Bridge dynamic interaction analysis method fusing monitoring data and hybrid substructure

Abstract

The invention discloses a bridge dynamic interaction analysis method fusing monitoring data and a mixed substructure. The method comprises the following steps: firstly, determining an interface constraint condition of a key area by establishing a mathematical equation by utilizing a mixed substructure technology; then, an expectation-oriented adaptive sparse Gaussian process algorithm is introduced, and parameter identification is performed on the key areas based on actual monitoring data. Furthermore, the invention provides a new theoretical framework which is used for realizing dynamic interaction between nonlinear analysis of the substructure and linear elasticity analysis of the main structure. The effectiveness of the method is verified by an application example on an actual concrete beam bridge. According to the method, a new way is opened up for deep fusion of the monitoring data and the finite element model, and the accuracy of substructure analysis and overall structure performance evaluation is remarkably enhanced.

Description

Technical Field

[0001] The present invention relates to the fields of structural safety evaluation and data processing technology, and in particular to a hybrid substructure bridge dynamic interaction analysis method that integrates detection and monitoring data. Background Art

[0002] Existing structures are prone to structural stress changes due to cracks and other defects and changes in material properties, which can affect the evaluation of structural service performance. Bridge structure health monitoring and detection technology can provide structural disease information in a timely and effective manner, making it possible to evaluate the structural condition. However, for structures with large size and complex stress states, it is difficult to truly and effectively analyze and evaluate the structural service performance using only structural detection and monitoring data. The development of finite element theory has facilitated the analysis of bridge structures, but the existing methods still have the following problems in the integration of inspection and monitoring data with finite element models:

[0003] First, existing rapid detection methods for external defects mainly focus on the qualitative identification, positioning, and quantification of structural defects based on detection data, and fail to establish an effective connection between the detection data and the local degradation performance of the structure.

[0004] Second, non-contact monitoring methods mainly focus on measuring structural deformation, while less research is done on the relationship between monitoring data and the intrinsic performance of the structure.

[0005] Third, the finite element model correction technology mainly achieves the consistency between the theoretical value and the monitoring value of the measuring point by modifying the internal parameters of the model, but these corrections often reflect the overall structural properties and fail to make full use of the inspection and monitoring data, resulting in large errors in the structural assessment results.

[0006] In summary, the finite element analysis method used in the existing structural safety performance assessment process has a very limited degree of integration of inspection and monitoring data, especially non-contact sensing data, which makes it easy for inaccuracies to occur in structural analysis and assessment. Summary of the invention

[0007] The present invention aims to provide a hybrid substructure bridge dynamic interaction analysis method that integrates detection and monitoring data to solve the problem that dynamic monitoring data and finite element models are difficult to effectively integrate in the prior art. By combining dynamic monitoring data with finite element models, this method can not only perform detailed analysis of key areas, but also feed back the analysis results of key areas to the main simplified finite element model, thereby improving the accuracy of structural evaluation.

[0008] In order to achieve the above technical objectives, the present invention adopts the following technical means:

[0009] The hybrid substructure bridge dynamic interaction analysis method integrating detection and monitoring data includes the following steps:

[0010] S1. According to the structural stress characteristics, the key areas are selected as substructures and refined finite element modeling is adopted;

[0011] S2, automatically segment cracks and calculate crack width based on deep learning model;

[0012] S3. Based on the above crack width data and corresponding damage mechanism, the weakened element method is used to consider the existing damage in the substructure;

[0013] S4, establish the monitoring data of other parts except the substructure and the mathematical equations of the large-scale finite element model and solve the dynamic boundary conditions of the substructure;

[0014] S5. Based on the substructure monitoring data, the expectation-oriented adaptive sparse Gaussian process method is used to identify the substructure parameters, thereby achieving independent calculation of the substructure;

[0015] S6. A dynamic interaction analysis method between the substructure and the main structure is proposed to convert the dynamic response of the substructure interface into the equivalent external dynamic load of the main structure, thereby realizing the dynamic coupling between the two and enabling the substructure information to be transmitted to the main structure. The specific steps include the following:

[0016] S61, taking out a certain part of the large-scale structure to establish a substructure model, and at the same time establishing a main structure model;

[0017] S62, by introducing nonlinear correction force to achieve the coordination of displacement boundary and force balance between the main structure and substructure;

[0018] S63, the substructure is modeled using solid elements, the main structure is modeled using beam elements, and the two are coupled through multi-point constraints;

[0019] S64, nonlinear analysis of substructure and linear elastic analysis of main structure;

[0020] S65, determining material properties through substructure monitoring data, and adjusting material constitutive parameters to update to corresponding parts of the main structure;

[0021] S66, apply the nonlinear correction force as an external load to the main structure;

[0022] S67. Ensure the displacement coordination between the main structure and the substructure by deriving simplified bridge structure formulas;

[0023] S68, solving the equilibrium equations between the main structure and the substructure by a non-iterative solution method;

[0024] S7. Feedback the results of the substructure nonlinear analysis to the main structure to achieve equivalent dynamic linear analysis.

