Bridge non-uniform deformation prediction method and system

Abstract

The application provides a bridge non-uniform deformation prediction method and system. The method comprises the following steps: S1, acquiring visual measurement point cloud data, load data and environmental parameter data of a target heavy-load railway bridge under different working conditions; S2, constructing a non-uniform deformation feature extraction method based on a robust parameterized model, and extracting global non-uniform deformation features of the bridge through point cloud preprocessing, curved surface parameterized modeling and multi-period point cloud registration; S3, constructing a global displacement correction method for the heavy-load railway bridge based on an RBF neural network, and realizing global displacement correction through influence factor analysis, a multi-scale deformation residual error objective function and a nonlinear mapping model; and S4, constructing a deformation prediction uncertainty parameter optimization method for the heavy-load railway bridge, optimizing the uncertainty parameters based on a non-Gaussian distribution assumption, an improved maximum likelihood estimation and an EM algorithm, and outputting a non-uniform deformation prediction result. This method can improve the accuracy, stability and reliability of deformation prediction of the heavy-load railway bridge.

Description

Technical Field

[0001] This application relates to the fields of bridge health monitoring and intelligent bridge operation and maintenance technology. More specifically, this invention relates to a method and system for predicting non-uniform deformation of heavy-load railway bridges based on visual measurement point cloud and global displacement correction. Background Technology

[0002] Heavy-haul railway bridges are prone to complex deformations such as longitudinal non-uniform bending, transverse non-uniform bending, and torsion under the long-term influence of multiple factors, including heavy train loads, temperature changes, stiffness loss, changes in boundary conditions, and pier settlement. Existing bridge deformation monitoring methods mostly rely on local or single-point measurement methods such as strain gauges, displacement gauges, total stations, leveling, and GPS. These methods can reflect the response changes at specified measurement points, but they cannot fully describe the global non-uniform deformation in unmonitored areas. Furthermore, the factors influencing the deformation of heavy-haul railway bridges are complex, and there is a strong nonlinear relationship between structural parameters and deformation response. Traditional model correction methods based on local measurement points suffer from problems such as insufficient data completeness, low computational efficiency, and low prediction reliability.

[0003] Visual measurement point cloud data has advantages such as being non-contact, having a wide coverage area, and being rich in spatial information, which can provide a more complete data foundation for the extraction of global deformation of bridges. However, field point cloud data is susceptible to noise, outliers, and registration errors, and direct point cloud comparison can easily reduce the accuracy of deformation extraction. Summary of the Invention

[0004] The objective of this invention is achieved through the following technical solutions.

[0005] This application proposes a method and system for predicting non-uniform deformation of heavy-haul railway bridges based on visual measurement point cloud and global displacement correction. The method extracts the global non-uniform deformation features of the bridge through a robust parametric model, establishes a nonlinear mapping relationship between structural parameters and global displacement response based on RBF neural network, and combines deformation prediction uncertainty parameter optimization to improve the accuracy, efficiency and reliability of non-uniform deformation prediction of heavy-haul railway bridges.

[0006] The purpose of this application is to provide a method and system for predicting non-uniform deformation of heavy-haul railway bridges based on visual measurement point clouds and global displacement correction. This method addresses the problems in existing heavy-haul railway bridge deformation prediction, such as the difficulty of local measurement points to fully characterize global non-uniform deformation, the impact of point cloud noise on the accuracy of deformation feature extraction, the difficulty of model correction to accurately reflect nonlinear relationships, and the high uncertainty in prediction.

[0007] To achieve the above objectives, this application proposes a method for predicting non-uniform deformation of heavy-haul railway bridges based on visual measurement point clouds and global displacement correction, comprising the following steps:

[0008] S1, acquire visual measurement point cloud data, load data and environmental parameter data of the target heavy-load railway bridge under different working conditions;

[0009] S2, Based on the robust parametric model, extract the non-uniform deformation features of the target heavy-load railway bridge to obtain global displacement information for model correction;

[0010] S3. Based on the RBF neural network, a global displacement correction model for the target heavy-load railway bridge is constructed, and a nonlinear mapping relationship between structural parameters and global displacement response is established.

