A test risk digital twin early warning method for multi-domain collaborative monitoring

Abstract

The application provides a test risk digital twin early warning method for multi-domain cooperative monitoring, and belongs to the technical field of virtual-real fusion test and digital twin. The steps are as follows: firstly, a fine finite element simulation model considering nonlinearity is established, which comprises a digital tooling system, a digital sensor and a test piece; secondly, nonlinear finite element simulation analysis is carried out, and a multi-level mechanical response field inversion reduced model is constructed; thirdly, complete sensor data set construction is completed through a data filling algorithm, and first-level structure failure judgment is carried out based on the same; finally, future loading stage sensor data prediction is carried out, and second-level structure failure judgment is completed; full-field mechanical response inversion and online real-time correction are carried out; response inversion of the concerned area is carried out, and third-level structure failure judgment is realized. The application can realize real-time dynamic monitoring and early warning of structure test risk, has high real-time performance, robustness and accuracy, and provides a powerful guarantee for the safety and reliability of structure test.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of virtual-real fusion test and digital twin technology, and relates to a test risk digital twin early warning method for multi-domain collaborative monitoring. BACKGROUND

[0002] With the increasing complexity and large-scale of modern engineering structures, structural tests play an irreplaceable role in verifying design theory and evaluating safety performance. In recent years, structural digital twin monitoring has become one of the key technologies to ensure the safety and reliability of structural tests. However, with the increasing complexity of engineering structures and the increasing demand for safety, traditional structural digital twin monitoring methods have gradually exposed many problems.

[0003] Under complex loads or extreme working conditions, structures often exhibit strong nonlinear behavior due to material nonlinearity, geometric nonlinearity or local damage accumulation, and even trigger sudden failure. Traditional structural test monitoring methods rely on linear assumptions or empirical thresholds, making it difficult to accurately capture the nonlinear evolution process, leading to delayed risk warning or misjudgment, which seriously threatens test safety and data reliability. At the same time, although nonlinear finite element simulation is an effective means to evaluate the nonlinear behavior of structures, it faces significant challenges in practical application: first, the idealized loading method of structures is often difficult to accurately simulate actual test conditions, especially under strong nonlinear conditions, and this simplification will lead to continuous accumulation and significant amplification of errors; second, high-fidelity nonlinear analysis requires fine meshing and complex constitutive models, resulting in exponential growth of computational load; third, the convergence problem in the iterative solution process further increases the computational cost. These bottlenecks seriously restrict the application effect of the single evaluation method of nonlinear finite element simulation in real-time monitoring.

[0004] At the same time, sensors in tests are easily affected by environmental interference or hardware failure, leading to missing or abnormal data of measuring points, which also brings challenges to the effective implementation of structural digital twin monitoring. The core goal of structural digital twin monitoring is to collect real-time response data of the structure through sensors and evaluate the health status of the structure based on these data. However, in practical applications, sensor failure, data loss or abnormality often leads to a decrease in the reliability of the monitoring system. Especially in large-scale tests of complex structures, the failure or abnormal data of some sensors may affect the overall risk assessment of the structure. Traditional methods lack real-time identification and repair mechanisms, which seriously affect the prediction reliability.

[0005] Li Ying et al. (Chinese Invention Patent CN118940569A) proposed a virtual experiment method based on support vector machine classification and regression training for target ship operating conditions. While this method trains a surrogate model by fusing batches of simulated data with a small amount of experimental data, the surrogate model only exhibits high prediction accuracy for the experimental conditions used in its training. Its prediction accuracy for unknown operating conditions remains to be tested. Fang Ming et al. (Chinese Invention Patent CN119004594A) reduced the computational workload by dividing large, complex structures into local nonlinear substructures and simultaneously achieved rapid online acquisition of analysis results using a reduction-order method. However, this method focuses on nonlinear seismic response analysis and does not incorporate real experimental data. Furthermore, none of the above methods consider the construction of a digital loading system or real-time identification and repair of fault sensors.

[0006] Therefore, to address the above challenges, a novel early warning method that integrates the advantages of physical mechanisms and data-driven approaches is urgently needed. This method should combine high-fidelity nonlinear modeling with efficient order reduction techniques to achieve dynamic and accurate prediction of risks in complex structural experiments. This invention proposes a multi-domain collaborative monitoring digital twin early warning method for experimental risks, aiming to overcome the aforementioned technical bottlenecks and provide an innovative solution for the safe and controllable conduct of structural experiments. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides a digital twin early warning method for test risks based on multi-domain collaborative monitoring. By combining measured physical strain gauge data and displacement sensor data from structural tests with the full-field response reconstruction results, real-time test risk early warning is achieved. Simultaneously, based on the measured data, the prediction model is dynamically updated and evolved, ultimately establishing a digital twin early warning system for test risks based on multi-domain collaborative monitoring, enabling precise monitoring and risk warning of strength tests.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A digital twin early warning method for experimental risks based on multi-domain collaborative monitoring includes the following steps:

[0010] The first step is to establish a detailed finite element simulation model that considers nonlinearity, including a digital tooling system, digital sensors, and the specimen. Specifically:

[0011] Step 1.1: Construct a finite element simulation model based on the geometric model of the structure, and divide the region into nonlinear regions according to the location of the hazardous area of ​​concern.

[0012] Specifically, the nonlinear region should be arranged at a position meeting the following requirements: ① The linear-nonlinear region interface should not be too close to the dangerous region where nonlinearity occurs (the minimum distance between the interface and the dangerous region should be greater than 2*local grid size or 3*structure thickness); ② The linear-nonlinear region interface should not be located in a region where the mechanical behavior changes significantly (stress gradient > 20% / mm). Commonly excluded regions include stress concentration zones, material property change zones, contact surfaces, or connection parts.

[0013] Further, for complex structures with excessive grid numbers, a simplified finite element model can also be constructed by substructure method or submodel method. The specific analysis process of the substructure method is as follows: first, a shell simplified model is established for the linear region of the structure that is not dangerous, and a local detailed finite element model is established for the structure part including the nonlinear region, and fine grid division is performed. The specific analysis process of the substructure method is as follows: the global shell model and the local detailed model of the structure are connected through binding, solid-shell coupling, and MPC beam interaction connection mode to realize effective load transfer. The specific analysis process of the submodel method is as follows: based on the simulation results of the global shell model, the simulation loading is realized by applying a submodel boundary to the local detailed model of the structure. If there are multiple levels of submodels, each level of submodel and the global model can be gradually increased based on this idea.

[0014] Step 1.2, based on the nonlinear region obtained in step 1.1, fine grid division is performed, and linear properties are applied to the linear region and nonlinear properties are applied to the nonlinear region.

[0015] The linear properties include material linear properties, and the nonlinear properties include material nonlinear properties and contact nonlinear properties. Material nonlinear properties include material plasticity and damage criteria, where material plasticity can be set by yield criteria, hardening model, flow rule, or porous metal plasticity model, and damage criteria can be set by ductile metal damage criteria, Johnson-Cook criteria, or Hashin damage initiation criteria. The contact nonlinear properties can be set by hard contact method, penalty function method, Lagrange multiplier method, and augmented Lagrange method.

