A substructure numerical model construction method based on equivalent force control and boundary dynamic updating

Abstract

The application discloses a kind of substructure numerical model construction method based on equivalent force control and boundary dynamic updating.The application constructs the double-layer nested iteration calibration framework of inner ring matching global load-displacement curve, outer ring checking substructure strain energy and deformation mode, overcomes the defect that traditional single index cannot guarantee the consistency of substructure failure mode, realizes the dual unity of numerical model and overall test in ultimate bearing capacity and buckling mode in fusion measured data.The method takes interface equivalent rotational stiffness, axial spring stiffness and eccentricity as calibration parameters, does not need interface force sensor, avoids the problem that continuum cannot measure load and sensor additional interference.Introduce strain energy relative error and modal confidence criterion as outer ring checking index, suppress iteration shock, improve convergence stability, and provide objective evaluation standard with layered quantitative weighted comprehensive criterion.The application can adapt to different substructure configurations, has good engineering adaptability and practical value.

Description

Technical Field

[0001] This invention relates to the field of cross-validation technology of structural strength testing and numerical simulation, specifically to a method for constructing a substructure numerical model based on equivalent force control and dynamic boundary updates. Background Technology

[0002] In the field of ultimate strength and failure mode assessment of large, complex continuous structures such as ships and offshore platforms, full-scale integral structural physical testing is subject to multiple constraints, including test site, loading equipment tonnage, test model processing cost, and measurement equipment range. This results in extremely high engineering implementation difficulty and testing costs. Furthermore, structures such as large thin-walled stiffened plates and compartment frames are prone to strong nonlinear failures such as buckling and plastic collapse. Pure numerical simulation alone cannot accurately reproduce the actual failure behavior under the coupling of initial material defects, welding residual stress, and local plastic evolution. Therefore, the industry has gradually developed substructure fusion testing technology: critical areas prone to plastic buckling and failure are cut into physical substructures for physical loading tests, while the remaining structure serves as the main structure for finite element numerical simulation. Through bidirectional data interaction and coupling of interface forces and displacements, this approach combines the advantages of realistic nonlinearity in physical testing with the efficient modeling capabilities of numerical simulation, and is currently the mainstream technical approach for predicting the ultimate strength of large structures.

[0003] The current boundary processing of numerical models that integrate measured data can be mainly divided into three categories: First, ideal constraints such as fixed support and simply supported are used, ignoring the actual coupling stiffness of the interface, resulting in significant deviations between the buckling mode and the ultimate load; Second, uniform springs are introduced for manual parameter adjustment, which can only match the elastic stiffness and cannot constrain local failure behavior; Third, the support reaction force is calibrated by measuring the interface sensor, but the continuum has no separation and it is impossible to set up force measuring equipment. Moreover, the sensor will introduce additional stiffness interference. All three types of solutions have boundary mismatch errors that are difficult to avoid.

[0004] Existing calibration algorithms rely solely on global load-displacement curves for optimization, resulting in significant technical shortcomings: a single global index cannot guarantee that the strain distribution and deformation modes within the substructure are consistent with actual experiments; multi-objective synchronous optimization is prone to iterative oscillations and convergence difficulties; cumulative damage to the substructure can cause singularities in the sensitivity matrix, rendering conventional differential iterations ineffective; furthermore, there is a lack of hierarchical quantitative convergence criteria, and calibration results lack a unified and objective evaluation standard.

