters and continuous evolution of the digital twin. This allows the diagnostic results to closely approximate the actual structural state, overcoming the inaccuracy of traditional static model assessments.

[0029] 10. Guided by functional objectives, the system quantitatively assesses the load-bearing capacity, stiffness evolution, and long-term performance degradation of beams and columns, and provides scientific decision-making basis for engineering maintenance, performance optimization, and secondary reinforcement by visually outputting diagnostic results and optimization suggestions. At the same time, it forms a full life-cycle data archive, realizes the accumulation of structural performance knowledge and intelligent management closed loop, and provides reliable data support for the long-term safety assurance, continuous performance evaluation, and future renovation of existing buildings, which has significant technological innovation and engineering practical value. Attached Figure Description

[0030] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0031] Figure 1 This is a flowchart of a physical constraint-driven intelligent diagnostic method for existing building beams and columns according to the present invention.

[0032] Figure 2 This is a schematic diagram of the structure of an intelligent diagnostic system for existing building beams and columns driven by physical constraints, according to the present invention. Detailed Implementation

[0033] The overall concept of the technical solution in this application is as follows: By deploying a multimodal sensing network and using edge computing for real-time data processing, continuous and high-frequency acquisition of key parameters such as beam and column strain, vibration, and cracks during the demolition and alteration process is achieved, thereby enhancing the transparency of the entire process. On this basis, a cloud-based, physically constrained, structure state cognitive intelligent agent integrates real-time data with built-in physical prior knowledge and uses neural operators for reasoning. When the deviation between prediction and measurement exceeds the limit, a self-evolutionary update mechanism is triggered to dynamically correct key state variables of the model, enabling the computational model to continuously evolve with the actual state of the structure, achieving a fundamental shift from static rigidity to dynamic correction. Finally, the real mechanical profile of the structure based on real-time updates is continuously judged against the pre-set functional target parameters, and quantitative evaluation results and optimization suggestions are output, forming a closed-loop intelligent evaluation system from perception, cognition to decision-making, thereby improving the transparency of the entire beam and column demolition and alteration process, the dynamic correction of the computational model, and the closed-loop intelligent evaluation capability of structural performance.

[0034] Please refer to Figures 1 to 2 As shown, a preferred embodiment of the physical constraint-driven intelligent diagnosis method for existing building beams and columns of the present invention includes the following steps: Step S1: Construct multimodal basic data including the original structural design data of existing buildings and the beam and column demolition and modification design data, and obtain the input functional target parameters encoded as target constraint vectors; Step S2: Identify the key stress areas of the existing building based on the multimodal basic data, and deploy a multimodal sensing network based on the key stress areas to collect on-site monitoring data. The on-site monitoring data includes strain parameters, vibration parameters, crack parameters, and environmental parameters. Step S3: The field monitoring data is preprocessed in real time through edge computing nodes, including at least data noise reduction, timestamp alignment (to ensure that different types of data have a consistent spatiotemporal correspondence under the same structural response event) and feature extraction (including strain change rate, structural main frequency change characteristics, crack width and its evolution index, environmental correction factor, etc.) to reduce data transmission load and improve the reliability of subsequent analysis, obtain standardized feature vectors and upload them to the cloud. In practice, the edge computing nodes first perform data quality control on the collected field monitoring data. They use a combination of time-domain filtering and frequency-domain analysis to identify and remove abnormal data caused by non-structural factors such as construction machinery vibration, accidental impacts, or environmental interference, thereby ensuring the authenticity and stability of the structural response data.

[0035] Step S4: Construct and run a physical constraint-driven structural state cognitive agent in the cloud. The structural state cognitive agent dynamically infers and self-evolves to update the structural state of the beams and columns after dismantling and modification by integrating prior physical knowledge with the feature vector, thereby generating a true mechanical profile of the structure. Step S5: Based on the functional target parameters and the actual mechanical profile of the structure, continuously evaluate and intelligently diagnose the beam and column demolition and modification effects, and output the functional target compliance judgment results. Step S6: Based on the functional objective compliance determination results, output visualized diagnostic results and optimization suggestions, and form a full lifecycle data archive covering the entire transformation process.

[0036] In step S1, the multimodal basic data includes the structural topology, geometric parameters, and initial mechanical parameters obtained by parsing the structural drawings of the CAD or BIM model. The structural drawings include the original structural design drawings of the existing building and the beam and column demolition and alteration design drawings. The drawing data includes at least the geometric dimensions, reinforcement form and material strength grade of the beams and columns before the renovation, the cross-sectional form, material parameters and connection structure of the newly added or reinforced components in the renovation plan, and detailed construction drawings of beam and column joints and interfaces between new and old components. The functional target parameters include at least one of the following: bearing capacity improvement ratio, maximum allowable displacement, and seismic performance requirements. The functional target parameters also include target bearing capacity level, deflection control index, appraisal report conclusion parameters, ductility improvement requirements, and performance stability or durability indicators during service. These functional target parameters are encoded into system-recognizable target constraint vectors, which serve as important target modal data in subsequent model calculations, threshold setting, and evaluation result determination.