[0025] Step S2 specifically includes the following features:

[0026] The crack images are automatically segmented using a deep learning model. The model adopts a U-shaped structure consisting of an encoder and a decoder, and introduces a group multi-axis Hadamard product attention module to enhance the segmentation ability.

[0027] This deep learning model introduces a group aggregation bridge module at each stage between the encoder and decoder, introduces mask information to fuse and process multi-scale features, and uses dilated convolution to better extract rich feature information from different resolutions, further enhancing the segmentation ability of the model.

[0028] The loss function consists of binary cross entropy and Dice loss, which is used to optimize model training and ensure the best effect in the segmentation task.

[0029] In step S3, the weakened unit method includes the following sub-steps:

[0030] S31. Determine the weakened tensile properties based on the bilinear mode I stress-crack width relationship, and construct a stress-strain relationship suitable for the weakened unit by calculating the area under the ultimate strain-maintaining stress-strain curve.

[0031] S32, by not imparting tensile properties to finite elements with pre-existing crack overlap based on an assumed bilinear mode I stress-crack width relationship;

[0032] S33. Based on the measured width of the crack, the tensile strength and the residual properties of the crack under stress-free conditions are derived using a bilinear model;

[0033] S34. In the finite element model of concrete, there is a cracked area. The residual properties derived above are introduced to consider the impact of the cracks on the overall performance of the structure.

[0034] In step S4, the monitoring data of other parts include: displacement, strain and inclination.

[0035] Specific implementation of step S4:

[0036] S41. Establish mathematical models based on overall measurement data and mechanical information of large-scale finite element models. In dynamics, Among them, F represents the external force load, M and C are the structural mass and structural damping respectively, and x represents the structural displacement vector;

[0037] S42. Solving the boundary condition of the substructure by an adaptive sparsity matching pursuit method, wherein the boundary is a dynamic boundary.

[0038] In step S5, the expectation-guided adaptive sparse Gaussian process substructure parameter identification method includes:

[0039] S51, based on the optimal Latin hypercube design, by randomly generating a set of sample points p (1) ,p (2) ,…,p (N) Cover the entire design space; the initial sample points are around the current optimal point P optimal,t-1 Update within the range of ±20% to ensure that the sampling points cover the entire design space while focusing on the local area near the optimal point;

[0040] S52, applying time history displacement u PS,0-t , calculate the parameter set formed by the above sample points and obtain the corresponding response data

[0041] S53, calculating the error index R based on the current response data, and constructing a sparse Gaussian process model based on these data;

[0042] S54, calculating the expected value EI of the candidate sampling points, and selecting the point with the maximum expected improvement as the next sampling point;

[0043] S55, updating the sparse Gaussian process model based on the updated sampling points;

[0044] S56, repeat the above steps until the convergence criterion is met;

[0045] S57 , when the maximum value of the expected value EI of the candidate sampling points approaches zero or reaches a preset upper limit of the sampling points, the optimization is stopped.

[0046] The specific steps of the adaptive sparse Gaussian process substructure parameter identification method and the dynamic calculation time history t are:

[0047] (1) Solve the structural dynamics at time t and convert the corresponding displacement u NS,t Send to NS;

[0048] (2) Read the structural response R at the tth time step PS Send to SPI-EASG module;

[0049] (3) Based on P optimal,t-1 Calculation error NRMSE (R TS,0-t ,R PS,0-t ), and determine whether it exceeds the tolerance. If it exceeds the tolerance, calculate And obtain the new optimal parameter P from the SPI-EASG module optimal,t , otherwise, let P optimal,t =P optimal,t-1 ;

[0050] (4) Change NS from u NS,t-1 Load to uNS,t , and use P optimal,t-1 Calculate the response R NS,t ;

[0051] (5) Repeat steps 1 to 4 until the convergence criterion is met;

[0052] (6) When the convergence criterion is reached, SPI-EASG will output the optimal parameter combination P optimal , as the final parameter identification result of the finite element model.