[0005] Currently, the industry lacks a standardized calibration system that does not rely on actual interface load measurements and takes into account both global strength and local failure mechanisms. There is also no double-layer nested iterative framework that matches the global load in the inner loop and verifies the strain energy and deformation modes of the substructure in the outer loop. This makes it impossible to achieve high-precision unification between numerical models that integrate measured data and overall physical tests in terms of ultimate bearing capacity and buckling failure modes, and it is difficult to meet the engineering specifications for quantitative strength prediction of ship structures. Summary of the Invention

[0006] To address the aforementioned technical issues, this invention quantitatively determines the experimental model design using finite element analysis, employs compensating boundaries and coaxial tooling to achieve equivalence to real boundaries, deploys multiple measurement points according to failure zones, and eliminates loading eccentricity using a limited four-column test platform. By combining pre-test defect scanning with iterative correction of simulation stiffness using test time-series data, a closed loop between physical test and numerical simulation data is established, creating a fusion numerical prediction model that matches the overall test failure characteristics and ultimate strength. This enables high-precision prediction of the ultimate strength of large-scale stiffened plates through low-cost substructure model testing, providing an effective test measurement system.

[0007] To achieve the above objectives, this invention provides a method for constructing a substructure numerical model based on equivalent control and dynamic boundary updates, comprising the following steps: S1. Construct the vector of parameters to be calibrated and assign initial values; S2. Run the numerical model based on the parameter vector to be calibrated and the fused measured data to obtain the global load-displacement curve and the strain energy of the substructure region. S3. Correct the parameter vector to be calibrated based on the deviation between the global load-displacement curve and the target curve; S4. Based on the deviation between the strain energy of the substructure region corresponding to the corrected parameters and the target strain energy, the parameter vector to be calibrated is corrected again. S5. Repeat steps S2 to S4 until the preset convergence criterion is met, and output the calibrated fused numerical prediction model.

[0008] Preferably, S1 includes: The virtual connection stiffness and load transfer characteristics of the interface between the substructure and the main structure are parameterized into a calibration vector Θ. The calibration vector includes at least the equivalent rotational stiffness Kr of the interface, the equivalent axial spring stiffness Ka, and the load distribution eccentricity e in the substructure region. Based on the measured global stiffness of the overall structure in the elastic stage, the initial value of the vector Θ to be calibrated is obtained, and the outer loop convergence tolerance and the inner loop convergence tolerance are set.

[0009] Preferably, S2 includes: After fixing the current outer ring parameters, run the numerical model composed of substructure physical experiments and main structure numerical simulations, which integrates measured data. Obtain the global load-displacement curve based on the running results of the numerical model with integrated measured data, and extract key feature points from the global load-displacement curve. Meanwhile, based on the full-field strain measurement data of the substructure region in the numerical model that integrates measured data, the total strain energy of the substructure region is obtained, and the principal deformation mode vector of the substructure region is extracted based on the full-field strain measurement data.

[0010] Preferably, S3 includes: Using the characteristic points of the global load-displacement curve measured by the physical test of the overall structure as the target value, an inner loop residual vector is constructed based on the deviation between the characteristic points of the global load-displacement curve of the numerical model that integrates the measured data and the target value. Apply perturbations to each parameter in the calibration vector, and obtain the sensitivity matrix of the global response based on the changes in the residual vector before and after applying the perturbations; The vector to be calibrated is updated using the Levenberg-Marquardt iterative scheme based on the sensitivity matrix and the inner loop residual vector. Repeat this step until the inner loop residual vector satisfies the inner loop convergence tolerance, obtain the parameters after the inner loop converges, and obtain the global peak load and substructure region strain energy of the numerical model that integrates the measured data at this time based on the parameters after the inner loop converges.

[0011] Preferably, S4 includes: The relative error of the total strain energy of the substructure region is obtained by comparing the strain energy of the substructure region corresponding to the parameters after the inner ring convergence with the target strain energy of the substructure region obtained by inversion from the physical test of the overall structure. The equivalent rotational stiffness is corrected by multiplicative increment based on the relative error to obtain the updated equivalent rotational stiffness; Based on the numerical model that integrates measured data and the principal deformation mode vectors of the overall structural physical test in the substructure region, the principal deformation mode confidence criterion value is obtained. If the principal deformation mode confidence criterion value is less than the preset threshold, the load distribution eccentricity is updated according to the deviation between the principal deformation mode confidence criterion value and the preset threshold.

