House crack life prediction method and system based on digital twinning and data fusion

Abstract

This invention discloses a method and system for predicting the remaining life of building cracks based on digital twins and data fusion, belonging to the field of structural health monitoring technology. The method includes acquiring explicit state data of the crack area, measured structural response data, and environmental and usage status data; constructing a digital twin model based on this; generating initial implicit state variables characterizing the internal state of the crack and configuring their probability distribution intervals by comparing the model's simulated response with the measured response; generating an evolution trajectory set of future states using environmental and usage status data; determining the prediction interval of the remaining life of the crack based on the temporal distribution of the trajectory set touching multi-criteria failure thresholds; verifying the prediction deviation using measured data, calculating the prediction reliability index, and correlating it with the prediction interval for output. This invention achieves interval-based risk assessment of the remaining life of cracks, improving the reliability of early warning for building structural safety.

Description

Technical Field

[0001] This invention relates to the field of structural health monitoring technology, specifically to a method and system for predicting the lifespan of building cracks based on digital twins and data fusion. Background Technology

[0002] Building structural safety monitoring and life prediction technologies are crucial for ensuring the safe operation of urban infrastructure. They aim to achieve early warning of risks and precise maintenance through continuous tracking and evolution analysis of structural defects. In recent years, with advancements in intelligent sensor technology and digital modeling methods, building crack analysis solutions based on monitoring data have gradually evolved towards digitalization and predictive capabilities.

[0003] Existing technologies primarily rely on deploying displacement gauges, crack gauges, or employing image recognition technology to continuously monitor the geometric parameters of crack surfaces, such as width and length. These methods then combine empirical formulas from structural mechanics to extrapolate the remaining life of the crack based on the changing trends of these surface parameters. While these methods can provide some reference under typical uniform degradation scenarios and possess basic real-time monitoring capabilities, they are not universally applicable.

[0004] However, for buildings in complex environments, crack deterioration is not merely a linear increase in surface size. Due to material heterogeneity and load randomness, the equivalent damage depth and other deep physical properties within the cracks (i.e., the implicit state) are dynamically evolving and accompanied by significant uncertainties. Stable surface parameters do not indicate stagnant internal damage; fluctuations in this implicit state lead to nonlinear transformations in the load-bearing capacity weakening mechanism, causing a disconnect between the actual deterioration rate and surface observation data. Existing technologies have failed to establish effective prediction mechanisms for this "unobservable" and "uncertain evolutionary pattern" of the implicit state of cracks. Relying solely on explicit surface data for lifetime estimation makes it difficult to capture the risk of sudden failure induced by accelerated internal implicit deterioration, easily leading to distorted lifetime prediction ranges or delayed warnings, increasing the safety hazard of sudden structural collapse. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for predicting the lifespan of building cracks based on digital twins and data fusion.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] In a first aspect, the present invention discloses a method for predicting the lifespan of a building based on digital twins and data fusion, comprising the following steps:

[0008] Acquire explicit crack status data, measured structural response data, and environmental and usage status data for the cracked area of ​​the building under test;

[0009] Based on the explicit crack state data and the measured structural response data, a digital twin model of the building cracks is constructed.

[0010] By comparing the simulated response output by the digital twin model with the measured structural response data, an initial implicit state variable characterizing the internal physical properties of the crack is generated, and an initial probability distribution interval is configured for the initial implicit state variable.

[0011] Using the environmental and usage status data as input, the initial implicit state variables and the initial probability distribution interval are driven to evolve in the digital twin model according to the time step, generating a set of implicit state evolution trajectories that characterize the future state of the crack.

[0012] Based on the temporal distribution of the implicit state evolution trajectory set reaching the preset failure threshold, the prediction range of the crack's remaining lifetime is determined.

[0013] Historical explicit crack state data measured in actual measurements are retrieved as verification data; the matching deviation between the verification data and the predicted crack state data of the implicit state evolution trajectory set at the corresponding historical time points is calculated; a prediction confidence index is calculated based on the matching deviation; and the prediction range of the crack remaining lifetime is correlated with the prediction confidence index and output.

[0014] Secondly, this invention discloses a building crack life prediction system based on digital twins and data fusion, which uses the aforementioned building crack life prediction method based on digital twins and data fusion, including:

[0015] The multi-source data acquisition module is used to acquire explicit crack status data, measured structural response data, and environmental and usage status data of the cracked area of ​​the building under test.

[0016] The twin modeling module is used to construct a digital twin model of building cracks based on the explicit crack state data and the measured structural response data.

[0017] The inversion generation module is used to invert and generate initial implicit state variables that characterize the internal physical properties of the crack by comparing the simulated response output by the digital twin model with the measured structural response data, and to configure an initial probability distribution interval for the initial implicit state variables.

[0018] An evolution-driven module is used to take the environmental and usage status data as input, drive the initial implicit state variables and the probability distribution interval to evolve in the digital twin model according to the time step, and generate a set of implicit state evolution trajectories representing the future state of the crack.

[0019] The prediction module is used to determine the prediction range of the remaining life of the crack based on the temporal distribution of the implicit state evolution trajectory set reaching the preset failure threshold.

[0020] The matching evaluation module is used to retrieve measured historical explicit crack state data as verification data; calculate the matching deviation between the verification data and the predicted crack state data of the implicit state evolution trajectory set at the corresponding historical time points; calculate the prediction confidence index based on the matching deviation; and associate the predicted range of the crack remaining lifetime with the prediction confidence index for output.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0022] 1. This invention, by inverting implicit state variables, deepens crack analysis from surface geometric features to internal physical properties, achieving a quantitative characterization of internal damage. This enables the model to accurately capture deep properties such as equivalent damage depth, establish a more precise evolution starting point, and overcome the initial assessment bias caused by neglecting internal degradation in traditional methods.

[0023] 2. This invention uses a digital twin model to simulate a set of temporal evolution trajectories under the interference of uncertain factors. The temporal evolution based on distribution intervals can fully cover the uncertainty propagation law during crack degradation, reflecting the probability distribution characteristics of future states near the failure threshold. The final determined remaining lifetime prediction interval, rather than a single numerical value, can more objectively reflect the discreteness of the structural degradation rate, thereby reducing the risk of lifetime misjudgment due to environmental fluctuations or model parameter sensitivity.