[0037] In step S2, the multimodal sensing network includes at least three of the following sensors: fiber optic strain sensor, vibrating wire stress gauge, triaxial MEMS accelerometer, biaxial tilt sensor, industrial-grade vision sensor, and temperature and humidity sensor. The multimodal sensing network adopts a dynamic trigger sampling mechanism, using a high sampling frequency during key construction processes and a low sampling frequency or threshold-triggered sampling during long-term service monitoring.

[0038] In practice, a unified data structure model is used to model the multimodal basic data, functional target parameters, and on-site monitoring data, and to establish indexes and relationships between the multimodal data, thereby providing a complete and unified data foundation for perception, calculation, and intelligent diagnosis during the beam and column demolition and modification process.

[0039] After completing the construction of multimodal basic data, the system integrates existing building renovation construction drawings, relies on the building knowledge base and fact set, and accurately identifies key stress areas in beam and column demolition and modification through inference engine. A multimodal perception network is then deployed to acquire the actual mechanical response of the structure throughout its service life. Based on design drawings and initial mechanical simulation results, the system automatically identifies the mid-span tension zone of beams, the shear zone at supports, the compression zones at the four corners of reinforced columns, the interface between new and old components, and easily tilted parts of the overall structure as key monitoring points. This enables effective perception of internal force redistribution, interface co-stress, and spatial positional changes.

[0040] Specifically, based on the imported structural design drawings and initial mechanical analysis results, key monitoring locations are identified, including the mid-span tension zone of the beam, the support shear zone, the four corner compression zones of the reinforced columns, and the interfaces between new and old components. This aims to ensure effective perception of the redistribution of internal forces and the collaborative stress state of interfaces. A multimodal sensing network is deployed at these locations, with the following specific configuration: ① Fiber optic strain sensor and vibrating wire stress gauge: Deployed together to obtain high-precision strain response and internal stress state; ② Triaxial MEMS accelerometer: used to collect the natural frequency and dynamic response of the structure; ③ Dual-axis tilt sensor: used to monitor the tilt angle and deformation trend of components or the overall structure in real time; ④ Industrial-grade vision sensor: Periodically captures images of beam and column surfaces, automatically identifies and quantifies crack characteristics; ⑤ Temperature and humidity sensor: Synchronously collects environmental disturbance factors to correct the data.

[0041] Various sensors perform synchronous sampling based on a unified time reference. The sampling frequency adopts a dynamic triggering mechanism: during critical processes such as construction unloading and reloading, the sampling frequency is increased to 50Hz-100Hz to capture instantaneous responses; during long-term service monitoring, the static sensor frequency is adjusted to once per hour (or sampling is triggered according to a threshold), thereby achieving continuous and reliable monitoring of the structure throughout its entire life cycle and providing real and complete data support for subsequent intelligent analysis and safety diagnosis.

[0042] In step S4, constructing and running the physically constrained structure state cognitive agent includes: A cognitive kernel is constructed for the structural state cognitive agent. The cognitive kernel has a built-in physical prior knowledge base. The physical prior knowledge base stores the basic constraint rules of structural mechanics (including nonlinear constitutive relations of materials, structural force equilibrium conditions, displacement compatibility relations and energy conservation criteria) as well as the physical prior knowledge of nonlinear mechanical constraint operators (bond-slip, interface shear degradation, etc.) for the connection interface between new and old components (the interface between new and old concrete, the anchoring zone of rebar and the connection interface of reinforced components in the demolition and renovation of existing buildings). The cognitive kernel does not rely on the discretization and step-by-step iterative solution of the stiffness matrix in traditional finite element analysis. Instead, it achieves rapid cognition and global inference of the service state of the structure by integrating physical prior constraints and data-driven reasoning. A neural operator inference engine is employed as the core of structural response calculation for the structural state cognitive agent. It learns the operator mapping relationship from structural parameters to structural response, replacing the traditional explicit finite element method (FEM) solution. Based on a network architecture of Fourier neural operators, the engine learns the operator mapping relationship between "structural geometry—load—material parameters" and "stress—displacement—damage distribution," transforming the traditionally explicitly solved partial differential equation problem into a forward inference process in operator space. Compared to conventional FEM analysis, this method is independent of mesh refinement, possesses resolution independence, and can quickly output the overall structural response results even when the structural topology undergoes local changes, thus significantly improving the real-time performance and continuity of structural risk assessment during the demolition and alteration construction phase. In the reasoning process of the structural state cognitive agent, the node force balance residual and energy conservation deviation are embedded as constraints in the loss function to ensure the physical consistency of the reasoning results. To ensure the physical consistency of the reasoning results of the structural state cognitive agent, a physical constraint embedding mechanism is introduced in the reasoning process, embedding physical quantities such as node force balance residual and energy conservation deviation as constraints in the loss function of the structural state cognitive agent. In this way, the structural state cognitive agent can still output reliable results that conform to the basic laws of structural mechanics even when multimodal data is incomplete or there is noise interference, fundamentally avoiding the "black box inference" and physical distortion problems that may occur in pure data-driven models. The feature vector is dynamically compared with the predicted state of the structural state cognitive agent. When the deviation exceeds a preset threshold, an adaptive update mechanism based on physical constraints is triggered to correct the key state variables that reflect the evolution of structural performance, and a true mechanical profile of the structure is generated based on the corrected key state variables. The adaptive update mechanism employs reinforcement learning or adversarial optimization strategies to learn the evolution of structural internal force redistribution, damage accumulation, and cooperative force relationships during the dismantling and modification process, while satisfying physical constraints.

