Building digital twin parameter self-calibration system based on creep inversion

Abstract

The application discloses a building digital twin parameter self-calibration system based on creep inversion, relates to the technical field of building engineering digital twin and structure simulation, and comprises successively connected conjugate anchor point generation units, a coupled gradient network construction unit, a working condition adaptive parameter generation unit and a network self-iteration verification unit.In the application, a gradient network architecture with three-way coupling of boundary conditions, parameters and residual errors is used, and a parameter generation logic with coupled coherence weighting is adopted, so that the parameter boundary dependency and residual error information of the original technology are converted into the core basis for calibration, the problem that the inversion parameters of the existing system are only adapted to specific working conditions, systematic prediction deviation easily occurs in cross-working conditions, and the power input and precision benefit are not matched is solved, adaptive and accurate adaptation of equivalent parameters in all working conditions is realized, and a passive identification effect of original structure abnormalities is derived.

Description

Technical Field

[0001] This invention relates to the field of digital twin and structural simulation technology for building engineering, specifically a self-calibration system for building digital twin parameters based on creep inversion. Background Technology

[0002] In the digital twin system of building engineering, the parameter self-calibration system based on creep inversion is a core technical module to ensure the accuracy of virtual-to-real mapping and support the long-term performance evaluation of the structure. Existing systems generally use on-site structural monitoring data and inversion algorithms to iteratively correct the calculation parameters related to concrete creep in order to reduce the deformation deviation between the twin model and the actual structure. This is a standard technical component for digital twins of large-span concrete structures and super high-rise buildings.

[0003] It is generally accepted in the industry that parameters such as creep coefficient, viscosity coefficient, and relaxation modulus obtained through self-calibration inversion correspond to the inherent physical properties of concrete materials and can be directly transferred and applied to full life cycle simulation calculations under different load conditions, environmental temperature and humidity, and structural age. Therefore, the optimization direction of existing technologies has always focused on improving the accuracy of inversion algorithms, densifying monitoring points, and increasing data sampling frequency. All technological iterations revolve around more accurately approximating the real material parameters.

[0004] However, both practical engineering applications and computational mechanics inverse problem theory show that all parameters obtained through self-calibration inversion are equivalent emergent values ​​formed by the coupling of finite element model assumptions, sensor observation dimensions, data preprocessing rules, and real material properties. They only possess fitting validity under specific observation boundaries and operating conditions during inversion execution. When structural load distribution, environmental constraints, or observation locations change, the calibrated parameters will exhibit systematic prediction biases, failing to guarantee the reliability of simulation calculations throughout the entire lifecycle. For a long time, this type of cross-condition prediction failure has been broadly attributed to insufficient accuracy of the inversion algorithm or inadequate monitoring data coverage, failing to recognize that the root cause lies in the inversion parameters themselves lacking independent inherent physical meaning. All existing self-calibration processes essentially only achieve optimal fitting of historical monitoring data, rather than identifying the true physical parameters of the material. This cognitive blind spot constitutes the fundamental bottleneck that current self-calibration technology struggles to overcome. This leads to existing systems continuously investing significant computing power and hardware costs to improve single-point inversion accuracy, yet failing to obtain matching full-condition prediction benefits, severely restricting the practical value of digital twin systems in scenarios such as long-term structural performance evaluation, operation and maintenance decision support, and disaster risk prediction. Therefore, a self-calibration system for building digital twin parameters based on creep inversion is provided to overcome the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a self-calibration system for building digital twin parameters based on creep inversion, so as to solve the problems mentioned in the background art.

[0006] To address the aforementioned technical issues, this invention provides a self-calibration system for building digital twin parameters based on creep inversion. This system integrates on-site structural monitoring data with a digital twin finite element simulation model. It corrects concrete creep-related calculation parameters through an inversion algorithm to reduce the deformation deviation between the twin model and the actual structure. The system includes a conjugate anchor point generation unit, a coupled gradient network construction unit, a working condition adaptation parameter generation unit, and a network self-iterative verification unit connected in sequence. The conjugate anchor point generation unit is used to extract the boundary condition vector under steady-state conditions, output the equivalent parameter set and extract the corresponding residual distribution vector, and bind the boundary condition vector, the equivalent parameter set and the residual distribution vector into conjugate anchor points for storage. The coupled gradient network building unit uses multiple sets of conjugate anchor points as base nodes and the dimension of the boundary condition vector as the coordinate space to construct a coupled gradient network in which the boundary conditions, parameters, and residuals are linked. The working condition adaptation parameter generation unit is used to extract the boundary condition vector of the target working condition, filter the neighboring anchor points in the coupled gradient network, calculate the coupling weighting coefficient by integrating the boundary distance difference and the residual distribution similarity, and generate the equivalent parameter set of the target working condition to be substituted into the simulation model. The network self-iterative verification unit is used to compare the actual residual distribution and the predicted residual distribution under the target working condition, and to complete the self-iterative update and structural anomaly determination of the coupled gradient network.