[0053] In step S6, the implementation of the interaction between the substructure and the main structure specifically includes the following steps:

[0054] SA, extract the motion displacement of the control nodes C and D in the nth time step of the main structure isolation area and transmit it to the substructure, the motion displacement of the control nodes C and D in the nth time step of the main structure isolation area is used as the first interface physical quantity, and the first interface physical quantity is used as the displacement w' of the substructure control nodes C' and D' (m,n) Iteration initial value;

[0055] After receiving the above-mentioned first interface physical quantity, the SB,substructure performs a dynamic analysis and transmits the calculated reaction force R' of the substructure control nodes C' and D' back to the main structure;

[0056] SC, the main structure receives the reaction force R', calculates the external force and loads it to the main structure control nodes C and D, and performs a dynamic analysis;

[0057] SD, extract the displacement w of the main structure control nodes C and D c,(m,n) And transmit it to the substructure. According to the displacement coordination, the movement is the displacement w' of the substructure control node C', D' (m+1,n) , the substructure controls the displacement w' of nodes C', D' (m+1,n) As the second interface physical quantity;

[0058] After receiving the above-mentioned second interface physical quantity, the SE,substructure performs a dynamic analysis and transmits the calculated reaction force R of the substructure control nodes C', D' back to the main structure;

[0059] SF and the main structure receive the reaction force R, calculate the external force and load it to the main structure control nodes C and D, and then perform a dynamic analysis;

[0060] SG, extract the displacement w of the main structure control nodes C and D c,(m+1,n) , and transfer it to the substructure.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] First, realize the effective integration of inspection and monitoring data and finite element models: Traditional structural inspection and health monitoring methods face many challenges when directly evaluating the structural status, especially how to accurately integrate data from different sources. This invention proposes an innovative framework that effectively combines inspection and monitoring perception data with finite element models through hybrid substructure technology. This framework not only improves the accuracy of structural analysis and evaluation, but also solves the long-standing problem of separation between inspection and monitoring data and finite element models, providing a solid foundation for the safe operation of complex structures such as bridges.

[0063] Second, improve the ability of refined analysis of key areas: In order to meet the needs of refined analysis of key areas, the present invention proposes a hybrid substructure independent dynamic interaction method that integrates inspection and monitoring data. This method first uses solid units to establish a finite element model of the key area to improve the accuracy of the analysis. Then, based on the detection data, the structural damage and its dynamic degradation model are introduced to reflect the performance changes of the structure during actual use. The interface constraints of the substructure are resolved through monitoring data, and the key area parameter identification is realized based on the expectation-guided sparse Gaussian process, which ultimately realizes independent dynamic analysis of the substructure and significantly improves the accuracy of local analysis.

[0064] Third, dynamic interaction analysis between substructure and main structure: In view of the problem of insufficient interaction analysis between substructure and main structure, the present invention derives a hybrid substructure dynamic interaction analysis theory. This theory converts the dynamic response of the substructure interface into an equivalent external dynamic load of the main structure, thereby realizing dynamic coupling between the two and enabling the substructure information to be transmitted to the main structure. Through this interaction mechanism, the local dynamic response can be more accurately reflected in the overall analysis, thereby providing a more reliable structural assessment under complex loads and environmental changes.

[0065] Fourth, interactive analysis is applied to actual bridge dynamic calculations: Compared with traditional static analysis, the method of the present invention focuses on dynamic analysis and applies it to actual bridges. By introducing dynamic monitoring data and nonlinear analysis, the present invention can more accurately simulate the behavior of bridges under actual traffic loads and impact vibrations. This not only improves the accuracy of the analysis, but also makes the structural assessment results closer to the actual situation. In addition, the effectiveness of the method was verified through field tests, further proving its feasibility and reliability in actual engineering applications.