[0024] 3. This invention utilizes the deviation calculation between measured explicit data and historically predicted crack state data from the trajectory set to achieve quantitative feedback on the reliability of the prediction model. This allows the output lifespan range to be correlated with indicators that reflect the model's accuracy in real time. The system can automatically identify the degree of deviation from the predicted path based on measured feedback, thus providing more reliable evidence for operation and maintenance decisions and ensuring the stability and interpretability of building structural safety assessment results under complex service conditions. Attached Figure Description

[0025] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0026] Figure 1 This is a flowchart of the steps of the present invention;

[0027] Figure 2 This is a flowchart illustrating the working principle of the present invention;

[0028] Figure 3 This is a flowchart illustrating the system module connection of the present invention. Detailed Implementation

[0029] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0030] In existing technologies, the prediction of building crack lifespan largely relies on periodic manual inspections or calculations based on single-point sensor data, making it difficult to balance the long-term reliability of the prediction results with real-time dynamic updates. Traditional methods treat cracks as static defects or rely on deterministic mechanical models, failing to quantify the unobservable nature of the crack's internal state and the evolutionary uncertainty caused by external environmental disturbances. This results in predictions that are mostly single numerical values, lacking quantitative risk assessment and failing to support refined operation and maintenance decisions. Existing methods are mostly limited to monitoring the cracks themselves, failing to effectively establish a quantitative and dynamic correlation between crack status and the overall building structure's safety status and remaining service life, leading to a lack of foresight and systematic approach in safety early warning.

[0031] To address the aforementioned issues, this study discovered a quantifiable probabilistic mapping relationship between the equivalent damage state (implicit state) within a crack and its resulting observable structural responses (such as displacement and strain), as well as with environmental loads. By defining the true crack state as a set of implicit state variables containing uncertainties and then reverse-identifying and probabilistically representing these variables in a digital twin model, a dynamic digital image of the crack state was constructed. Further findings revealed that multi-path evolution simulation based on this probabilistic state in a twin environment can statistically yield a predicted range for the remaining lifetime, rather than a single deterministic value. Simultaneously, by comparing subsequent measured data with the predicted trajectory, a closed-loop evaluation of the reliability of the prediction model itself can be formed.

[0032] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0033] Example 1:

[0034] like Figure 1 As shown, the method for predicting the lifespan of a building based on digital twins and data fusion includes the following steps:

[0035] Acquire explicit crack status data, measured structural response data, and environmental and usage status data for the cracked area of ​​the building under test;

[0036] A digital twin model of building cracks is constructed based on explicit crack state data and measured structural response data.

[0037] By comparing the simulated response output by the digital twin model with the measured structural response data, initial implicit state variables characterizing the internal physical properties of the crack are generated, and an initial probability distribution interval is configured for the initial implicit state variables.

[0038] Using environmental and usage data as input, the implicit state variables and initial probability distribution intervals are driven to evolve in the digital twin model according to the time step, generating a set of implicit state evolution trajectories that characterize the crack state change process.

[0039] Based on the temporal distribution of the implicit state evolution trajectory set reaching the preset failure threshold in the future time period, the prediction range of the crack's remaining lifetime is determined.

[0040] Historical explicit crack state data measured in actual measurements are retrieved as validation data. The matching deviation between the validation data and the predicted crack state data at the corresponding historical time points is calculated. The prediction confidence index is calculated based on the matching deviation, and the prediction range of the crack remaining lifetime is correlated with the prediction confidence index and output.

[0041] like Figure 2 The diagram shown illustrates the working principle of this application. This application primarily aims to predict cracks in building structures that may pose potential risks to the structural bearing capacity and safety performance due to stress, environmental factors, and material deterioration. Specifically, it includes the following types of cracks:

[0042] Stress cracks: such as cracks that occur on the surface or inside of components subjected to bending, shear, tension, etc., with particular attention paid to the relationship between their depth propagation and the decrease in load-bearing capacity.

[0043] Environmental factors causing cracks: such as cracks caused by temperature stress, humidity changes, freeze-thaw cycles, etc.

[0044] Fatigue cracks: cracks that gradually expand under cyclic loading (such as traffic or mechanical vibration).

[0045] Complex cracks: These are complex cracks caused by the combined effects of multiple factors, resulting in both surface geometric expansion and internal accumulation of hidden damage.

[0046] The working principle of this application is as follows: The system first acquires multi-source monitoring data through a smart sensor network deployed on-site. Explicit crack status data is acquired by a high-definition CCD (charge-coupled device) industrial camera, crack width gauge, or laser scanner, capturing the geometric morphology of the crack in real time; measured structural response data is acquired by vibration sensors, strain gauges, inclinometers, and dynamic characteristic acquisition devices deployed on load-bearing components, reflecting the macroscopic mechanical performance of the structure under load; environmental and usage status data are acquired in real time through a small weather station (temperature, humidity, and wind speed sensors) and load sensors (such as weighing sensors or simulated traffic flow monitoring devices). The above hardware is connected to the edge computing unit through an industrial gateway to ensure synchronous transmission of data with a unified timestamp.

[0047] After obtaining the measured data, the system constructs a high-fidelity digital twin model of the cracked area of ​​the building using finite element simulation software or a BIM (Building Information Modeling) platform. This digital twin model includes not only the building's geometric topology and material mechanical parameters (such as elastic modulus and Poisson's ratio), but also the explicit geometric definition of the crack at the current moment. By assigning the measured explicit crack data to the digital twin model, the system achieves initial geometric synchronization between the physical entity and the digital model, establishing a mechanical simulation platform that can be driven by external forces.

[0048] For physical properties inside the crack that cannot be directly observed, the system enters the parameter inversion processing stage. By inputting boundary conditions consistent with the measured environment into the digital twin model, the model outputs a simulated response sequence. Using inverse analysis algorithms (such as genetic algorithms, particle swarm optimization, or neural network inversion), with the objective function of minimizing the residual between the simulated response and the measured structural response, implicit state variables are derived through inversion. These variables contain equivalent parameters characterizing the degree of damage inside the crack. Subsequently, the system statistically monitors the measurement noise of the monitoring hardware and the sensitivity of the model parameters to configure probability distribution intervals (such as Gaussian or beta distributions) for these implicit state variables, completing the quantification initialization of uncertainties.

[0049] The system's driving engine receives environmental and usage status data and converts it into time-varying load excitations recognizable by the digital twin model. Within a preset time step, the digital twin model calculates the evolution trend of implicit state variables under the coupled effects of periodic loads and environmental corrosion using explicit iteration or implicit integration algorithms. Utilizing Monte Carlo sampling or Latin hypercube sampling techniques, the system performs multi-sample random evolution calculations within a configured probability interval, thereby generating a series of divergent implicit state evolution trajectories. These implicit state evolution trajectories constitute a complete dataset characterizing the possible future states of the cracks, covering the entire process from the current moment to the predicted future endpoint.

[0050] The system performs threshold exceedance judgment on the generated trajectory set. It has pre-set digital failure criteria and automatically compares the relationship between the instantaneous state of each path in the trajectory set and the safety threshold. When a trajectory line crosses the failure boundary, the corresponding time point is recorded. Through statistical analysis of the lifetime endpoints of all paths in the trajectory set, the system extracts prediction intervals that meet a specific confidence level, serving as a preliminary quantitative result of the remaining lifespan of the building cracks.