[0012] Preferably, S5 includes: Replace the current outer loop parameters in S2 with the updated parameters in S4, and repeat S2 to S4 until the convergence criteria are met simultaneously. The convergence criteria include: the relative error of the global ultimate load is less than the first threshold, the relative error of the total strain energy of the substructure region is less than the second threshold, and the value of the modal confidence criterion is greater than the third threshold. The weighted composite index is obtained by numerically calculating the relative error of the global ultimate load, the relative error of the total strain energy of the substructure region, and the modal confidence criterion. When the weighted composite index is less than the preset threshold, it is determined that the numerical model fused with the measured data and the physical test of the overall structure have reached consistency in failure behavior, and the calibrated fused numerical prediction model is output.

[0013] Preferably, S3 further includes: Before performing the inner loop iterative calculation, it is necessary to pre-determine whether the physical substructure has accumulated plastic damage; If cumulative plastic damage is determined to exist, the sensitivity matrix is ​​obtained based on the Bayesian Gaussian process regression surrogate model; If it is determined that there is no cumulative plastic damage, the sensitivity matrix is ​​obtained by the finite difference method.

[0014] Preferably, the relaxation coefficient used in the multiplicative incremental correction is obtained by segmented adaptive value based on the current number of outer loop iterations: the first relaxation coefficient value is taken in the first 3 outer loop iterations, and the second relaxation coefficient value is taken in the 4th iteration and thereafter; when matching substructure models with different configurations, the relaxation coefficient is adjusted according to the selection difference of the substructure model.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a two-layer nested iterative calibration framework, with an inner loop matching the global load-displacement curve and an outer loop verifying the strain energy and deformation modes of the substructure region. This overcomes the shortcomings of traditional single global indices, which cannot guarantee the consistency of failure modes within the substructure. It achieves dual unification of the numerical model integrating measured data and the overall physical test in terms of ultimate bearing capacity and buckling modes. This method uses the equivalent rotational stiffness of the interface, axial spring stiffness, and load distribution eccentricity as the parameters to be calibrated, eliminating the need for interface force sensors and avoiding the difficulties of measuring interface loads in continuous structures and the interference of additional sensor stiffness. Simultaneously, the relative error of strain energy and modal confidence criteria are introduced as outer-loop verification indices, effectively suppressing iterative oscillations and improving convergence stability. Furthermore, a layered quantified weighted comprehensive criterion is used to provide an objective evaluation standard for the calibration results. This invention is adaptable to different substructure configurations and has good engineering adaptability and practical value. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram illustrating the identification of key failure substructure regions in the overall structure of the orthogonal stiffening plate of the hull according to an embodiment of the present invention. Figure 2 This invention provides three types of adaptable physical substructure test model design schemes. Figure 3 This is a schematic diagram of the four-column limiting coaxial loading physical substructure test device according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the construction principle of the substructure model in an embodiment of the present invention; Figure 5 This is a flowchart illustrating the entire data flow of the substructure model construction system according to an embodiment of the present invention. Detailed Implementation

[0018] 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.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] Example 1 S1. Construct the vector of parameters to be calibrated and assign initial values.

[0021] The virtual connection stiffness and load transfer characteristics at the interface between the substructure and the main structure are parameterized into a vector to be calibrated. Θ The vector to be calibrated includes at least the equivalent rotational stiffness of the interface. Kr Equivalent axial spring stiffness Ka and the load distribution eccentricity in the substructure region e Based on the measured global stiffness of the overall structure in the elastic stage according to the physical test, the vector to be calibrated is... Θ Assign initial values, set the iteration step number n=0, and set the outer loop convergence tolerance. ε 2 and inner ring convergence tolerance ε 1.