[0007] Furthermore, the conjugate anchor point generation unit adopts a rolling window verification mechanism to determine the stable working condition. The criteria for determining the stable working condition are that the structural load fluctuation is less than 5% within 24 consecutive hours, the ambient temperature and humidity change is less than 3%, and there is no temporary construction load interference. The boundary condition vector includes six quantitative dimensions: vertical load distribution coefficient, horizontal load direction, mean ambient temperature, mean ambient relative humidity, structural stress age, and normalized values ​​of spatial coordinates of observation points. Each dimension has been normalized without dimension. The residual distribution vector contains the relative deformation deviation value and deviation direction of all observation points. The relative deviation value is the difference between the model calculated value and the monitored value divided by the total deformation of the corresponding measurement point.

[0008] Furthermore, the coupled gradient network uses a KD tree topology to complete the spatial organization of anchor points. The KD tree is constructed by recursively dividing nodes using the maximum variance partitioning principle. All anchor points retain their original calculated values ​​and are not averaged or smoothed. After the coupled gradient network building unit completes the validity verification of the conjugate anchor points, it calculates the boundary change rate and parameter change rate of adjacent anchor points, and marks the region where the parameter change rate exceeds three times the boundary change rate as the coupling characteristic transformation node.

[0009] Furthermore, the coupled gradient network building unit uses a global sensitivity analysis method to complete the self-learning of the coupling weights of each boundary dimension. The values ​​of the other five boundary dimensions are fixed, and the average rate of change of the creep coefficient when the value of a single dimension changes is statistically analyzed. The proportion of the rate of change of each dimension to the total rate of change is used as the coupling weight of the corresponding dimension, and the sum of all weights equals 1. The coupling weights are dynamically updated as the number of anchor points in the network increases, adapting to the parameter sensitivity characteristics of different structures.

[0010] Furthermore, the working condition adaptation parameter generation unit uses a KD tree search to filter the five sets of conjugate anchor points that are closest to the target working condition boundary to form a set of neighboring anchor points; The working condition adaptation parameter generation unit first calculates the boundary normalized distance between the target working condition and each neighboring anchor point, then calculates the residual distribution coherence coefficient between the target working condition and each neighboring anchor point, merges the boundary normalized distance and the residual distribution coherence coefficient to calculate the coupling weighting coefficient of each neighboring anchor point, and obtains the equivalent parameter set of the target working condition by weighted summation based on the coupling weighting coefficient.

[0011] Furthermore, the network self-iterative verification unit uses the residual fitting determination coefficient to quantify the degree of agreement between the actual residual distribution and the predicted residual distribution; When the residual fitting determination coefficient is higher than the set threshold, the corresponding boundary condition vector, equivalent parameter set, and true residual distribution vector form a new conjugate anchor point, which is automatically added to the coupled gradient network. The network synchronously performs anchor point deduplication operation. For multiple sets of anchor points with boundary condition vector differences less than the set threshold, only the set with the highest residual fitting determination coefficient is retained.

[0012] Furthermore, when the residual fitting determination coefficient is lower than the set threshold, the network self-iterative verification unit determines that the structural deformation deviates from the normal creep evolution law and triggers a structural anomaly prompt. The anomaly prompt includes the location of the measurement point with the largest deviation, the deviation magnitude, and the deviation direction. Structural anomaly determination does not require additional damage identification algorithms or dedicated monitoring hardware; it is accomplished through consistency judgment of normal creep evolution patterns.

[0013] Furthermore, when the addition or removal of monitoring points in the structure leads to a change in the observation dimension, the system only needs to supplement the conjugate anchor points under the corresponding observation dimension to complete the adaptation, and the original anchor point data remains valid without the need for overall recalibration. Coupled gradient networks do not presuppose the functional form and variation law of parameter evolution. They are based entirely on the conjugate anchor points accumulated in actual engineering, and are suitable for different types of concrete structures and finite element models of different precision.

[0014] Compared with the prior art, the beneficial effects of the present invention are: Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Addressing the fundamental issue of cross-condition failure prediction at its root. The system completely abandons the traditional approach of using general material parameters, generating a dedicated set of equivalent parameters for each type of operating condition. Parameter adaptability is unaffected by load migration, environmental changes, age increases, or adjustments to observation locations. Through a parameter generation logic that couples boundary conditions and residuals in two dimensions, the equivalent parameter set automatically compensates for the inherent deviations between the finite element model and the observation system. This fundamentally improves the reliability of simulation calculations throughout the entire lifecycle, significantly reducing the virtual-to-real deviation in cross-condition predictions compared to existing technologies. This completely breaks through the industry dilemma of existing technologies that are extremely accurate in fitting historical data but have significant deviations in predicting future operating conditions.

[0015] 2. The overall computational power consumption of the system is significantly reduced. This system does not optimize the accuracy of the inversion algorithm itself, and does not require a large investment of computing power to pursue the ultimate accuracy of single-point inversion. It only needs to ensure the accuracy of the boundary conditions and parameter correspondence of the conjugate anchor points. At the same time, the self-iterative process of the coupled gradient network does not need to repeat the complete inversion calculation. The conjugate anchor points can be supplemented simply by comparing the prediction results with the monitoring data. The overall computational power consumption is significantly reduced compared with existing high-precision inversion systems, which can significantly reduce the hardware configuration threshold and operating cost of digital twin systems.