[0011] S4. Based on the deformation prediction uncertainty parameter optimization method, the global displacement correction model is optimized, and the non-uniform deformation prediction result of the target heavy-load railway bridge is output.

[0012] Further, in step S2, the visual measurement point cloud data is preprocessed to obtain the point cloud of the main structure of the bridge; a surface parameterized model is established based on the characteristics of the point cloud of the main structure of the bridge; a robust noise suppression term is constructed by combining the noise characteristics of the point cloud; a deformation registration model is established based on multi-period point cloud data, and the full-field displacement matrix is ​​obtained through the gridded mapping mechanism of the control points of the surface parameterized model, thereby extracting the non-uniform deformation characteristics of the target heavy-load railway bridge.

[0013] Further, in step S3, a finite element model of the target heavy-load railway bridge is established and the structural parameters to be corrected are determined; the correlation between the vertical settlement at the pier bottom, the inclination angle at the pier bottom, the overall temperature rise and fall, the solar radiation temperature, the stiffness reduction factor, and the non-uniform deformation of the bridge is analyzed; structural parameter samples are generated based on the experimental design method, and the corresponding displacement response values ​​are obtained through finite element simulation calculation; a global displacement correction objective function is constructed by combining the multi-scale characteristics of bridge deformation; and an RBF neural network is used to approximate the nonlinear mapping relationship between the deformation response characteristics and the design parameters to form a global displacement correction model.

[0014] Further, in step S4, the sources of uncertainty in the deformation prediction of the target heavy-load railway bridge are analyzed; a log-likelihood model between the measured deformation response value and the structural parameter distribution parameter is established based on the sensitive deformation response characteristics and their statistical distribution law; random samples are generated according to the probability assumption of the structural parameter t distribution, and the corresponding response value is calculated using the modified model; the response probability density function is established through the kernel density estimation method, and the maximum likelihood solution of the parameters is solved by combining the EM optimization solution algorithm to realize uncertainty quantification and parameter optimization; based on the optimized model parameters, the non-uniform deformation prediction result of the target heavy-load railway bridge is output.

[0015] To achieve the above objectives, this application also proposes a system for predicting non-uniform deformation of heavy-haul railway bridges based on visual measurement point clouds and global displacement correction, comprising:

[0016] The data acquisition module is used to acquire visual measurement point cloud data, load data, and environmental parameter data of the target heavy-load railway bridge under different working conditions;

[0017] The non-uniform deformation feature extraction module is used to extract the global non-uniform deformation features of the target heavy-load railway bridge from visual measurement point cloud data based on a robust parametric model.

[0018] The global displacement correction module is used to establish a nonlinear mapping relationship between structural parameters and global displacement response based on the RBF neural network, and to construct a global displacement correction model for the target heavy-load railway bridge.

[0019] The uncertainty parameter optimization module is used to identify and optimize the uncertainty parameters of the deformation prediction model for heavy-haul railway bridges.

[0020] The deformation prediction output module is used to output the non-uniform deformation prediction results of the target heavy-load railway bridge.

[0021] In summary, the technical effects of this application are as follows: obtaining global deformation information of bridges through visual measurement point clouds, thus improving the problem of incomplete local measurement point data; improving the stability of non-uniform deformation feature extraction under complex noise conditions through robust parameterization models; enhancing the ability of global displacement correction models to express nonlinear relationships through RBF neural networks; and improving the reliability of non-uniform deformation prediction results for heavy-haul railway bridges through uncertainty parameter optimization. Attached Figure Description

[0022] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0023] Figure 1 This is a technical roadmap for the research on high-precision deformation prediction of heavy-haul railway bridges in this application.