[0016] Step 1.3, parameterized scripts are used to complete the assembly of digital tooling systems, digital sensors, and digital specimens, and a fine finite element model including digital tooling systems, digital sensors, and specimens is established. The internal connection of the digital tooling system and the connection between the digital tooling system, the digital sensor, and the specimen are achieved through binding, MPC beam, or coupling. Specifically:

[0017] Furthermore, the digital tooling system includes loading levers, connectors, and fixing fixtures; the digital sensors include strain sensors, displacement sensors, temperature sensors, force sensors, and acceleration sensors. The binding between components within the digital tooling system includes loading levers to loading levers and loading levers to connectors; the binding between the digital tooling system and the specimen includes connectors to the specimen, loading levers to the specimen, and fixing fixtures to the specimen. The parameterized script can move the digital sensors to the patch positions and complete the binding with the specimen according to the test patch scheme; the patch position coordinates and angles can be arbitrarily modified. The test patch scheme includes sensor type, number, patch position, and patch angle. The load of the finite element model is applied to the loading coupling points of the aforementioned digital tooling system through the parameterized script.

[0018] The second step involves performing nonlinear finite element simulation analysis based on the refined finite element simulation model built in the first step, sampling the simulation results, extracting the mechanical response results, constructing a corresponding dataset, and building a multi-level mechanical response field inversion and order reduction model. Specifically:

[0019] Step 2.1 involves performing simulation analysis on the detailed finite element simulation model considering nonlinearity constructed in Step 1. Specifically, calculations are performed using static, explicit dynamic, implicit dynamic, or Riks' method analysis steps to obtain the finite element simulation analysis results of the model.

[0020] Step 2.2: Based on the obtained finite element simulation analysis results, perform sensitivity analysis based on the model parameters.

[0021] Specifically, the sensitivity analysis process is as follows: for the model shown in equation (1) ,have:

[0022] (1)

[0023] in, For model parameters, The total number of parameters. For using parameters The finite element results obtained from the calculation. Then the local sensitivity of the parameters. for:

[0024] (2)

[0025] Approximating formula (2) using the central difference method, we obtain:

[0026] (3)

[0027] in, Represents model parameters The increase.

[0028] Specifically, the model parameters include material elastic modulus, Poisson's ratio, plasticity parameters, load size, position and direction.

[0029] Step 2.3, selecting the highest sensitivity The group model parameters generate a plurality of sampling points in the design space, and interval simulation sampling is performed to obtain finite element model sampling results.

[0030] Further, the design space is a load design domain based on a plurality of test working conditions, and the interval simulation sampling method includes Latin hypercube sampling, uniform random sampling or stratified sampling.

[0031] Step 2.4, constructing a sample set according to the finite element model sampling results obtained in step 2.3, and assembling a snapshot matrix.

[0032] The constructed snapshot matrix The number of rows of the matrix represents the total number of finite element nodes of the test piece, and the number of columns of the matrix represents the total number of frames of all simulation results, and the input matrix The number of rows of the matrix is the number of digital sensor channels. At the same time, in the process of carrying out the strength static verification test, the material is in the elastic stage under low loading level, and the strength mechanical response of the test piece changes linearly, but as the loading level increases, the material enters the plastic stage or even breaks down, at this time, the strength mechanical response of the test piece presents obvious nonlinear change, therefore, it is necessary to construct a snapshot sub-matrix in layers to reduce the nonlinear influence. Specifically:

[0033] The method for constructing the layered snapshot matrix includes: carrying out simulation analysis according to the to-be-conducted test working condition, setting the output frame according to the test loading step, extracting the mechanical response field of each frame to construct the snapshot matrix. For each loading level, a window of 5%-15% frames is expanded forward and backward based on the loading time, and the outputs of all frames in the window are recombined into a snapshot sub-matrix; or the snapshot matrix is divided based on correlation, first, the Pearson correlation coefficient matrix between the frames of all sampling results is calculated, then the hierarchical clustering or K-means clustering algorithm is used to cluster based on the Pearson correlation coefficient, and finally the corresponding columns are distributed into each sub-matrix according to the clustering result to create a snapshot sub-matrix; or the snapshot matrix is divided based on principal component analysis, first, the snapshot matrix is centralized and standardized, second, the principal component analysis is performed on the processed matrix to obtain a load matrix, then the load matrix is grouped using a simple threshold method, a K-means clustering method or a rotated principal component method, and finally a snapshot sub-matrix is created according to the grouping result.

[0034] At the same time, based on the layered results of the snapshot matrix, the input matrix The same layering is performed according to the corresponding frame number to obtain a snapshot sub-matrix of the input matrix.

[0035] In step 2.5, the reduced-order method is used to perform eigenvalue decomposition on each snapshot sub-matrix after layering, extract the corresponding reduced-order basis vectors and reduced-order basis coefficients, and realize the truncation screening of the reduced-order basis vectors and the corresponding coefficients by setting an energy threshold, and finally obtain the reduced-order basis and the corresponding reduced-order basis coefficients. At the same time, the reduced-order basis coefficient prediction proxy model of each reduced-order model is trained based on the input matrix and the reduced-order coefficient set. Specifically:

[0036] The constructed multi-level mechanical response field inversion reduced-order model includes a linear region reduced-order model and a nonlinear region reduced-order model. The sample input of the linear region reduced-order model is the digital displacement and strain sensor simulation result, and the output is the full-field mechanical response of the linear region; the sample input of the nonlinear region reduced-order model is the digital displacement, strain sensor simulation result and mechanical response of the linear region on the interface, and the output is the full-field mechanical response field of the nonlinear region.

[0037] Further, the reduced-order method includes intrinsic orthogonal decomposition, Krylov method, dynamic modal decomposition, random projection, non-negative matrix decomposition or canonical multivariate decomposition.

[0038] Further, the screening method of the reduced-order basis vector and the corresponding coefficient is that the energy threshold used for truncation screening is not less than 99%.

[0039] Further, the method for constructing the reduced-order basis coefficient prediction proxy model includes Gaussian process regression, random forest, support vector machine, deep neural network or gradient boosting tree.

[0040] In the third step, a data feature determination criterion is established to screen the sensor data, and at the same time, abnormal measurement point data is filled in through a data filling algorithm to obtain a complete sensor data set; at the same time, a first level structure failure determination is performed based on the complete sensor data set. Specifically:

[0041] In step 3.1, during the test, the data feature determination criterion is used to screen the sensor data failure to screen and exclude the sensors with faults. Specifically:

[0042] The data feature determination criterion includes a physical rationality criterion, an outlier criterion, a strain change trend criterion, a data stability criterion, a white noise fault criterion and a zero point drift criterion; the data set used for determination includes a historical sensor data set composed of historical test data and a current sensor data set composed of current test data. Specifically as follows:

[0043] (1) Physical rationality criterion:

[0044] The physical rationality criterion is that the strain value of a single strain gauge channel should be within the reasonable range of the material's mechanical properties, which can be expressed as:

[0045] (4)

[0046] wherein, ε is the strain value of the current sensor data set; εallow is the allowable strain value of the material.