[0016] 3. Possesses native passive structural anomaly identification capabilities. This capability is not achieved through the addition of extra functional modules, but rather is a technical effect naturally derived from the coupled gradient network system, requiring no additional damage identification algorithms or dedicated monitoring hardware. The system can detect structural deformation anomalies caused by non-creep factors simply by comparing the coupled gradient patterns of normal creep evolution. This achieves a native fusion of parameter self-calibration and structural safety early warning functions, expanding the functional boundaries of the parameter self-calibration system. This capability cannot be obtained through conventional technological iterations under existing technical systems.

[0017] 4. It possesses the ability to smoothly adapt to changes in the observation system. When monitoring points are added or removed from the structure, or when the observation scheme is adjusted, resulting in changes in the observation dimensions, the existing system requires complete recalibration, rendering the original parameters completely invalid. This system only needs to supplement the conjugate anchor points under the corresponding observation dimensions to complete the adaptation. The original anchor point data can continue to be used normally without the need for overall recalibration, completely avoiding the calibration cost losses caused by changes in the observation system, and adapting to the actual needs of dynamic adjustments to the observation scheme throughout the entire life cycle of the project.

[0018] 5. Possesses extremely strong structural generalization ability. The system does not rely on a specific creep constitutive model, nor does it make any prior assumptions about the evolution of parameters. It operates entirely based on the conjugate anchor points accumulated from actual monitoring data, making it adaptable to different forms of concrete structures such as long-span beam bridges, super high-rise buildings, and large-space shells. It can also adapt to finite element models of different accuracy levels and monitoring systems with different deployment schemes, without requiring customized modifications for specific scenarios, and has broad engineering application value. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the self-calibration system for building digital twin parameters based on creep inversion, as per the present invention. Detailed Implementation

[0020] 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.

[0021] Please see Figure 1 The present invention provides a technical solution: See Figure 1 As shown, an embodiment of a self-calibration system for building digital twin parameters based on creep inversion is presented: This embodiment is a self-calibration system for building digital twin parameters based on creep inversion. Addressing the fundamental bottleneck of cross-condition prediction failure commonly found in existing technologies, it completely departs from existing technologies that focus on approximating the inherent physical parameters of concrete creep. From the root of computational logic, it acknowledges that the inversion parameters are equivalent emergent values ​​coupled with finite element model assumptions, the observation system, and boundary conditions. It transforms the parameter boundary dependence, condition discreteness, and residual fluctuations—considered error defects in existing technologies—into core, effective information supporting system operation. By constructing a gradient network system with three-dimensional coupling of boundary conditions, parameters, and residuals, it achieves adaptive and accurate generation of equivalent parameters under different conditions, eliminating the systematic bias in the cross-condition application of general parameters. The system does not aim to obtain independent true material parameters; all parameters exist only as a prerequisite for corresponding boundary conditions and residual characteristics. The parameter calibration process simultaneously completes the full-domain calibration of the normal creep evolution law of the structure, resulting in a native structural anomaly identification capability that was impossible with existing technologies.

[0022] The system consists of a conjugate anchor point generation unit, a coupled gradient network construction unit, a working condition adaptation parameter generation unit, and a network self-iterative verification unit, connected sequentially. These units communicate and interact logically to jointly complete a full self-calibration process from basic data acquisition to parameter generation and then to closed-loop iteration. The specific implementation process and technical details of each unit are as follows: The first step involves the conjugate anchor point generation unit performing basic load condition calibration. This unit simultaneously connects to real-time data from the structural site monitoring system and the finite element simulation model of the digital twin platform, initiating the parameter inversion process only when the structure is in a stable load condition. A rolling window verification mechanism is used to determine the stable load condition. The unit collects monitoring data from all points every hour, continuously calculating the structural load fluctuation amplitude and environmental temperature and humidity changes over the past 24 hours. If the threshold requirements are met for three consecutive collection cycles, and no temporary construction loads, sudden external forces, or other disturbances are recorded during the period, the structure is considered to have entered a stable load condition, and the parameter inversion calculation officially begins. The criteria for determining a stable load condition are that the structural load fluctuation amplitude is less than 5% and the environmental temperature and humidity change amplitude is less than 3% over 24 consecutive hours, ensuring that the inverted parameters correspond to the steady-state characteristics of the current load condition.

[0023] Before inversion calculations begin, the unit first extracts the boundary condition vector for the current operating condition. This vector contains six quantized dimensions, each of which is normalized to eliminate dimensional differences. All calculation results are dimensionless values. Specifically, the vertical load distribution coefficient is the spatial variation coefficient of the measured load values ​​from several load zones divided into sections on the current floor, reflecting the uniformity of load distribution. It is calculated as the ratio of the standard deviation of the load values ​​to the arithmetic mean of the load values, with a normalization interval of 0 to 1. The horizontal load orientation is the cosine of the angle between the resultant horizontal force and the longitudinal principal axis of the structure, reflecting the direction of the horizontal load, with a normalization interval of -1 to 1. The mean ambient temperature is the daily average ambient temperature for the current period, linearly normalized using the historical extreme maximum and minimum temperatures during the structure's service life as upper and lower limits, mapped to the 0 to 1 interval. The mean ambient relative humidity is the daily average relative humidity for the current period, directly normalized by dividing the percentage value by 100%, mapped to the 0 to 1 interval. The structural stress age is calculated by taking the number of days from the completion of the structure's construction to the current moment, and normalizing it using a natural logarithmic transformation. The normalized value is then calculated using the natural logarithmic value of the structure's design service life in days as the denominator, matching the nonlinear evolution characteristics of creep over age. The normalized values ​​of the spatial coordinates of the observation points are taken as the three-dimensional relative coordinates of each measuring point in the overall structural coordinate system, and normalized using the total dimensions of the structure in the corresponding direction as a reference, reflecting the influence of the observation location on the inversion results.