[0051] To enhance the evidentiary value of the prediction results, the system executes a feedback calibration procedure. It retrieves historical explicit crack state data recorded in memory, using it as a true value reference. The system extracts simulated crack state values ​​corresponding to historical moments from the implicit trajectory set and calculates the matching deviation between the simulated and true values ​​using mathematical metrics such as Euclidean distance or Mahalanobis distance. Finally, the deviation is converted into a confidence index between 0 and 1 using a preset confidence function (such as exponential decay mapping), and the lifetime prediction range is associated and packaged with this index, then output to management decision-makers in real time via a visualization terminal.

[0052] This application achieves a two-way mapping between physical entities and digital space through a digital twin model, effectively solving the technical challenge of the unmeasurability of implicit damage parameters. Employing probabilistic evolution methods and multi-sample trajectory simulation, it quantifies the uncertainty risks in the prediction process, overcoming the shortcomings of traditional deterministic prediction methods in terms of insufficient accuracy under complex working conditions. Through a backward verification mechanism using historical observation data, it provides a clear confidence level assessment for the prediction results, ensuring the objectivity and reliability of the output results, and providing technical support for the refined operation and maintenance and risk early warning of building structures.

[0053] This application further proposes a step for inverting and generating initial implicit state variables characterizing the internal physical properties of cracks. After acquiring measured structural response data, the system does not directly perform geometric comparison, but rather performs noise reduction and feature extraction based on the measured structural response data to calculate the stiffness degradation characteristic value of the structure. This characteristic value is typically expressed as the offset of the structure's first or higher frequency, the modal correlation coefficient, or the rate of change of strain modes. This is achieved by comparing the stiffness with the baseline stiffness under the original healthy state of the structure. Measured stiffness under current damage condition The stiffness reduction factor can be calculated. This process provides a macroscopic mechanical basis for subsequent implicit state variable inversion, enabling internal damage that was originally not directly observable to be quantitatively characterized through external mechanical response features.

[0054] To achieve accurate identification of crack depth and morphology, a theoretical response database containing multi-dimensional physical parameters is pre-constructed for the digital twin model. By pre-setting different crack depths d (e.g., values ​​ranging from 0mm to 200mm with a step size of 5mm) and different morphological parameters (such as crack angle and number of branches), a series of corresponding theoretical response sequences of the digital twin model under different pre-set crack depths and morphologies are obtained using finite element simulation.

[0055] The system employs nonlinear least squares or surrogate model optimization algorithms to construct the residual function. The formula for calculating the residual between the measured structural response data and the theoretical response sequence is:

[0056] ;

[0057] in, The measured structural response at the i-th measurement point;

[0058] For digital twin models in parameter vectors Theoretical response under the following conditions;

[0059] parameter vector It includes the equivalent damage depth and the weakened area of ​​the tensile zone.

[0060] By iteratively searching for... By minimizing the global optimal solution, the system can accurately identify the equivalent damage depth and the weakened area of ​​the tensile zone inside the crack from the external response and determine them as the initial implicit state variables.

[0061] For the identification of complex structures, this embodiment also relates to an auxiliary identification method using a lightweight neural network model. This model is generated offline using a digital twin model. The neural network was trained using simulated data from at least one set of samples. It consisted of an input layer (10 nodes), two hidden layers (LSTM layers with 50 and 20 nodes respectively), and an output layer (2 nodes). Training conditions were set as follows: a learning rate of 0.001, an Adam optimizer, and a loss function of mean squared error (MSE). The number of training epochs was no less than 500. Overfitting was prevented using a dropout rate of 0.2 and early stopping. The trained model can instantly output the corresponding predicted damage depth value based on the real-time input stiffness degradation feature value, improving the response speed of online monitoring.

[0062] This embodiment achieves a deep understanding of the health status of building cracks by establishing a mapping relationship between macroscopic stiffness degradation characteristics and microscopic crack internal parameters. Through a residual minimization identification mechanism, the interference of environmental noise on identification accuracy is effectively eliminated, ensuring the physical authenticity of the initial implicit state variables. This technique not only improves the synchronization accuracy of the digital twin model and the physical entity in terms of mechanical performance, but also lays a reliable parameter foundation for the accurate generation of subsequent time-series evolution trajectories, enhancing the objectivity and robustness of predicting the remaining life of building cracks under complex load environments.

[0063] This application further proposes that, in configuring the initial probability distribution interval for the initial implicit state variables, the configuration logic of the initial probability distribution interval is defined in depth.

[0064] The system first uses a signal processing unit to monitor and statistically analyze the explicit crack state data transmitted from the front-end intelligent sensors in real time, aiming to isolate the instrument's measurement noise and random deviations caused by external environmental interference. It statistically analyzes the deviations caused by instrument measurement noise and environmental interference during the explicit crack state data acquisition process. Specifically, the system performs Gaussian fitting on the zero-point drift data of the intelligent sensors in a static environment, and combines this with the effects of environmental temperature and humidity fluctuations on refraction and creep caused by optical lenses or strain gauges, to calculate the comprehensive deviation value. This deviation is typically between 1% and 5% of the original measurement.

[0065] The system invokes a digital twin model to investigate the influence weights of physical parameter fluctuations on the structural response. Based on the physical parameter sensitivity of the digital twin model, the mapping weights of the initial implicit state variables to the deviations are calculated. Specifically, by applying small perturbations to the material stiffness, boundary constraints, and crack geometry parameters in the digital twin model, the rate of change of the model's output response is observed, thereby calculating the mapping weights of the initial implicit state variables to the aforementioned deviations. Its calculation formula is usually expressed as:

[0066] ;

[0067] Where S represents the structural response parameters output by the model;

[0068] These are the implicit state variables to be inverted;

[0069] This is the baseline response value.

[0070] Based on this mapping weight The system utilizes kernel density estimation (KDE) or Monte Carlo stochastic simulation to construct a probability envelope surface characterizing the discreteness of the variable distribution within the parameter space of the implicit state variables. This envelope surface defines the initial discrete boundary of the evolution trajectory set. By using the probability envelope surface as the initial probability distribution interval of the implicit state evolution trajectory set, it ensures that the initial uncertainty in the prediction process is entirely determined by the combined experimental accuracy and the characteristics of the physical model.

[0071] To further improve the efficiency of weight calculation, this embodiment employs a pre-trained random forest regression model to replace real-time finite element sensitivity calculation. This model is trained using 5000 sets of parameter perturbation samples obtained through offline simulation. The training environment is set with 100 decision trees and a maximum tree depth between 15 and 20. The model learns the nonlinear relationship between parameter perturbation and response bias through historical samples. After receiving real-time bias values, the trained model can output mapped weights within milliseconds, supporting the dynamic updating of the probability envelope.