[0022] S2. Run the numerical model based on the parameter vector to be calibrated and the fused measured data to obtain the global load-displacement curve and the strain energy of the substructure region. Fixed current outer loop parameters Θ (n) Run the numerical model that integrates measured data from substructure physical experiments and main structure numerical simulations, and record the global load-displacement curves of the numerical model that integrates measured data. P Hybrid (Δ, Θ (n) And extract the key feature points on the curve, including the slope of the elastic segment. K 0 Hybrid Peak load P u Hybrid and peak displacement Δ u Hybrid .

[0023] Simultaneously, the full-field strain data of the substructure region Ω in the numerical model, which integrates measured data, is measured using the digital image correlation (DIC) method, and the total strain energy of the substructure region is calculated through stress-strain integration.E Ω Hybrid ( Θ (n) And extract the principal deformation mode vectors of the substructure region. ϕ Hybrid .

[0024] S3. Correct the parameter vector to be calibrated based on the deviation between the global load-displacement curve and the target curve; This step is the inner loop iteration—parameter correction based on global curve feature points, using global load-displacement curve feature points measured in the overall structural physical experiment. K 0 Exp , P u Exp , Δ u Exp As the target value, construct the inner loop residual vector: in, These parameters represent the virtual connection stiffness and load transfer characteristics at the interface between the substructure and the main structure. , , This represents the stiffness matrix, load vector, and displacement increment of the numerical model that integrates measured data.

[0025] For the i A perturbation is applied to each parameter to be calibrated. The sensitivity matrix of the global response is calculated using the finite difference method. S Its column vectors are determined by the following formula: in, Represents the inner loop residual vector. This indicates the eccentricity of the load distribution in the substructure region.

[0026] The parameters to be calibrated are updated using the Levenberg-Marquardt iterative method, and the calculation formula is as follows: In the formula, λ This is the adaptive damping factor. Repeat this step until the || condition is met. R Global (n) ||2< ε 1. Obtain the parameters after the inner loop converges. Θ (n+1 / 2) And record the global peak load of the numerical model that integrates the measured data at this time. Pu Hybrid With the strain energy of the substructure region E Ω Hybrid .

[0027] S4. Based on the deviation between the strain energy of the substructure region corresponding to the corrected parameters and the target strain energy, the parameter vector to be calibrated is corrected again. This step is the outer loop iteration—parameter verification based on the strain energy and failure mode of the substructure region.

[0028] The relative error in calculating the total strain energy of the substructure region is used as an indicator of failure strength: in, This refers to the total strain energy at the corresponding location in the substructure region, obtained through inversion from the overall structural physical test. This represents the total strain energy of the numerical model that integrates measured data.

[0029] Based on the relative error, the equivalent rotational stiffness Kr Perform multiplicative increment correction as follows: In the formula, β The relaxation coefficient ranges from 0.15 to 0.3.

[0030] The numerical model that integrates measured data and the confidence criteria for the principal deformation modes in the substructure region based on the overall structural physical experiment ( MAC Value: in, This represents the principal deformation obtained from physical experimental measurements of the substructure. This represents the principal deformation of the numerical model that integrates measured data.

[0031] like MAC If the offset is less than the preset threshold of 0.85, update the load distribution eccentricity. e : In the formula, γ This is the step size coefficient, ranging from 0.02 to 0.05; gradient sign. Determined through two finite difference calculations.

[0032] S5. Repeat steps S2 to S4 until the preset convergence criterion is met, and output the calibrated fused numerical prediction model.

[0033] Parameters updated via outer loop iteration Θ (n+1) substitution Θ(n), return to S2 for the next iteration. Repeat S2 to S4 until the following convergence criterion is met simultaneously: (1) The relative error of the global ultimate load is less than 2%; (2) The relative error of the total strain energy in the substructure region is less than 5%; (3) Modal confidence criterion MAC The value is greater than 0.90.

[0034] Define weighted composite index C as follows: in, w 1. w 2. w 3 is the weighting coefficient, and satisfies w 1+ w 2+ w 3 = 1.