[0066] Fifth, improve the efficiency and accuracy of overall structural analysis: The feasibility of the proposed method is demonstrated through experimental verification of prestressed concrete beam bridges. The experimental results show that the proposed method can accurately evaluate the dynamic stress state of existing structures and significantly improve the accuracy of structural performance analysis of large and complex bridges. This method can effectively analyze the status of the overall structure and key areas while using a small number of sensors, thereby achieving a comprehensive assessment of important infrastructure such as bridges. The fusion of structural disease detection data significantly improves the accuracy of local analysis, and at the same time, the analysis efficiency of the overall structure is further improved through the dynamic interaction analysis between the substructure and the main structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is the basic idea of ​​the bridge dynamic interaction analysis method proposed by the present invention that integrates inspection and monitoring data with hybrid substructures;

[0068] Figure 2 The automatic crack segmentation algorithm based on deep learning proposed by the present invention;

[0069] Figure 3 This is a correlation diagram between crack width and weakening unit attribute proposed by the present invention;

[0070] Figure 4 This is a flow chart of the substructure analysis method for fusing measurement data and finite element models proposed in the present invention;

[0071] Figure 5 A simplified schematic diagram of the SPI-EASG proposed by the present invention;

[0072] Figure 6 Flow chart of the SPI-EASG proposed by the present invention;

[0073] Figure 7 It is a schematic diagram of the principle of the dynamic interaction analysis between the substructure and the main knot proposed in the present invention;

[0074] Figure 8 The concrete soil continuous box girder bridge mentioned as an example includes (a) elevation view; (b) photo of the actual bridge; (c) cross-sectional view;

[0075] Fig. 9 The schematic diagram of the sensor layout proposed for the example includes: (a) the overall situation of sensor layout; (b) the specific situation of accelerometer layout;

[0076] Fig.10 The crack identification results based on the deep learning model mentioned in the example;

[0077] Fig.11 Substructure analysis procedures for integrated detection and monitoring data;

[0078] Fig.12 The substructure parameter identification results include: (a) sampling samples; (b) impact load identification results; (c) vehicle dynamic load identification results;

[0079] Fig.13 The calculation results of the substructure proposed for the example;

[0080] Fig.14 The calculation results of the impact vibration strain mentioned in the example;

[0081] Fig.15 The calculation results of the dynamic load of the vehicle mentioned in the example. DETAILED DESCRIPTION

[0082] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0083] Some of the definitions involved in this application are as follows:

[0084] Non-contact sensing data: refers to the detection and monitoring data obtained by non-contact measurement technology, specifically including intelligent crack detection data based on optical cameras and structural deformation monitoring data based on laser scanning technology.

[0085] Substructure and main structure: The main structure is the overall finite element model of the structure, and the substructure is the finite element model of the local area.

[0086] like Figure 1 As shown, a hybrid substructure bridge dynamic interaction analysis method integrating detection and monitoring data includes the following steps:

[0087] S1. According to the structural stress characteristics, the key areas are selected as substructures and refined finite element modeling is adopted;

[0088] S2, automatically segment cracks and calculate crack width based on deep learning model;

[0089] S3. Based on the above crack width data and corresponding damage mechanism, the weakening unit theory is used to consider the existing damage in the substructure;

[0090] S4, establishing the monitoring data of other parts except the substructure and the mathematical equations of the large-scale finite element model and solving the dynamic boundary conditions of the substructure, wherein the monitoring data of other parts include displacement, strain and inclination;

[0091] S5. Based on the substructure monitoring data, the expectation-oriented adaptive sparse Gaussian process method is used to identify substructure parameters, thereby achieving independent analysis of substructures.

[0092] S6. Propose a dynamic interaction analysis theory between the substructure and the main structure, which specifically includes the following sub-steps:

[0093] S61. Decompose the large-scale structure into multiple fine substructure models and a relatively rough main structure model;

[0094] S62, by introducing nonlinear correction force to achieve the coordination of displacement boundary and force balance between the main structure and substructure;

[0095] S63, the substructure is modeled using solid elements, the main structure is modeled using beam elements, and the two are coupled through multi-point constraints;

[0096] S64, nonlinear analysis of substructure and linear elastic analysis of main structure;

[0097] S65, determining material properties through substructure monitoring data, and adjusting material constitutive parameters to update to corresponding parts of the main structure;

[0098] S66, apply the nonlinear correction force as an external load to the main structure;

[0099] S67. Ensure the displacement coordination between the main structure and the substructure by deriving simplified bridge structure formulas;

[0100] S68. Solve the equilibrium equations between the main structure and the substructure through non-iterative solution.

[0101] S7. Feedback the results of the substructure nonlinear analysis to the main structure to achieve equivalent dynamic linear analysis.

[0102] like Figure 2 As shown, the automatic crack segmentation based on the deep learning model in step S2 has the following characteristics:

[0103] (1) The crack images are automatically segmented using a deep learning model. The model adopts a U-shaped structure consisting of an encoder and a decoder, and introduces a group multi-axis Hadamard product attention module to enhance the segmentation ability.