[0047] (2) Abnormal value criterion:

[0048] The abnormal value criterion uses the 3σ principle to detect the abnormal value of a single sensor, wherein σ is the standard deviation of the data. This principle is based on the assumption of normal distribution, and considers that the probability of data points falling within the range of mean ± 3 times standard deviation is 99.7%, and data points exceeding this range can be regarded as abnormal values. The criterion formula is:

[0049] (5)

[0050] wherein, R is the response value of the current sensor data set; Rmean is the mean value of the historical sensor data set; Rstd is the standard deviation of the historical sensor data set.

[0051] (3) Strain change trend criterion:

[0052] The strain change trend criterion means that the strain change trend of adjacent two loading steps should be consistent with the theoretical or historical data, and should not show an increasing or decreasing phenomenon regardless of the external load size. The criterion formula uses the least squares regression slope between the response data and the time sequence :

[0053] (6)

[0054] wherein, n is the loading step sequence of the current sensor data set; R is the response data sequence of the current sensor data set; N is the number of data points in .

[0055] (4) Data stability criterion:

[0056] The data stability criterion means that the strain fluctuation of adjacent two loading steps should be within a reasonable range. Stable data should have a small fluctuation range, while data with excessive fluctuation may indicate that the sensor has a fault or is disturbed by external interference. The criterion formula is:

[0057] (7)

[0058] wherein, is the response value of the current sensor data set under the loading level; is the response value of the current sensor data set under the loading level; is the standard deviation of the historical sensor data set; is a threshold coefficient, usually in the range of 2-3.

[0059] (5) White noise failure criterion:

[0060] White noise failure refers to the fact that the sensor reading is composed of only background noise, that is, the sensor value fluctuates around 0, and its criterion formula is:

[0061] (8)

[0062] wherein, is white noise, that is, a random signal with unknown mean and covariance.

[0063] (6) Zero drift criterion:

[0064] Zero drift refers to the fact that the zero strain value of a single sensor should not drift significantly. Its criterion formula is:

[0065] (9)

[0066] wherein, is the zero sensor value of the current sensor data set; is the maximum value of the allowed zero sensor value.

[0067] In the process of judging the failure of the sensor, the structural failure criterion is determined simultaneously, and the strain change trend criterion and the data stability criterion are used to further determine the sudden slope mutation in the loading process, and the first level failure criterion is completed.

[0068] Step 3.2, the sensor data filling model is constructed by the data filling algorithm offline, and the actual sensor data is mapped online to reconstruct the missing data of the failed sensor, realize the data filling of the failed sensor, and construct a complete sensor data set, while the first level failure criterion is determined.

[0069] Further, the data filling algorithm refers to the process of reasonably estimating and reconstructing the missing part of the overall data by using mathematical modeling or statistical methods. The data filling algorithm used includes linear interpolation method, singular value decomposition method, interval proper orthogonal decomposition method, random forest method or generative adversarial network.

[0070] Fourthly, the full-field mechanical response inversion and online real-time correction are performed by the multi-level mechanical response field inversion model established in the second step and the complete sensor data set established in the third step; the sensor data of the future loading step is predicted by combining the time sequence prediction model established based on the historical loading data, the second-level structural failure is determined, the response inversion of the concerned area is performed, the future risk warning is performed in real time through the calculation of the nonlinear regional damage factor, and the third-level structural failure is determined. Specifically:

[0071] Step 4.1, input the complete sensor data set obtained in step 3.2 online, obtain the reduced basis coefficients of the constructed reduced-order models based on the reduced basis coefficient prediction proxy model constructed in step 2.5, and complete the upscaling inversion of the mechanical response field by multiplying the reduced basis coefficients with the reduced basis to obtain the full-field mechanical response field results.

[0072] Step 4.2, based on the complete sensor data set and the numerical calculation of the sensor measurement point position, the residual error is calculated, and the full-field response correction proxy model is constructed to perform online real-time correction on the mechanical response field of the specimen after upscaling inversion. Specifically:

[0073] According to the coordinates of the sensors in the test patch scheme, the response values of the corresponding coordinates in the specimen mechanical response field predicted in step 4.1 are extracted, the deviation values between the predicted sensor response values and the test sensor measured values are calculated, and the full-field response correction proxy model is constructed based on the relationship between the sensor deviation values and the corresponding coordinates. The global correction deviation value obtained by superposition and the specimen mechanical response field obtained in step 4.1 are used to realize online dynamic correction of the global mechanical response field.

[0074] Further, the method for training the full-field response correction proxy model includes radial basis function, Gaussian process regression, random forest, transfer learning neural network or incremental learning gradient boosting tree.

[0075] Step 4.3, the sensor time sequence prediction model is trained by establishing a historical loading database to predict the sensor data of the future loading step and complete the second-level failure determination. Specifically:

[0076] The historical loading database includes sample data at the measurement points and historical working condition loading sensor data. The sensor time sequence data model training method adopts long short-term memory network, autoregressive and moving average model, recurrent neural network, convolutional neural network, Transformer model, support vector machine or random forest algorithm.

[0077] In the sensor time series prediction model training process, according to the historical loading database time series data, the time series characteristics of the influence of load change on the response of the measuring point are captured through the time series prediction method, and the time series prediction with load change is carried out. Through the time series dependence characteristics of the time series prediction algorithm, the short-term and long-term dependence relationship in the time series data is identified, so that the sensor time series prediction model can still accurately predict the response of the measuring point when facing complex load changes.

[0078] In the online phase, based on the established sensor time series prediction model, the sensor data set composed of the collected normal sensor data and the replaced fault sensor data in step 3.2 is input to realize real-time sensor future prediction, and the second-level failure judgment is realized by comparing the prediction value with the first-level failure judgment method in step 3.1.

[0079] Further, after the test is completed, the collected normal sensor data and the replaced fault sensor data in step 3.2 are loaded as historical working condition sensor data into the historical loading database to retrain and update the sensor time series prediction model.

[0080] Step 4.4, based on the multi-level mechanical response field inversion reduced order model constructed in the second step, the future loading step measuring point data is input to realize real-time inversion of the future mechanical response field. Among them, the measuring point data includes strain and displacement data; the mechanical response field includes strain, stress, plastic strain field and displacement field.

[0081] Further, after the test is completed, the input measuring point data and the inverted mechanical response field are used as snapshot matrix elements to update the snapshot matrix and re-reduce to realize dynamic updating of the reduced order model.

[0082] Step 4.5, based on the inverted mechanical response field, the full-field damage factor value corresponding to the damage criterion used is calculated to obtain the full-field damage of the structure and perform the third-level failure judgment, and the loading level at which the damage factor reaches the failure threshold is predicted. Specifically:

[0083] The commonly used damage criteria for calculating damage factors include: ductile metal damage criterion, Johnson-Cook criterion, Hashin damage initiation criterion; the damage factor value range is 0-1, and the failure threshold is usually taken as 1. When the damage factor value is equal to 0, it can be considered that the structure is undamaged and the stiffness is not degraded; when the damage factor is greater than 0 and less than 1, it can be considered that the structure is partially damaged and the stiffness begins to be reduced; when the damage factor is greater than 1, the material is completely failed and loses the carrying capacity.

