A segment deformation prediction method based on comparison of scanning data and simulation data
By comparing scanned data with simulation data, setting control points, and utilizing a neural network model, the problem of deviation between bridge segment simulation data and actual deformation was solved, enabling more accurate deformation prediction and construction guidance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SICHUAN STEEL STRUCTURE INTELLIGENT MFG CO LTD
- Filing Date
- 2026-04-10
- Publication Date
- 2026-07-21
AI Technical Summary
In the existing bridge segment assembly process, there is a significant discrepancy between simulation data and actual deformation, which increases the difficulty of construction.
By comparing scanned and simulated data, control points are set to obtain measured and simulated deformations. A neural network model is then used for training to generate a deformation prediction model for predicting the deformation of bridge segments.
It improves the accuracy of simulation results for bridges in various postures, guides the process, reduces component deformation, and lowers construction difficulty.
Smart Images
Figure CN122021070B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of bridge engineering technology, and in particular to a method for predicting segment deformation based on the comparison of scanned data and simulation data. Background Technology
[0002] Large and complex bridges require extremely high precision in matching and aligning the dimensions of multiple segments during actual assembly and hoisting. However, due to their large cross-sectional dimensions, under different stress states and after heat treatment, local and overall deformations occur. Some of these deformations are recoverable, while others are plastic deformations. Simulation data can guide the process, helping to reduce deformation or utilizing relevant deformation amounts to achieve smooth assembly and docking, significantly reducing the difficulty of docking construction. However, existing simulation input designs are often too idealistic, deviating significantly from actual deformation.
[0003] Therefore, a prediction method that can relatively accurately reflect the deformation of segmental components under various states is needed. Summary of the Invention
[0004] The purpose of this invention is to overcome the problems of large deformation prediction deviation in the existing bridge segment assembly process, and to provide a segment deformation prediction method based on the comparison of scanning data and simulation data.
[0005] This invention provides a method for predicting segmental deformation based on the comparison of scanned data and simulation data, comprising:
[0006] S1. Set control points based on the surface model of the bridge segment, perform deformation analysis on the surface model according to the set process steps, and obtain the simulated deformation amount of the control points in the corresponding process steps.
[0007] S2. Based on the point cloud data of the bridge segments in the corresponding process steps, obtain the measured deformation of the control point and the three-dimensional coordinate data of the control point;
[0008] S3. Repeat S1 to S2 to obtain several simulated deformations, measured deformations and three-dimensional coordinate data corresponding to the control points, and generate a dataset.
[0009] S4. Input the dataset into the constructed neural network model for training to obtain a trained deformation prediction model;
[0010] S5. Based on the deformation prediction model, predict the deformation of the bridge segment to be predicted.
[0011] According to a specific implementation, in the above deformation prediction method, the process of setting the process steps includes:
[0012] Based on the key process control points in the manufacturing and construction process, the process steps that need to be controlled are set.
[0013] According to a specific implementation, in the above deformation prediction method, step S2 specifically includes:
[0014] A coarse localization transformation is performed on the point cloud data and the patch model to obtain a rotation transformation matrix between the point cloud data and the patch model. The point cloud data is then transformed according to the rotation transformation matrix to obtain coarsely localized point cloud data.
[0015] Calculate the point closest to each control point in the coarsely located point cloud data, obtain the vector from that point to the control point, and obtain the measured deformation.
[0016] According to one specific implementation, in the above deformation prediction method, the coarse positioning transformation adopts the ICP method.
[0017] According to a specific implementation, in the above deformation prediction method, the point cloud data is obtained by scanning the bridge segments in the corresponding process steps using three-dimensional scanning technology, and the three-dimensional scanning technology includes at least one of laser scanning, oblique photography and total station scanning.
[0018] According to a specific implementation, in the above deformation prediction method, the control points include key control points and auxiliary control points; the key control points include the corner points of bridge segments, the upper and lower edge points at intervals, and the centroid of the plates; the auxiliary control points include points generated according to the uv values of the patches in the patch model or according to a preset mesh interval.
[0019] According to a specific implementation, in the above deformation prediction method, the deformation prediction model adopts a hierarchical prediction structure, including two multilayer perceptrons; wherein the multilayer perceptron F θ1 Used to extract implicit features from the dataset to obtain the actual deformation of the auxiliary control points; Multilayer Perceptron F θ2 The implicit features and features of the dataset are used to fuse the actual deformation of the key control points.
[0020] According to a specific implementation, in the above deformation prediction method, step S5 specifically includes:
[0021] After setting control points for the bridge segment to be predicted, deformation simulation is performed to obtain the simulated predicted deformation of the control points.
[0022] The simulated predicted deformation is input into the deformation prediction model to obtain the deformation prediction results of the key control points.
