Historical building safety performance evaluation method and system

Abstract

The invention relates to the technical field of historical building structure health monitoring, and discloses a historical building safety performance evaluation method and system. The method comprises the steps of obtaining a structure response signal through monitoring of a multi-position sensor, and generating a fusion feature map through a multi-scale feature sensing network. And performing three-dimensional reconstruction based on the atlas, adaptively dividing difference precision entity units according to feature density, and constructing the high-fidelity digital twinborn body. And driving the twin to perform time-history evolution under simulation of long-term environment excitation, and extracting damage evolution behavior fingerprints. And performing multi-round iterative comparison and reverse traceability analysis on the behavior fingerprint and a safety baseline containing material time-varying degradation and component interaction influence, identifying a failure mode and an evolution path, calculating a residual bearing capacity margin of a key component, and generating a structural life map in combination with a component interaction network. According to the method, high-fidelity modeling based on monitoring data and reverse intelligent diagnosis of dynamic damage evolution are realized.

Description

Technical Field

[0001] This invention relates to the field of structural health monitoring technology for historical buildings, specifically a method and system for assessing the safety performance of historical buildings. Background Technology

[0002] Current safety performance assessments of historical buildings primarily rely on periodic manual inspections, empirical judgments, and finite element static or dynamic analyses based on simplified assumptions. Manual inspections are inherently lagging and subjective, making it difficult to capture the initiation and dynamic evolution of damage within the structure. Traditional finite element analysis methods typically use models built from design drawings or simple inspection data, and the model meshes are often homogenized or locally refined based on fixed empirical rules, failing to accurately reflect the spatial heterogeneity of the actual mechanical properties of historical buildings caused by material aging and uneven damage distribution. This results in a significant discrepancy between the calculated model and the actual response of the physical structure, limiting the fidelity of the assessment results.

[0003] Some existing technologies attempt to combine sensor monitoring with digital models, but model construction is often disconnected from real-time monitoring data, or uses fixed levels of granularity, leading to wasted computational resources or insufficient analysis of critical components. At the safety diagnostic level, existing methods largely rely on direct comparisons between monitoring data and preset thresholds, or on single-condition simulations based on intact state models. These methods can only determine "whether it is abnormal," but cannot explain "why it is abnormal," and are even less capable of tracing the origin and evolution mechanism of potential microscopic damage from macroscopic phenomena. They lack the predictability and source tracing capabilities for the dynamic development paths of complex failure modes. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for assessing the safety performance of historical buildings, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for assessing the safety performance of historical buildings, the method comprising:

[0006] Sensor arrays were deployed at multiple pre-selected locations on the historical building to collect raw monitoring data on vibration, deformation, and the environment.

[0007] The original monitoring stream is synchronously calibrated and anomaly cleaned to generate a structural response signal with a unified spatiotemporal reference.

[0008] The structural response signal is input into a pre-trained multi-scale feature perception network to obtain a fused feature map.

[0009] Based on the fused feature map, the historical building is digitally reconstructed in three dimensions. Based on the feature distribution density, the reconstruction model is adaptively divided into sets of entity units with different discrete precisions to form a high-fidelity digital twin structure.

[0010] The high-fidelity digital twin of the structure is driven to perform time-history evolution simulation under simulated long-term environmental excitation, and the response history of each entity unit set during the evolution process is recorded. Behavioral fingerprints that can characterize the evolution law of structural damage are extracted from the response history.

[0011] Establish a safety baseline that includes a material time-varying degradation model and a component interaction model;

[0012] The behavioral fingerprint is compared with the security baseline through multiple rounds of iterative comparison and reverse tracing analysis to identify potential failure modes and their evolution paths;

[0013] Based on the evolution path, calculate the residual bearing capacity margin of each key component of the historical building under the set service objectives;

[0014] By combining the residual load-bearing capacity margin of each key component and the interaction network between components, a structural life map reflecting the overall structural safety status is generated.

[0015] Preferably, the step of synchronously calibrating and cleaning up anomalies in the original monitoring stream to generate a structural response signal with a unified spatiotemporal reference includes: receiving the original monitoring stream from the sensor array; performing timestamp alignment processing on the sampling times of different sensors in the original monitoring stream to eliminate deviations caused by transmission delays and generate a time synchronization signal; filtering the time synchronization signal to remove signal singularities caused by instantaneous environmental interference or equipment failures and generate a preliminary cleaning signal; comparing the preliminary cleaning signal with the known structural dynamic characteristics of the historical building to correct the signal amplitude and phase shifts caused by sensor installation errors, and finally generating the structural response signal with a unified spatiotemporal reference.

[0016] Preferably, the step of inputting the structural response signal into a pre-trained multi-scale feature perception network to obtain a fused feature map includes:

[0017] The multi-scale feature perception network uses hierarchical convolution and pooling operations to parse and obtain a fused feature map containing local damage features and global modal features.

[0018] The multi-scale feature perception network is constructed, comprising a shallow feature extraction branch and a deep feature extraction branch connected sequentially. The structural response signal is simultaneously input to both the shallow and deep feature extraction branches. The shallow feature extraction branch extracts local subtle fluctuation features of the structural response using small-scale convolutional kernels, while the deep feature extraction branch extracts the overall macroscopic modal features of the structural response using large-scale convolutional kernels and downsampling. The local subtle fluctuation features output by the shallow feature extraction branch and the overall macroscopic modal features output by the deep feature extraction branch are concatenated and fused along the channel dimension to generate the fused feature map.