[0024] After the boundary condition vector is extracted, the unit initiates creep parameter inversion calculation using the current monitoring data, outputting an equivalent parameter set including the creep coefficient, viscosity coefficient, and relaxation modulus. The inversion calculation is implemented using a conventional iterative optimization algorithm. This system does not optimize the accuracy of the inversion algorithm itself; the core lies in the organization and application logic of the inversion results.

[0025] After the inversion calculation is completed, the unit further extracts the residual distribution vector between the finite element model calculation values ​​and the field monitoring values ​​under the current equivalent parameter set. The residual distribution vector contains the deformation deviation values ​​and directions of all observation points. The residual value of each measuring point is in the form of relative deviation, calculated by dividing the difference between the model calculation value and the monitoring value by the total deformation of that measuring point, thus eliminating the influence of differences in deformation magnitude between different measuring points. The residual distribution vector is not used as an evaluation index to judge the accuracy of the inversion, but rather as an information carrier of inherent systematic deviations such as finite element model assumption biases, observation system errors, and boundary condition simplification errors. It is an indispensable component of the anchor point system.

[0026] The unit ultimately binds the boundary condition vector, equivalent parameter set, and residual distribution vector into a set of conjugate anchor points for storage. These three elements together constitute the smallest effective unit of the calibration system. There are no effective parameters that exist independently of the boundary conditions and residual characteristics. This step overturns the industry-default assumption that parameters have independent physical meaning from the data level, integrating the previously fragmented operating conditions, calculation parameters, and calculation errors into a unified conjugate whole, providing the basic data unit for the subsequent construction of coupled gradient networks.

[0027] The second step involves the coupled gradient network construction unit, which completes the topology building of the calibration system. Using multiple sets of conjugate anchor points output by the conjugate anchor point generation unit as base nodes, and the six dimensions of the boundary condition vector as the coordinate space, the coupled gradient network constructs a network linking the boundary condition parameters and residuals. The network employs a KD-tree topology to organize the anchor points spatially, improving the retrieval efficiency for subsequent working condition matching. All anchor points retain their original calculated values ​​without any form of mean or smoothing. The KD-tree construction uses the maximum variance partitioning principle, selecting the boundary dimension with the largest variance in the current node set as the partitioning axis, and taking the median anchor point under that dimension as the partitioning node. This process recursively divides the left and right child node sets until all anchor points are stored in a tree-like index. Each tree node synchronously stores the boundary condition vector, equivalent parameter set, and residual distribution vector of the corresponding conjugate anchor point, supporting direct retrieval of complete data during retrieval.

[0028] During network construction, the unit first verifies the validity of all conjugate anchor points, eliminating invalid anchor points that fail to meet stability standards or have missing monitoring data, ensuring that all nodes in the network correspond to real and stable operating conditions. After verification, the unit calculates the boundary change rate and parameter change rate between each group of adjacent anchor points. When the parameter change rate of adjacent anchor points exceeds three times the boundary change rate, the region is marked as a coupling characteristic transition node. Coupling characteristic transition nodes correspond to critical operating conditions where the structural stress system undergoes a fundamental change, including scenarios such as load distribution exceeding the stress blind zone and temperature stress offsetting part of the self-weight stress. The existence of such nodes directly reflects the nonlinear coupling between parameters and boundaries.

[0029] The coupled gradient network synchronously completes the self-learning calculation of coupling weights for each boundary dimension, providing a weighting basis for subsequent parameter generation. The coupling weight self-learning employs a global sensitivity analysis method. Based on data from all effective anchor points within the network, it calculates the creep parameter change magnitude corresponding to independent changes in each boundary dimension, using the proportion of parameter sensitivity for each dimension as the corresponding coupling weight. Specifically, the calculation process involves fixing the values ​​of the other five boundary dimensions, calculating the average rate of change of the creep coefficient when the value of a single dimension changes, and repeating this calculation for all six boundary dimensions. The proportion of each dimension's change rate to the total change rate of the six dimensions is used as the corresponding coupling weight, with the sum of all weights equal to 1. The coupling weights are dynamically updated as the number of anchor points in the network increases, adapting to the parameter sensitivity characteristics of different structures without requiring manual adjustment. The physical basis of the coupling weights is that the sensitivity of the creep parameter to different boundary factors inherently differs; dimensions with higher sensitivity have a greater impact on parameter adaptation results. Therefore, weighted distance is used instead of ordinary Euclidean distance to improve the accuracy of parameter matching.