[0023] According to a specific implementation, in the above deformation prediction method, the deformation prediction result is used to indicate the assembly of bridge segments under hoisting deformation; and to compare the process deformation of bridge segments after heat treatment; and to straighten and install bridge segments.
[0024] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0025] The segment deformation prediction method provided by this invention, which compares scanned data with simulation data, improves the accuracy of simulation results for bridges under various postures by correcting simulation results with measured data. It provides an important reference for recording process deformation to guide the process and using simulation tools to utilize and avoid component deformation. Attached Figure Description
[0026] Figure 1 This is a flowchart illustrating a segment deformation prediction method based on the comparison of scanned data and simulation data, provided in an embodiment of the present invention. Detailed Implementation
[0027] The present invention will now be described in further detail with reference to specific embodiments. However, this should not be construed as limiting the scope of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.
[0028] This invention addresses the problem of significant discrepancies between actual and simulated deformation data of existing bridge segments. It proposes a segment deformation prediction method based on the comparison of scanned data and simulated data. By comparing and correcting CAE simulation predictions with actual on-site scanned data, a set of simulation data correction methods is formed.
[0029] Please refer to Figure 1 The diagram illustrates a flowchart of a segment deformation prediction method based on the comparison of scanned data and simulation data, provided by an embodiment of the present invention. The method includes:
[0030] S1. Based on the surface model of the bridge segment, control points are set, and deformation analysis is performed on the surface model according to the set process steps to obtain the simulated deformation amount of the control points in the corresponding process steps.
[0031] In this step, the process of setting the process steps includes setting the process steps that need to be controlled according to the key process control points of the manufacturing and construction process.
[0032] Specifically, based on the key process control points in the manufacturing and construction process, the critical process steps S that need to be controlled are set. i Input the ideal design model M0, and set n key control points P on the patch model M0 according to the process feature control points. n0(Generally, key control points mainly include the corner points of segments, the upper and lower edge points at intervals, and the centroid of the plates, etc.) At the same time, m auxiliary control points P are generated according to rules (such as according to the UV values of the patches or according to the preset mesh intervals). m0 Preset mesh spacing refers to a method of regularly sampling a series of auxiliary control points along the surface of a 3D patch model (such as a triangular mesh or a quadrilateral mesh) according to a fixed spatial or topological step size. These points do not depend on the geometric features of the model (such as corners or edges), but are automatically generated through uniform or semi-uniform mesh division, and are used to provide additional spatial constraint information for the neural network.
[0033] The S was obtained through analysis using finite element software. i The workpiece deformation triangular patch model M1 under deformation. Following the same rules, n key control points P under deformation are obtained. n1 m auxiliary control points P m1 Establish the deformation vector V of the control points in the simulation data. ni V mi .
[0034] S2. Based on the point cloud data of the bridge segments in the corresponding process steps, obtain the measured deformation of the control point and the three-dimensional coordinate data of the control point.
[0035] In this step, the point cloud data is obtained by scanning the bridge segments in the corresponding process steps using three-dimensional scanning technology, which includes at least one of laser scanning, oblique photography, and total station scanning.
[0036] Specifically, using laser scanning (including total station scanning) or oblique photography and other equivalent 3D scanning techniques, S is scanned. i The actual workpiece in the given state obtains a point cloud P0. Preferably, to obtain higher measurement accuracy, a total station is used for global positioning, and a laser scanner is used as a local scanning tool to obtain laser point cloud scanning results based on the global coordinates of the total station.
[0037] Further, a coarse localization transformation is performed on the point cloud data and the patch model to obtain a rotation transformation matrix between the point cloud data and the patch model. The point cloud data is then transformed according to the rotation transformation matrix to obtain coarsely localized point cloud data.
[0038] Calculate the point closest to each control point in the coarsely located point cloud data, obtain the vector from that point to the control point, and obtain the measured deformation.
[0039] In one possible implementation, calculating the coarse localization transformation includes performing a coarse localization transformation on the scanned point cloud P0 and the surface M0 to obtain the rotation transformation matrix M between the point cloud P0 and the surface M0.R Common methods, such as ICP, are used to obtain the coarsely located point cloud P1 by following this rotation transformation.
[0040] Preferably, in this embodiment of the invention, after the initial ICP rotation transformation results, the nearest point P between each point in the scanned point cloud P0 and the triangular facet M0 is calculated. i To obtain P0 and P i ICP transformation matrix M Ri After applying this transformation, repeat the above steps until P0 and P... i The average distance no longer decreases, and the final transformation matrix M is obtained. result .