[0019] Preferably, the step of performing digital 3D reconstruction of the historical building based on the fused feature map, and adaptively dividing the reconstructed model into sets of entity units with different discretization accuracies according to feature distribution density to form a high-fidelity structural digital twin, includes: calling the original design drawings and point cloud scan data of the historical building, combining them with the current structural morphology features represented in the fused feature map, and reconstructing a 3D geometric model to obtain an initial geometric model; superimposing the fused feature map on the surface of the initial geometric model, calculating the gradient magnitude of the fused feature map of each region on the model surface, and marking regions with gradient magnitudes exceeding a preset threshold as feature-rich regions; discretizing the marked feature-rich regions using a dense mesh and discretizing the non-feature-rich regions using a sparse mesh to generate the sets of entity units with different discretization accuracies; assigning each entity unit in the set of entity units with material properties and initial state parameters mapped from the fused feature map, thus completing the construction of the high-fidelity structural digital twin.

[0020] Preferably, the step of driving the high-fidelity structural digital twin to perform time-history evolution simulation under simulated long-term environmental excitation includes: constructing a long-term environmental excitation sample library covering wind load, temperature load, seismic motion, and crowd live load; extracting excitation sequences from the long-term environmental excitation sample library and applying them to the high-fidelity structural digital twin in chronological order; solving the dynamic equilibrium equations of the high-fidelity structural digital twin at each loading step, updating the stress, strain, and displacement states of each entity element, and recording the state update data; continuously solving the loading steps until the simulation of the entire long-term environmental excitation sequence is completed, thereby obtaining the response history.

[0021] Preferably, the step of extracting a behavioral fingerprint that can characterize the evolution of structural damage from the response history includes: performing time-frequency analysis on the response history to extract the natural frequency change trajectory and damping ratio change trajectory of each entity unit set under different excitation stages; analyzing the stress amplitude spectrum and cycle number of the stress concentration region in the response history; extracting a feature vector sequence that evolves with loading time from the natural frequency change trajectory, damping ratio change trajectory, stress amplitude spectrum and cycle number, and defining the feature vector sequence as the behavioral fingerprint.

[0022] Preferably, the establishment of a safety baseline comprising a material time-varying degradation model and a component interaction model includes: obtaining performance degradation data of the main building materials of the historical building under different environmental stresses through accelerated aging tests in the laboratory and on-site sampling and testing; fitting the material time-varying degradation model based on the performance degradation data to describe the changes in material strength and elastic modulus with service time and external environment; establishing the component interaction model reflecting the relationship between internal force redistribution and damage transmission among components by analyzing the internal force transmission path of the high-fidelity structural digital twin during the simulation process; and coupling the material time-varying degradation model and the component interaction model to form the safety baseline.

[0023] Preferably, the step of performing multiple rounds of iterative comparison and reverse tracing analysis between the behavioral fingerprint and the safety baseline to identify potential failure modes and their evolution paths includes: inputting the behavioral fingerprint into the safety baseline to positively predict the structural state evolution curve under the constraints of the safety baseline; comparing the positively predicted structural state evolution curve with the actual evolution trend observed from the behavioral fingerprint to calculate the residual sequence; based on the residual sequence, reversely adjusting the key parameters in the safety baseline and performing a new round of positive prediction and comparison, iterating through multiple rounds until the residual converges; in the converged state, analyzing the activated failure criteria and key transmission paths in the safety baseline and identifying them as the potential failure modes and their evolution paths.

[0024] Preferably, the step of generating a structural life map reflecting the overall structural safety status by integrating the residual bearing capacity margin of each key component and the interaction network between components includes: calculating, based on the evolution path, the maximum additional load allowed for each key component not to enter the potential failure mode within a set service target life, wherein the ratio of the maximum additional load to the current load is defined as the residual bearing capacity margin; constructing an interaction network with key components as nodes and the internal force transmission relationship between components as edges; in the interaction network, using the residual bearing capacity margin as the node weight, simulating the transmission effect of local node weight changes in the entire network, and evaluating the robustness of the overall structure; and integrating the residual bearing capacity margin of each node with the robustness evaluation results of the network to generate the structural life map visualized on a two-dimensional or three-dimensional structural model.

[0025] Preferably, the present invention also includes a historical building safety performance assessment system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the historical building safety performance assessment method described above.

[0026] Compared with the prior art, the beneficial effects of the present invention are:

[0027] Based on the fused feature map generated by the multi-scale feature-aware network, the digital 3D reconstruction model is adaptively divided into sets of entity units with different discretization precisions according to the feature distribution density. Feature-dense regions represent areas with complex structural responses or concentrated potential damage, and are automatically discretized using high-precision units, while feature-sparse regions are discretized using lower-precision units. This data-driven adaptive discretization strategy enables the final high-fidelity structural digital twin to precisely focus limited computational resources on key structural parts and sensitive areas. Under the premise of controllable overall computational load, it improves the accuracy and analytical capability of the digital model in representing the real complex mechanical state of the structure, especially the evolution of local damage.

[0028] Behavioral fingerprints characterizing structural damage evolution are extracted from the time-history evolution simulation of a high-fidelity digital twin under long-term environmental excitation. These fingerprints are then subjected to multiple rounds of iterative comparison and reverse tracing analysis with a safety baseline that integrates the time-varying degradation of materials and the interaction between components. The behavioral fingerprints reflect the inherent patterns of the structural system's dynamic evolution. Through repeated iterative calibration and reverse simulation with theoretical and empirical safety baselines, the initial damage location, type, and propagation and amplification path within the component network can be traced back from observed anomalies or development trends. This process represents a leap from "state judgment" to "mechanism tracing," enabling early identification of potential failure modes and revealing their complete chain from inception to development, thus providing precise targets for intervention measures. Attached Figure Description

[0029] Figure 1 This is a schematic diagram illustrating the working principle of the historical building safety performance assessment method described in this invention.