[0104] (2) The deep learning model introduces a group aggregation bridge module at each stage between the encoder and decoder. It introduces mask information to fuse and process multi-scale features and uses dilated convolution to better extract rich feature information from different resolutions, further enhancing the segmentation ability of the model.

[0105] (3) The loss function consists of binary cross entropy and Dice loss, which is used to optimize model training and ensure the best effect in the segmentation task.

[0106] (4) Combined with image processing methods, quantitative information such as crack length and width can be quickly extracted.

[0107] like Figure 3As shown, step S3, based on the crack width data and the corresponding damage mechanism, the existing damage is considered in the substructure through the weakening unit theory, and the weakening unit method includes:

[0108] (1) Determine the weakened tensile properties based on the bilinear mode I stress-crack width relationship;

[0109] (2) Based on the measured crack width, the tensile strength and the residual opening of the crack under stress-free conditions are derived using a bilinear model;

[0110] (3) construct a stress-strain relationship applicable to the weakened unit by calculating the area under the ultimate strain-maintaining stress-strain curve equal to the ratio of the residual fracture energy to the crack bandwidth;

[0111] (4) The influence of cracks on the overall performance of the structure is simulated by introducing crack weakening units, and the mechanical properties of the material are adjusted to simulate the weakening effect of cracks on the structure.

[0112] like Figure 4 As shown, the specific implementation of step S4 is:

[0113] S41. Establish mathematical models based on overall measurement data and mechanical information of large-scale finite element models. In dynamics, Among them, F represents the external force load, M and C are the structural mass and structural damping respectively, and x represents the structural displacement vector;

[0114] S42. The boundary conditions of the substructure are solved by the adaptive sparsity matching pursuit method, where the boundary is the dynamic boundary.

[0115] like Figure 5 As shown, in step S5, the expectation-guided adaptive sparse Gaussian process substructure parameter identification method includes:

[0116] (1) Based on the Optimal Latin Hypercube Design (OLHD), this method randomly generates a set of sample points p (1) ,p (2) ,…,p (N) Cover the entire design space. The initial sample points are around the current optimal point P optimal,t-1 Update within the range of ±20% to ensure that the sampling points cover the entire design space while focusing on the local area near the optimal point;

[0117] (2) Applying time history displacement u PS,0-t , calculate the parameter set formed by the above sample points and obtain the corresponding response data

[0118] (3) Calculate the error index R based on the current response data, and build a sparse Gaussian process model based on these data;

[0119] (4) Calculate the expected value (EI) of the candidate sampling points and select the point with the largest expected improvement as the next sampling point;

[0120] (5) updating the sparse Gaussian process model based on the updated sampling points;

[0121] (6) Repeat the above steps until the convergence criterion is met;

[0122] (7) When the maximum value of EI approaches zero or reaches the preset upper limit of sampling points, the optimization is stopped.

[0123] like Figure 6 As shown, the specific steps of the expectation-oriented adaptive sparse Gaussian process substructure parameter identification method at time t are described as follows:

[0124] (1) Read the structural response R at the tth time step PS Send to SPI-EASG module;

[0125] (2) Based on P optimal,t-1 Calculation error NRMSE (R TS,0-t ,R PS,0-t ), and determine whether it exceeds the tolerance. If it exceeds the tolerance, calculate And obtain the new optimal parameter P from the SPI-EASG module optimal,t , otherwise, let P optimal,t =P optimal,t-1 ;

[0126] (3) Change NS from u NS,t-1 Load to u NS,t , and use P optimal,t-1 Calculate the response R NS,t ;

[0127] (4) Repeat steps 1 to 4 until the convergence criterion is met;

[0128] (5) When the convergence criterion is reached, SPI-EASG will output the optimal parameter combination P optimal , as the final parameter identification result of the finite element model. These identified parameters can be used for model calibration and improve the accuracy of structural response prediction.