[0035] When the weighted composite index C When the value is less than the preset threshold, the numerical model that integrates the measured data and the overall structural physical test are finally determined to be consistent in terms of failure behavior, and the calibrated fused numerical prediction model is output.

[0036] Example 2 Combined with appendix Figure 1 According to the appendix Figure 1 The theoretical constraint stiffness of the interface is extracted from the finite element pre-analysis results of the integral stiffened plate, and the vector to be calibrated is completed. Θ Initialize: K r (0) Take the rotational stiffness of the elastic constraint in the finite element method. K a (0) Take the theoretical stiffness of the axial tension at the interface and the initial eccentricity. e (0) =0; Inner loop convergence tolerance ε 1 The preferred value is 1%.

[0037] Combined with appendix Figure 4 The bidirectional interactive interface integrates measured data. During the iterative operation of the numerical model, the physical substructure outputs the nonlinear FU restoring force, which is fed back to the numerical master structure. The numerical master structure solves the interface displacement and sends it to the loading actuator to update the substructure boundary constraints. The bidirectional data collaborative interaction interface has a transmission latency of less than 50ms, ensuring synchronous updates of dynamic boundaries.

[0038] Combined with appendix Figure 5 Multi-field synchronous measurement unit, total strain energy of substructure region E ΩExp The full-field displacement field was obtained through a 3D-DIC system for overall testing. The strain field was then solved using geometric equations and combined with material constitutive integrals. The 3D-DIC equipment is described in the attached... Figure 3 The image is positioned on the front of the specimen, with the shooting angle between the angle of view and the specimen normal ≤10° to eliminate out-of-plane deformation measurement errors. The speckle image, fiber strain, and load displacement signal are all synchronously sampled at a 1ms interval. After the data is transmitted to the computer, the time axis is aligned first, and then it is substituted into the iterative calculation to eliminate time sequence misalignment residual deviations.

[0039] Before the inner loop iterative calculation, it is pre-determined whether the physical substructure has accumulated plastic damage from multiple loadings; if the specimen exhibits local buckling or plastic indentation damage, the sensitivity matrix is ​​pre-determined. S The method is unique in that it directly uses a Bayesian Gaussian process regression surrogate model to replace the finite difference method for calculating the sensitivity matrix; it uses the finite difference method for normal solution under non-destructive conditions; the number of sampling points in the surrogate model is no less than twice the number of current iteration steps, and the sampling interval is fully covered. Kr , Ka , e All parameter correction ranges.

[0040] Relaxation coefficient during outer loop iteration β Piecewise adaptive value selection is adopted: the outer loop (n≤3) in the first 3 iterations is selected as ( β =0.3) accelerates boundary stiffness convergence; after 4 iterations, take ( β =0.15) Suppress stiffness oscillations; Adaptor attachment Figure 2 Differential adjustments for the three substructure models: For the vertical multi-span scheme, the overall adjustment is reduced by 0.05 to weaken boundary over-correction; (Appendix) Figure 2 Option 2, which adds boundary aggregate, remains constant throughout. β =0.2), to avoid iterative oscillations caused by sudden changes in the stiffness of the boundary aggregate.

[0041] Matching Figure 3 The spherical indenter has a coaxial loading structure with an eccentricity. e The total range of a single correction is limited to ±5mm. If the calculated value exceeds the threshold, the system will automatically trigger a coaxial reset prompt for the loading fixture to avoid extra boundary stiffness interference caused by out-of-range eccentricity.

[0042] The weighted composite index weights can be adaptively matched with the appendix. Figure 2 Three types of substructure models: Optimal weighting of basic working conditions ( w 1 = 0.4 w 2 = 0.3 w 3=0.3), with the ultimate bearing capacity having the highest weight; for the attached Figure 2 Scheme 3 for vertical multi-span substructure, the weights are adjusted to ( w 1 = 0.35 w2 = 0.35 w 3=0.3), strengthening the deformation mode constraints of multi-span continuous structures; comprehensive index C The preferred convergence threshold is 0.05.