[0084] Step 4.6, after the damage factor value is calculated, whether the failure occurs is determined by comparing with the failure threshold, and based on the time series predicted loading strain curve, the loading level at which the damage factor value reaches the failure threshold is predicted, and a warning is given.

[0085] The beneficial effects of the present application are:

[0086] (1) The present application can realize real-time monitoring of the test process through nonlinear modeling and data-driven technology.

[0087] (2) In view of the problem that it is difficult to accurately simulate the actual test conditions, a digital tooling system is established; in view of the problem of large amount of nonlinear analysis calculation, fine modeling and nonlinear attribute assignment are performed on the dangerous area, which effectively controls the calculation scale while ensuring the calculation accuracy; in view of the problem of sensor failure that may occur in the test process, an abnormal measurement point data reconstruction mechanism is established based on the data filling method; at the same time, combined with the non-invasive reduced-order model and the time series prediction algorithm, a multi-level failure judgment method from the whole field to the dangerous local area, from the current loading state to the future loading condition is constructed, realizing the test risk digital twin early warning of multi-domain collaborative monitoring.

[0088] In summary, the present application combines the advantages of nonlinear modeling and data-driven technology, improves the characterization ability of the nonlinear behavior of the structure while ensuring the calculation efficiency, and provides a technical approach for continuous updating and optimization of the model after the test, and provides a new solution for the safety monitoring of complex structure test. BRIEF DESCRIPTION OF DRAWINGS

[0089] Figure 1 It is a flow chart for realizing the test risk digital twin early warning method of multi-domain collaborative monitoring;

[0090] Figure 2 It is a schematic diagram of the assembly of the stiffened cylinder structure test system;

[0091] Figure 3 It is a schematic diagram of the stiffened cylinder structure test patch scheme;

[0092] Figure 4 It is a schematic diagram of the assembly effect and simulation effect of the digital sensor; Figure 4 a in the figure is a digital sensor assembly effect, Figure 4 b in the figure is a digital sensor simulation effect schematic diagram;

[0093] Figure 5 It is a schematic diagram of the full-field von Mises stress inversion result of the stiffened cylinder structure;

[0094] Figure 6 It is a schematic diagram of the plastic strain field inversion result of the nonlinear region of the stiffened cylinder structure; Figure 6 a in the figure is a PE11 full-field cloud map, Figure 6 b in the figure is a PE22 full-field cloud map, Figure 6 c in the figure is a PE33 full-field cloud map, Figure 6 d in the figure is a PE12 full-field cloud map,Figure 6 e is a PE13 full-field cloud map in the figure, Figure 6 f is a PE23 full-field cloud map in the figure;

[0095] Figure 7 is a future loading step failure judgment and early warning schematic diagram of the stiffened cylinder structure test; Figure 7 a is a displacement-load curve schematic diagram of the stiffened cylinder structure test in the figure, Figure 7 b is a sensor data prediction schematic diagram of the future loading step of the stiffened cylinder structure in the figure;

[0096] Figure 8 is a full-field damage factor calculation result schematic diagram of the nonlinear region of the stiffened cylinder structure;

[0097] In the figure: 1 loading joint, 2 stiffened cylinder, 3 loading lever, 4 electric push rod, 5 base table, 6 tooling beam, 7 foundation table. DETAILED DESCRIPTION

[0098] The following will be combined with a specific implementation example, and the technical solutions involved in the present application will be clearly and completely described. The specific embodiments described herein are only used to explain the present application, and not to limit the present application. In addition, it should be noted that, in order to facilitate description, only the parts related to the present application are shown in the drawings, not all the contents. Based on the examples in the present application, all other examples obtained by those skilled in the art without creative labor are within the scope of the present application.

[0099] Figure 1 is a flow chart of the technical route realized by the present application example. According to the steps described in the flow chart, the technical solution of the present application example is realized, including:

[0100] First, a fine finite element simulation model considering nonlinearity is established, which includes digital tooling system, digital sensor and test piece. Specifically:

[0101] Step 1.1, based on the geometric model of the structure, the finite element simulation model is constructed, and the nonlinear region is divided according to the location of the dangerous concern area. In this embodiment:

[0102] The digital tooling system, digital sensor and test piece finite element model are established for the stiffened cylinder test piece. The digital tooling system includes loading compression rod, loading joint, electric push rod, base table, tooling beam and foundation table. The digital sensor is a digital strain sensor. The radius of the stiffened cylinder test piece is , the height is , the radius of the base table is , and the height is The digital tooling system is modeled using C3D8 elements, and the material is high-strength alloy steel, with an elastic modulus of 210,000 MPa, a Poisson's ratio of 0.3, and a density of 7.8 g / cm 3 ; the test specimen material is 2024 aluminum alloy, with an elastic modulus of 85,413 MPa, a Poisson's ratio of 0.33, and a density of 2.78 g / cm 3 The digital strain sensor is modeled using a shell S4R element, with a size of 2x1 mm and a thickness of 0.2 mm. The material properties are consistent with the properties of the patch at the location. For the local dangerous detail part, the material linear-nonlinear region interface is divided, and the material and contact nonlinear region is set.

[0103] Step 1.2, based on the nonlinear region obtained in step 1.1, fine meshing is performed, and linear properties are applied to the linear region, and nonlinear properties are applied to the nonlinear region. In this embodiment:

[0104] For the contact nonlinear region, the screw hole surface and the screw surface are set with contact, where the tangential behavior is set as a penalty function, the friction coefficient is set to 0.1, and the normal behavior is set as hard contact. For the material nonlinear region, in addition to applying metal elastic properties (consistent with the linear region of the test specimen), the material nonlinear properties are obtained based on the uniaxial tensile test of a standard 2024 aluminum alloy rod specimen with a diameter of 5.991 mm at room temperature, and the ductile metal damage criterion is selected for application. The specific properties are as follows: yield stress is 336 MPa, ultimate stress is 550 MPa, corresponding plastic strain is 0.04933 mm, stress triaxiality is 0.33, strain ratio is 0.0067, and fracture strain is 0.04933 mm.

[0105] Step 1.3, parameterized script is used to complete the assembly of the digital tooling system, digital sensor and digital test specimen, and a fine finite element model including the digital tooling system, digital sensor and test specimen is established. In this embodiment:

[0106] For the assembly of the digital tooling system and the test specimen, the connection between the loading compression rod and the loading compression rod, the loading compression rod and the loading adapter, the loading adapter and the test specimen, the loading adapter and the tooling beam, the tooling beam and the foundation platform, the base platform and the test specimen, the loading compression rod and the test specimen, the base platform and the foundation platform, the electric push rod and the loading compression rod, and the electric push rod and the foundation platform is completed, the connection form is binding, and a face set is established at the top end of the loading lever. The assembled test system is shown in Figure 2 .