[0030] In the existing technical system, the discrete differences in parameters obtained under different operating conditions were always considered as inversion errors and needed to be eliminated through mean filtering, regression fitting, function approximation, etc., ultimately yielding a set of so-called true material parameters. This system completely abandons this approach, using the parameter discreteness and gradient variation characteristics under different operating conditions as a true representation of the coupling effect between boundary conditions and the model system, directly using them as topological features of the gradient network. The error defects that the original technology sought to eliminate are transformed into the core basis for accurate adaptation to operating conditions in this system, a technical logic completely opposite to the industry's conventional thinking.

[0031] Coupled gradient networks do not presuppose any functional form of parameter evolution, nor do they make any prior assumptions about the variation of parameters with boundary conditions. They rely solely on conjugate anchor points accumulated in actual engineering to truly reflect the coupling relationship between parameters and boundary conditions. This construction method completely breaks away from the limitations of existing technologies that rely on creep constitutive models. It does not require pre-assuming the variation of parameters, adapts to calibration needs under different structural forms, material properties, and observation systems, and has extremely strong generalization capabilities.

[0032] The third step involves the dynamic parameter adaptation unit for the target working condition. When the digital twin system initiates a simulation calculation request for the target working condition, the working condition adaptation parameter generation unit first extracts the boundary condition vector corresponding to the target working condition and maps it to the six-dimensional coordinate space of the coupled gradient network. The unit first uses a KD-tree search to select the five sets of conjugate anchor points closest to the boundary of the target working condition, forming a set of neighboring anchor points as the basic reference nodes for parameter generation.

[0033] To achieve dual-dimensional coupling and adaptation of boundary conditions and residual characteristics, the working condition adaptation parameter generation unit adopts a coupled coherent weighted parameter generation logic. It completes the accurate generation of equivalent parameters through hierarchical progressive quantization calculations. The specific calculation logic is as follows: First, the normalized boundary distance between the target working condition and each neighboring anchor point is calculated to quantify the degree of boundary difference between working conditions. The normalized boundary distance is not a simple Euclidean distance, but rather incorporates coupling weights for each dimension to reflect the difference in the influence of different boundary dimensions on the creep parameter. The quantification expression is as follows: ; In the formula, For the target working condition and the first The boundary normalized distance of the neighboring anchor points is dimensionless. For the target working condition Normalized values ​​of each boundary dimension, with dimensions being dimensionless; For the first The nearest anchor point Normalized values ​​of each boundary dimension, with dimensions being dimensionless; For the first The coupling weights of each boundary dimension are dimensionless, and the sum of all weights satisfies... The physical meaning of this formula is that, in a six-dimensional boundary condition space, the weighted spatial distance between the target working condition and its neighboring anchor points is calculated using parameter sensitivity as the weight. The smaller the distance, the higher the similarity of the boundary conditions between the two. The physical basis of the formula is that the creep parameter has different sensitivities to different boundary factors. Dimensions with higher sensitivity have a greater weight in influencing parameter adaptation, and the weighted distance more accurately reflects the degree of parameter reference between working conditions.

[0034] After obtaining the boundary normalized distance, the element first uses the inverse distance weighting method to perform preliminary parameter interpolation, obtaining the preliminary equivalent parameter set for the target working condition. This set is then substituted into the creep calculation module of the finite element model to run steady-state creep simulation calculations, obtaining the predicted deformation values ​​at each observation point. The relative deviation between the predicted values ​​and the corresponding monitoring benchmark values ​​is used as the preliminary prediction residual, forming the preliminary prediction residual distribution for the target working condition. Preliminary interpolation is only an intermediate calculation step used to obtain the initial predicted values ​​of the residual distribution; it is not the final parameter output result.

[0035] Based on this, the coherence coefficient of the residual distribution between the target working condition and each adjacent anchor point is calculated to quantify the similarity of the system deviation distribution under different working conditions. The residual distribution carries the inherent deviation information of the model and the observation system. The higher the distribution similarity, the closer the system deviation characteristics between the working conditions are, and the stronger the parameter transferability. The quantification expression is: ; In the formula, For the target working condition and the first The coherence coefficient of the residual distribution of the neighboring anchor points is dimensionless and ranges from -1 to 1. The closer the value is to 1, the higher the similarity of the residual distribution. For the $i$-th neighboring anchor point at the th The residual values ​​of each observation point are dimensionless. For the first The average residual of all measuring points of a neighboring anchor point is dimensionless. The first result obtained from the preliminary simulation calculation of the target working condition The predicted residual values ​​for each observation point are dimensionless. The average predicted residual of all measuring points under the target working condition is dimensionless. This represents the total number of observation points deployed on the structure. The physical meaning of this formula is to calculate the spatial correlation of two sets of residual distributions using the Pearson correlation coefficient, measuring the degree of consistency in the distribution of system deviations under different operating conditions. The physical basis of the formula is that the assumed deviations of the finite element model and the measurement deviations of the observation system possess spatial distribution stability; the spatial distribution of residuals under similar boundary conditions exhibits high consistency; and a higher correlation coefficient indicates that the system deviation characteristics of the operating conditions are closer, resulting in better deviation compensation after parameter adaptation.