[0041] In the aforementioned calculation of the nearest point between the point cloud and the facet, an octree of the M0 triangular facet can be constructed for acceleration. Preferably, SVD decomposition can be used to accelerate the calculation of the rotation matrix between the original point cloud and the nearest point pair. Furthermore, edges and corners can be further optimized using line fitting or facet fitting separately to obtain higher accuracy matching results.
[0042] Furthermore, calculate the distance P in P1. n1 With P m1 The nearest point P j1 and obtain P j1 To P n1 P m1 Vector V of corresponding point nj1 V mj1 The measured deformation was obtained. Then, a total station was used to measure the deformation at n key control points P. n and m auxiliary control points P m In S i The corresponding point positions of the actual workpiece under the condition are measured to obtain the three-dimensional coordinate data of the key control points and auxiliary control points, that is, the corresponding measured deformation.
[0043] S3. Repeat S1 to S2 to obtain several simulated deformation quantities, measured deformation quantities, and three-dimensional coordinate data corresponding to the control points, and generate a dataset. In this step, k vector relationship groups between simulation results and actual measurement results can be obtained.
[0044] S4. Input the dataset into the constructed neural network model for training to obtain a trained deformation prediction model.
[0045] Specifically, the deformation prediction model adopts a hierarchical prediction structure, including two multilayer perceptrons; wherein the multilayer perceptron F θ1 Used to extract implicit features from the dataset to obtain the actual deformation of the auxiliary control points; Multilayer Perceptron F θ2The implicit features and features of the dataset are used to fuse the actual deformation of the key control points.
[0046] In one possible implementation, a control point compensator based on a neural network is established, employing a hierarchical prediction architecture: the main input is n key control points P obtained from simulation calculations. n m auxiliary control points P m and S i Explicit environmental parameters of the actual workpiece under the condition (including V) ni V mi , and P n P m (3D coordinate data of the point) passes through a multilayer perceptron. Obtain implicit features This implicitly fits the actual deformation of the auxiliary control points measured by the total station. This forms a secondary branch:
[0047]
[0048] in, For dataset; For the secondary branch loss function Network parameters; These are the n key control points obtained from simulation calculations; These are the m auxiliary control points obtained from simulation calculations; Let V be the vector set of the i-th critical process step, including the deformation vector V of the control point. ni V mi and key control point P n and auxiliary control point P m 3D coordinate data; This refers to the measured deformation of auxiliary control points. It can be understood that explicit environmental parameters are physical quantities input into the neural network that represent the actual processing / measurement state of the workpiece. Here, the environment refers to the physical state or external conditions of the workpiece, providing the neural network with supplementary information about the current state of the actual workpiece. Additionally, The output is the actual deformation of the auxiliary control point. .
[0049] Furthermore, the implicit features extracted from the secondary branches , with the main sensor F θ2 Feature fusion is performed to train and predict the actual deformation of key control points. This forms a main branch:
[0050]
[0051] in, Main branch loss function Network parameters; The measured deformation at key control points; The output is the actual deformation of the key control points. .
[0052] The total loss function can be expressed as:
[0053]
[0054] in, Total loss function Network parameters, For the secondary branch loss function The weighting coefficients.
[0055] Furthermore, adversarial training is employed, introducing a gradient inversion layer (GRL). Adversarial learning reduces the dependence on simulation-specific environment parameters and error control points, thereby improving generalization. The adversarial loss function... It can be represented as:
[0056]
[0057] in, This represents taking the expectation over all datasets; Indicates the distribution of the dataset; Total loss function The weighting coefficients are used to control the mixing ratio of conventional loss and adversarial loss. This indicates the search for the total loss function. The direction of the largest perturbation (i.e., the worst-case adversarial dataset). To combat noise, This is the upper limit of the noise level.
[0058] Preferably, model iteration and mapping data storage are performed simultaneously. When the segment data volume k reaches a certain scale, the branch architecture network structure can be further replaced by combining physical constraints and hybrid modeling: a graph neural network is introduced to model the topological relationship between control points, treating each control point as a graph node, and the physical association between control points as edge weights, using a message passing mechanism to enhance spatial correlation learning; preferably, a generative model (such as a conditional variational autoencoder, CVAE) can be combined to generate control point compensation parameters that conform to physical constraints through joint distribution modeling of simulation-measured data pairs. The multilayer perceptron structure can still maintain high generalization ability under limited measured data, while the introduction of graph structure and generative methods in large-scale data scenarios can further reduce the prediction error of control points and support the dynamic calibration of segment digital twin systems.
[0059] S5. Based on the deformation prediction model, predict the deformation of the bridge segment to be predicted. Specifically, this includes:
[0060] After setting control points for the bridge segments to be predicted, deformation simulation is performed to obtain the simulated deformation of the control points.