[0024] Figure 2 This is a schematic diagram of a research scheme for non-uniform deformation feature extraction based on a robust parametric model;

[0025] Figure 3 This is a schematic diagram of a research scheme for global displacement correction of heavy-haul railway bridges based on RBF neural networks. Detailed Implementation

[0026] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0027] This application provides a method for predicting non-uniform deformation of heavy-load railway bridges based on visual measurement point clouds and global displacement correction, specifically including the following steps:

[0028] S1, acquire visual measurement point cloud data, load data, and environmental parameter data of the target heavy-load railway bridge under different working conditions. The visual measurement point cloud data may be derived from actual bridge visual measurement tests, on-site monitoring conditions, or preset verification conditions of the target heavy-load railway bridge.

[0029] Furthermore, full-field point cloud data of the target heavy-load railway bridge is acquired using a 3D laser scanner, photogrammetry equipment, or other visual measurement equipment. The point cloud data may include bridge point cloud data before, during, or after the train load is applied. The load data includes train load, load location, load application time, and operating condition information. The environmental parameter data includes temperature, sunlight conditions, humidity, and wind speed.

[0030] S2. Based on the robust parametric model, the non-uniform deformation features of the target heavy-load railway bridge are extracted to obtain global displacement information for model correction.

[0031] Furthermore, the visual measurement point cloud data is preprocessed. Based on the analysis of the bridge's geometric features, multi-view methods, voxel methods, or filtering methods are used to remove the surrounding scene data of the bridge, obtaining the point cloud data of the bridge's main structure.

[0032] Furthermore, the point spacing, bridge dimensions, component geometric features, and finite element model mesh generation characteristics of the bridge point cloud are analyzed to determine the number of control points and order parameters of the surface parameterization model, thus establishing a surface parameterization model of the bridge's main structure. Specifically, the surface parameterization model can adopt a B-spline surface model; by setting the number of control points, surface order parameters, and node vectors, the discrete and irregular bridge point cloud data is converted into a continuous surface expression with a unified parameter domain, providing a foundation for subsequent deformation registration, full-field displacement matrix acquisition, and non-uniform deformation feature extraction.

[0033] Furthermore, the multi-component characteristics of point cloud noise and their relationship with the model fitting residuals are analyzed. Noise suppression weight vectors are constructed for strong noise and random errors, and a robust noise suppression term is established based on the Huber function. This robust noise suppression term is used to reduce the impact of outliers, occlusion points, and random errors on the model parameter solution during the surface parameterization model fitting process, thereby improving the stability of the surface representation of the bridge's main structure.

[0034] Furthermore, combining the connection characteristics of bridge components and multi-phase point cloud data, a deformation registration model of the overall bridge structure is established. Specifically, in the multi-phase point cloud deformation registration process, the Iterative Closest Point Algorithm (ICP) algorithm can be used. Using the point cloud of the main bridge structure before loading or its B-spline surface parameterized model as a reference, spatial registration is performed on the point clouds during or after loading. Through nearest point search, spatial transformation parameter solving, and iterative minimization of registration error, the correspondence of multi-phase point clouds in a unified coordinate system is achieved. To reduce the impact of on-site noise and local anomalies on the registration results, the connection relationships of bridge components, boundary constraints, and Huber robust noise suppression terms are incorporated into the ICP registration process to reduce the weight of anomalies and strong noise points.

[0035] Furthermore, after completing the parametric modeling of the B-spline surface and the multi-stage point cloud deformation registration, the spatial differences between the surface control points or corresponding points in the parameter domain under different working conditions are converted into a full-field displacement matrix through the mesh mapping mechanism of the surface parametric model control points. The full-field displacement matrix is ​​used to characterize the overall displacement trend, local curvature changes, key section displacements, boundary region displacements, and torsional deformation characteristics of the bridge's main structure under different loads and environmental conditions, and extracts longitudinal non-uniform bending characteristics, transverse non-uniform bending characteristics, torsional characteristics, key section displacement characteristics, and local curvature change characteristics from it.

[0036] Furthermore, to verify the rationality of the full-field displacement matrix and non-uniform deformation characteristics, they can be compared and analyzed with traditional measurement point displacement results, finite element calculation results, or deformation trends under preset verification conditions to determine whether the extracted deformation characteristics can reflect the actual global deformation state of the target heavy-load railway bridge.