[0072] By introducing a bias mapping weight and probability envelope surface construction mechanism, this application effectively solves the problem that traditional deterministic prediction methods cannot assess the limitations of monitoring accuracy. By quantifying the influence of instrument noise on implicit state variables, the subsequently generated evolution trajectory set can include various potential states caused by the limitations of actual measurement conditions.

[0073] This application further proposes a step for generating an implicit set of state evolution trajectories characterizing the crack state change process. The system first extracts the core dynamic parameters affecting crack evolution, namely the temperature stress amplitude, from environmental and operational data. The load cycle number N. The system uses a built-in fracture mechanics mapping algorithm to map the temperature stress amplitude from environmental and operational data. The load cycle number N is calculated and transformed into microscopic parameters acting on the crack tip in the digital twin model.

[0074] In practice, the system calculates the increment of the equivalent energy release rate based on the current crack geometry and constraints. Increment in equivalent energy release rate This describes the energy required to propagate a unit area of ​​crack under a specific load cycle step. Its calculation logic follows a modified energy balance equation:

[0075] ;

[0076] in, The energy release ratio coefficient related to the material is determined through laboratory fracture mechanics tests and dynamically adjusted according to "structure type, service life, and safety level".

[0077] This represents the amplitude of the external load stress.

[0078] 'a' represents the crack length at the current step size;

[0079] This is the function that determines the effect of the number of cycles on energy accumulation.

[0080] Within each preset time step (e.g., 30 days or...), (Each cycle is one step), the system will calculate the increment of the equivalent energy release rate. As the driving input, parameters characterizing crack propagation (such as crack depth increment) in the implicit state variables are updated in real time. Or forward-propelled displacement).

[0081] To reflect the random fluctuations in the prediction process, the system combines the initial probability distribution interval with parallel computation using multi-sample random sampling simulation techniques (such as Markov Chain Monte Carlo (MCMC)). A large number of sample points are extracted within the probability envelope as the starting point for evolution. Each path updates its state based on the increment of the equivalent energy release rate with random perturbations as it progresses over time. With the accumulation of time steps, multiple implicit state evolution paths diverge temporally, generating multiple implicit state evolution paths. Due to differences in sampling points and local stress responses among the paths, the originally concentrated parameter points gradually form a cluster of paths that diverge temporally, thus constituting the implicit state evolution trajectory set. This implicit state evolution trajectory set, through thousands or even tens of thousands of evolution paths, completely covers all the potential probability space of the crack's evolution from its current damage state to its future failure state.

[0082] By converting macroscopic environmental stress into microscopic energy release rate increments, this application reveals the physical mechanism of crack propagation, avoiding the distortion problem of traditional empirical formulas when dealing with complex environmental coupling effects. Through iterative updates of the time step and multi-sample random sampling, the system can not only simulate the average trend of crack evolution, but also quantitatively describe the prediction dispersion caused by material inhomogeneity and environmental load fluctuations, ensuring the technical rigor of the implicit state trajectory set in long-term prediction.

[0083] This application further proposes that the system monitors each path in the evolution trajectory set in real time within each time step and compares its state parameters with a preset failure threshold. To ensure that the prediction results comply with mandatory standards for building structural safety, the preset failure threshold is not a single value, but rather encompasses both intuitive geometric representation and underlying mechanical essence. Based on the temporal distribution of the implicit state evolution trajectory set reaching the preset failure threshold in future time periods, the preset failure threshold is defined using a multi-criteria fusion approach.

[0084] Geometric failure threshold This geometric failure threshold is directly related to the explicit crack state data. It is used to characterize whether the crack width in the explicit crack state data has reached a preset maximum safety limit. In specific implementation, the system determines the preset maximum safety limit for crack width based on the civil building reliability assessment standards or relevant industry technical specifications. For example, for concrete flexural members, this limit is typically set between 0.3 mm and 0.5 mm. This applies to the crack width component in the evolution trajectory. When the preset maximum safety limit is reached, the system determines that the geometric safety of the path has been lost.

[0085] Mechanical failure threshold Used to characterize the structural bearing capacity reduction coefficient calculated in real time by the digital twin model. The structure reaches the preset ultimate limit state boundary of its bearing capacity. The load-bearing capacity reduction factor of this structure... The calculation logic is as follows: This is used to quantify the reduction in effective cross-sectional strength caused by crack propagation.

[0086] ;

[0087] in, The remaining bearing capacity is calculated in real time by the model based on the equivalent damage depth d and the weakened area A of the tension zone in the current implicit state variables;

[0088] The limit state boundary value for the bearing capacity of the structure.

[0089] Mechanical failure threshold The structural importance coefficient is typically set between 0.85 and 0.95. The system evaluates in real time whether the structural resistance corresponding to each evolution path crosses this limit boundary.

[0090] The decision-making logic employs the "barrel principle" of logical OR operation. That is, at any time step in the evolution trajectory set, if the parameters of any path in the implicit state evolution trajectory set trigger either the geometric failure threshold or the mechanical failure threshold, the system immediately terminates the iterative calculation of that path, thus determining that the lifetime of that path has ended, and recording that moment as the lifetime termination point of that path. By statistically summarizing the temporal distribution of tens of thousands of evolution path termination points, the system automatically generates a probability density curve for lifetime prediction, extracts its mean and confidence interval, and finally determines the prediction range of the crack's remaining lifetime.

[0091] By integrating geometric and mechanical criteria, this application overcomes the one-sidedness of traditional life prediction methods that rely solely on the width of visible cracks while ignoring the degradation of the underlying load-bearing capacity. The geometric failure threshold ensures the building's serviceability limit state, while the mechanical failure threshold locks in the structure's ultimate load-bearing capacity state. The synergistic determination of these two criteria provides multiple safety guarantees for life prediction. This failure determination method, based on real-time calculation using a digital twin model, can accurately capture the critical point where the structure undergoes a qualitative change from a quantitative one. This ensures that the output life prediction range not only has the support of physical mechanisms but also meets the compliance requirements of engineering safety standards.

[0092] This application further proposes that, after completing the lifetime evolution simulation and failure determination, a backward verification mechanism is introduced to quantitatively assess the reliability of the prediction results. After generating an implicit state evolution trajectory set covering historical and future time periods, the system initiates a matching evaluation, aiming to "anchor" and calibrate the complex physical simulation model using known observational facts. The matching deviation between the verification data and the predicted crack state data of the implicit state evolution trajectory set at corresponding historical time points is calculated, and the prediction reliability index is calculated based on the matching deviation. The system extracts the characteristic values ​​of its probability distribution by performing cross-sectional analysis on each discrete time point of the implicit state evolution trajectory set within the historical time series.

[0093] In practice, the computing unit statistically analyzes the state parameters of all evolution paths at a given moment, extracts the center value of the probability distribution of the implicit state evolution trajectory set at the corresponding time point, and connects these center values ​​from historical periods in a time series to form a distribution center trajectory line that represents the expected trend predicted by the model. This distribution center trajectory line reflects the optimal estimate of the crack historical evolution logic by the digital twin model under the current parameter configuration.