[0043] Appendix Figure 2 The physical substructure model has selection rules: only considering the local buckling of a single-cell plate, the supplementary model is selected. Figure 2 Option 1; For options with high boundary constraint stiffness matching requirements and the possibility of boundary aggregate plastic failure, an additional option is selected. Figure 2 Option 2; Selection of Appendix for Overall Instability Prediction of Multi-Span Continuously Stiffened Plates Figure 2 Option 3: Unified adaptation of three types of models Figure 3 The four-column limiting loading device is centered in the loading span to eliminate end boundary effects.

[0044] After iterative convergence and consensus determination, the system synchronously outputs a calibration completion message. Θ Parameter set, appendix Figure 4 The numerical model shown integrates measured data, including the global load-displacement curve and the buckling / plastic collapse failure modes of the substructure, and serves as a prediction model for the ultimate strength of ship structures.

[0045] Appendix Figure 3 The physical substructure loading device adopts a four-column limiting rigid loading frame. The four columns are symmetrically arranged to limit the horizontal lateral displacement of the specimen. The spherical indenters are coaxially connected to eliminate the eccentric bending moment of the loading surface and avoid the additional boundary stiffness error introduced by eccentric loading.

[0046] Example 3 This embodiment selects Scheme 2 (edge ​​plate reinforcement + boundary rib configuration), and is compatible with... Figure 3 Four-column limiting coaxial loading test device; the device has a built-in spherical indenter to eliminate loading eccentricity, distributed optical fibers are laid on the surface, and a 3D-DIC measurement device is mounted on the front. All measurement equipment is as per the attached specifications. Figure 5 The data flow rules implement 1ms synchronous sampling. After collection, the timing is aligned before being substituted into the iterative calculation.

[0047] Taking a stiffened slab structure as an example, a complete structural physical test of the stiffened slab is first carried out to collect global load-displacement curves. Then, the full-field deformation data of the substructure region is recorded using a 3D-DIC system, and the target feature points are obtained after processing. P u Exp and strain energy E Ω Exp .

[0048] Based on the overall structural elastic stage stiffness, set K r (0) This is the theoretical value of the elastic constraint stiffness.K a (0) For axial tensile stiffness, eccentricity e (0) = 0. Set the inner loop tolerance. ε 1=1%, outer ring tolerance limit load error 2%, strain energy error 5%. MAC Threshold 0.90.

[0049] The main structure is modeled using ABAQUS as a finite element model, and a co-simulation interface is used to connect it with the physical substructure. Figure 4 The bidirectional force-displacement closed-loop interaction is shown, performing a double-layer nested iterative calibration: The boundary stiffness parameters of the inner ring are corrected using the Levenberg-Marquardt algorithm. In this embodiment, the specimen has no obvious plastic damage, and the sensitivity matrix is ​​solved directly using the finite difference method. If buckling damage occurs after multiple loading of the specimen, the Bayesian Gaussian process regression surrogate model is used to avoid matrix singular failure.

[0050] The outer loop iteration uses a piecewise adaptive relaxation coefficient. β The first three rounds use 0.3 to accelerate convergence, and the subsequent rounds use 0.15 to suppress oscillation. In this embodiment, Model 2 with boundary aggregate is selected, and β=0.2 is kept constant throughout. When the MAC value is lower than 0.85, the eccentricity e is corrected, and the eccentricity correction range is limited to ±5mm. If the limit is exceeded, the tooling will be automatically reset.

[0051] When iterating to the 5th round of the outer loop, the relative error of the global ultimate load is 2.8%, the relative error of the substructure strain energy is 3.0%, and the MAC is 0.94; using weights... w 1 = 0.4, w 2 = 0.3, w The weighted comprehensive index C is calculated to be less than 0.05, which satisfies all convergence criteria. Therefore, the numerical model that integrates the measured data is consistent with the overall experimental failure behavior.