[0030] Figure 2 A flowchart for generating structural response signals;

[0031] Figure 3 A flowchart for generating the fused feature map;

[0032] Figure 4 A statistical analysis chart of stress amplitude and cycle number in stress concentration areas of historical buildings;

[0033] Figure 5 This is a diagram showing the residual bearing capacity margin and load analysis of key components of a historical building. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Please see Figure 1 This invention provides a method for assessing the safety performance of historical buildings. The method includes: deploying sensor arrays at multiple predetermined locations on the historical building to collect raw monitoring data on vibration, deformation, and the environment; synchronously calibrating and cleaning anomalies from the raw monitoring data to generate a structural response signal with a unified spatiotemporal reference; inputting the structural response signal into a pre-trained multi-scale feature perception network to obtain a fused feature map; and performing digital 3D reconstruction of the historical building based on the fused feature map, adaptively dividing the reconstructed model into sets of entity units with different discrete precisions according to feature distribution density to form a high-fidelity structural digital twin.

[0036] The high-fidelity structural digital twin is driven to perform time-history evolution simulation under simulated long-term environmental excitation, recording the response history of each entity unit set during the evolution process. Behavioral fingerprints characterizing the structural damage evolution are extracted from the response history. A safety baseline is established, incorporating a material time-varying degradation model and a component interaction influence model. The behavioral fingerprints and the safety baseline are iteratively compared and reverse-engineered to identify potential failure modes and their evolution paths. Based on the evolution paths, the residual load-bearing capacity margin of each key component of the historical building is calculated under a set service target. By integrating the residual load-bearing capacity margin of each key component and the interaction network between components, a structural life map reflecting the overall structural safety status is generated.

[0037] Example 1: See Figure 2 The system receives the raw monitoring stream from the sensor array, timestamps the sampling times of different sensors in the raw monitoring stream to eliminate deviations caused by transmission delays, and generates a time synchronization signal. The time synchronization signal is then filtered to remove signal singularities caused by transient environmental interference or equipment malfunctions, generating a preliminary cleaning signal. This preliminary cleaning signal is compared with the known structural dynamic characteristics of the historical building to correct signal amplitude and phase shifts caused by sensor installation errors, ultimately generating the structural response signal with a unified spatiotemporal reference.

[0038] In practice, the system receives raw monitoring streams from sensor arrays deployed at key locations within the historical building. These streams include vibration signals collected by accelerometers, deformation signals collected by strain sensors, and environmental signals collected by temperature and humidity sensors. The sampling times of different sensors in the raw monitoring stream are timestamped. This process uses a unified time server to attach a globally coordinated time stamp (GCS) accurate to milliseconds to each data record, eliminating transmission delays caused by internal sensor clock drift and network transmission queues, thus generating a time synchronization signal. The timestamp alignment process employs a delay estimation method based on maximum cross-correlation calculation to calculate the relative delays between different sensor signals. Interpolation and resampling are then used to unify all signals to the same time reference, generating the time synchronization signal.

[0039] In practical implementation, the time synchronization signal is filtered to remove signal singularities caused by transient environmental interference or equipment failure. For example, a threshold denoising method based on wavelet transform is used to identify and remove transient pulse components with amplitudes exceeding a preset threshold, generating a preliminary cleaning signal. It can be understood that the filtering process can also employ an adaptive Kalman filter, dynamically updating the filtering parameters based on the historical statistical characteristics of the signal to smooth random noise. In practical implementation, the preliminary cleaning signal is compared with the known structural dynamic characteristics of the historical building. These known structural dynamic characteristics originate from the finite element modal analysis results of the building or the low-order natural frequencies and mode shapes identified in previous vibration tests. The correction process involves calculating the amplitude scaling factor and phase difference between the theoretical response at the sensor installation location and the actual preliminary cleaning signal. This scaling factor and phase difference are used to globally scale and rotate the preliminary cleaning signal, correcting the signal amplitude and phase shifts caused by sensor installation azimuth deviations or imperfect coupling, ultimately generating a structural response signal with a unified spatiotemporal reference. An example of a scaling factor used to calculate amplitude correction is as follows:

[0040]

[0041] Where: proportionality coefficient This represents the scaling factor for amplitude correction. This represents the response amplitude predicted by the theoretical model for a specific frequency component. This represents the measured response amplitude at the corresponding frequency component extracted from the initial cleaned signal. In some embodiments, the phase shift correction is achieved by calculating the cross-power spectrum phase angle between the theoretical response and the measured signal, and then compensating in the frequency domain. It can be understood that the signal after amplitude and phase correction is a structural response signal with a unified spatiotemporal reference that is strictly synchronized in the time dimension and aligned with the actual motion state of the structure in the spatial dimension.