[0094] The system retrieves pre-stored historical explicit crack state data from memory as the verification true value and compares it with the distribution center trajectory line through spatial projection. To eliminate the influence of different physical dimensions on the deviation calculation, the system calculates the spatial distance between the verification data and the distribution center trajectory line. Among them, spatial distance Either Euclidean distance or Mahalanobis distance is used. In the specific implementation, the system provides two optional distance operators: when the components are independent, Euclidean distance is used for linear measurement; when considering the correlation between multiple characteristic parameters such as crack width and length, Mahalanobis distance is used, and its calculation logic is as follows:

[0095] ;

[0096] in, These are historical measured vectors;

[0097] The distribution center value vector;

[0098] S is the covariance matrix.

[0099] The calculated spatial distance D reflects the degree of deviation of the simulated trajectory from the actual physical process. The system substitutes the spatial distance D into a preset confidence decay function. The output value between 0 and 1 is used as a prediction confidence index. Credibility decay function Negative exponential functions or Gaussian decay functions are typically used, for example: ,in This is a preset attenuation coefficient (typically ranging from 0.1 to 0.5). This function ensures that when the spatial distance D is 0 (i.e., a perfect fit), the output confidence index is [value missing]. The reliability index is 1; as the deviation increases, the reliability index decreases. Approaching 0.

[0100] By establishing a comparative mapping between "measured" and "predicted" data over historical periods, and through the extraction of the distribution center trajectory and the calculation of multi-dimensional spatial distance, this application can objectively identify the fitting accuracy of the digital twin model under current operating conditions. Transforming the matching deviation into an intuitive credibility index and correlating it with the output not only enables technicians to determine the reference value of the prediction results but also provides a decision-making basis for subsequently triggering automatic model correction or manual intervention, fundamentally solving the technical risk of "blind output" by prediction models in the field of structural monitoring.

[0101] This application further proposes that the method also includes abnormal disturbance identification. The system performs high-frequency real-time monitoring of environmental and usage status data through a data acquisition terminal, focusing on capturing pulse-type load data that do not conform to the conventional statistical distribution. It monitors in real time whether pulse-type load data exceeding a preset range appears in the environmental and usage status data. In specific implementation, the system presets a dynamic range threshold. This threshold is typically set to 3 to 5 times the standard live load of the structural design or the average historical load (for example, for a typical residential floor, the pulse load threshold can be set to 10 kN). The system uses a sliding window algorithm to monitor instantaneous stress or impact force. Once it detects that the amplitude in the data packet exceeds the preset range in a very short time (e.g., within 0.1 s), it determines that pulse load data has been identified.

[0102] Once an anomalous perturbation is triggered, the computational engine immediately intervenes in the current evolutionary process. Because pulsed load databases (such as sudden strong earthquakes, object impacts, or illegal overloading) can cause unpredictable damage to the microstructure within cracks, the system performs an "instantaneous discretization" operation. This covers this uncertainty by instantaneously increasing the probability distribution range of the implicit state evolution trajectory set. Numerically, the system adjusts the coefficients of the random perturbation terms in the evolution equations or directly expands the covariance matrix of the initial probability distribution interval (e.g., by increasing its variance). (Instantaneous expansion of 1.5 to 2 times) causes the trajectory set to exhibit a greater divergence in time.

[0103] Subsequently, based on this expanded probability space, the crossing determination for the failure threshold is re-executed, thereby updating and redetermining the predicted range of the crack's remaining lifetime.

[0104] Furthermore, when the abnormal disturbance identification unit detects pulsed load data, it not only instantaneously increases the probability distribution range (i.e., variance) of the implicit state variables, but also dynamically corrects the prior distribution mean of the implicit state variables based on the amplitude and duration of the pulsed load. Specifically, based on a preset structural impact response model, the pulsed load is converted into an equivalent damage increment, which is then injected into the Bayesian update framework as prior knowledge to adjust the distribution mean of implicit state variables such as crack depth and damage factor, thereby reflecting the abrupt state change characteristics of the structure under transient strong disturbances. For example, when an impact acceleration signal with an amplitude exceeding a preset threshold is detected, the potential microcrack propagation amount is estimated based on empirical formulas or finite element calibration curves, and this value is used as the offset of the initial predicted mean of the implicit state variables to avoid prediction distortion due to ignoring state transitions.

[0105] By introducing an abnormal disturbance identification and dynamic probability interval adjustment mechanism, this application can realistically simulate the accelerating effect of sudden loads on structural damage evolution. By increasing the width of the prediction interval, it reflects the risk premium brought about by abnormal disturbances. This dynamic correction logic effectively avoids the risk of underreporting caused by the model's "overconfidence," providing necessary redundancy and technical buffer for building safety monitoring under extreme conditions.

[0106] This application further proposes that the system also includes a closed-loop feedback mechanism for the continuous evolution of the digital twin model, aiming to address the prediction bias caused by material aging or operating condition drift during long-term model operation. In specific implementation, the system sets a preset monitoring period (e.g., every 30 calendar days or every 5000 sets of data collected) and periodically acquires the latest explicit crack state data and measured structural response data from the field sensor array. This latest data is defined as the observation evidence set and used to correct the model's existing evolution parameters in real time.

[0107] The system employs a Bayesian update algorithm to perform closed-loop feedback, incorporating key physical parameters controlling crack propagation speed in the digital twin model (including the energy release ratio coefficient in the energy release rate calculation formula). The function f(N) affecting the number of iterations, the decay coefficient λ, and the coverage coefficient k are treated as random variables, and the current weight distribution is taken as the prior distribution. The key physical parameters in the digital twin model were corrected using the latest explicit crack state data and measured structural response data.

[0108] When the latest observational evidence is obtained Then, using the likelihood function This describes the probability of observing the current actual crack state under different parameter values. The posterior distribution of the parameters is calculated using Bayes' theorem, as follows:

[0109] ;

[0110] The computational unit uses Markov chain Monte Carlo (MCMC) sampling or variational inference techniques to solve for the optimal estimate of the posterior distribution. Based on the updated posterior distribution, the system instantaneously corrects key physical parameters in the digital twin model, making the model's evolution behavior in the next step more closely reflect the latest physical properties of the building. For example, if the recent crack propagation rate is faster than the model predicts, the system will automatically increase the energy release ratio coefficient. The value of is used to reflect the trend of escalating fatigue damage in materials.