[0129] like Figure 7 As shown, in step S6, the dynamic interaction analysis method between the substructure and the main structure has the following characteristics:

[0130] Steps to implement substructure interaction:

[0131] (1) Extract the displacement of the control nodes C and D in the isolation zone of the main structure at the nth time step and transmit it to the substructure, and use the interface physical quantity as the displacement w' of the control nodes C' and D' of the substructure (m,n) Iteration initial value;

[0132] (2) After receiving the above interface physical quantities, the substructure performs a dynamic analysis and transmits the calculated reaction force R' of the substructure control nodes C' and D' back to the main structure;

[0133] (3) The main structure receives the reaction force R', and the external force is calculated according to formula (8) and loaded to the main structure control nodes C and D to perform a dynamic analysis;

[0134] (4) Extract the displacement w of the main structure control nodes C and D c,(m,n) And it will be transmitted to the substructure. According to the displacement coordination, the movement amount is the displacement w' of the substructure control node C', D' (m+1,n) ;

[0135] (5) After receiving the above interface physical quantities, the substructure performs a dynamic analysis and transmits the calculated reaction force R' of the substructure control nodes C' and D' back to the main structure;

[0136] (6) The main structure receives the reaction force R', and the external force is calculated according to formula (8) and loaded to the main structure control nodes C and D, and then a dynamic analysis is performed;

[0137] (7) Extract the displacement w of the main structure control nodes C and D c,(m+1,n) , and transfer it to the substructure.

[0138] The solution of this embodiment is described below using an actual bridge scenario.

[0139] like Figure 8 As shown in the figure, the test object is a three-span prestressed concrete variable-section single-box single-chamber continuous box girder bridge, the upper structure size is 53 meters + 85 meters + 53 meters, and the box girder height gradually increases from 2.4 meters at 1 meter from the mid-span to 5.0 meters at 1.5 meters from the pier center, presenting a quadratic parabola shape. Fig. 9As shown in the figure, in order to fully monitor the response of the bridge under different loads, a variety of sensors are installed throughout the bridge: a total of 25 long-gauge strain sensors are used to monitor the strain response of the structure under load; a total of 15 inclinometers are arranged to capture the angular changes during loading; 25 acceleration sensors are also set up to record the dynamic response of the structure under dynamic loads. This test is divided into two phases. The first phase is a truck dynamic load test, which simulates the load effect under actual traffic conditions. A truck is driven at a speed of 40km / h and finally stops on lane 3 in the middle of the bridge span; the second phase involves impact vibration testing, using an exciter to apply excitation at specified locations (excitation points 18, 19 and 20) to simulate the impact vibration conditions that the bridge may encounter.

[0140] like Fig.10 As shown in the figure, based on the deep learning model, the crack images collected in the experiment are input to achieve automatic segmentation of cracks. Fig.10 The original image of the crack in the frame structure and the result of its automatic segmentation by the deep learning model are shown. The deep learning model used can simultaneously identify the high-level global semantic features and low-level detail information of the crack, thereby improving the accuracy and efficiency of crack segmentation. The image processing method is further combined to perform pixel-level threshold segmentation of the crack, determine the total number of crack pixels, and extract the crack skeleton by combining area calculation and the medial axis method, thereby realizing the automatic measurement of the average width of the crack.

[0141] According to the existing crack width and the corresponding element properties of the weakened constitutive relation, the cracks are taken into account in the refined substructure model through the weakened element method. The overall size of the substructure is set to 0.5m, while the mesh in the critical area is refined to 0.1m, and the geometric nonlinear analysis option is enabled. Fig.11 As shown in step S4, the boundary conditions of the sub-solution substructure are solved based on the mathematical and physical model. Finally, the SPI-EASG algorithm is used to complete the identification of the key parameters of the substructure, including the update of the elastic modulus of concrete and the adjustment of the elastic modulus and yield strength of the steel bars. The results of parameter identification are shown in Fig.12 .

[0142] like Fig.13 As shown in the figure, the strain and stress results of the substructure under impact vibration and vehicle dynamic load are displayed. Through the proposed substructure interaction method, the key areas can be analyzed independently, which is convenient for considering the actual monitoring data, and then the local stress state can be evaluated more accurately and the results can be fed back to the main structure, finally realizing the information transmission between "overall large-scale model-local fine model".

[0143] like Fig.14As shown in the figure, the strain response results of the structure under the impact vibration load are displayed. By selecting the corresponding strain sensor signals below the positions of acceleration sensors A18 and A19 for comparison, the effectiveness of the method proposed in this paper is verified. Under the action of impact vibration, compared with the simple model updating method, the substructure dynamic interaction method proposed in this paper is closer to the measured strain response and can effectively evaluate the stress state of the structure.

[0144] like Fig.15 As shown in the figure, the structural strain and displacement response results under the vehicle dynamic load are displayed. By selecting the measurement signals of the acceleration sensors A2, A18, and A19 for comparison, the structural response calculated by the method proposed in this paper is closer to the measured response, which can more accurately reflect the dynamic response of the structure and is expected to improve the accuracy of structural safety assessment.