[0107] For the parameterized assembly of the digital strain sensor, the Figure 3The test patch scheme shown is based on the getByBoundingSphere detection parameterized algorithm developed by Abaqus to locate the target unit. Specifically, taking the coordinates of the sensor in the global coordinate system as the center, the search radius is gradually expanded until the patch unit is found. Subsequently, the FaceToFace fitting algorithm is applied to parameterize the geometric constraint relationship to ensure accurate fitting of the sensor and the target unit surface. Then, the RadialInstancePattern rotation replication function is used to generate a three-way digital strain sensor. Finally, the connection relationship between the digital strain sensor and the specimen is established. By establishing the binding of the strain sensor and the specimen unit surface and specifying the binding tolerance size, the above process is repeated to sequentially parameterize and assemble all digital strain sensors according to the test point layout scheme. Taking No. 6 strain gauge as an example, the parameterized and assembled digital strain sensor and the maximum principal strain cloud of the digital strain sensor and the unit where the patch is located are shown. Figure 4

[0108] Second, based on the fine finite element simulation model considering nonlinearity constructed in the first step, nonlinear finite element simulation analysis is performed, and based on the simulation results, the mechanical response results are extracted and the corresponding data set is established to construct a multi-level mechanical response field inversion reduced order model. Specifically:

[0109] Step 2.1, the fine finite element simulation model considering nonlinearity constructed in the first step is simulated and analyzed. Specifically, the model finite element simulation analysis results are obtained by calculating through static, explicit dynamics, implicit dynamics or Riks analysis step.

[0110] In this example, the fine finite element simulation model considering nonlinearity constructed in the first step is simulated and analyzed by static analysis step, and the model finite element simulation analysis results are obtained.

[0111] Step 2.2, based on the model finite element simulation analysis results obtained, sensitivity analysis is performed based on the model parameters.

[0112] The sensitivity analysis method is used to analyze the material elastic modulus, Poisson's ratio, plasticity parameters and load parameters.

[0113] Step 2.3, select the Generate a number of sampling points in the design space for the model parameters with the highest sensitivity, and perform interval simulation sampling to obtain finite element model sampling results.

[0114] ​The two groups of model parameters with the highest sensitivity, i.e., the load size and the Young's modulus, are determined as the sampling parameters through the sensitivity calculation of step 2.2. The Latin hypercube sampling method is used to randomly extract, in the interval of 0.7-1.2 times the Young's modulus and the load size, 36 groups of simulation results through the parameterized loading program and the Abaqus finite element simulation software.

[0115] In step 2.4, the sample set is constructed according to the finite element model sampling results obtained in step 2.3, the snapshot matrix is assembled, and each snapshot sub-matrix after layering is obtained.

[0116] The multi-level mechanical response field inversion reduced-order model constructed in this example includes a linear region reduced-order model and a nonlinear region reduced-order model. The sample input of the linear region reduced-order model is the digital strain sensor result, and the output is the linear region full-field von Mises stress response; the sample input of the nonlinear region reduced-order model is the digital strain sensor result and the linear-nonlinear region interface von Mises stress result, and the output is the full-field plastic strain PE11, PE22, PE33, PE12, PE23, and PE13 values.

[0117] According to the test loading step, the simulation result frames are divided into 0-50% (10% interval division, a total of 5 frames), 50%-70% (5% division, a total of 4 frames), and 70%-100% (2% division, a total of 15 frames), and based on the snapshot matrix layering method ①, each loading step is expanded by 10% before and after the division window, the nonlinear region full-field plastic strain PE11, PE22, PE33, PE12, PE23, and PE13 values, the digital strain sensor result, the linear region full-field von Mises stress response, and the linear-nonlinear region interface von Mises stress result in the window are extracted from the simulation result, and each snapshot sub-matrix after layering is constructed according to the divided frame number.

[0118] In step 2.5, the snapshot sub-matrix after layering is subjected to eigenvalue decomposition by using the reduced-order method, the corresponding reduced-order basis vectors and reduced-order basis coefficients are extracted, the reduced-order basis vectors and the corresponding coefficients are screened by setting an energy threshold, and finally the reduced-order basis and the corresponding reduced-order basis coefficients are obtained. At the same time, the input matrix and the reduced-order coefficient set are used to train the reduced-order basis coefficient prediction proxy model of each reduced-order model.

[0119] The proper orthogonal decomposition (POD) technique is selected to perform non-invasive reduction on the input and output snapshot sub-matrices of each reduced-order model, and the reduced-order basis and the corresponding reduced-order basis coefficients are obtained. The specific principle is as follows:

[0120] The sample set is assembled into a snapshot matrix:

[0121] (10)

[0122] where, is the snapshot matrix, is the snapshot vector.

[0123] After obtaining the snapshot matrix, the snapshot matrix constructed in the first step is processed using the POD model reduction method to obtain the basis and calculate the reduced basis coefficients corresponding to the field data at each stage . Based on the snapshot matrix , the correlation matrix is calculated:

[0124] (11)

[0125] The eigenvalues and eigenvectors of the correlation matrix are solved, where the number of eigenvalues is , and the eigenvalues are arranged in descending order:

[0126] (12)

[0127] The minimum value of is selected, where satisfies , which is the truncation threshold, generally taking a value of 99% or a number closer to 1. It represents the energy occupied by the first order POD modal eigenvalue.

[0128] (13)

[0129] According to the first order eigenvectors, the reduced basis matrix is established:

[0130] (14)

[0131] In the formula, is the reduced basis vector that constitutes the reduced basis matrix.

[0132] Based on the reduced basis matrix , the reduced basis coefficients corresponding to the field data at each stage are calculated .

[0133] (15)

[0134] Finally, based on the input matrix and the reduced basis coefficient set, the Gaussian process regression method is used to train the reduced basis coefficient prediction proxy model of each reduced order model.

[0135] Thirdly, the data feature judgment criterion is established to screen the sensor data, and the abnormal data of the sensor is filled by the data filling algorithm to obtain the complete sensor data set. Meanwhile, the first level structure failure judgment is made based on the complete sensor data set. In step 3.1, during the test process, the data feature judgment criterion is used to screen the sensor data failure to screen and exclude the faulty sensors. Specifically:

[0136] Secondly, in the online stage, the sensor data is screened according to the established criterion, and the screening result shows that the 1st, 5th and 8th measuring points appear abnormal collection, among which the 1st and 5th measuring points do not meet the physical rationality criterion, and the 8th measuring point does not meet the white noise failure criterion. No measuring point meets the strain change trend criterion and data stability criterion.

[0137] In step 3.2, the sensor data filling model is constructed by the data filling algorithm offline, and the missing data of the failed sensor is reconstructed by mapping the actual sensor data online to realize the data filling of the faulty sensor and construct the complete sensor data set, and the first level failure judgment is made.

[0138] According to the digital strain sensor measurement value obtained by sampling, the sensor data filling model is constructed based on the data filling method in step 3.2. In this example, the interval proper orthogonal decomposition technology is selected for construction. The construction principle is as follows:

[0139] Based on the digital sensor, the sensor measurement value simulation data is obtained, and a snapshot matrix composed of all digital sensor simulation data is constructed :

[0140] (16)

[0141] wherein, is a vector composed of all sensor readings under the th working condition, is the number of strain sensors, is the total number of working conditions sampled. Offline, based on the POD reduction method described above, the snapshot matrix is subjected to POD decomposition, and the first order main mode is extracted by truncation screening to construct a reduced basis matrix , forming a reduced model.