[0111] Through periodic Bayesian updates, this application achieves the transformation of the digital twin model from static simulation to dynamic evolution. This closed-loop feedback mechanism effectively overcomes potential biases in the initial modeling parameters and can capture changes in the evolution rate of the structure due to factors such as material degradation and environmental corrosion accumulation during long-term use. By continuously fine-tuning the evolution weights using new observation data, the system ensures that the implicit state evolution trajectory set is always anchored on the actual physical evolution path, guaranteeing the continuous accuracy and adaptability of long-term prediction results from the technical source.

[0112] This application further proposes that, after completing the Bayesian evolution and verification of the digital twin model, this application ensures the scientific rigor and accuracy of the lifetime prediction results in terms of expression by establishing a dynamic adjustment mechanism at the output end. The prediction interval of the crack's remaining lifetime is linked to the prediction confidence index in the output, aiming to map the aforementioned calculated prediction confidence index to the physical representation of the prediction interval, thereby intuitively conveying the predicted risk through changes in the interval width.

[0113] In practice, the system receives a prediction confidence index C, ranging from 0 to 1, output by the matching evaluation. The system has a pre-set interval adjustment algorithm that adjusts the confidence boundary of the lifetime prediction interval in real time based on this prediction confidence index C, thereby adjusting the width. The lifetime prediction interval represents the highly probable range of crack failure time, while the confidence index C represents the quality of the model's fit to the current physical properties.

[0114] When performing width adjustment, the system calculates the interval expansion coefficient based on the reliability index C. When the credibility index is high (e.g. When the prediction confidence index (C) is low, the system maintains the prediction interval width calculated from the original evolution trajectory set, indicating that the current model is highly synchronized with the physical entity, and the prediction results have extremely high reference value. However, when the confidence index decreases due to model bias or interference factors, the system automatically initiates an interval expansion procedure. If the prediction confidence index C falls below a preset threshold (ranging from 0.5 to 1.5, for example, a preset threshold of 0.70), the coverage of the prediction interval is automatically expanded. The adjustment logic is as follows:

[0115] ;

[0116] in, This is the adjusted lifetime prediction range;

[0117] This represents the mean value at the end of the lifespan.

[0118] k is the coverage coefficient, used to control the confidence level in the calculation of lifetime prediction intervals. Its value range is usually from 1.5 to 3.0, for example, k=2.0;

[0119] denoted as the standard deviation of the evolutionary trajectory set.

[0120] By using the confidence index C as the denominator, the half-width of the interval increases significantly as C decreases. This automatic expansion mechanism reflects the uncertainty premium in prediction in both visual and data dimensions, forcibly widening the potential failure window.

[0121] Finally, the system pushes the adjusted life prediction range and the current confidence index to the monitoring terminal simultaneously. This correlated output method not only provides technicians with a quantitative conclusion on "how long the crack can last," but also indicates "how confident this conclusion is" through the dynamic changes in the range width. If the range is too wide and the confidence level continues to decline, the system will automatically trigger maintenance suggestions, prompting manual intervention for inspection or sensor calibration.

[0122] This application constructs a closed-loop system for predicting the lifespan of building cracks based on both physical mechanisms and data-driven constraints. It addresses the lack of underlying information regarding unobservable internal damage through "digital twin inversion," quantifies uncertainty risks using "multi-sample random evolution," and achieves compliant safety assessments by combining "multi-criteria thresholds." Through "backward matching verification" and "dynamic adjustment of the prediction interval," the system possesses the ability to self-evaluate its prediction results. Combined with "abnormal disturbance identification" and "Bayesian closed-loop update," lifespan prediction is no longer an isolated static deduction, but an intelligent monitoring process that can dynamically correct itself with environmental impulses and material evolution, providing real-time feedback on confidence levels. In complex service environments, this system not only improves the accuracy of lifespan prediction but also effectively avoids decision-making biases caused by single model predictions through adaptive adjustment of the interval width, providing highly reliable technical support for the full life-cycle safety management of building structures.

[0123] By combining implicit damage quantification, future evolution simulation, and multi-criteria failure determination, this application achieves a comprehensive analysis from local crack phenomena to overall structural safety risks, providing key technical support for preventive maintenance, safety risk classification and control, and precise decision-making intervention for building structures, and promoting the transformation of structural health management from a passive response to an active intelligent early warning model.

[0124] The following is a specific example of a method for predicting the lifespan of buildings based on digital twins and data fusion:

[0125] A minor crack appeared in the load-bearing beam of a historical brick-concrete building. The maintenance team activated a building crack life prediction system based on digital twins and data fusion to conduct monitoring and assessment. The system used a high-definition industrial camera (12 megapixel resolution, sampling frequency 1 time / hour) to capture the crack's initial width of 0.15mm, extending to one-third of the beam's span. A piezoelectric accelerometer (500Hz sampling frequency) measured the beam's first-order natural frequency at 12Hz, and a fiber optic strain gauge (1 microstrain accuracy) collected the maximum tensile strain at mid-span at 85 microstrain. An integrated meteorological monitoring station simultaneously recorded environmental data: daily average temperature fluctuation range of -5℃ to 18℃, daily average relative humidity of 65%, and maximum instantaneous wind speed of 12m / s.

[0126] Hardware-level synchronization of multi-source data (time error <0.5ms) was achieved via the IEEE 1588 PTP protocol. The twin modeling module imported the beam's geometric parameters (section size 300mm×500mm, span 4.8m) based on the BIM model, configured the C30 concrete material properties (elastic modulus 30GPa, Poisson's ratio 0.2), and constructed a high-fidelity finite element model (mesh size 20mm). The inversion generation module, using measured frequency and strain data as targets, iteratively calculated using a particle swarm optimization algorithm to identify an equivalent damage depth of 125mm (1 / 4 of the beam height) inside the crack, with the weakened area in the tension zone reaching 18% of the cross-sectional area. The system statistically analyzed the camera measurement noise (±0.02mm) and model parameter sensitivity to calculate the initial probability distribution interval of the implicit state variables: damage depth [118mm, 132mm], weakened area [16%, 20%]. Based on this initial probability distribution range, the evolution-driven module uses the Latin hypercube sampling method to generate 1000 sets of initial implicit state variable samples that conform to this distribution, which serve as the initial starting point for 1000 evolution trajectories.

[0127] The evolution-driven module converts environmental data into temperature stress amplitude ( The increment of the equivalent energy release rate was calculated using fracture mechanics formulas, taking into account the number of equivalent load cycles (365 cycles / year based on daily temperature fluctuations). ,in , (Live load stress), a = 0.15m (crack length). (Correction factor for the number of iterations), calculated as follows Monte Carlo sampling technique was used to generate 1000 evolutionary trajectories, and the implicit state variables were iteratively updated within a time step (30 days).