[0036] After obtaining the boundary normalized distance and the coherence coefficient of the residual distribution, the coupling weighting coefficient for each neighboring anchor point is further calculated. This fuses the information from the two dimensions of boundary distance difference and residual distribution similarity to obtain the contribution weight of each anchor point to the target parameter. The quantification expression for the coupling weighting coefficient is as follows: ; In the formula, For the first The coupling weighting coefficients of the neighboring anchor points are dimensionless, and the sum of all weights satisfies ; The filtered set of neighboring anchor points; This is a distance-inverse weighting term, reflecting the basic principle that the closer the boundary distance, the higher the reference value of the anchor point. The dimension is dimensionless. The coherence adjustment factor maps the coherence coefficient from -1 to 1 to 0 to 1, reflecting the principle that the more similar the residual distribution, the higher the reference value of the anchor point. It is dimensionless. The two terms are multiplied and then normalized to obtain the final coupling weighting coefficient. The physical meaning of this formula is to integrate information from two dimensions: boundary condition similarity and residual distribution similarity, to calculate the contribution weight of each neighboring anchor point to the calculation of the target operating condition parameters. The physical basis of the formula is that similar boundary conditions ensure the consistency of the overall trend of parameter evolution, and similar residual distributions ensure the consistency of the compensation effect for system deviations. Both together determine the reference value of the anchor point; the dual-dimensional weighting simultaneously considers the accuracy of parameter trends and the compensation effect for system deviations.

[0037] Finally, based on the coupling weighting coefficient, the parameter values ​​of each neighboring anchor point are weighted and summed to obtain the final equivalent parameter values ​​under the target working condition. The quantification expression is as follows: ; In the formula, The target equivalent parameter values ​​under the target working conditions correspond to three types of parameters: creep coefficient, viscosity coefficient, and relaxation modulus. The dimensions of each parameter are consistent with the inherent dimensions of the corresponding creep parameter. For the first The corresponding parameter values ​​for each neighboring anchor point have the same dimensions as the target parameters. By performing the above calculations on the three types of parameters respectively, a complete set of equivalent parameters for the target working condition can be obtained, which can then be directly substituted into the finite element model to complete the simulation calculation.

[0038] The goal of this step is not to obtain universally applicable true physical parameters of the material, but rather to obtain optimal equivalent parameters that perfectly match the current model system, observation system, and boundary conditions. Equivalent parameters naturally carry compensation for inherent biases between the model and the observation system, resulting in a significantly higher degree of virtual-to-real matching in simulation calculations compared to those based on true material parameters. This process completely breaks free from the technical inertia of pursuing true parameter values, fundamentally avoiding systematic prediction biases when applying general parameters across different operating conditions.

[0039] Existing parameter interpolation methods rely solely on boundary or operating conditions, never incorporating residual distribution characteristics into the interpolation weighting system. This is because existing technologies treat residuals as errors, failing to recognize their role in carrying valid information. This system introduces the residual distribution coherence coefficient as a second-dimensional weight, achieving a two-dimensional adaptation between boundary differences and system bias.

[0040] The fourth step involves the network self-iterative verification unit completing the closed-loop update of the calibration system. After each target working condition simulation calculation is completed and the corresponding time period's entity structure monitoring data is obtained, the network self-iterative verification unit automatically starts the verification process, comparing the actual residual distribution obtained from the monitoring with the residual distribution predicted by the coupled gradient network to complete the parameter validity verification and network update.

[0041] The validation process uses the residual fitting determination coefficient as the core criterion to quantify the degree of agreement between the actual residual distribution and the network's predicted distribution. The quantification expression is as follows: ; In the formula, The coefficient of determination for residual fitting is the anomaly detection index. It is dimensionless and has an upper limit of 1. The closer the value is to 1, the higher the degree of agreement between the predicted residual and the actual residual. The first one obtained from actual monitoring The true residual value of each measuring point is dimensionless. The first prediction obtained by the coupled gradient network The residual values ​​at each measuring point are dimensionless. The coefficient of determination (COP) represents the average of the true residuals at all measuring points, and is dimensionless. The physical meaning of this formula is to measure the degree to which the predicted residuals explain the true residuals, reflecting whether the deformation pattern under the current working condition conforms to the normal creep evolution pattern represented by the coupled gradient network. The physical basis of the formula is that under normal creep evolution, the residual distribution follows the coupled change pattern of boundary conditions and parameters, and the predicted values ​​have a high degree of agreement with the true values; when non-creep deformation factors exist, the residual distribution deviates from the normal pattern, and the agreement decreases significantly.

[0042] When the residual fitting determination coefficient is higher than a set threshold, the structural deformation under the current operating condition is determined to conform to the normal creep evolution law. The corresponding boundary condition vector, equivalent parameter set, and true residual distribution vector constitute new conjugate anchor points, which are automatically added to the coupled gradient network to achieve self-growth and self-iteration of the network coverage. The network synchronously performs anchor point deduplication. For multiple sets of anchor points with boundary condition vector differences less than a set threshold, only the set with the highest residual fitting determination coefficient is retained to avoid redundant nodes in the network and ensure retrieval efficiency and calculation accuracy. This process does not require manual intervention or additional full inversion calculation. It can be completed simply by comparing the prediction results with the monitoring data, and the computational power consumption is far lower than the traditional successive inversion calibration mode. As the service time increases, the number of anchor points in the coupled gradient network will continue to increase, the coverage of operating conditions will continue to expand, and the parameter adaptation accuracy will also improve synchronously, forming a positive self-reinforcing closed loop.