[0043] During operation, the structural state cognitive agent continuously receives feature vectors uploaded by edge computing nodes and dynamically compares them with the internal physical reasoning results (predicted state) as external perception input. When a significant deviation occurs between the monitored feature vector and the agent's current predicted state, the agent will automatically determine that the structural state has evolved and trigger an adaptive update mechanism.

[0044] The adaptive update mechanism is not simply a numerical correction of model parameters, but rather an autonomous update of key state variables reflecting the evolution of structural performance under physical constraints. These variables include the degradation parameter of elastic modulus, the equivalent shear stiffness of the interface between new and old components, and connection constraints. By introducing reinforcement learning or adversarial optimization strategies, it can gradually learn the evolution of internal force redistribution, damage accumulation, and cooperative force relationships during the demolition and modification process, while satisfying physical constraints. This allows the description of the internal structural state to continuously approximate the actual structural behavior as the actual engineering process progresses.

[0045] After completing the self-evolution of the structural state, the structural state cognitive agent uses variational Bayesian inference to probabilistically represent the current structural state, forming a "real mechanical portrait of the structure" that includes descriptions of uncertainty.

[0046] Through the above mechanism, the structural state cognitive intelligent agent constructed by this invention realizes a closed-loop operation mode from "multimodal monitoring - state cognition - self-evolutionary reasoning - decision feedback", so that the structural safety status of existing building beams and columns during the demolition and modification process and service stage is always within a perceptible, explainable and predictable controllable range.

[0047] In step S5, the continuous assessment and intelligent diagnosis specifically include: By combining the revised digital twin model with on-site monitoring data, the load-bearing capacity reserve, stiffness change and structural dynamic performance of beams and columns are calculated, and the actual load-bearing capacity and safety margin of the structure under different loads and service conditions are fully quantified. Based on the above indicators, long-term damage accumulation and structural stability analysis are conducted. By continuously monitoring the changes in natural frequency, displacement and strain evolution trends, accumulated damage, stiffness degradation or instability risks can be identified, and it can be determined whether the beams and columns have entered a stable service state, thereby obtaining the evaluation results. This continuous and dynamic evaluation mechanism can capture performance degradation that is not visible in the short term and provide a basis for early warning. The evaluation results are compared with the functional target parameters to complete the functional target compliance determination; By determining whether beam and column alterations have achieved preset goals such as increased load-bearing capacity, deflection control, seismic performance, or durability, the system can provide clear decision-making basis for engineering maintenance, performance optimization, and secondary reinforcement, forming a closed-loop functional goal-oriented evaluation system, thus possessing both innovativeness and practical engineering application value.

[0048] In step S6, the optimization suggestions include secondary grouting, adding carbon fiber reinforcement, or adjusting the construction plan.

[0049] After completing the functional goal-oriented continuous assessment, the diagnostic results are presented in a visual manner, outputting the structural safety status level after beam and column modifications, and displaying the evolution trends of bearing capacity, stiffness, and performance, enabling intuitive monitoring and dynamic tracking of the structural health status. When the diagnostic results fail to meet the preset functional goals or detect the risk of structural performance degradation, targeted reinforcement or maintenance optimization suggestions are automatically generated based on the digital twin model and historical data, including but not limited to secondary grouting, adding carbon fiber reinforcement, or adjusting the construction plan, providing a scientific basis for engineering decisions.

[0050] Furthermore, multimodal basic data, functional target parameters, on-site monitoring data, and diagnostic results are uniformly archived to form a full lifecycle data archive covering the entire renovation process and service life. This full lifecycle data archive not only records the evolution of structural performance but also serves as an important reference for future maintenance, secondary renovations, or design optimization, achieving a closed loop of knowledge accumulation and intelligent management, and providing long-term support for the safety assessment and continuous performance assurance of existing buildings.

[0051] A preferred embodiment of the present invention, a physically constraint-driven intelligent diagnostic system for existing building beams and columns, includes the following modules: The initialization module is used to construct multimodal basic data, including the original structural design data and beam and column demolition and modification design data of existing buildings, and to obtain the input functional target parameters encoded as target constraint vectors. The on-site monitoring data acquisition module is used to identify key stress areas of existing buildings based on the multimodal basic data, and to deploy a multimodal sensing network based on the key stress areas to collect on-site monitoring data, including strain parameters, vibration parameters, crack parameters and environmental parameters. The on-site monitoring data preprocessing module is used to perform real-time preprocessing on the on-site monitoring data through edge computing nodes, including at least data noise reduction, timestamp alignment (to ensure that different types of data have a consistent spatiotemporal correspondence under the same structural response event), and feature extraction (including strain change rate, structural main frequency change characteristics, crack width and its evolution index, environmental correction factor, etc.), so as to reduce the data transmission load and improve the reliability of subsequent analysis, obtain standardized feature vectors, and upload them to the cloud. In practice, the edge computing nodes first perform data quality control on the collected field monitoring data. They use a combination of time-domain filtering and frequency-domain analysis to identify and remove abnormal data caused by non-structural factors such as construction machinery vibration, accidental impacts, or environmental interference, thereby ensuring the authenticity and stability of the structural response data.