[0128] The prediction module sets dual failure thresholds: geometric failure threshold. With a crack width of 0.3 mm, the mechanical failure threshold is... Bearing capacity reduction factor =0.85 (ratio of remaining bearing capacity to design bearing capacity). Statistically, 95% of the trajectories trigger the failure threshold within the 240-310 day range, preliminarily determining the remaining life prediction range as [240, 310] days. The matching evaluation module retrieves measured crack data from the past 3 months, calculates the Mahalanobis distance D=0.12 between the distribution center trajectory line and the measured data, and substitutes it into the confidence decay function. The prediction reliability index was obtained as C=0.96.

[0129] The system's final output: "Predicted remaining lifespan is 240 to 310 days, with a confidence level of 96%." The current prediction confidence level is higher than the threshold (0.8), maintaining the original range width. The system recommends on-site verification on day 200. If the measured crack width exceeds 0.22 mm, a reinforcement plan must be initiated.

[0130] The system subsequently detected a transient wind pressure pulse (amplitude 15 kN, duration 0.2 s), which was identified as a pulsed load. The variance of the trajectory set probability distribution was immediately increased by a factor of 1.8, and the lifespan interval was recalculated from [240, 310] days to [220, 330] days. Based on measured data from the past 30 days, the system adjusted the energy release ratio coefficient. from Revised to This reflects the accelerating trend of material fatigue.

[0131] This embodiment achieves accurate prediction of the remaining lifespan of cracks through the dynamic evolution of digital twin models and the fusion of multi-source data, providing a basis for preventive maintenance of historical buildings.

[0132] In summary, this embodiment not only demonstrates the specific process of crack life prediction but also reflects the comprehensive value of this invention in improving the safety management level of building structures. By transforming discrete monitoring data into continuous knowledge of the evolution of safety status, it outputs the lifespan range and reliability index of building cracks, essentially providing a 'health status forecast' and 'risk barometer' for building structures. This is of great significance for protecting people's lives and property and realizing intelligent operation and maintenance of infrastructure.

[0133] Example 2:

[0134] like Figure 3 As shown, the building crack life prediction system based on digital twin and data fusion uses the aforementioned building crack life prediction method based on digital twin and data fusion, including:

[0135] The multi-source data acquisition module is used to acquire explicit crack status data, measured structural response data, and environmental and usage status data of the cracked area of ​​the building under test.

[0136] The twin modeling module is used to construct a digital twin model of building cracks based on explicit crack state data and measured structural response data.

[0137] The inversion generation module is used to invert and generate initial implicit state variables that characterize the internal physical properties of the crack by comparing the simulated response output by the digital twin model with the measured structural response data, and to configure the initial probability distribution interval for the initial implicit state variables.

[0138] The evolution-driven module is used to take environmental and usage status data as input, drive the implicit state variables and probability distribution intervals to evolve in the digital twin model according to the time step, and generate a set of implicit state evolution trajectories that characterize the crack state change process.

[0139] The prediction module is used to determine the prediction range of the remaining life of the crack based on the temporal distribution of the implicit state evolution trajectory set reaching the preset failure threshold in the future period.

[0140] The matching evaluation module is used to retrieve measured historical explicit crack state data as verification data; calculate the matching deviation between the verification data and the predicted crack state data at the corresponding historical time points of the implicit state evolution trajectory set; calculate the prediction confidence index based on the matching deviation; and output the prediction range of the crack remaining lifetime in association with the prediction confidence index.

[0141] Specifically, the multi-source data acquisition module integrates a high-definition industrial camera, a piezoelectric accelerometer, a fiber optic strain gauge, and an integrated meteorological monitoring station on the hardware side. The industrial camera is fixed above the crack area by a bracket to acquire explicit state data on the crack width and length; the strain gauge and accelerometer are bonded to key stress-bearing parts of the building beams and columns with structural adhesive to acquire measured structural response data; the meteorological monitoring station is used to collect real-time environmental temperature, humidity, and wind load. All hardware interfaces are unified and converged to an industrial edge computing gateway, which transmits the structured raw data stream to the server via wireless or wired networks.

[0142] The digital twin modeling module is configured in a high-performance computing center and runs on a cluster of general-purpose computing units (CPUs) and graphics processing units (GPUs). This module parametrically assembles the acquired explicit crack parameters and structural material parameters by calling the 3D modeling engine and finite element analysis kernel, generating a digital twin model with high-fidelity mechanical properties.

[0143] The inversion generation module, evolution-driven module, and prediction module constitute the core algorithm engine of the system. The inversion generation module utilizes a parallel computing framework to perform residual minimization calculations between the simulated and measured responses, thereby determining implicit physical parameters; the evolution-driven module, based on a numerical integration step algorithm, drives the crack model to evolve over time on a daily or load cycle basis; and the prediction module, through a statistical processing unit, estimates the density function at the moment when the evolution trajectory crosses the threshold, calculating the remaining lifetime interval.

[0144] The matching and evaluation module is associated with a historical database storage unit, which is used to retrieve monitoring records at historical moments. It calculates spatial distance through a logical comparison unit, and finally displays the prediction interval and credibility index on a graphical display terminal for visualization.

[0145] Taking the monitoring of cracks in the load-bearing beams of a historical brick-concrete building as an example: The system first identifies a crack width of 0.15mm using a high-definition camera (explicit data), while the accelerometer measures the beam's first-order frequency as 12Hz (measured response). The inversion generation module compares this with the theoretical frequency of 12.5Hz predicted by the digital twin model, and uses the minimum residual method to identify that the equivalent damage depth inside the crack has reached 1 / 4 of the beam height (implicit state variable).

[0146] Subsequently, the system input environmental forecast data (environmental data) for the next year, and the evolution-driven module generated 1000 random evolution trajectories. Within the interval of day 240 to day 310, the calculated width of 950 trajectories exceeded the geometric failure threshold of 0.3 mm. Therefore, the prediction module initially determined the remaining lifetime prediction range to be [240, 310] days. Finally, the matching evaluation module retrieved the measured crack width values ​​from the past three months and found that the Mahalanobis distance between the measured growth curve and the model distribution center trajectory line was extremely small, resulting in a prediction confidence index of 0.92. The system ultimately output to the operations and maintenance team: "Predicted remaining lifetime is 240 to 310 days, conclusion confidence level is 92%", with a suggestion to conduct on-site verification on day 200.

[0147] This system, through high-level integration of hardware and software, achieves a technological leap from passive monitoring to proactive early warning of building cracks. The hardware perception layer ensures real-time acquisition of explicit physical indicators and environmental stimuli, while the algorithm processing layer reveals the physical logic between the microscopic evolution and macroscopic failure within cracks through implicit state variable inversion and stochastic evolution trajectory generation. The system's integrated matching and evaluation module effectively utilizes historical data to dynamically validate the prediction model, solving the industry challenge of quantifying the predictive reliability of digital twin technology in engineering applications. The overall solution, by correlating the output prediction interval and credibility index, provides a high-precision, high-reliability decision-making basis for the preventative maintenance of buildings.