[0061] The simulated deformation is input into the deformation prediction model to obtain the deformation prediction results of the key control points.
[0062] Specifically, for a control point given by a simulation that has not been measured, the deformation vector V ni and V mi Through the prediction in step S4, the corrected deformation prediction result is obtained, and the vector is obtained. .
[0063] In one possible implementation, embodiments of the present invention also provide a vector of deformation prediction results. The utilization of the deformation prediction results is used to indicate the assembly of bridge segments under hoisting deformation; and to compare the process deformation of bridge segments after heat treatment; and to straighten and install bridge segments.
[0064] Based on the above technical solution, the segment deformation prediction method by comparing scanned data and simulation data provided by the present invention improves the accuracy of simulation results of bridges under various postures by correcting simulation results with measured data. It provides an important reference for recording process deformation to guide the process and using simulation tools to utilize and avoid component deformation.
[0065] It should be noted that the above method embodiments can be applied to a processor, or implemented by a processor. A processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by integrated logic circuits in the processor's hardware or by software instructions.
[0066] The aforementioned processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.
[0067] The steps of the method disclosed in the embodiments of this application can be directly manifested as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in mature storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.
[0068] In embodiments of the present invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0069] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.
[0070] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.
[0071] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.
[0072] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), sync link dynamic random access memory (SLDRAM), and direct memory bus RAM (DRRAM).
[0073] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.
[0074] Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.
[0075] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for predicting segmental deformation based on the comparison of scanned data and simulation data, characterized in that, include: S1. Set control points based on the surface model of the bridge segment, perform deformation analysis on the surface model according to the set process steps, and obtain the simulated deformation amount of the control points in the corresponding process steps. S2. Based on the point cloud data of the bridge segments in the corresponding process steps, obtain the measured deformation of the control point and the three-dimensional coordinate data of the control point; S3. Repeat S1 to S2 to obtain several simulated deformations, measured deformations and three-dimensional coordinate data corresponding to the control points, and generate a dataset. S4. Input the dataset into the constructed neural network model for training to obtain a trained deformation prediction model; S5. Based on the deformation prediction model, predict the deformation of the bridge segment to be predicted. The control points include key control points and auxiliary control points; the key control points include the corner points of bridge segments, the upper and lower edge points at intervals, and the centroid of the slabs; the auxiliary control points include points generated according to the uv values of the patches in the patch model or at certain mesh intervals. The deformation prediction model adopts a hierarchical prediction structure, including two multilayer perceptrons; wherein the multilayer perceptron F θ1 Used to extract implicit features from the dataset to obtain the actual deformation of the auxiliary control points; Multilayer Perceptron F θ2 The implicit features and features of the dataset are used to fuse the actual deformation of the key control points.
2. The segment deformation prediction method based on the comparison of scanned data and simulation data according to claim 1, characterized in that, The process of setting the process steps includes: Based on the key process control points in the manufacturing and construction process, the process steps that need to be controlled are set.
3. The segment deformation prediction method based on the comparison of scanned data and simulation data according to claim 1, characterized in that, S2 specifically includes: A coarse localization transformation is performed on the point cloud data and the patch model to obtain a rotation transformation matrix between the point cloud data and the patch model. The point cloud data is then transformed according to the rotation transformation matrix to obtain coarsely localized point cloud data. Calculate the nearest point position of the control point in the coarsely located point cloud data, obtain the vector from the nearest point position to the control point, and obtain the measured deformation.
4. The segment deformation prediction method based on the comparison of scanned data and simulation data according to claim 3, characterized in that, The coarse positioning transformation employs the ICP method.
5. The segment deformation prediction method based on the comparison of scanned data and simulation data according to claim 3, characterized in that, The point cloud data is obtained by scanning the bridge segments in the corresponding process steps using three-dimensional scanning technology, which includes laser scanning, oblique photography, and total station scanning.
6. The segment deformation prediction method based on the comparison of scanned data and simulation data according to claim 1, characterized in that, S5 specifically includes: After setting control points for the bridge to be predicted, deformation simulation is performed to obtain the simulated deformation of the control points. The simulated deformation is input into the deformation prediction model to obtain the deformation prediction results of the key control points.
7. The segment deformation prediction method based on the comparison of scanned data and simulation data according to claim 6, characterized in that, The deformation prediction results are used to indicate the assembly of bridge segments under hoisting deformation; and to compare the process deformation of bridge segments after heat treatment. And, the straightening and installation of bridge segments.
8. An electronic device, characterized in that, The device includes a memory and a processor; The memory is used to store computer programs; the processor is used to call and execute the computer programs so that the device performs a segment deformation prediction method based on the comparison of scan data and simulation data as described in any one of claims 1 to 7.