[0052] The structural real mechanics profile generation module is used to build and run a physical constraint-driven structural state cognitive intelligent agent in the cloud. The structural state cognitive intelligent agent dynamically infers and self-evolves to update the structural state of beams and columns after dismantling and modification by integrating physical prior knowledge with the feature vector, thereby generating a structural real mechanics profile. The intelligent diagnostic module is used to continuously evaluate and intelligently diagnose the effects of beam and column demolition and modification based on the functional target parameters and the actual mechanical profile of the structure, and output the functional target compliance judgment result. The results output module is used to output visualized diagnostic results and optimization suggestions based on the functional target compliance judgment results, and to form a full life cycle data archive covering the entire transformation process.

[0053] In the initialization module, the multimodal basic data includes the structural topology, geometric parameters, and initial mechanical parameters obtained by parsing the structural drawings of the CAD or BIM model. The structural drawings include the original structural design drawings of the existing building and the beam and column demolition and alteration design drawings. The drawing data includes at least the geometric dimensions, reinforcement form and material strength grade of the beams and columns before the renovation, the cross-sectional form, material parameters and connection structure of the newly added or reinforced components in the renovation plan, and detailed construction drawings of beam and column joints and interfaces between new and old components. The functional target parameters include at least one of the following: bearing capacity improvement ratio, maximum allowable displacement, and seismic performance requirements. The functional target parameters also include target bearing capacity level, deflection control index, appraisal report conclusion parameters, ductility improvement requirements, and performance stability or durability indicators during service. These functional target parameters are encoded into system-recognizable target constraint vectors, which serve as important target modal data in subsequent model calculations, threshold setting, and evaluation result determination.

[0054] In the field monitoring data acquisition module, the multimodal sensing network includes at least three of the following sensors: fiber optic strain sensor, vibrating wire stress gauge, triaxial MEMS accelerometer, biaxial tilt sensor, industrial-grade vision sensor, and temperature and humidity sensor. The multimodal sensing network adopts a dynamic trigger sampling mechanism, using a high sampling frequency during key construction processes and a low sampling frequency or threshold-triggered sampling during long-term service monitoring.

[0055] In practice, a unified data structure model is used to model the multimodal basic data, functional target parameters, and on-site monitoring data, and to establish indexes and relationships between the multimodal data, thereby providing a complete and unified data foundation for perception, calculation, and intelligent diagnosis during the beam and column demolition and modification process.

[0056] After completing the construction of multimodal basic data, the system integrates existing building renovation construction drawings, relies on the building knowledge base and fact set, and accurately identifies key stress areas in beam and column demolition and modification through inference engine. A multimodal perception network is then deployed to acquire the actual mechanical response of the structure throughout its service life. Based on design drawings and initial mechanical simulation results, the system automatically identifies the mid-span tension zone of beams, the shear zone at supports, the compression zones at the four corners of reinforced columns, the interface between new and old components, and easily tilted parts of the overall structure as key monitoring points. This enables effective perception of internal force redistribution, interface co-stress, and spatial positional changes.

[0057] Specifically, based on the imported structural design drawings and initial mechanical analysis results, key monitoring locations are identified, including the mid-span tension zone of the beam, the support shear zone, the four corner compression zones of the reinforced columns, and the interfaces between new and old components. This aims to ensure effective perception of the redistribution of internal forces and the collaborative stress state of interfaces. A multimodal sensing network is deployed at these locations, with the following specific configuration: ① Fiber optic strain sensor and vibrating wire stress gauge: Deployed together to obtain high-precision strain response and internal stress state; ② Triaxial MEMS accelerometer: used to collect the natural frequency and dynamic response of the structure; ③ Dual-axis tilt sensor: used to monitor the tilt angle and deformation trend of components or the overall structure in real time; ④ Industrial-grade vision sensor: Periodically captures images of beam and column surfaces, automatically identifies and quantifies crack characteristics; ⑤ Temperature and humidity sensor: Synchronously collects environmental disturbance factors to correct the data.

[0058] Various sensors perform synchronous sampling based on a unified time reference. The sampling frequency adopts a dynamic triggering mechanism: during critical processes such as construction unloading and reloading, the sampling frequency is increased to 50Hz-100Hz to capture instantaneous responses; during long-term service monitoring, the static sensor frequency is adjusted to once per hour (or sampling is triggered according to a threshold), thereby achieving continuous and reliable monitoring of the structure throughout its entire life cycle and providing real and complete data support for subsequent intelligent analysis and safety diagnosis.