[0037] S3. Based on the RBF neural network, a global displacement correction model for the target heavy-load railway bridge is constructed, and a nonlinear mapping relationship between structural parameters and global displacement response is established.

[0038] Furthermore, a finite element model of the target heavy-haul railway bridge was established, and the structural parameters to be corrected were determined. These parameters include material parameters, dimensional parameters, boundary conditions, stiffness reduction parameters, vertical settlement parameters at the pier base, pier base inclination angle parameters, overall temperature rise and fall parameters, and solar radiation temperature parameters.

[0039] Furthermore, for the structural parameters, load parameters, and environmental parameters to be corrected, variance analysis, sensitivity analysis, or correlation analysis methods are used to analyze the influence of each parameter on the non-uniform deformation response of the bridge and determine the main influencing factors. Based on the main influencing factors, structural parameter samples are generated using experimental design methods, and the corresponding full-field displacement response and multi-scale deformation response characteristics are obtained through finite element forward calculation.

[0040] Furthermore, by combining the multi-scale characteristics of bridge deformation, a global displacement correction objective function is constructed to reflect the difference between the calculated displacement response and the measured displacement response. The multi-scale characteristics include overall displacement trend, local curvature change, key section displacement, boundary region displacement, sub-region deformation response, and torsional deformation characteristics; wherein, the sub-region deformation response refers to the local deformation characteristics extracted after dividing the main structure of the bridge according to the mid-span region, the region near the support, the boundary region, or the region of key components.

[0041] Furthermore, based on the finite element simulation results, the actual deformation features extracted from the visual measurement point cloud, and the full-field displacement matrix, training samples for the RBF neural network are constructed. The RBF neural network takes at least one of the structural parameters, load parameters, and environmental parameters as input, and takes the full-field displacement response, non-uniform bending characteristics, torsional characteristics, or multi-scale deformation residuals as output, approximating the nonlinear mapping relationship between structural parameters and the non-uniform deformation response of the bridge.

[0042] Furthermore, the actual deformation data extracted from the visual measurement point cloud is input into the aforementioned nonlinear mapping relationship to correct the structural parameters of the target heavy-load railway bridge, forming a global displacement correction model. During model training and objective function solving, genetic algorithms (GA), particle swarm optimization (PSO), or a combination of both can be used to optimize the RBF neural network parameters, structural parameter corrections, or the global displacement correction objective function. The optimization strategy that meets the requirements of computational efficiency and correction accuracy is determined through algorithm performance comparison.

[0043] Furthermore, the optimized structural parameters are input into the corrected finite element model to calculate the corrected full-field deformation response, which is then compared with the measured full-field displacement matrix extracted from the visual measurement point cloud. The global displacement correction results are verified through deformation correlation analysis, correction error analysis, correction value range verification, comparative analysis with traditional methods, or verification under preset working conditions. When the error between the calculated response and the measured response of the corrected model meets the preset requirements, the RBF neural network global displacement correction model is obtained; otherwise, the structural parameter samples, RBF neural network parameters, or optimization algorithm parameters are readjusted, and the model is corrected and the results are verified again.

[0044] S4. Based on the deformation prediction uncertainty parameter optimization method, the global displacement correction model is optimized, and the non-uniform deformation prediction result of the target heavy-load railway bridge is output.

[0045] Furthermore, the sources of uncertainty in the deformation prediction of the target heavy-load railway bridge are analyzed. These sources of uncertainty include visual measurement errors, point cloud registration errors, load parameter errors, environmental parameter disturbances, finite element model errors, and structural parameter discreteness.

[0046] Furthermore, based on the characteristics of the sensitive deformation response and its statistical distribution law, a log-likelihood model is established between the measured deformation response value and the structural parameter distribution parameters:

[0047] (1)

[0048] In the formula, This represents the measured response value. This represents the structural parameters and distribution parameters, where N represents the sample size. This represents the probability density function of the measured response; through the analysis of... Solve, so that It has reached its maximum value.