[0042] Example 2: See Figure 3 The multi-scale feature perception network comprises a shallow feature extraction branch and a deep feature extraction branch connected sequentially. The structural response signal is simultaneously input to both the shallow and deep feature extraction branches. The shallow feature extraction branch extracts local subtle fluctuation features of the structural response using small-scale convolutional kernels, while the deep feature extraction branch extracts the overall macroscopic modal features of the structural response using large-scale convolutional kernels and downsampling. The local subtle fluctuation features output from the shallow feature extraction branch and the overall macroscopic modal features output from the deep feature extraction branch are concatenated and fused along the channel dimension to generate the fused feature map. The original design drawings and point cloud scan data of the historical building are used, combined with the current structural morphology features represented in the fused feature map, to reconstruct a three-dimensional geometric model, obtaining an initial geometric model. The fused feature map is superimposed on the surface of the initial geometric model, and the gradient magnitude of the fused feature map in each region of the model surface is calculated. Regions with gradient magnitudes exceeding a preset threshold are marked as feature-rich areas. The marked feature-rich areas are discretized using a dense grid, while non-feature-rich areas are discretized using a sparse grid, generating the set of entity units with different discretization accuracies. Each entity in the entity unit set is assigned material properties and initial state parameters mapped from the fused feature map, thereby completing the construction of the high-fidelity structural digital twin.

[0043] In practice, the structural response signal with a unified spatiotemporal reference is input into a pre-trained multi-scale feature perception network. This network comprises a shallow feature extraction branch and a deep feature extraction branch connected sequentially. The structural response signal with the unified spatiotemporal reference is simultaneously input into both branches for parallel processing. Specifically, the shallow feature extraction branch extracts local, subtle fluctuation features of the structural response using small-scale convolutional kernels. For example, a 3x3 convolutional kernel is used to convolve the structural response signal with the unified spatiotemporal reference layer by layer to capture high-frequency, localized minor fluctuation patterns caused by damage crack initiation or material delamination. The deep feature extraction branch extracts the overall macroscopic modal features of the structural response using large-scale convolutional kernels and downsampling. For example, a 7x7 convolutional kernel is used in conjunction with a pooling layer with a stride of 2 to progressively expand the receptive field and reduce the spatial resolution of the feature map, thereby capturing the low-order vibration modes and slow time-varying characteristics of the overall structure.

[0044] In practical implementation, the local subtle fluctuation features output from the shallow feature extraction branch and the overall macroscopic modal features output from the deep feature extraction branch are spliced ​​and fused along the channel dimension to generate a fused feature map. The fusion process first concatenates the two sets of feature maps along the channel dimension, then performs channel dimensionality reduction and feature recalibration through a 1x1 convolutional layer, finally outputting a fused feature map integrating local damage information and overall modal information. Feature fusion can also be achieved by element-wise addition or weighted averaging. In practical implementation, the original design drawings of historical buildings and point cloud scan data obtained through laser scanning are used, combined with the current structural morphology features represented in the fused feature map, to reconstruct a three-dimensional geometric model, obtaining an initial geometric model, which is represented as a triangular mesh surface. In practical implementation, the fused feature map is superimposed on the surface of the initial geometric model, and the gradient magnitude of the fused feature map in each region of the model surface is calculated. Regions with gradient magnitudes exceeding a preset threshold are marked as feature-rich areas. The gradient magnitude of the fused feature map can be calculated using the following formula:

[0045]

[0046] in: Indicates the gradient magnitude. This represents the scalar field of the fused feature map superimposed on the model surface. and These represent the spatial partial derivatives of the fused feature map along the x and y directions of the local coordinate system on the model surface, respectively. Optionally, the preset threshold can be set to the upper quartile of the overall gradient magnitude distribution. It can be understood that the feature-rich region typically corresponds to areas of drastic change in the fused feature map, such as damage edges, material property boundaries, or abrupt geometric changes.

[0047] In practical implementation, dense meshes are used to discretize the marked feature-rich regions, while sparse meshes are used to discretize the non-feature-rich regions, generating sets of solid units with different discretization accuracies. For example, hexahedral solid units with a side length of 10 cm are used for meshing in feature-rich regions, while hexahedral solid units with a side length of 50 cm are used for meshing in non-feature-rich regions. Optionally, the meshing process is implemented using an adaptive mesh subdivision algorithm, which uses the gradient magnitude field of the fused feature map. As the basis for mesh subdivision, the aim is to generate a dense mesh in feature-rich regions and maintain a sparse mesh in non-feature-rich regions. The execution of the adaptive mesh subdivision algorithm begins with a relatively sparse initial background mesh covering the entire initial geometry model. The cell size of the initial background mesh is set to a baseline value. In its implementation, the algorithm iteratively subdivides the initial background mesh and its derived meshes. Each iteration includes the following steps: calculating the gradient magnitude field within the geometric region corresponding to each cell in the current mesh. Statistics, such as those within a computing unit The maximum or average value. The statistics for each unit are compared with a preset segmentation threshold. Compare them. If the statistic of a unit exceeds the subdivision threshold... If a cell is identified as a cell to be subdivided, a mesh subdivision operation is performed on all identified cells. The mesh subdivision operation breaks down a large cell into multiple geometrically regular, smaller sub-cells. In some embodiments, for a hexahedral cell, a single subdivision operation may uniformly divide it into eight sub-hexahedral cells of half their original size.

[0048] Subdivision operations must ensure mesh node compatibility between newly generated sub-elements and adjacent unsubdivided elements. This is typically achieved by introducing transition layer elements or performing necessary secondary subdivisions on adjacent elements. After completing all subdivision operations in this round, the element and node information of the entire computational mesh is updated. Based on the new mesh, the gradient magnitude field is recalculated through interpolation. The values ​​in each new cell are determined, and a new round of iterative judgment and subdivision begins. The adaptive mesh subdivision algorithm terminates iteration when any of the following conditions are met: the gradient magnitude statistics of all cells are not higher than the subdivision threshold. Or the size of the grid cells has reached the preset minimum size limit. In practice, the final generated mesh is a set of solid cells with different discretization precisions. Regions that have undergone multiple subdivision iterations correspond to feature-rich areas, and their cell sizes are significantly smaller than the baseline value. Regions that maintain their initial size or undergo only a few subdivisions correspond to non-feature-rich regions.