[0145] Through the above detailed experimental design and result analysis, the effectiveness of the hybrid substructure bridge dynamic interaction analysis method proposed in this invention that integrates detection and monitoring data has been verified. This method not only solves the problem of difficulty in integrating detection and monitoring data with finite element models, but also greatly improves the accuracy of structural safety assessment. In addition, it also demonstrates the potential for wide application in the field of safety assessment of large and complex structures.

[0146] In summary, the successful demonstration of the specific embodiment shows that the scheme proposed by the present invention is feasible and effective. Experiments have proved that the finite element model established by integrating inspection and monitoring data can significantly improve the analysis efficiency and accuracy of the overall structure, which is of great significance for simulating the stress conditions of existing structures, especially in improving the accuracy of large-scale structural performance analysis under complex stress conditions.

[0147] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or substitutions within the technical scope disclosed by the present invention, and these modifications or substitutions should all be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A hybrid substructure bridge dynamic interaction analysis method integrating detection and monitoring data, characterized in that: The steps include: S1. According to the structural stress characteristics, the key areas are selected as substructures and refined finite element modeling is adopted; S2, automatically segment cracks and calculate crack width based on deep learning model; S3. Based on the above crack width data and corresponding damage mechanism, the weakened element method is used to consider the existing damage in the substructure; S4, establish the monitoring data of other parts except the substructure and the mathematical equations of the large-scale finite element model and solve the dynamic boundary conditions of the substructure; S5. Based on the substructure monitoring data, the expectation-oriented adaptive sparse Gaussian process method is used to identify the substructure parameters, thereby achieving independent calculation of the substructure; S6. A dynamic interaction analysis method between the substructure and the main structure is proposed to convert the dynamic response of the substructure interface into the equivalent external dynamic load of the main structure, thereby realizing the dynamic coupling between the two and enabling the substructure information to be transmitted to the main structure. The specific steps include the following: S61, taking out a certain part of the large-scale structure to establish a substructure model, and at the same time establishing a main structure model; S62, by introducing nonlinear correction force to achieve the coordination of displacement boundary and force balance between the main structure and substructure; S63, the substructure is modeled using solid elements, the main structure is modeled using beam elements, and the two are coupled through multi-point constraints; S64, nonlinear analysis of substructure and linear elastic analysis of main structure; S65, determining material properties through substructure monitoring data, and adjusting material constitutive parameters to update to corresponding parts of the main structure; S66, apply the nonlinear correction force as an external load to the main structure; S67. Ensure the displacement coordination between the main structure and the substructure by deriving simplified bridge structure formulas; S68, solving the equilibrium equations between the main structure and the substructure by a non-iterative solution method; S7. Feedback the results of the substructure nonlinear analysis to the main structure to achieve equivalent dynamic linear analysis.

2. The hybrid substructure bridge dynamic interaction analysis method of fusion detection and monitoring data according to claim 1 is characterized in that: Step S2 specifically includes the following features: The crack images are automatically segmented using a deep learning model. The model adopts a U-shaped structure consisting of an encoder and a decoder, and introduces a group multi-axis Hadamard product attention module to enhance the segmentation ability. This deep learning model introduces a group aggregation bridge module at each stage between the encoder and decoder, fuses and processes multi-scale features by introducing mask information, and further enhances the segmentation ability of the model by using dilated convolution to better extract rich feature information from different resolutions. The loss function consists of binary cross entropy and Dice loss, which is used to optimize model training and ensure the best effect in the segmentation task.

3. The hybrid substructure bridge dynamic interaction analysis method of fusion detection and monitoring data according to claim 1 is characterized in that: In step S3, the weakened unit method includes the following sub-steps: S31. Determine the weakened tensile properties based on the bilinear mode I stress-crack width relationship, and construct a stress-strain relationship suitable for the weakened unit by calculating the area under the ultimate strain-maintaining stress-strain curve. S32, by not imparting tensile properties to finite elements with pre-existing crack overlap based on an assumed bilinear mode I stress-crack width relationship; S33. Based on the measured width of the crack, the tensile strength and the residual properties of the crack under stress-free conditions are derived using a bilinear model; S34. In the finite element model of concrete, there is a cracked area. The residual properties derived above are introduced to consider the impact of the cracks on the overall performance of the structure.