[0059] In the structural real-mechanical profile generation module, the constructed and run physical constraint-driven structural state cognitive agent includes: A cognitive kernel is constructed for the structural state cognitive agent. The cognitive kernel has a built-in physical prior knowledge base. The physical prior knowledge base stores the basic constraint rules of structural mechanics (including nonlinear constitutive relations of materials, structural force equilibrium conditions, displacement compatibility relations and energy conservation criteria) as well as the physical prior knowledge of nonlinear mechanical constraint operators (bond-slip, interface shear degradation, etc.) for the connection interface between new and old components (the interface between new and old concrete, the anchoring zone of rebar and the connection interface of reinforced components in the demolition and renovation of existing buildings). The cognitive kernel does not rely on the discretization and step-by-step iterative solution of the stiffness matrix in traditional finite element analysis. Instead, it achieves rapid cognition and global inference of the service state of the structure by integrating physical prior constraints and data-driven reasoning. A neural operator inference engine is employed as the core of structural response calculation for the structural state cognitive agent. It learns the operator mapping relationship from structural parameters to structural response, replacing the traditional explicit finite element method (FEM) solution. Based on a network architecture of Fourier neural operators, the engine learns the operator mapping relationship between "structural geometry—load—material parameters" and "stress—displacement—damage distribution," transforming the traditionally explicitly solved partial differential equation problem into a forward inference process in operator space. Compared to conventional FEM analysis, this method is independent of mesh refinement, possesses resolution independence, and can quickly output the overall structural response results even when the structural topology undergoes local changes, thus significantly improving the real-time performance and continuity of structural risk assessment during the demolition and alteration construction phase. In the reasoning process of the structural state cognitive agent, the node force balance residual and energy conservation deviation are embedded as constraints in the loss function to ensure the physical consistency of the reasoning results. To ensure the physical consistency of the reasoning results of the structural state cognitive agent, a physical constraint embedding mechanism is introduced in the reasoning process, embedding physical quantities such as node force balance residual and energy conservation deviation as constraints in the loss function of the structural state cognitive agent. In this way, the structural state cognitive agent can still output reliable results that conform to the basic laws of structural mechanics even when multimodal data is incomplete or there is noise interference, fundamentally avoiding the "black box inference" and physical distortion problems that may occur in pure data-driven models. The feature vector is dynamically compared with the predicted state of the structural state cognitive agent. When the deviation exceeds a preset threshold, an adaptive update mechanism based on physical constraints is triggered to correct the key state variables that reflect the evolution of structural performance, and a true mechanical profile of the structure is generated based on the corrected key state variables. The adaptive update mechanism employs reinforcement learning or adversarial optimization strategies to learn the evolution of structural internal force redistribution, damage accumulation, and cooperative force relationships during the dismantling and modification process, while satisfying physical constraints.

[0060] During operation, the structural state cognitive agent continuously receives feature vectors uploaded by edge computing nodes and dynamically compares them with the internal physical reasoning results (predicted state) as external perception input. When a significant deviation occurs between the monitored feature vector and the agent's current predicted state, the agent will automatically determine that the structural state has evolved and trigger an adaptive update mechanism.

[0061] The adaptive update mechanism is not simply a numerical correction of model parameters, but rather an autonomous update of key state variables reflecting the evolution of structural performance under physical constraints. These variables include the degradation parameter of elastic modulus, the equivalent shear stiffness of the interface between new and old components, and connection constraints. By introducing reinforcement learning or adversarial optimization strategies, it can gradually learn the evolution of internal force redistribution, damage accumulation, and cooperative force relationships during the demolition and modification process, while satisfying physical constraints. This allows the description of the internal structural state to continuously approximate the actual structural behavior as the actual engineering process progresses.

[0062] After completing the self-evolution of the structural state, the structural state cognitive agent uses variational Bayesian inference to probabilistically represent the current structural state, forming a "real mechanical portrait of the structure" that includes descriptions of uncertainty.

[0063] Through the above mechanism, the structural state cognitive intelligent agent constructed by this invention realizes a closed-loop operation mode from "multimodal monitoring - state cognition - self-evolutionary reasoning - decision feedback", so that the structural safety status of existing building beams and columns during the demolition and modification process and service stage is always within a perceptible, explainable and predictable controllable range.

[0064] In the intelligent diagnostic module, the continuous assessment and intelligent diagnosis specifically include: By combining the revised digital twin model with on-site monitoring data, the load-bearing capacity reserve, stiffness change and structural dynamic performance of beams and columns are calculated, and the actual load-bearing capacity and safety margin of the structure under different loads and service conditions are fully quantified. Based on the above indicators, long-term damage accumulation and structural stability analysis are conducted. By continuously monitoring the changes in natural frequency, displacement and strain evolution trends, accumulated damage, stiffness degradation or instability risks can be identified, and it can be determined whether the beams and columns have entered a stable service state, thereby obtaining the evaluation results. This continuous and dynamic evaluation mechanism can capture performance degradation that is not visible in the short term and provide a basis for early warning. The evaluation results are compared with the functional target parameters to complete the functional target compliance determination; By determining whether beam and column alterations have achieved preset goals such as increased load-bearing capacity, deflection control, seismic performance, or durability, the system can provide clear decision-making basis for engineering maintenance, performance optimization, and secondary reinforcement, forming a closed-loop functional goal-oriented evaluation system, thus possessing both innovativeness and practical engineering application value.

[0065] In the result output module, the optimization suggestions include secondary grouting, adding carbon fiber reinforcement, or adjusting the construction plan.

[0066] After completing the functional goal-oriented continuous assessment, the diagnostic results are presented in a visual manner, outputting the structural safety status level after beam and column modifications, and displaying the evolution trends of bearing capacity, stiffness, and performance, enabling intuitive monitoring and dynamic tracking of the structural health status. When the diagnostic results fail to meet the preset functional goals or detect the risk of structural performance degradation, targeted reinforcement or maintenance optimization suggestions are automatically generated based on the digital twin model and historical data, including but not limited to secondary grouting, adding carbon fiber reinforcement, or adjusting the construction plan, providing a scientific basis for engineering decisions.