[0049] Each entity element in the set of entity elements is assigned material properties and initial state parameters mapped from the fused feature map. The material properties include elastic modulus, Poisson's ratio, and material density, while the initial state parameters include initial strain and initial stress, thus completing the construction of a high-fidelity structural digital twin. A high-fidelity structural digital twin is a finite element computational model with realistic geometry, non-uniform material property distribution and initial state, and mesh accuracy matching the feature distribution.

[0050] Example 3: Constructing a long-term environmental excitation sample library covering wind load, temperature load, seismic motion, and crowd live load. Excitation sequences are extracted from the long-term environmental excitation sample library and applied to the high-fidelity structural digital twin in chronological order. At each loading step, the dynamic equilibrium equations of the high-fidelity structural digital twin are solved, the stress, strain, and displacement states of each entity element are updated, and the state update data is recorded. The loading steps are continuously solved until the simulation of the entire long-term environmental excitation sequence is completed, obtaining the response history. Time-frequency analysis is performed on the response history to extract the natural frequency variation trajectory and damping ratio variation trajectory of each entity element set under different excitation stages. The stress amplitude spectrum and cycle number in the stress concentration region of the response history are analyzed. From the natural frequency variation trajectory, damping ratio variation trajectory, stress amplitude spectrum, and cycle number, a feature vector sequence evolving with loading time is extracted, and this feature vector sequence is defined as the behavioral fingerprint.

[0051] In practice, a long-term environmental excitation sample library is constructed, encompassing wind load, temperature load, seismic motion, and crowd live load. This library is built based on historical meteorological data, seismic hazard analysis reports, and building usage surveys of the target historical building's site. A 50-year excitation sequence is extracted from this library, arranged chronologically. This sequence includes annual seasonal temperature cycles, randomly occurring wind load events, seismic motion time histories conforming to probabilistic seismic hazard models, and crowd live loads simulated based on building opening schedules. The extracted excitation sequence is then applied chronologically to a high-fidelity structural digital twin. In some embodiments, the excitation sequence is extracted using Monte Carlo simulation to generate statistically consistent random load time series.

[0052] In practice, at each loading step, the dynamic equilibrium equations of the high-fidelity structural digital twin are solved, the stress, strain, and displacement states of each solid element are updated, and the state update data is recorded. The dynamic equilibrium equations are solved using either the explicit central difference method or the implicit Newmark-β method for numerical integration. In practice, loading steps are performed continuously. The time step size of each loading step is determined based on the highest frequency component of the excitation sequence and the smallest solid element size of the high-fidelity structural digital twin to ensure numerical stability. The solution process continues until the simulation of the entire long-term environmental excitation sequence is completed, yielding the response history. The response history is a database containing stress, strain, displacement, velocity, and acceleration information for all solid elements at all time steps. It can be understood that the solution process is performed on a high-performance computing cluster to handle the massive computational demands arising from the potentially hundreds of thousands or even millions of solid elements contained in the high-fidelity structural digital twin.

[0053] In practical implementation, time-frequency analysis is performed on the response history to extract the natural frequency and damping ratio variation trajectories of each entity element set under different excitation stages. The time-frequency analysis employs a sliding time window Fast Fourier Transform or a random subspace identification method to identify structural modal parameters from the acceleration response data at various time intervals in the response history. For a specific entity element set, the variation of its first-order vertical bending frequency over a 50-year simulation period is recorded as a curve starting from the initial value... In the beginning, over time The trajectory of evolution In practical implementation, the stress amplitude spectrum and cycle number of stress concentration regions in the response history are analyzed. These stress concentration regions are determined based on the initial stress analysis before simulation, for example, a set of solid elements for a beam-column joint. The analysis process uses the rainflow counting method to process the von Mises stress time history of the solid elements in this region, statistically analyzing different stress amplitudes. The corresponding number of loops .

[0054] In practical implementation, a sequence of feature vectors evolving with loading time is extracted from the natural frequency variation trajectory, damping ratio variation trajectory, stress amplitude spectrum, and cycle number. The feature vector sequence is constructed by applying the sequence at equally spaced simulation time points. The above features are then summarized into a multi-dimensional vector. For example, at a time point... eigenvectors It can be represented as:

[0055]

[0056] Where: eigenvectors Indicates a point in time Extracted feature vectors, Indicates a point in time The first-order natural frequency was identified. Indicates a point in time The identified corresponding damping ratio, Indicates the deadline. The maximum stress amplitude obtained statistically from the stress concentration region. Indicates the deadline. The equivalent number of cycles is calculated using the Miner linear cumulative damage criterion. Optionally, the eigenvector can also include more modal frequencies or stress parameters at different locations. This indicates a transpose operation. It can be understood as applying the transpose to all time points. eigenvectors Arranged chronologically, these constitute a behavioral fingerprint characterizing the evolution of structural damage. The behavioral fingerprint, in the form of a time series, depicts the coordinated evolution of structural dynamic characteristics and local stress history.

[0057] See Figure 4 This is a statistical analysis chart of stress amplitude and cycle count in stress concentration areas of historical buildings. The highest number of cycles corresponds to moderate stress amplitudes (30-60 MPa), reflecting the most frequent stress fluctuations in this range during normal structural service. The equivalent cycle count increases non-linearly with increasing stress amplitude, indicating that a small number of cycles at high stress amplitudes contribute more to structural damage. This chart aids in assessing the degree of localized fatigue damage in structures: the equivalent cycle count quantifies the damage contribution of different stress ranges, providing data support for predicting the residual load-bearing capacity of components and identifying potential failure modes.