4. The hybrid substructure bridge dynamic interaction analysis method of fusion detection and monitoring data according to claim 1 is characterized in that: In step S4, the monitoring data of other parts include: displacement, strain and inclination.

5. The hybrid substructure bridge dynamic interaction analysis method of fusion detection and monitoring data according to claim 1 is characterized in that: Specific implementation of step S4: S41. Establish mathematical models based on overall measurement data and mechanical information of large-scale finite element models. In dynamics, Among them, F represents the external force load, M and C are the structural mass and structural damping respectively, and x represents the structural displacement vector; S42. The boundary conditions of the substructure are solved by the adaptive sparsity matching pursuit method, where the boundary is the dynamic boundary.

6. The hybrid substructure bridge dynamic interaction analysis method of fusion detection and monitoring data according to claim 1 is characterized in that: In step S5, the expectation-guided adaptive sparse Gaussian process substructure parameter identification method includes: S51, based on the optimal Latin hypercube design, by randomly generating a set of sample points p (1) ,p (2) ,…,p (N) Cover the entire design space; the initial sample points are around the current optimal point P optimal,t-1 Update within the range of ±20% to ensure that the sampling points cover the entire design space while focusing on the local area near the optimal point; S52, applying time history displacement u PS,0-t , calculate the parameter set formed by the above sample points and obtain the corresponding response data S53, calculating the error index R based on the current response data, and constructing a sparse Gaussian process model based on these data; S54, calculating the expected value EI of the candidate sampling points, and selecting the point with the maximum expected improvement as the next sampling point; S55, updating the sparse Gaussian process model based on the updated sampling points; S56, repeat the above steps until the convergence criterion is met; S57 , when the maximum value of the expected value EI of the candidate sampling points approaches zero or reaches a preset upper limit of the sampling points, the optimization is stopped.

7. The hybrid substructure bridge dynamic interaction analysis method of fusion detection and monitoring data according to claim 5 is characterized in that: The specific steps of the adaptive sparse Gaussian process substructure parameter identification method and the dynamic calculation time history t are: (1) Solve the structural dynamics at time t and convert the corresponding displacement u NS,t Send to NS; (2) Read the structural response R at the tth time step PS Send to SPI-EASG module; (3) Based on P optimal,t-1 Calculation error NRMSE (R TS,0-t ,R PS,0-t ), and determine whether it exceeds the tolerance. If it exceeds the tolerance, calculate And obtain the new optimal parameter P from the SPI-EASG module optimal,t , otherwise, let P optimal,t =P optimal,t-1 ; (4) Change NS from u NS,t-1 Load to u NS,t , and use P optimal,t-1 Calculate the response R NS,t ; (5) Repeat steps 1 to 4 until the convergence criterion is met; (6) When the convergence criterion is reached, SPI-EASG will output the optimal parameter combination P optimal , as the final parameter identification result of the finite element model.

8. The hybrid substructure bridge dynamic interaction analysis method of fusion detection and monitoring data according to claim 1 is characterized in that: In step S6, the implementation of the interaction between the substructure and the main structure specifically includes the following steps: SA, extract the motion displacement of the control nodes C and D in the nth time step of the main structure isolation area and transmit it to the substructure, the motion displacement of the control nodes C and D in the nth time step of the main structure isolation area is used as the first interface physical quantity, and the first interface physical quantity is used as the displacement w' of the substructure control nodes C' and D' (m,n) Iteration initial value; After receiving the above-mentioned first interface physical quantity, the SB,substructure performs a dynamic analysis and transmits the calculated reaction force R' of the substructure control nodes C' and D' back to the main structure; SC, the main structure receives the reaction force R', calculates the external force and loads it to the main structure control nodes C and D, and performs a dynamic analysis; SD, extract the displacement w of the main structure control nodes C and D c,(m,n) And transmit it to the substructure. According to the displacement coordination, the movement is the displacement w' of the substructure control node C', D' (m+1,n) , the substructure controls the displacement w' of nodes C', D' (m+1,n) As the second interface physical quantity; After receiving the above-mentioned second interface physical quantity, the SE,substructure performs a dynamic analysis and transmits the calculated reaction force R of the substructure control nodes C', D' back to the main structure; SF and the main structure receive the reaction force R, calculate the external force and load it to the main structure control nodes C and D, and then perform a dynamic analysis; SG, extract the displacement w of the main structure control nodes C and D c,(m+1,n) , and transfer it to the substructure.

Images