[0142] Online, the data of the 1st, 5th and 8th measuring points is replaced with null values, and the data of the remaining normal test measuring points is used as the input of the interval proper orthogonal decomposition model to obtain the replacement values of the 1st, 5th and 8th measuring points in real time. The specific principle of interval proper orthogonal decomposition online data filling is as follows:

[0143] Online, define the corresponding normal sensor vector All normal sensor indices are contained in the mask matrix Each row is a unit vector corresponding to the normal sensor index. Normal sensor data as input:

[0144] (17)

[0145] Where is the number of normal sensors. First, center the data:

[0146] (18)

[0147] Extract the reduced basis submatrix Solve the reduced basis coefficients by least squares :

[0148] (19)

[0149] We get:

[0150] (20)

[0151] Using the reduced basis coefficients obtained Reconstruct the data:

[0152] (21)

[0153] De-center to get the predicted sensor data:

[0154] (22)

[0155] Replace the missing abnormal sensor data with Combine with normal sensor data to form a complete sensor data set, complete the process of sensor missing data prediction and replacement based on interval eigenvalue orthogonal decomposition. The results show that compared with the collected normal data, the average relative error of the replacement value is only 3.5%.

[0156] In the test, based on the above complete sensor data set, the first hierarchical structure failure judgment is carried out, and the maximum sensor strain value is , which is less than the material failure strain determined by the standard 2024 aluminum alloy rod-shaped test specimen uniaxial tensile test before the test, and the first hierarchical structure failure judgment result is not failed.

[0157] Step 4.1, input the complete sensor dataset obtained in step 3.2 online, obtain the reduced basis coefficients of each reduced model based on the reduced basis coefficient prediction surrogate model constructed in step 2.5, and multiply the reduced basis coefficients by the reduced basis to complete the upscaling inversion of the mechanical response field of the specimen, to obtain the full-field mechanical response field results. Specifically:

[0158] Input the complete sensor dataset (30% loading level) obtained in step 3 into the full-field reduced model established, and upscale to obtain the full-field von Mises stress results.

[0159] Step 4.2, based on the complete sensor dataset and the numerical calculation of the sensor measurement point position residual, a full-field response correction surrogate model is constructed to correct the mechanical response field of the specimen online in real time. Specifically:

[0160] An RBF bridge function is used to establish an online full-field von Mises stress response correction surrogate model to correct the full-field von Mises stress response, as shown in Figure 5 The obtained full-field von Mises stress response prediction results are compared with the test results using leave-one-out validation, i.e., using 26 test point data from 27 measurement points to construct the surrogate model each time, and predicting the data of the remaining 1 point, and repeating 27 times. The relative root mean square error RRMSE and the coefficient of determination R 2 are used as accuracy evaluation criteria, and their formulas are as follows:

[0161] (23)

[0162] (24)

[0163] The leave-one-out validation results show that the RRMSE is 0.096, and the R 2 is 0.9557.

[0164] Step 4.3, offline establishment of a historical loading database to train a sensor time series prediction model, online prediction of sensor data at future loading steps, and completion of the second level of failure determination. Specifically:

[0165] The historical loading database composed of the measurement values of the digital strain sensors of each loading stage is established offline based on simulation data before the test, and the time sequence relationship between the loading stage and the measurement values of the digital strain sensors is trained through a time sequence prediction algorithm. In this example, the LSTM algorithm is used for the time sequence prediction algorithm, and the specific principle is as follows:

[0166] For the time sequence data of the historical loading database, in each time step, the LSTM network training will perform the following processes:

[0167] (1) Forget gate: decides which information is forgotten.

[0168] (25)

[0169] wherein, is a sigmoid activation function, is the weight of the forget gate, is a bias term, is the output of the previous time, is the current input.

[0170] (2) Input gate and candidate cell state: decide which new information to write to the cell state.

[0171] (26)

[0172] (27)

[0173] wherein, tanh is the hyperbolic tangent function, and are the weights of the input gate and the candidate cell state, respectively.

[0174] (3) Update cell state : update the cell state, forget the state of the previous time through the forget gate, and update the current state through the input gate.

[0175] (28)

[0176] (4) Output gate : decides how much information the current cell state outputs.

[0177] (29)

[0178] The current output can be calculated as:

[0179] (30)

[0180] Then, based on the established sensor time series prediction model, the complete sensor data set obtained in the third step (30% loading level) is input online to predict the sensor data at the 80% loading level. Taking the 14th strain channel as an example, the prediction effect is shown in Figure 7 . The predicted maximum von Mises stress is 252.1 MPa, which is less than the material yield limit, and the second hierarchical structure failure judgment is not failed. Compared with the data obtained by normal loading, the average error of 81 strain channels is 5.1%.

[0181] Step 4.4, based on the multi-level mechanical response field inversion reduced order model constructed in the second step, the future loading step test point data is input to realize the real-time inversion of the future mechanical response field. Specifically:

[0182] Then, based on the established multi-level mechanical response field inversion reduced order model, the predicted 80% loading level sensor data is input online to obtain the predicted values of the full-field plastic strain PE11, PE22, PE33, PE12, PE23, and PE13 in the nonlinear region at the 80% loading level, as shown in Figure 6 , and the time consumption is 0.0073s.

[0183] Step 4.5, based on the inverted mechanical response field, the full-field damage factor value corresponding to the damage criterion used is calculated to obtain the full-field damage of the structure, and the third-level failure judgment is performed, and the loading level at which the damage factor reaches the failure threshold is predicted. Specifically:

[0184] In this example, the damage factor calculation method corresponding to the ductile metal damage criterion is used to calculate the full-field damage factor based on the full-field plastic strain inversion result, and the calculation process is as follows:

[0185] First, the equivalent plastic strain is calculated, and the plastic strain tensor satisfies the incompressibility of volume. The equivalent plastic strain increment calculation formula is:

[0186] (31)

[0187] wherein, the plastic strain increment components in each direction, is the plastic shear strain increment in the xy plane.

[0188] Secondly, the cumulative equivalent plastic strain is calculated:

[0189] (32)

[0190] Finally, according to the cumulative equivalent plastic strain and the fracture strain, the damage factor is calculated:

[0191] (33)

[0192] wherein, is the material fracture strain.

[0193] The final calculation of the full-field damage factor of the nonlinear region of the local detail model at 80% loading level is shown in Figure 8 .

[0194] Step 4.6, after the calculation of the damage factor value, determine whether the failure is by comparing with the failure threshold, and based on the time sequence predicted loading strain curve, predict the loading level at which the damage factor value reaches the failure threshold, and give a warning.

[0195] The maximum value of the calculated full-field damage factor is 0.963, which indicates that the structure has entered a partially damaged state at this loading level, but has not yet failed. The third failure determination is not failed. According to the prediction, at 86% loading level, the maximum value of the damage factor reaches the threshold 1, and the structure will fail. The failure determination warning effect is shown in Figure 7 .