[0058] Example 4: In the specific implementation, a safety baseline was established, including a material time-varying degradation model and a component interaction model. Performance degradation data of the main building materials of the historical building under different environmental stresses were obtained through accelerated aging tests in the laboratory and on-site sampling and testing. The main building materials included blue bricks, fir wood, and lime mortar. The accelerated aging tests in the laboratory were conducted in an environmental test chamber, simulating wet-dry cycles, freeze-thaw cycles, and salt spray erosion conditions. On-site sampling and testing involved non-destructive or minimally destructive core sampling from non-load-bearing parts of the building to test their current mechanical properties. Performance degradation data recorded the changes in material strength and elastic modulus with equivalent accelerated aging time. Refer to Table 1, which shows a segment of performance degradation data for a blue brick material under accelerated salt spray erosion stress.

[0059] Table 1: Performance Degradation Data of Blue Brick Materials under Accelerated Salt Spray Erosion

[0060] Equivalent service years (years) Average compressive strength (MPa) Standard deviation of compressive strength (MPa) Average elastic modulus (GPa) 0 25.6 1.2 10.5 10 23.1 1.4 9.8 20 20.3 1.7 9.0 30 17.8 2.0 8.2

[0061] In practical implementation, a time-varying degradation model describing the changes in material strength and elastic modulus with service time and external environment is obtained based on performance degradation data. For example, an exponential decay model or a power function model is used to perform regression analysis on the data in the table. In some embodiments, the time-varying degradation model for the compressive strength of blue bricks can be expressed as:

[0062]

[0063] in: This indicates the equivalent service years of the blue bricks. Predicted compressive strength at that time This indicates the initial compressive strength of the blue brick. This represents the material degradation rate coefficient determined by the type and intensity of environmental stress. It can be understood that different building materials and different combinations of environmental stresses will correspond to different forms and parameters of time-varying material degradation models.

[0064] In practical implementation, a component interaction influence model reflecting the relationship between internal force redistribution and damage transmission among components is established by analyzing the internal force transmission paths of the high-fidelity structural digital twin during the simulation process. The analysis process is based on the stress flow distribution of the high-fidelity structural digital twin under simulated loads, identifying the force transmission paths and their relative contribution weights among the main load-bearing components. The component interaction influence model can be characterized as an influence degree matrix between components, with matrix elements... Quantified components Stiffness degradation or failure of components The proportion of change in the internal forces borne. Optionally, the analysis of the internal force transmission path can be achieved by calculating the internal force distribution of the structure under unit load or by performing modal strain energy analysis. In specific implementation, the material time-varying degradation model and the component interaction influence model are coupled to form a safety baseline. The coupling method is to substitute the material property parameters that evolve over time into a high-fidelity structural digital twin, and define the transmission rules of component state changes through the component interaction influence model. The safety baseline constitutes a predictive framework that can simulate the long-term performance evolution of the structure under the dual constraints of material property time-varying and component interaction.

[0065] In specific implementations, behavioral fingerprints are input into the safety baseline to positively predict the structural state evolution curve under the constraints of the safety baseline. For example, the feature vector sequence extracted from the behavioral fingerprint is used as initial conditions or intermediate observations to drive the safety baseline model to perform an evolution simulation for the same duration as in Example 3, starting from the initial state, predicting the changes in state variables such as modal frequencies and key stresses at future time points. In some embodiments, the positively predicted structural state evolution curve is compared with the actual evolution trend observed from the behavioral fingerprint, and a residual sequence is calculated. The residual sequence is the difference vector between the predicted value and the observed value retrieved from the behavioral fingerprint at each time point. It can be understood that the residual sequence characterizes the prediction bias of the safety baseline model.

[0066] In practice, key parameters in the safety baseline are adjusted in reverse based on the residual sequence, and a new round of positive prediction and comparison is performed. Key parameters include the degradation rate coefficient in the material time-varying degradation model. and the influence degree matrix elements in the component interaction influence model After multiple iterations until the residuals converge, the iterative process employs an optimization algorithm to minimize the norm of the residual sequence. In the convergent state, the activated failure criteria and critical transmission paths in the safety baseline are analyzed and identified as potential failure modes and their evolution paths. Failure criteria are predefined judgment conditions in the safety baseline, such as when the stress on a component exceeds its time-degrading strength, or when the overall structural displacement exceeds a limit. Critical transmission paths are identified by analyzing connections in the component interaction matrix where weights change significantly during the failure process. These criteria and paths together describe the most likely types of structural failure and their sequential development among components.

[0067] Example 5: Based on the evolution path, calculate the maximum additional load allowed for each critical component to avoid entering the potential failure mode within the set service target life. The ratio of the maximum additional load to the current load is defined as the residual bearing capacity margin. Construct an interaction network with critical components as nodes and the internal force transmission relationship between components as edges. In the interaction network, using the residual bearing capacity margin as the node weight, simulate the transmission effect of local node weight changes in the entire network to evaluate the robustness of the overall structure. Integrate the residual bearing capacity margin of each node with the robustness evaluation results of the network to generate the structural life map, which is visualized on a two-dimensional or three-dimensional structural model.