[0067] Furthermore, multimodal basic data, functional target parameters, on-site monitoring data, and diagnostic results are uniformly archived to form a full lifecycle data archive covering the entire renovation process and service life. This full lifecycle data archive not only records the evolution of structural performance but also serves as an important reference for future maintenance, secondary renovations, or design optimization, achieving a closed loop of knowledge accumulation and intelligent management, and providing long-term support for the safety assessment and continuous performance assurance of existing buildings.

[0068] In summary, the advantages of this invention are: 1. By constructing multimodal foundational data including the original structural design data and beam-column alteration design data of existing buildings, functional target parameters encoded as target constraint vectors are obtained as input. Then, based on the multimodal foundational data, key stress areas of the existing buildings are identified, and a multimodal sensing network is deployed based on these key stress areas to collect on-site monitoring data. Next, the on-site monitoring data is preprocessed in real time using edge computing nodes to obtain standardized feature vectors, which are then uploaded to the cloud. The cloud uses a structural state cognitive agent to fuse prior physical knowledge with feature vectors, dynamically inferring and self-evolving the structural state of the altered beams and columns, generating a realistic structural mechanical profile. Based on the functional target parameters and the realistic structural mechanical profile, the alteration effect of beams and columns is continuously evaluated and intelligently diagnosed, outputting a functional target compliance judgment result. Finally, based on the functional target compliance judgment result, a visualized diagnostic result and optimization suggestions are output, forming a full lifecycle data archive covering the entire renovation process. This involves deploying a multimodal sensing network and leveraging edge computing... The system performs real-time data processing, enabling continuous and high-frequency acquisition of key parameters such as beam and column strain, vibration, and cracks during the demolition and alteration process, thereby enhancing the transparency of the entire process. Based on this, a cloud-based, physically constrained, structure-state cognitive intelligent agent integrates real-time data with built-in physical prior knowledge and uses neural operators for reasoning. When the deviation between prediction and measurement exceeds limits, it triggers a self-evolutionary update mechanism, dynamically correcting key state variables of the model. This allows the computational model to continuously evolve with the actual structural state, achieving a fundamental shift from static rigidity to dynamic correction. Finally, the system continuously assesses the conformity between the real-time updated structural mechanical profile and pre-set functional target parameters, outputting quantitative evaluation results and optimization suggestions. This forms a closed-loop intelligent evaluation system from perception and cognition to decision-making, overcoming the limitations of traditional methods that are lagging and disconnected. Ultimately, this significantly improves the transparency of the entire beam and column demolition and alteration process, the dynamic correction of the computational model, and the closed-loop intelligent evaluation capabilities of structural performance.

[0069] 2. By integrating the original structural design data of existing buildings, beam and column demolition and modification design data, and on-site monitoring data (such as strain, vibration, cracks, and environmental parameters), a multimodal basic data system was constructed. This multi-source data fusion method overcomes the limitations of traditional diagnostic methods that rely on a single data source, and can more comprehensively capture changes in structural state. In particular, during the demolition and modification process, by analyzing CAD or BIM models to obtain structural topological relationships, geometric parameters, etc., the accuracy and reliability of diagnosis are enhanced, thus providing a solid data foundation for subsequent intelligent diagnosis.

[0070] 3. Edge computing nodes are used to perform real-time preprocessing of on-site monitoring data, including data noise reduction, timestamp alignment, and feature extraction. After generating standardized feature vectors, the data is uploaded to the cloud. This design significantly reduces data transmission latency and cloud computing burden, achieving near real-time data processing capabilities. It is particularly suitable for dynamic monitoring scenarios in the renovation of existing buildings, avoiding the diagnostic lag caused by data transmission bottlenecks in traditional methods, and improving the overall system response speed and efficiency.

[0071] 4. By constructing a physical constraint-driven structural state cognitive agent in the cloud, physical prior knowledge (such as basic constraint rules of structural mechanics and constraint operators of nonlinear mechanics) is fused with feature vectors to perform dynamic reasoning and self-evolutionary updates. This physical constraint embedding mechanism ensures the physical rationality of the reasoning results and avoids non-physical errors that may be generated by pure data-driven methods. At the same time, the neural operator reasoning engine is used to replace the traditional finite element solution, which improves the computational efficiency and makes the diagnostic process more intelligent and adaptable.

[0072] 5. The structural state cognitive agent has an adaptive update mechanism. When the deviation between the monitored data and the predicted state exceeds a threshold, it triggers a correction strategy based on reinforcement learning or adversarial optimization to update key state variables. This feature enables the system to continuously learn the evolution laws of structural internal force redistribution and damage accumulation, enhancing the robustness and long-term stability of diagnosis. It is particularly suitable for monitoring the entire life cycle of existing building renovations and can promptly identify accumulated damage or instability risks.

[0073] 6. By covering the entire process from data collection to diagnostic output and forming a full lifecycle data archive, including visualized diagnostic results and optimization suggestions, this systematic management approach supports long-term tracking and retrospection of renovation projects, improves data traceability and management efficiency, provides decision-makers with a complete chain of evidence, helps optimize subsequent maintenance strategies, and is in line with the development trend of modern building information management.