[0196] Finally, it should be particularly noted that the above-mentioned embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the scope of protection. For those skilled in the art, on the basis of a full understanding of the foregoing embodiments, the technical solutions can be reasonably adjusted or some technical features can be replaced with equivalent ones. The modified or replaced scheme shall still belong to the protection scope defined by the claims of the present application.

Claims

1. A test risk digital twin early warning method of multi-domain collaborative monitoring, characterized in that, The test risk digital twin early warning method comprises the following steps: First, a fine finite element simulation model considering nonlinearity is established, which includes a digital tooling system, digital sensors and a test piece; Second, based on the fine finite element simulation model established in the first step, nonlinear finite element simulation analysis is performed, sampling is performed based on the simulation results, mechanical response results are extracted and a corresponding data set is established, and a multi-level mechanical response field inversion reduced model is constructed; Third, a data feature judgment criterion is established to screen sensor data, and abnormal measurement point data is reconstructed through a data filling algorithm to obtain a complete sensor data set; meanwhile, the first level structure failure judgment is performed based on the complete sensor data set; Fourth, through the multi-level mechanical response field inversion reduced model established in the second step and the complete sensor data set established in the third step, full-field mechanical response inversion and online real-time correction are performed, the second level full-field dangerous area judgment is performed based on the full-field inversion data; and a time sequence prediction model established based on historical loading data is used to predict sensor data of a future loading stage, response inversion of a concerned area is performed, real-time future risk warning is performed through calculation of a nonlinear region damage factor, and the third level structure failure judgment is realized.

2. The test risk digital twin early warning method of multi-domain cooperative monitoring according to claim 1, characterized in that, The first step specifically comprises: Step 1.1, based on the geometric model of the structure, a finite element simulation model is constructed, and according to the location of the dangerous concerned area, nonlinear regions are divided; Step 1.2, based on the nonlinear regions obtained in step 1.1, fine meshing is performed, and linear properties are applied to linear regions and nonlinear properties are applied to nonlinear regions; Step 1.3, parameterized scripts are used to complete the assembly of the digital tooling system, digital sensors and digital test pieces, and a fine finite element model including the digital tooling system, digital sensors and test pieces is established; wherein the digital tooling system, the digital tooling system, the digital sensors and the test pieces are connected through binding, MPC beam or coupling.

3. The test risk digital twin early warning method of multi-domain cooperative monitoring according to claim 2, characterized in that, In the first step: In step 1.1, the nonlinear regions should be arranged at positions meeting the following requirements: ① the linear-nonlinear region interface should not be too close to the dangerous region where nonlinearity occurs; ② the linear-nonlinear region interface should not be located in a region where the mechanical behavior changes significantly, with a stress gradient > 20% / mm; For complex structures with many divided meshes, substructure method or submodel method can also be used for finite element model construction; In step 1.2, the linear properties include material linear properties, and the nonlinear properties include material nonlinear properties and contact nonlinear properties; the material nonlinear properties include material plasticity and damage criteria, wherein the material plasticity is set by yield criterion, hardening model, flow rule or porous metal plasticity model, and the damage criterion is set by ductile metal damage criterion, Johnson-Cook criterion or Hashin damage initiation criterion; the contact nonlinear properties are set by hard contact method, penalty function method, Lagrange multiplier method or augmented Lagrange method.

4. The test risk digital twin early warning method of multi-domain cooperative monitoring according to claim 1, characterized in that, The second step specifically comprises: Step 2.1, simulation analysis is performed on the fine finite element simulation model considering nonlinearity constructed in the first step; specifically, calculation is performed through static force, explicit dynamics, implicit dynamics or Riks analysis step to obtain model finite element simulation analysis results; Step 2.2, sensitivity analysis is performed based on model parameters according to the obtained finite element simulation analysis results; Step 2.3, select the two groups of model parameters with the highest sensitivity to generate a plurality of sampling points in the design space, perform interval simulation sampling to obtain finite element model sampling results; Step 2.4, construct a sample set according to the finite element model sampling results obtained in step 2.3, assemble a snapshot matrix, and obtain each snapshot sub-matrix after layering; Step 2.5, perform eigenvalue decomposition on each snapshot sub-matrix after layering by using a reduction method, extract the corresponding reduced basis vectors and reduced basis coefficients, and realize the truncation screening of the reduced basis vectors and the corresponding coefficients by setting an energy threshold, and finally obtain the reduced basis and the corresponding reduced basis coefficients; meanwhile, train the reduced basis coefficient prediction proxy model based on the input matrix and the reduced coefficient set to construct a multi-level mechanical response field inversion reduced model.

5. The test risk digital twin early warning method of multi-domain cooperative monitoring according to claim 4, characterized in that, In the second step: In step 2.2: The specific procedure for sensitivity analysis is as follows: for the model as shown in equation (1) has: （1） wherein, is a model parameter, is the total number of parameters, is a parameter used the finite element results calculated; the local sensitivity of the parameters is: （2） The central difference method is used to approximate formula (2) to obtain: （3） wherein, represents a model parameter an increase in the amount of The model parameters include material elastic modulus, Poisson's ratio, plasticity parameters, load size, position and direction; In step 2.3, the design space is a load design domain based on multiple test conditions, and the interval simulation sampling method includes Latin hypercube sampling, uniform random sampling or stratified sampling; In step 2.4, the snapshot matrix constructed is , where the number of matrix rows represents the total number of finite element nodes of the test piece, and the number of matrix columns represents the total number of frames of all simulation results, and the input matrix , where the number of matrix rows is the number of digital sensor channels; meanwhile, the snapshot matrix is processed in layers, and based on the snapshot matrix layering result, the input matrix is layered according to the corresponding frame number to obtain the snapshot sub-matrix of the input matrix; In step 2.5, the reduction method includes eigenvalue orthogonal decomposition, Krylov method, dynamic modal decomposition, random projection, non-negative matrix factorization or canonical polyadic decomposition; In step 2.5, the multi-level mechanical response field inversion reduced model includes a linear region reduced model and a nonlinear region reduced model; the sample input of the linear region reduced model is the digital displacement and strain sensor simulation result, and the output is the full-field mechanical response of the linear region; the sample input of the nonlinear region reduced model is the digital displacement, strain sensor simulation result and mechanical response of the linear region on the interface, and the output is the full-field mechanical response field of the nonlinear region; In step 2.5, the screening method of the reduced basis vectors and the corresponding coefficients is that the energy threshold used for truncation screening is not less than 99%; In step 2.5, the construction method of the reduced basis coefficient prediction proxy model includes Gaussian process regression, random forest, support vector machine, deep neural network or gradient boosting tree.