[0052] The calibrated numerical model integrating measured data was used for ultimate strength prediction. The repeatability error between the ultimate bearing capacity and the overall solid test was controlled within 4.0%, and the buckling modes of the substructure perfectly matched the 3D-DIC observation results. If the attached... Figure 2 Vertical multi-span scheme three model, the weights can be adjusted as follows: w 1 = 0.35, w 2 = 0.35, w 3=0.3 strengthens deformation mode constraints, adapting to instability prediction scenarios of multi-span continuous stiffened plates.

[0053] This embodiment fully covers the basic double-layer nested iterative calibration process, and verifies the engineering feasibility of all optimized schemes. It can flexibly switch between different hull stiffening plate configurations. Figure 2Three types of substructure models and corresponding iterative control parameters.

[0054] Example 4 This embodiment also provides a substructure model construction process based on equivalent force control and boundary dynamic update, which is used to support the implementation of the method in the above embodiment. It includes: a key failure area identification module, a physical substructure loading test device, a multi-field synchronous measurement unit, a numerical main structure simulation module, a two-way data collaborative interaction interface, and a double-layer nested iterative calibration calculation unit.

[0055] (1) Key failure area identification module: buckling pre-analysis is carried out on the overall stiffened plate, the plate grid that is easy to buckle and has concentrated plasticity is defined as physical substructure, and the remaining structure is numerical main structure. The interface between the substructure and the main structure is parameterized into three types of parameters to be calibrated: equivalent rotational stiffness Kr, equivalent axial spring stiffness Ka, and load distribution eccentricity e. (2) Physical substructure loading test device: a four-column rigid bearing frame equipped with an electro-hydraulic servo coaxial loading unit, which can apply static loads to the physical substructure test model; (3) Multi-field synchronous measurement unit: synchronously acquires load, displacement, and full-field strain and deformation image data of the test model; (4) Numerical main structure simulation module: Finite element simulation of the main structure, with built-in interactive subroutines to achieve interface equilibrium solution; (5) Two-way data collaborative interaction interface: realizes two-way real-time closed-loop transmission of physical substructure output restoring force and numerical main structure output interface displacement; (6) Double-layer nested iterative calibration calculation unit: It has built-in inner loop global load optimization algorithm and outer loop strain energy and deformation mode verification algorithm, and outputs consistency judgment index to complete parameter calibration.

[0056] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for constructing a substructure numerical model based on equivalent control and dynamic boundary updates, characterized in that, Includes the following steps: S1. Construct the vector of parameters to be calibrated and assign initial values; S2. Run the numerical model based on the parameter vector to be calibrated and the fused measured data to obtain the global load-displacement curve and the strain energy of the substructure region. S3. Correct the parameter vector to be calibrated based on the deviation between the global load-displacement curve and the target curve; S4. Based on the deviation between the strain energy of the substructure region corresponding to the corrected parameters and the target strain energy, the parameter vector to be calibrated is corrected again. S5. Repeat S2 to S4 until the preset convergence criterion is met, and output the calibrated numerical prediction model with fused measured data.

2. The method for constructing a substructure numerical model based on equivalent control and dynamic boundary updates according to claim 1, characterized in that, S1 includes: The virtual connection stiffness and load transfer characteristics of the interface between the substructure and the main structure are parameterized into a calibration vector Θ. The calibration vector includes at least the equivalent rotational stiffness Kr of the interface, the equivalent axial spring stiffness Ka, and the load distribution eccentricity e in the substructure region. Based on the measured global stiffness of the overall structure in the elastic stage, the initial value of the vector Θ to be calibrated is obtained, and the outer loop convergence tolerance and the inner loop convergence tolerance are set.