[0074] 7. Based on functional target parameters and the actual mechanical profile of the structure, the system automatically performs continuous evaluation and intelligent diagnosis, outputting the functional target compliance judgment results and optimization suggestions (such as secondary grouting or carbon fiber reinforcement). This automated process reduces manual intervention, improves diagnostic efficiency and accuracy, and lowers the professional threshold through visual output, enabling non-professionals to quickly understand the results, thereby enhancing the practicality and promotional value of the technology.

[0075] 8. The multimodal sensing network adopts a dynamic trigger sampling mechanism, using a high sampling frequency during critical construction processes and a low sampling or threshold-triggered sampling during long-term service. This intelligent scheduling method optimizes the use of sensor resources and reduces energy consumption and data storage costs. This not only improves economic efficiency but also conforms to the concept of green building. By reducing unnecessary monitoring activities, it achieves energy-saving and environmental protection effects.

[0076] 9. By constructing an intelligent assessment system that integrates "end-cloud-model" collaboration, continuous monitoring and intelligent diagnosis of the entire process of demolition and renovation of existing building beams and columns, as well as their service life, have been achieved, significantly improving the accuracy and reliability of the renovation quality assessment. A multimodal sensing network is used to collect structural strain, vibration, cracks, and environmental parameters. Combined with design drawings and functional target data, real-time preprocessing and feature extraction are performed through edge computing. This forms a dual-drive closed loop with the cloud-based structural state cognitive intelligent agent, enabling dynamic adaptive correction of model parameters and continuous evolution of the digital twin. This allows the diagnostic results to closely approximate the actual structural state, overcoming the inaccuracy of traditional static model assessments.

[0077] 10. Guided by functional objectives, the system quantitatively assesses the load-bearing capacity, stiffness evolution, and long-term performance degradation of beams and columns, and provides scientific decision-making basis for engineering maintenance, performance optimization, and secondary reinforcement by visually outputting diagnostic results and optimization suggestions. At the same time, it forms a full life-cycle data archive, realizes the accumulation of structural performance knowledge and intelligent management closed loop, and provides reliable data support for the long-term safety assurance, continuous performance evaluation, and future renovation of existing buildings, which has significant technological innovation and engineering practical value.

[0078] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A physically constrained intelligent diagnostic method for existing building beams and columns, characterized in that: Includes the following steps: Step S1: Construct multimodal basic data including the original structural design data of existing buildings and the beam and column demolition and modification design data, and obtain the input functional target parameters encoded as target constraint vectors; Step S2: Identify the key stress areas of the existing building based on the multimodal basic data, and deploy a multimodal sensing network based on the key stress areas to collect on-site monitoring data. The on-site monitoring data includes strain parameters, vibration parameters, crack parameters, and environmental parameters. Step S3: Perform real-time preprocessing on the field monitoring data through edge computing nodes, including at least data noise reduction, timestamp alignment, and feature extraction, to obtain standardized feature vectors and upload them to the cloud; Step S4: Construct and run a physical constraint-driven structural state cognitive agent in the cloud. The structural state cognitive agent dynamically infers and self-evolves to update the structural state of the beams and columns after dismantling and modification by integrating prior physical knowledge with the feature vector, thereby generating a true mechanical profile of the structure. Step S5: Based on the functional target parameters and the actual mechanical profile of the structure, continuously evaluate and intelligently diagnose the beam and column demolition and modification effects, and output the functional target compliance judgment results. Step S6: Based on the functional objective compliance determination results, output visualized diagnostic results and optimization suggestions, and form a full lifecycle data archive covering the entire transformation process.

2. The intelligent diagnostic method for existing building beams and columns driven by physical constraints as described in claim 1, characterized in that: In step S1, the multimodal basic data includes the structural topology, geometric parameters, and initial mechanical parameters obtained by parsing the structural drawings of the CAD or BIM model. The functional target parameters include at least one of the following: bearing capacity improvement ratio, maximum allowable displacement, and seismic performance requirements.

3. The intelligent diagnostic method for existing building beams and columns driven by physical constraints as described in claim 1, characterized in that: In step S2, the multimodal sensing network includes at least three of the following sensors: fiber optic strain sensor, vibrating wire stress gauge, triaxial MEMS accelerometer, biaxial tilt sensor, industrial-grade vision sensor, and temperature and humidity sensor. The multimodal sensing network adopts a dynamic trigger sampling mechanism, using a high sampling frequency during key construction processes and a low sampling frequency or threshold-triggered sampling during long-term service monitoring.

4. The intelligent diagnostic method for existing building beams and columns driven by physical constraints as described in claim 1, characterized in that: In step S4, constructing and running the physically constrained structure state cognitive agent includes: The cognitive kernel of the structure state cognitive agent is constructed. The cognitive kernel has a built-in physical prior knowledge base, which stores the physical prior knowledge of the basic constraint rules of structural mechanics and the nonlinear mechanical constraint operators for the connection interface between new and old components. A neural operator inference engine is used as the core of the structural response calculation of the structural state cognitive agent. It learns the operator mapping relationship from structural parameters to structural response to replace the traditional explicit solution of finite element method. In the reasoning process of the structure state cognitive agent, the node force balance residual and energy conservation deviation are embedded as constraints in the loss function to ensure the physical consistency of the reasoning results. The feature vector is dynamically compared with the predicted state of the structural state cognitive agent. When the deviation exceeds a preset threshold, an adaptive update mechanism based on physical constraints is triggered to correct the key state variables that reflect the evolution of structural performance, and a true mechanical profile of the structure is generated based on the corrected key state variables. The adaptive update mechanism employs reinforcement learning or adversarial optimization strategies to learn the evolution of structural internal force redistribution, damage accumulation, and cooperative force relationships during the dismantling and modification process, while satisfying physical constraints.