6. The test risk digital twin early warning method of multi-domain cooperative monitoring according to claim 5, characterized in that, In step 2.4, the method for constructing the snapshot sub-matrix includes: According to the test conditions to be carried out, simulation analysis is carried out, and output frames are set according to the test loading step length, and the intensity mechanical response field of each frame is extracted to construct a snapshot matrix; for each loading level, the window is expanded by 5% frames forward and backward based on the loading time, and all output frames in the window are recombined into a snapshot sub-matrix; Or snapshot matrix segmentation based on correlation, first calculate the Pearson correlation coefficient matrix between the results of all sampling results, then use hierarchical clustering or K-means clustering algorithm, clustering based on Pearson correlation coefficient, finally according to the clustering results corresponding column is assigned to each sub-matrix, create snapshot sub-matrix; Or snapshot matrix segmentation based on principal component analysis, first centralization and standardization processing of snapshot matrix, secondly, principal component analysis is carried out on the processed matrix to obtain the load matrix, then using simple threshold method, K-means clustering method or rotating principal component method to group the load matrix, finally according to the grouping result to create snapshot sub-matrix.

7. The test risk digital twin early warning method of multi-domain cooperative monitoring according to claim 6, characterized in that, The third step is specifically: Step 3.1, in the test process, using data feature judgment criterion to screen sensor data failure, to screen and exclude the sensor with fault; The data feature judgment criterion includes physical rationality criterion, abnormal value criterion, strain change trend criterion, data stability criterion, white noise fault criterion and zero drift criterion; The data set used for judgment includes historical sensor data set composed of historical test data and current sensor data set composed of current test data; Step 3.2, offline sensor data filling model is constructed by data filling algorithm, online missing data of failed sensor is reconstructed by using actual sensor data for mapping, data filling of fault sensor is realized, complete sensor data set is constructed, and first level failure judgment is carried out.

8. The test risk digital twin early warning method of multi-domain cooperative monitoring according to claim 7, characterized in that, In the third step: The step 3.1 is specifically as follows: (1) the formula of physical rationality criterion is: （4） wherein, is a strain value of the current sensor data set; is a maximum allowed strain value of the material; is a minimum allowed strain value of the material; (2) the criterion formula of abnormal value criterion is: （5） wherein, is a response value for the current sensor data set; is a mean value for the historical sensor data set; is a standard deviation for the historical sensor data set; (3) the criterion formula of strain change trend criterion is: （6） wherein, is a load step sequence for the current sensor data set; is a response data sequence for the current sensor data set; is a is the number of data points in the sequence. (4) the criterion formula of data stability criterion is: （7） wherein, is the response value of the current sensor data set at the loading level; is the response value of the current sensor data set at the loading level; is the standard deviation of the historical sensor data set; is a threshold coefficient, typically in the range of 2-3; (5) the criterion formula of white noise fault criterion is: （8） wherein is a white noise, i.e. a random signal with unknown mean and covariance; (6) the criterion formula of zero drift criterion is: （9） wherein, is a zero point sensor value for the current sensor data set; is a maximum value of the allowed zero point sensor values; In the process of judging sensor failure, structure failure judgment is carried out at the same time, the sudden slope mutation in the loading process is further judged through the strain change trend criterion and the data stability criterion in the criterion, and the first level failure judgment is completed; In the step 3.2, the data filling algorithm refers to the process of reasonably estimating and reconstructing the missing part of the overall data by using mathematical modeling or statistical method; The data filling algorithm includes linear interpolation method, singular value decomposition method, interval proper orthogonal decomposition method, random forest method or generative adversarial network.

9. The test risk digital twin early warning method of multi-domain cooperative monitoring according to claim 8, characterized in that, The fourth step is specifically: Step 4.1, input the complete sensor data set obtained in step 3.2 online, obtain the reduced order basis coefficients of each reduced order model based on the reduced order basis coefficient prediction proxy model constructed in step 2.5, complete the upscaling inversion of the strength mechanical response field by multiplying the reduced order basis coefficients with the reduced order basis, and obtain the full field mechanical response field result; Step 4.2, based on the complete sensor data set and the numerical calculation of sensor measuring point position, the residual error is calculated, and the full field response correction proxy model is constructed to correct the strength mechanical response field of the test piece after upscaling inversion online in real time; Specifically: According to the coordinates of the sensors in the test patch scheme, the response values at the corresponding coordinates in the specimen strength mechanical response field predicted in step 4.1 are extracted, the deviation values between the predicted response values of the sensors and the measured values of the test sensors are calculated, a full-field response correction surrogate model is constructed based on the relationship between the deviation values of each sensor and the corresponding coordinates, and the global correction deviation values obtained by superposition are combined with the specimen strength mechanical response field obtained in step 4.1 to realize online dynamic correction of the global strength mechanical response field; In step 4.3, a historical loading database is established, a sensor time series prediction model is trained, sensor data at future loading steps are predicted, and second-level failure determination is completed; the historical loading database includes simulation sample data at measuring points and historical working condition loading sensor data; After the test is completed, the normal sensor data and the replaced fault sensor data in step 3.2 are added to the historical loading database as historical working condition loading sensor data, and the sensor time series prediction model is retrained and updated; In step 4.4, based on the multi-level mechanical response field inversion reduction model constructed in the second step, the future mechanical response field is realized in real time by inputting the predicted measuring point data at the future loading steps; the measuring point data includes strain and displacement data; the mechanical response field includes strain, stress and plastic strain field; After the test is completed, the input measuring point data and the inverted mechanical response field are used as snapshot matrix elements to update the snapshot matrix and re-reduce, so as to realize dynamic updating of the reduction model; In step 4.5, based on the inverted mechanical response field, the full-field damage factor values corresponding to the damage criteria are calculated, the full-field damage of the structure is obtained, the third-level failure determination is performed, and the loading level at which the damage factor reaches the failure threshold is predicted; In step 4.6, after the damage factor value is calculated, whether the failure occurs is determined by comparing with the failure threshold, and based on the time series predicted loading strain curve, the loading level at which the damage factor reaches the failure threshold is predicted, and a warning is given.

10. The test risk digital twin early warning method of multi-domain cooperative monitoring according to claim 9, characterized in that, In the fourth step: In step 4.2, the method for training the full-field response correction surrogate model includes radial basis function, Gaussian process regression, random forest, transfer learning neural network or incremental learning gradient boosting tree; In step 4.3, the sensor time series prediction model is trained, the time series characteristics of the influence of load change on the response of the measuring point are captured through the time series prediction method according to the time series data of the historical loading database, the time series prediction is performed with the load change, the short-term and long-term dependence relationships in the time series data are identified through the time series dependence characteristics of the time series prediction algorithm, so that the sensor time series prediction model can still accurately predict the response of the measuring point when facing complex load changes; In the online stage, based on the established sensor time series prediction model, the sensor data set composed of the normal sensor data and the replaced fault sensor data in step 3.2 is input to realize real-time sensor future prediction, and the first-level failure determination method in step 3.1 is combined with the predicted value to realize second-level failure determination. ​ In step 4.5, the damage criterion for calculating the full-field damage factor includes: a ductile metal damage criterion, a Johnson-Cook criterion, and a Hashin damage initiation criterion; The damage factor ranges from 0 to 1, and the failure threshold is usually 1; when the damage factor value is equal to 0, it is considered that the structure is undamaged and the stiffness is not degraded; when the damage factor is greater than 0 and less than 1, it is considered that the structure is partially damaged and the stiffness begins to be reduced; when the damage factor is greater than 1, the material completely fails and loses the carrying capacity.

Images