[0043] When the residual fitting determination coefficient falls below a set threshold, the current structural deformation is determined to deviate from the normal creep evolution pattern, triggering a structural anomaly alert. The anomaly alert includes the location of the measurement point with the largest deviation, the magnitude of the deviation, and the direction of the deviation, assisting maintenance personnel in quickly locating the abnormal area. This anomaly identification capability is not an additional functional module added to the system, nor is it equipped with dedicated damage identification algorithms and monitoring hardware. Rather, it is a naturally occurring derivative effect of the coupled gradient network, achieved solely through consistency judgment of the normal creep evolution pattern. When the structure experiences deformations not caused by creep, such as microcrack propagation, localized damage accumulation, or internal force redistribution, the residual distribution will deviate from the normal coupled gradient pattern, thus being passively identified by the system.

[0044] The iterative logic of this step also differs from the calibration approach of existing technologies. Existing technologies always aim to reduce the deviation of single-point parameters in their iterative optimization, while the iteration of this system aims to expand the working condition coverage of the coupled gradient network. It does not pursue the ultimate accuracy of a single parameter, but rather improves the adaptability to all working conditions by expanding the network size. The optimization directions of the two are fundamentally different.

[0045] Summarize: First, it fundamentally solves the underlying problem of cross-condition failure prediction in existing systems. The system completely abandons the traditional approach of using general material parameters, generating a dedicated set of equivalent parameters for each type of operating condition. Parameter adaptability is unaffected by load migration, environmental changes, age increases, or adjustments to observation locations. Through a parameter generation logic coupled with boundary residuals in two dimensions, the equivalent parameters automatically compensate for the inherent deviations between the model and the observation system. This significantly improves the reliability of simulation calculations throughout the entire lifecycle, reducing the virtual-to-real deviation in cross-condition predictions compared to existing technologies. It completely breaks through the industry dilemma of existing technologies that are extremely accurate in fitting historical data but have significant deviations in predicting future operating conditions.

[0046] Second, the overall computing power consumption of the system is significantly reduced. This system does not optimize the accuracy of the inversion algorithm itself, and does not need to invest a lot of computing power to pursue the ultimate accuracy of single-point inversion. It only needs to ensure the accuracy of the boundary and parameter correspondence of the conjugate anchor points. At the same time, the network's self-iterative process does not need to repeat the complete inversion calculation. Anchor point supplementation can be completed simply by comparing the prediction results with the monitoring data. The overall computing power consumption is lower than that of existing high-precision inversion systems, which greatly reduces the hardware configuration threshold and operating cost of digital twin systems.

[0047] Third, it possesses a native passive structural anomaly identification capability. This capability is not achieved through the addition of extra functional modules, but rather is a technical effect naturally derived from the coupled gradient network system, requiring no additional damage identification algorithms or dedicated monitoring hardware. The system can detect structural deformation anomalies caused by non-creep factors simply by comparing the coupled gradient patterns of normal creep evolution. This achieves a native fusion of parameter self-calibration and structural safety early warning functions, expanding the functional boundaries of the parameter self-calibration system. This capability cannot be obtained through conventional technological iterations under existing technical systems.

[0048] Fourth, it possesses the ability to smoothly adapt to changes in the observation system. When monitoring points are added or removed from the structure, or when the observation scheme is adjusted, resulting in changes in the observation dimensions, the existing system needs to be completely recalibrated, rendering the original parameters invalid. This system only needs to supplement the conjugate anchor points under the corresponding observation dimensions, while the original anchor point data continues to be used normally without the need for overall recalibration. This completely avoids the calibration cost losses caused by changes in the observation system and adapts to the actual needs of dynamic adjustments to the observation scheme throughout the entire life cycle of the project.

[0049] Fifth, it possesses extremely strong structural generalization capabilities. The system does not rely on a specific creep constitutive model, nor does it make any prior assumptions about the evolution of parameters. It operates entirely based on the conjugate anchor points accumulated from actual monitoring data, making it suitable for various forms of concrete structures such as long-span beam bridges, super high-rise buildings, and large-space shells. It is also compatible with finite element models of different accuracy levels and monitoring systems with different deployment schemes. No customized modifications are required for specific scenarios, making it valuable for a wide range of engineering applications.