[0068] In practice, based on the identified potential failure modes and their evolution paths, the maximum additional load allowed for each critical component to avoid entering a potential failure mode within a set service target lifespan is calculated. The set service target lifespan is determined by the assessment requirements, such as fifty or one hundred years. The maximum additional load is determined by iteratively adjusting the load applied to the high-fidelity structural digital twin until the target component's state at the end of the simulation period meets the failure criteria corresponding to the potential failure mode. In practice, the ratio of the maximum additional load to the current load is defined as the residual bearing capacity margin. The current load refers to the typical load value borne by the component under normal service conditions, obtained statistically from a long-term environmental excitation sample library. Residual bearing capacity margin For the The calculation of each key component can be expressed as follows:

[0069]

[0070] in: Indicates the first The residual load-bearing capacity margin of each key component This indicates the first [number] years of service within the set target service period. The maximum additional load allowed for a critical component to not enter a potential failure mode. Indicates the first The current load borne by each key component.

[0071] In practical implementation, an interaction network is constructed, with key components as nodes and the internal force transmission relationships between components as edges. The node set covers all structural components identified as key components, such as load-bearing columns, main beams, and arches. Edges are established by analyzing the internal force transmission paths of a high-fidelity structural digital twin under standard load conditions. If there is a direct and significant internal force transmission between two components, an edge is established between the corresponding nodes. The interaction network can be understood as a directed weighted graph, where the direction of the edges represents the main direction of internal force transmission, and the initial weight of the edges can be assigned based on the proportion of transmitted internal forces or the geometric connection stiffness between components.

[0072] In practical implementation, within an interacting network, the residual carrying capacity margin is used as the node weight. The propagation effect of local node weight changes throughout the network is simulated to assess the robustness of the overall structure. The simulation process begins with a predetermined decay in the weight of one or a group of nodes. Based on the connection relationships and weights of edges in the network, the indirect impact of this decay on the weights of all other nodes is calculated using a linear or nonlinear network propagation model. Finally, the overall performance index of the entire network is evaluated after being subjected to a local shock. In some embodiments, the network propagation model can use the following linear impact propagation formula to calculate the node... The weight is affected by the node Change in the effect of attenuation :

[0073]

[0074] in: Represents a node The change in weight, Represents a node For nodes Influence factors (usually based on the influence degree matrix elements in the component interaction model) (or define the weight of the edge). Represents a node The initial weight decay. Optionally, a more complex nonlinear model or a graph-based robustness metric can be used to quantify the evaluation.

[0075] In practical implementation, the residual bearing capacity margin of each node is integrated with the robustness assessment results of the network to generate a structural life map that is visualized on a two-dimensional or three-dimensional structural model. The integration process incorporates the residual bearing capacity margin of each key component. The mapping is applied to the color or transparency of the corresponding location on the model, for example, using a gradient color spectrum from green (high margin) to red (low margin), while the robustness assessment results are presented next to the model in the form of a legend or global scale. In essence, the structural life map ultimately appears as a color cloud map or contour map overlaid on the three-dimensional geometric model of the building, visually reflecting the remaining safety reserves of each area and the overall structural vulnerability.

[0076] See Figure 5 This is a diagram showing the residual bearing capacity margin and load analysis of key components of a historical building. The residual bearing capacity margins of different components vary significantly (e.g., load-bearing column 4 and arch 1 have higher margins), reflecting the imbalance between the degree of damage and the bearing potential of the components. The residual bearing capacity margin is positively correlated with the "maximum additional load / current load." Components with higher margins can withstand more additional loads and are the core supporting units for structural safety. This visually displays the bearing redundancy of each key component, helping technicians identify "weak components" in the structure (such as load-bearing column 2 and main beam 2), providing a basis for subsequent reinforcement or load adjustment strategies.

[0077] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0078] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for assessing the safety performance of historical buildings, characterized in that, The method includes: Sensor arrays were deployed at multiple pre-selected locations on the historical building to collect raw monitoring data on vibration, deformation, and the environment. The original monitoring stream is synchronously calibrated and anomaly cleaned to generate a structural response signal with a unified spatiotemporal reference. The structural response signal is input into a pre-trained multi-scale feature perception network to obtain a fused feature map. Based on the fused feature map, the historical building is digitally reconstructed in three dimensions. Based on the feature distribution density, the reconstruction model is adaptively divided into sets of entity units with different discrete precisions to form a high-fidelity digital twin structure. The high-fidelity digital twin of the structure is driven to perform time-history evolution simulation under simulated long-term environmental excitation, and the response history of each entity unit set during the evolution process is recorded. Behavioral fingerprints that can characterize the evolution law of structural damage are extracted from the response history. Establish a safety baseline that includes a material time-varying degradation model and a component interaction model; The behavioral fingerprint is compared with the security baseline through multiple rounds of iterative comparison and reverse tracing analysis to identify potential failure modes and their evolution paths; Based on the evolution path, calculate the residual bearing capacity margin of each key component of the historical building under the set service objectives; By combining the residual load-bearing capacity margin of each key component and the interaction network between components, a structural life map reflecting the overall structural safety status is generated.

2. The method for assessing the safety performance of historical buildings as described in claim 1, characterized in that, The process of synchronously calibrating and cleaning up anomalies in the original monitoring stream to generate a structural response signal with a unified spatiotemporal reference includes: receiving the original monitoring stream from the sensor array; performing timestamp alignment on the sampling times of different sensors in the original monitoring stream to eliminate deviations caused by transmission delays and generate a time synchronization signal; filtering the time synchronization signal to remove signal singularities caused by transient environmental interference or equipment failures and generate a preliminary cleaning signal; comparing the preliminary cleaning signal with the known structural dynamic characteristics of the historical building to correct signal amplitude and phase shifts caused by sensor installation errors, and finally generating the structural response signal with a unified spatiotemporal reference.