[0148] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for predicting the lifespan of buildings with cracks based on digital twins and data fusion, characterized in that, Includes the following steps: Acquire explicit crack status data, measured structural response data, and environmental and usage status data for the cracked area of ​​the building under test; Based on the explicit crack state data and the measured structural response data, a digital twin model of the building cracks is constructed. By comparing the simulated response output by the digital twin model with the measured structural response data, initial implicit state variables characterizing the internal physical properties of the crack are generated, and an initial probability distribution interval is configured for the initial implicit state variables. Using the environmental and usage status data as input, the implicit state variables and the initial probability distribution interval are driven to evolve in the digital twin model according to the time step, generating a set of implicit state evolution trajectories that characterize the crack state change process. Based on the temporal distribution of the implicit state evolution trajectory set reaching the preset failure threshold in the future time period, the prediction range of the crack's remaining lifetime is determined. Historical explicit crack state data measured in actual measurements were retrieved as verification data; Calculate the matching deviation between the verification data and the predicted crack state data of the implicit state evolution trajectory set at the corresponding historical time point, calculate the prediction confidence index based on the matching deviation, and output the prediction range of the crack remaining lifetime in association with the prediction confidence index.

2. The method for predicting the lifespan of buildings based on digital twins and data fusion according to claim 1, characterized in that, The specific steps for inverting and generating initial implicit state variables characterizing the internal physical properties of the crack are as follows: Based on the measured structural response data, the stiffness degradation characteristic value of the structure is calculated. Obtain the theoretical response sequence of the digital twin model under different preset crack depths and morphologies; By minimizing the residual between the measured structural response data and the theoretical response sequence, the equivalent damage depth and the weakened area of ​​the tensile zone inside the crack are identified as initial implicit state variables.

3. The method for predicting the lifespan of buildings based on digital twins and data fusion according to claim 1, characterized in that, The initial probability distribution interval is configured for the initial implicit state variables, and the configuration logic for the initial probability distribution interval is as follows: The deviations in instrument measurement noise and environmental interference during the acquisition of explicit crack state data were statistically analyzed. Based on the physical parameter sensitivity of the digital twin model, calculate the mapping weights of the initial implicit state variables to the deviation; Based on the mapping weights, a probability envelope surface representing the degree of dispersion of the variable distribution is generated, which serves as the initial probability distribution interval of the implicit state evolution trajectory set.

4. The method for predicting the lifespan of buildings based on digital twins and data fusion according to claim 1, characterized in that, The specific steps for generating the implicit state evolution trajectory set characterizing the crack state change process include: The temperature stress amplitude and load cycle number in the environmental and usage data are calculated and converted into incremental data of the equivalent energy release rate acting on the crack tip. At each time step, the incremental data of the equivalent energy release rate is used as input to update the parameters characterizing crack propagation in the implicit state variables; Based on the initial probability distribution interval, multiple implicit state evolution paths that diverge in time are generated through multi-sample random sampling simulation, forming the implicit state evolution trajectory set.

5. The method for predicting the lifespan of buildings based on digital twins and data fusion according to claim 1, characterized in that, Based on the temporal distribution of the implicit state evolution trajectory set reaching a preset failure threshold in future time periods, the preset failure threshold is defined using a multi-criteria fusion approach, specifically including: The geometric failure threshold is used to characterize the crack width in the explicit crack state data reaching a preset maximum safety limit. The mechanical failure threshold is used to characterize the boundary of the structural bearing capacity limit state when the structural bearing capacity weakening coefficient calculated in real time by the digital twin model reaches the preset bearing capacity limit state boundary. If any path in the set of implicit state evolution trajectories reaches any of the above thresholds, the lifetime corresponding to that path is determined to have ended.

6. The method for predicting the lifespan of buildings based on digital twins and data fusion according to claim 1, characterized in that, The specific steps for calculating the matching deviation between the verification data and the predicted crack state data of the implicit state evolution trajectory set at the corresponding historical time points, and calculating the prediction confidence index based on the matching deviation, include: Extract the center value of the probability distribution of the implicit state evolution trajectory set at the corresponding time point to form the distribution center trajectory line; Calculate the spatial distance between the verification data and the distribution center trajectory line; wherein the spatial distance is Euclidean distance or Mahalanobis distance; Substituting the spatial distance into a preset confidence decay function, the output value between 0 and 1 is used as the prediction confidence index.

7. The method for predicting the lifespan of buildings based on digital twins and data fusion according to claim 1, characterized in that, The method also includes abnormal disturbance identification, with specific steps including: Real-time monitoring of the environment and usage status data to check for pulsed load data exceeding the preset range; If pulsed load data is identified, the probability distribution range of the implicit state evolution trajectory set is increased, and the prediction range of the crack's remaining lifetime is redefined.

8. The method for predicting the lifespan of buildings based on digital twins and data fusion according to claim 1, characterized in that, The method also includes a closed-loop feedback mechanism for the continuous evolution of the digital twin model, with specific steps including: Periodically acquire the latest explicit crack condition data and measured structural response data; A Bayesian update algorithm is used to correct the key physical parameters in the digital twin model using the latest explicit crack state data and measured structural response data.

9. The method for predicting the lifespan of buildings based on digital twins and data fusion according to claim 1, characterized in that, The specific steps for correlating the predicted range of the remaining crack lifetime with the prediction confidence index include: The prediction range of the remaining life of the crack is corrected in real time based on the prediction confidence index. If the prediction confidence index is lower than a preset threshold, the coverage of the prediction interval will be automatically expanded.

10. A building crack life prediction system based on digital twin and data fusion, characterized in that, The method for predicting building crack life based on digital twin and data fusion as described in any one of claims 1 to 9 includes: The multi-source data acquisition module is used to acquire explicit crack status data, measured structural response data, and environmental and usage status data of the cracked area of ​​the building under test. The twin modeling module is used to construct a digital twin model of building cracks based on the explicit crack state data and the measured structural response data. The inversion generation module is used to invert and generate initial implicit state variables that characterize the internal physical properties of the crack by comparing the simulated response output by the digital twin model with the measured structural response data, and to configure an initial probability distribution interval for the initial implicit state variables. An evolution-driven module is used to take the environmental and usage status data as input, drive the implicit state variables and the probability distribution interval to evolve in the digital twin model according to the time step, and generate an implicit state evolution trajectory set that characterizes the crack state change process. The prediction module is used to determine the prediction range of the remaining life of the crack based on the temporal distribution of the implicit state evolution trajectory set reaching a preset failure threshold in the future time period. The matching evaluation module is used to retrieve measured historical explicit crack state data as verification data; calculate the matching deviation between the verification data and the predicted crack state data of the implicit state evolution trajectory set at the corresponding historical time points; calculate the prediction confidence index based on the matching deviation; and associate the predicted range of the crack remaining lifetime with the prediction confidence index for output.

Images