Claims

1. A self-calibration system for building digital twin parameters based on creep inversion, which integrates on-site structural monitoring data and a digital twin finite element simulation model, corrects concrete creep-related calculation parameters through an inversion algorithm to reduce the deformation deviation between the twin model and the solid structure, characterized in that... The system includes a conjugate anchor point generation unit, a coupled gradient network construction unit, a working condition adaptation parameter generation unit, and a network self-iterative verification unit connected in sequence. The conjugate anchor point generation unit is used to extract the boundary condition vector under steady-state conditions, output the equivalent parameter set and extract the corresponding residual distribution vector, and bind the boundary condition vector, the equivalent parameter set and the residual distribution vector into conjugate anchor points for storage. The coupled gradient network building unit uses multiple sets of conjugate anchor points as base nodes and the dimension of the boundary condition vector as the coordinate space to construct a coupled gradient network in which the boundary conditions, parameters, and residuals are linked. The working condition adaptation parameter generation unit is used to extract the boundary condition vector of the target working condition, filter the neighboring anchor points in the coupled gradient network, calculate the coupling weighting coefficient by integrating the boundary distance difference and the residual distribution similarity, and generate the equivalent parameter set of the target working condition to be substituted into the simulation model. The network self-iterative verification unit is used to compare the actual residual distribution and the predicted residual distribution under the target working condition, and to complete the self-iterative update and structural anomaly determination of the coupled gradient network.

2. The self-calibration system for building digital twin parameters based on creep inversion as described in claim 1, characterized in that: The conjugate anchor point generation unit uses a rolling window verification mechanism to determine the stable working condition. The criteria for determining the stable working condition are that the structural load fluctuation is less than 5% within 24 consecutive hours, the ambient temperature and humidity change is less than 3%, and there is no temporary construction load interference. The boundary condition vector includes six quantitative dimensions: vertical load distribution coefficient, horizontal load direction, mean ambient temperature, mean ambient relative humidity, structural stress age, and normalized values ​​of spatial coordinates of observation points. Each dimension has been normalized without dimension. The residual distribution vector contains the relative deformation deviation value and deviation direction of all observation points. The relative deviation value is the difference between the model calculated value and the monitored value divided by the total deformation of the corresponding measurement point.

3. The self-calibration system for building digital twin parameters based on creep inversion as described in claim 1 or 2, characterized in that: The coupled gradient network uses a KD tree topology to organize the anchor points spatially. The KD tree is constructed by recursively dividing the nodes according to the maximum variance partitioning principle. All anchor points retain their original calculated values ​​and are not averaged or smoothed. After the coupled gradient network building unit completes the validity verification of the conjugate anchor points, it calculates the boundary change rate and parameter change rate of adjacent anchor points, and marks the region where the parameter change rate exceeds three times the boundary change rate as the coupling characteristic transformation node.

4. The self-calibration system for building digital twin parameters based on creep inversion as described in claim 3, characterized in that: The coupled gradient network building unit uses a global sensitivity analysis method to complete the self-learning of the coupling weights of each boundary dimension. The values ​​of the other five boundary dimensions are fixed, and the average rate of change of the creep coefficient when the value of a single dimension changes is statistically analyzed. The proportion of the rate of change of each dimension to the total rate of change is used as the coupling weight of the corresponding dimension, and the sum of all weights equals 1. The coupling weights are dynamically updated as the number of anchor points in the network increases, adapting to the parameter sensitivity characteristics of different structures.

5. The self-calibration system for building digital twin parameters based on creep inversion as described in claim 1 or 2, characterized in that: The working condition adaptation parameter generation unit uses a KD tree search to filter the five sets of conjugate anchor points that are closest to the boundary of the target working condition, forming a set of neighboring anchor points; The working condition adaptation parameter generation unit first calculates the boundary normalized distance between the target working condition and each neighboring anchor point, then calculates the residual distribution coherence coefficient between the target working condition and each neighboring anchor point, merges the boundary normalized distance and the residual distribution coherence coefficient to calculate the coupling weighting coefficient of each neighboring anchor point, and obtains the equivalent parameter set of the target working condition by weighted summation based on the coupling weighting coefficient.

6. The self-calibration system for building digital twin parameters based on creep inversion as described in claim 5, characterized in that: The network self-iterative verification unit uses the residual fitting determination coefficient to quantify the degree of agreement between the actual residual distribution and the predicted residual distribution; When the residual fitting determination coefficient is higher than the set threshold, the corresponding boundary condition vector, equivalent parameter set, and true residual distribution vector form a new conjugate anchor point, which is automatically added to the coupled gradient network. The network synchronously performs anchor point deduplication operation. For multiple sets of anchor points with boundary condition vector differences less than the set threshold, only the set with the highest residual fitting determination coefficient is retained.

7. The self-calibration system for building digital twin parameters based on creep inversion as described in claim 6, characterized in that: When the residual fitting determination coefficient is lower than the set threshold, the network self-iterative verification unit determines that the structural deformation deviates from the normal creep evolution law and triggers a structural anomaly prompt. The anomaly prompt includes the location of the measurement point with the largest deviation, the deviation magnitude, and the deviation direction. Structural anomaly determination does not require additional damage identification algorithms or dedicated monitoring hardware; it is accomplished through consistency judgment of normal creep evolution patterns.

8. The self-calibration system for building digital twin parameters based on creep inversion as described in claim 1, characterized in that: When the addition or removal of monitoring points in the structure causes a change in the observation dimension, the system only needs to supplement the conjugate anchor points under the corresponding observation dimension to complete the adaptation. The original anchor point data remains valid and there is no need for overall recalibration. Coupled gradient networks do not presuppose the functional form and variation law of parameter evolution. They are based entirely on the conjugate anchor points accumulated in actual engineering, and are suitable for different types of concrete structures and finite element models of different precision.

Images