3. The method for constructing a substructure numerical model based on equivalent control and dynamic boundary updates according to claim 1, characterized in that, S2 includes: After fixing the current outer ring parameters, run the numerical model composed of substructure physical experiments and main structure numerical simulations, which integrates measured data. Obtain the global load-displacement curve based on the running results of the numerical model with integrated measured data, and extract key feature points from the global load-displacement curve. Meanwhile, based on the full-field strain measurement data of the substructure region in the numerical model that integrates measured data, the total strain energy of the substructure region is obtained, and the principal deformation mode vector of the substructure region is extracted based on the full-field strain measurement data.

4. The method for constructing a substructure numerical model based on equivalent control and dynamic boundary updates according to claim 1, characterized in that, S3 includes: Using the characteristic points of the global load-displacement curve measured by the physical test of the overall structure as the target value, an inner loop residual vector is constructed based on the deviation between the characteristic points of the global load-displacement curve of the numerical model that integrates the measured data and the target value. Apply perturbations to each parameter in the calibration vector, and obtain the sensitivity matrix of the global response based on the changes in the residual vector before and after applying the perturbations; The vector to be calibrated is updated using the Levenberg-Marquardt iterative scheme based on the sensitivity matrix and the inner loop residual vector. Repeat this step until the inner loop residual vector satisfies the inner loop convergence tolerance, obtain the parameters after the inner loop converges, and obtain the global peak load and substructure region strain energy of the numerical model that integrates the measured data at this time based on the parameters after the inner loop converges.

5. The method for constructing a substructure numerical model based on equivalent control and dynamic boundary updates according to claim 1, characterized in that, S4 includes: The relative error of the total strain energy of the substructure region is obtained by comparing the strain energy of the substructure region corresponding to the parameters after the inner ring convergence with the target strain energy of the substructure region obtained by inversion from the physical test of the overall structure. The equivalent rotational stiffness is corrected by multiplicative increment based on the relative error to obtain the updated equivalent rotational stiffness; Based on the numerical model that integrates measured data and the principal deformation mode vectors of the overall structural physical test in the substructure region, the principal deformation mode confidence criterion value is obtained. If the principal deformation mode confidence criterion value is less than the preset threshold, the load distribution eccentricity is updated according to the deviation between the principal deformation mode confidence criterion value and the preset threshold.

6. The method for constructing a substructure numerical model based on equivalent control and dynamic boundary updates according to claim 1, characterized in that, S5 includes: Replace the current outer loop parameters in S2 with the updated parameters in S4, and repeat S2 to S4 until the convergence criteria are met simultaneously. The convergence criteria include: the relative error of the global ultimate load is less than the first threshold, the relative error of the total strain energy of the substructure region is less than the second threshold, and the value of the modal confidence criterion is greater than the third threshold. The weighted composite index is obtained by numerically calculating the relative error of the global ultimate load, the relative error of the total strain energy of the substructure region, and the modal confidence criterion. When the weighted composite index is less than the preset threshold, it is determined that the numerical model fused with the measured data and the physical test of the overall structure have reached consistency in failure behavior, and the calibrated fused numerical prediction model is output.

7. The method for constructing a substructure numerical model based on equivalent control and dynamic boundary updates according to claim 4, characterized in that, S3 further includes: Before performing the inner loop iterative calculation, it is necessary to pre-determine whether the physical substructure has accumulated plastic damage; If cumulative plastic damage is determined to exist, the sensitivity matrix is ​​obtained based on the Bayesian Gaussian process regression surrogate model; If it is determined that there is no cumulative plastic damage, the sensitivity matrix is ​​obtained by the finite difference method.

8. The method for constructing a substructure numerical model based on equivalent control and dynamic boundary updates according to claim 5, characterized in that, The relaxation coefficient used in the multiplicative incremental correction is obtained by segmented adaptive value based on the current outer loop iteration round: the first relaxation coefficient value is taken in the first 3 outer loop iterations, and the second relaxation coefficient value is taken in the 4th round and thereafter; when matching substructure models with different configurations, the relaxation coefficient is adjusted according to the selection difference of the substructure model.

Images