3. The method for assessing the safety performance of historical buildings as described in claim 2, characterized in that, The step of inputting the structural response signal into a pre-trained multi-scale feature perception network to obtain a fused feature map includes: The multi-scale feature perception network uses hierarchical convolution and pooling operations to parse and obtain a fused feature map containing local damage features and global modal features. The multi-scale feature perception network is constructed, comprising a shallow feature extraction branch and a deep feature extraction branch connected sequentially. The structural response signal is simultaneously input to both the shallow and deep feature extraction branches. The shallow feature extraction branch extracts local subtle fluctuation features of the structural response using small-scale convolutional kernels, while the deep feature extraction branch extracts the overall macroscopic modal features of the structural response using large-scale convolutional kernels and downsampling. The local subtle fluctuation features output by the shallow feature extraction branch and the overall macroscopic modal features output by the deep feature extraction branch are concatenated and fused along the channel dimension to generate the fused feature map.

4. The method for assessing the safety performance of historical buildings as described in claim 3, characterized in that, The process of digitally reconstructing the historical building based on a fused feature map, and adaptively dividing the reconstructed model into sets of entity units with different discretization accuracies according to feature distribution density to form a high-fidelity structural digital twin, includes: calling the original design drawings and point cloud scan data of the historical building, combining them with the current structural morphology features represented in the fused feature map, and reconstructing a three-dimensional geometric model to obtain an initial geometric model; superimposing the fused feature map on the surface of the initial geometric model, calculating the gradient magnitude of the fused feature map in each region of the model surface, and marking regions with gradient magnitudes exceeding a preset threshold as feature-rich regions; discretizing the marked feature-rich regions using a dense mesh and discretizing the non-feature-rich regions using a sparse mesh to generate the sets of entity units with different discretization accuracies; and assigning each entity unit in the set of entity units with material properties and initial state parameters mapped from the fused feature map to complete the construction of the high-fidelity structural digital twin.

5. The method for assessing the safety performance of historical buildings as described in claim 4, characterized in that, The process of driving the high-fidelity structural digital twin to perform time-history evolution simulation under simulated long-term environmental excitation includes: constructing a long-term environmental excitation sample library covering wind load, temperature load, seismic motion, and crowd live load; extracting excitation sequences from the long-term environmental excitation sample library and applying them to the high-fidelity structural digital twin in chronological order; solving the dynamic equilibrium equations of the high-fidelity structural digital twin at each loading step, updating the stress, strain, and displacement states of each entity element, and recording the state update data; and continuously solving the loading steps until the simulation of the entire long-term environmental excitation sequence is completed to obtain the response history.

6. The method for assessing the safety performance of historical buildings as described in claim 5, characterized in that, The step of extracting a behavioral fingerprint that can characterize the evolution of structural damage from the response history includes: performing time-frequency analysis on the response history to extract the natural frequency change trajectory and damping ratio change trajectory of each entity unit set under different excitation stages; analyzing the stress amplitude spectrum and cycle number of the stress concentration region in the response history; extracting the feature vector sequence that evolves with loading time from the natural frequency change trajectory, damping ratio change trajectory, stress amplitude spectrum and cycle number, and defining the feature vector sequence as the behavioral fingerprint.

7. The method for assessing the safety performance of historical buildings as described in claim 1, characterized in that, The establishment of a safety baseline comprising a material time-varying degradation model and a component interaction model includes: obtaining performance degradation data of the main building materials of the historical building under different environmental stresses through accelerated aging tests in the laboratory and on-site sampling and testing; fitting the material time-varying degradation model based on the performance degradation data to describe the changes in material strength and elastic modulus with service time and external environment; establishing the component interaction model reflecting the relationship between internal force redistribution and damage transmission among components by analyzing the internal force transmission path of the high-fidelity structural digital twin during the simulation process; and coupling the material time-varying degradation model and the component interaction model to form the safety baseline.

8. The method for assessing the safety performance of historical buildings as described in claim 7, characterized in that, The step of performing multiple rounds of iterative comparison and reverse tracing analysis between the behavioral fingerprint and the safety baseline to identify potential failure modes and their evolution paths includes: inputting the behavioral fingerprint into the safety baseline to positively predict the structural state evolution curve under the constraints of the safety baseline; comparing the positively predicted structural state evolution curve with the actual evolution trend observed from the behavioral fingerprint to calculate the residual sequence; based on the residual sequence, reversely adjusting the key parameters in the safety baseline and performing a new round of positive prediction and comparison, iterating through multiple rounds until the residual converges; in the converged state, analyzing the activated failure criteria and key transmission paths in the safety baseline and identifying them as the potential failure modes and their evolution paths.

9. The method for assessing the safety performance of historical buildings as described in claim 8, characterized in that, The method of generating a structural life map reflecting the overall structural safety status by integrating the residual bearing capacity margin of each key component and the interaction network between components includes: calculating the maximum additional load allowed for each key component to not enter the potential failure mode within a set service target life, based on the evolution path; defining the ratio of the maximum additional load to the current load as the residual bearing capacity margin; constructing an interaction network with key components as nodes and the internal force transmission relationship between components as edges; simulating the transmission effect of local node weight changes in the entire network using the residual bearing capacity margin as node weights in the interaction network to evaluate the robustness of the overall structure; and integrating the residual bearing capacity margin of each node with the robustness evaluation results of the network to generate the structural life map visualized on a two-dimensional or three-dimensional structural model.

10. A safety performance assessment system for historical buildings, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the historical building safety performance assessment method according to any one of claims 1 to 9.

Images