[0049] Furthermore, based on the probability assumption of the t-distribution of structural parameters, random samples satisfying the t-distribution are generated using the Monte Carlo method, and the corresponding response values ​​are calculated using the modified model. .

[0050] The response probability density function is established using the kernel density estimation method:

[0051] (2)

[0052] In the formula The kernel function can be obtained by setting the bandwidth H, and M represents the number of samples of the corrected model response. This represents the j-th corrected model response value.

[0053] Furthermore, by substituting formula (2) into formula (1), the maximum likelihood model of the structural parameters is simplified, and the maximum likelihood solution of the parameters is solved by combining the EM optimization solution algorithm, so as to realize uncertainty quantification and parameter optimization.

[0054] Furthermore, based on the optimized model parameters, the predicted results of non-uniform deformation of the target heavy-load railway bridge are output. The predicted results of non-uniform deformation include longitudinal non-uniform bending deformation, transverse non-uniform bending deformation, and torsional deformation.

[0055] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting non-uniform deformation of bridges, based on visual measurement point clouds and global displacement correction, characterized in that, Includes the following steps: S1, acquire visual measurement point cloud data, load data and environmental parameter data of the target heavy-load railway bridge under different working conditions; S2, Based on the robust parametric model, extract the non-uniform deformation features of the target heavy-load railway bridge to obtain global displacement information for model correction; S3. Based on the RBF neural network, a global displacement correction model for the target heavy-load railway bridge is constructed, and a nonlinear mapping relationship between structural parameters and global displacement response is established. S4, the global displacement correction model is optimized based on the deformation prediction uncertainty parameter optimization method, and the non-uniform deformation prediction result of the target heavy-load railway bridge is output. In step S4, the optimization of the deformation prediction uncertainty parameters includes the following steps: S41, Analyze the sources of uncertainty in the deformation prediction of the target heavy-load railway bridge. The sources of uncertainty include visual measurement error, point cloud registration error, load parameter error, environmental parameter disturbance, finite element model error, and structural parameter discreteness. S42. Based on the characteristics and statistical distribution of the sensitive deformation response, a log-likelihood model is established between the measured deformation response values ​​and the structural parameter distribution parameters: （1） In the formula, This represents the measured response value. This represents the structural parameters and distribution parameters, where N represents the sample size. This represents the probability density function of the measured response; through the analysis of... Solve, so that Reaching the maximum value; S43, based on the probability assumption of the t-distribution of structural parameters, random samples satisfying the t-distribution are generated using the Monte Carlo method, and the corresponding response values ​​are calculated using the modified model. ; S44, The response probability density function is established using the kernel density estimation method: （2） In the formula This represents the kernel function, obtained by setting the bandwidth H, where M represents the number of samples for the corrected model response. This represents the j-th corrected model response value; S45, Substitute formula (2) into formula (1) to simplify the maximum likelihood model of structural parameters, and combine it with the EM optimization solution algorithm to solve the maximum likelihood solution of parameters, thereby realizing uncertainty quantification and parameter optimization; S46. Based on the optimized model parameters, output the prediction results of non-uniform deformation of the target heavy-load railway bridge. The prediction results of non-uniform deformation include longitudinal non-uniform bending deformation, transverse non-uniform bending deformation and torsional deformation.

2. The method according to claim 1, characterized in that, In step S2, the extraction of the non-uniform deformation features includes the following steps: S21, preprocess the visual measurement point cloud data, filter out the surrounding scene data of the bridge, and obtain the point cloud of the main structure of the bridge. S22. Based on the characteristics of the point cloud of the main bridge structure, determine the number of control points and order parameters of the surface parameterization model, and establish the surface parameterization model of the main bridge structure. S23, analyze the multi-component characteristics of point cloud noise and its relationship with the model fitting residuals, construct a robust noise suppression term based on the Huber function to reduce the impact of strong noise and random errors on the parameterized model; S24. Based on the connection characteristics of bridge components and multi-phase point cloud data, a registration model of the overall structural deformation of the bridge is established. S25, the full-field displacement matrix is ​​obtained through the gridding mapping mechanism of the control points of the surface parameterized model, and the non-uniform deformation features of the target heavy-load railway bridge are extracted from the full-field displacement matrix.