5. The intelligent diagnostic method for existing building beams and columns driven by physical constraints as described in claim 1, characterized in that: In step S5, the continuous assessment and intelligent diagnosis specifically include: Calculate the load-bearing capacity reserve, stiffness variation, and structural dynamic performance indicators of beams and columns; Based on the above indicators, long-term damage accumulation and structural stability analysis are conducted to identify accumulated damage, stiffness degradation or instability risks, and thus obtain the assessment results. The evaluation results are compared with the functional target parameters to complete the functional target compliance determination; In step S6, the optimization suggestions include secondary grouting, adding carbon fiber reinforcement, or adjusting the construction plan.

6. A physically constrained intelligent diagnostic system for existing building beams and columns, characterized in that: Includes the following modules: The initialization module is used to construct multimodal basic data, including the original structural design data and beam and column demolition and modification design data of existing buildings, and to obtain the input functional target parameters encoded as target constraint vectors. The on-site monitoring data acquisition module is used to identify key stress areas of existing buildings based on the multimodal basic data, and to deploy a multimodal sensing network based on the key stress areas to collect on-site monitoring data, including strain parameters, vibration parameters, crack parameters and environmental parameters. The on-site monitoring data preprocessing module is used to perform real-time preprocessing on the on-site monitoring data through edge computing nodes, including at least data noise reduction, timestamp alignment and feature extraction, to obtain standardized feature vectors and upload them to the cloud. The structural real mechanics profile generation module is used to build and run a physical constraint-driven structural state cognitive agent in the cloud. The structural state cognitive agent dynamically infers and self-evolves to update the structural state of beams and columns after dismantling and modification by integrating physical prior knowledge with the feature vector, thereby generating a structural real mechanics profile. The intelligent diagnostic module is used to continuously evaluate and intelligently diagnose the effects of beam and column demolition and modification based on the functional target parameters and the actual mechanical profile of the structure, and output the functional target compliance judgment result. The results output module is used to output visualized diagnostic results and optimization suggestions based on the functional target compliance judgment results, and to form a full life cycle data archive covering the entire transformation process.

7. The intelligent diagnostic system for existing building beams and columns driven by physical constraints as described in claim 6, characterized in that: In the initialization module, the multimodal basic data includes the structural topology, geometric parameters, and initial mechanical parameters obtained by parsing the structural drawings of the CAD or BIM model. The functional target parameters include at least one of the following: bearing capacity improvement ratio, maximum allowable displacement, and seismic performance requirements.

8. The intelligent diagnostic system for existing building beams and columns driven by physical constraints as described in claim 6, characterized in that: In the field monitoring data acquisition module, the multimodal sensing network includes at least three of the following sensors: fiber optic strain sensor, vibrating wire stress gauge, triaxial MEMS accelerometer, biaxial tilt sensor, industrial-grade vision sensor, and temperature and humidity sensor. The multimodal sensing network adopts a dynamic trigger sampling mechanism, using a high sampling frequency during key construction processes and a low sampling frequency or threshold-triggered sampling during long-term service monitoring.

9. The intelligent diagnostic system for existing building beams and columns driven by physical constraints as described in claim 6, characterized in that: In the structural real-mechanical profile generation module, the constructed and run physical constraint-driven structural state cognitive agent includes: The cognitive kernel of the structure state cognitive agent is constructed. The cognitive kernel has a built-in physical prior knowledge base, which stores the physical prior knowledge of the basic constraint rules of structural mechanics and the nonlinear mechanical constraint operators for the connection interface between new and old components. A neural operator inference engine is used as the core of the structural response calculation of the structural state cognitive agent. It learns the operator mapping relationship from structural parameters to structural response to replace the traditional explicit solution of finite element method. In the reasoning process of the structure state cognitive agent, the node force balance residual and energy conservation deviation are embedded as constraints in the loss function to ensure the physical consistency of the reasoning results. The feature vector is dynamically compared with the predicted state of the structural state cognitive agent. When the deviation exceeds a preset threshold, an adaptive update mechanism based on physical constraints is triggered to correct the key state variables that reflect the evolution of structural performance, and a true mechanical profile of the structure is generated based on the corrected key state variables. The adaptive update mechanism employs reinforcement learning or adversarial optimization strategies to learn the evolution of structural internal force redistribution, damage accumulation, and cooperative force relationships during the dismantling and modification process, while satisfying physical constraints.

10. The intelligent diagnostic system for existing building beams and columns driven by physical constraints as described in claim 6, characterized in that: In the intelligent diagnostic module, the continuous assessment and intelligent diagnosis specifically include: Calculate the load-bearing capacity reserve, stiffness variation, and structural dynamic performance indicators of beams and columns; Based on the above indicators, long-term damage accumulation and structural stability analysis are conducted to identify accumulated damage, stiffness degradation or instability risks, and thus obtain the assessment results. The evaluation results are compared with the functional target parameters to complete the functional target compliance determination; In the result output module, the optimization suggestions include secondary grouting, adding carbon fiber reinforcement, or adjusting the construction plan.

Images