3. The method according to claim 1, characterized in that, In step S3, the construction of the global displacement correction model includes the following steps: S31. Establish the finite element model of the target heavy-load railway bridge and determine the structural parameters to be corrected. S32. Based on vertical settlement, pier bottom inclination angle, overall temperature rise and fall, solar radiation temperature and stiffness reduction factors, analyze the correlation between each factor and the non-uniform deformation of the bridge, and determine the main influencing factors. S33, based on the experimental design method, generates structural parameter samples and obtains the corresponding displacement response values ​​through finite element simulation calculation; S34. Combining the multi-scale characteristics of bridge deformation, a global displacement correction objective function is constructed to reflect the difference between the calculated displacement response and the measured displacement response. S35, based on non-uniform deformation response characteristics and structural parameter sample data, uses RBF neural network to approximate the nonlinear mapping relationship between deformation response characteristics and design parameters; S36, The actual deformation data extracted from the visual measurement point cloud is input into the nonlinear mapping relationship to correct the structural parameters of the target heavy-load railway bridge and form a global displacement correction model.

4. The method according to claim 1, characterized in that, In step S1, the visual measurement point cloud data is multi-period point cloud data, including point cloud data acquired before, during and after the train load is applied; the load data includes the train load, load location and load application time; the environmental parameter data includes temperature, sunshine conditions, humidity and wind speed.

5. The method according to claim 2, characterized in that, In step S23, the construction of a robust noise suppression term based on the Huber function includes: constructing a noise suppression weight vector for strong noise and random errors in the point cloud, and introducing the noise suppression weight vector into the fitting residual minimization process of the surface parameterized model to reduce the impact of point cloud noise on the parameterized model.

6. The method according to claim 2, characterized in that, In step S25, the non-uniform deformation features extracted from the full-field displacement matrix include longitudinal non-uniform bending features, transverse non-uniform bending features, torsional features, key section displacement features, and local curvature change features.

7. The method according to claim 3, characterized in that, In step S34, the multi-scale features include overall displacement trend features, local curvature change features, key section displacement features, boundary region displacement features, and torsional deformation features.

8. A bridge non-uniform deformation prediction system, using the method described in any one of claims 1-7, characterized in that, Based on visual measurement of point clouds and global displacement correction, including: The data acquisition module is used to acquire visual measurement point cloud data, load data, and environmental parameter data of the target heavy-load railway bridge under different working conditions; The non-uniform deformation feature extraction module is used to extract the global non-uniform deformation features of the target heavy-load railway bridge from visual measurement point cloud data based on a robust parametric model. The global displacement correction module is used to establish a nonlinear mapping relationship between structural parameters and global displacement response based on the RBF neural network, and to construct a global displacement correction model for the target heavy-load railway bridge. The uncertainty parameter optimization module is used to identify and optimize the uncertainty parameters of the deformation prediction model for heavy-haul railway bridges. The deformation prediction output module is used to output the non-uniform deformation prediction results of the target heavy-load railway bridge.

9. The system according to claim 8, characterized in that, The non-uniform deformation feature extraction module specifically includes: The preprocessing unit is used to filter out the scene data around the bridge and obtain the point cloud of the main structure of the bridge. The parametric modeling unit is used to establish a surface parametric model based on the point cloud characteristics of the main bridge structure, and to construct a robust noise suppression term based on the Huber function. The registration unit is used to establish a registration model of the overall structural deformation of the bridge by combining the connection characteristics of bridge components and multi-phase point cloud data. The feature extraction unit is used to obtain the full-field displacement matrix through the gridded mapping of control points of the surface parameterized model, and to extract non-uniform deformation features from the full-field displacement matrix.

Images