Finite element model updating method for transmission tower based on static force synergy and residual driving

Abstract

The application discloses a power transmission tower finite element model correction method based on static force cooperation and residual driving, and the correction method comprises the following steps: S1, constructing a shape-preserving reduced-order truss-beam hybrid initial model; S2, establishing a synchronous response field of multi-working-condition parameterized simulation and real-type test; S3, defining an equivalent elastic modulus correction vector with explicit physical meaning; S4, constructing a residual driving type deep learning neural network inversion model and iteratively correcting; the method realizes high-fidelity correction of the tower finite element model through the fusion of three technologies of hybrid modeling, static test synchronous cooperation and deep network residual driving inversion, has the technical advantages of high correction precision and strong adaptability, and can provide reliable simulation model support for power transmission line tower structure safety evaluation, design optimization and operation and maintenance guarantee, and has high practicability, adaptability and engineering application value.

Description

Technical Field

[0001] This invention relates to the field of structural health monitoring and finite element numerical simulation analysis of transmission line towers, and in particular to a method for correcting finite element models of transmission towers based on static synergy and residual driving. Background Technology

[0002] As the core supporting structure of the power system backbone, the structural stability of transmission line towers directly determines the safety and reliability of power transmission. With the development of transmission lines towards higher voltage, longer distances, and larger capacities, tower structures are becoming increasingly complex, and the service environment is becoming more severe, leading to continuously increasing demands on the design accuracy and safety assessment of tower structures. Finite element simulation technology, with its advantages of high efficiency, economy, and strong repeatability, has become a core tool for the structural design, mechanical property analysis, and safety assessment of transmission line towers. The accuracy of its simulation results directly affects the rationality of tower design optimization and the scientific nature of operation and maintenance decisions.

[0003] However, current finite element simulation modeling and correction methods for power transmission towers still have the following significant technical shortcomings:

[0004] (1) It is difficult to optimize modeling accuracy and computational efficiency in a coordinated manner. Existing modeling methods mostly use a single element type. If all beam elements are used for modeling, although the bending, torsion and bending-torsion coupling effects of the components can be accurately simulated, the number of degrees of freedom is huge. For a power transmission tower composed of hundreds of angle steel components, the calculation scale increases sharply and the solution efficiency is low. If all truss elements are used for modeling, although the calculation is fast, the contribution of secondary bending moments in the main material node area is ignored, resulting in a lower simulated value of the overall stiffness of the tower and a systematic deviation between the displacement calculation results and the actual situation.

[0005] (2) There is a fundamental difference between the model material parameters and the actual structural state. Existing methods generally directly adopt the nominal elastic modulus value from material handbooks or design standards, failing to consider the material property dispersion of steel during rolling, welding, hot-dip galvanizing, and long-term field service, and also failing to account for the stiffness reduction effect caused by the non-ideal rigid connection of the gusset plate. This fundamental deviation between the "ideal model" and the "real structure" seriously restricts the credibility and guiding value of the simulation results in engineering practice.

[0006] (3) Traditional model correction methods suffer from ambiguous physical meaning and insufficient extrapolation and generalization capabilities. Some correction methods rely on a small amount of measurement data under a single load condition for trial adjustments, lacking a systematic and comprehensive consideration of the overall structural response field under multiple load conditions. Matrix-based correction methods based on sensitivity analysis directly modify the stiffness or mass matrix elements of the finite element model. The corrected model parameters lack clear physical interpretability and may even destroy the positive definiteness and banded characteristics of the matrix. Furthermore, the correction effect is heavily dependent on the number and layout of measurement points. The prediction accuracy of the model corrected by the above methods is difficult to guarantee under load conditions that are not involved in the correction, and the generalization ability is poor.

[0007] In summary, there is an urgent need to develop a method for correcting the finite element model of transmission towers that can systematically integrate high-fidelity hybrid modeling strategies, multi-condition real-world static tests, and advanced intelligent inversion algorithms, in order to obtain a high-confidence simulation model that can truly reflect the actual mechanical behavior of the towers. Summary of the Invention

[0008] To overcome the above problems, the purpose of this invention is to provide a method for correcting the finite element model of transmission towers based on static coordination and residual driving. This correction method adopts a three-in-one technical approach of "hybrid modeling of trusses and beams to preserve form and reduce order, synchronous coordination of multi-condition static tests, and deep network residual driving inversion" to accurately correct the equivalent elastic modulus parameters of the tower members. This makes the simulation response of the corrected model highly consistent with the actual test response, effectively solving the technical defects of insufficient accuracy of existing models, poor specificity of correction methods, and weak extrapolation and generalization ability.

[0009] The technical solution adopted in this invention is:

[0010] A method for correcting the finite element model of transmission towers based on static synergy and residual driving includes the following steps:

[0011] S1: Construct a form-preserving, reduced-order hybrid initial model of the truss and beam. Use beam elements to simulate the main tower body and crossarm main materials, and use rod elements to simulate the diagonal members, cross diaphragms, and auxiliary materials. Establish the initial finite element model of the transmission tower. Maintain consistency with the real tower at the geometric topology level, and achieve order reduction and simplification at the mechanical degree of freedom level.

[0012] S2: Establish a synchronous response field for multi-condition parametric simulation and real-world testing. Apply differentiated load spectra to the initial finite element model, extract the simulated displacement field and stress field of key measuring points, apply the same load spectrum to the real-world tower, and collect the measured displacement field and stress field corresponding to the simulated measuring points one by one through a high-precision sensor network to form a "simulation-measurement" response field.

[0013] S3: Define an equivalent elastic modulus correction vector with clear physical meaning. Based on the strength grade differences of the steel used in the tower components, define the elastic modulus of the members corresponding to different strength grades of steel as mutually independent parameters to be corrected, thus forming a multi-maintenance positive vector.

[0014] S4: Construct a residual-driven deep learning neural network inversion model and iteratively correct it.

[0015] As a further description of the present invention, the construction principle of the form-preserving and order-reduced truss hybrid initial model in S1 is as follows:

[0016] For secondary members with a slenderness ratio greater than 80 or whose end-node constraints are hinged, rod elements are used for simulation.

[0017] For main members with a slenderness ratio not greater than 80 and subjected to significant bending moments, beam elements considering shear deformation are used for simulation.

[0018] The initial model refines the mesh at the tower slope change nodes, crossarm roots, and tower foot plate areas to reduce the mesh distortion rate to less than 5%.

[0019] As a further description of the present invention, the selection of key measuring points in S2 is determined based on the results of sensitivity analysis, with priority given to nodes that rank in the top 20% of the sensitivity of displacement to changes in elastic modulus and surface areas of high-stress components with stress levels exceeding 50% of the material yield limit.

[0020] As a further description of the present invention, the high-precision sensing network in S2 includes an array of stress sensors and displacement sensors arranged in layers along the height of the tower. The data sampling of the stress sensors, displacement sensors and strain gauges is triggered by the same clock to ensure the time synchronization of the displacement field and the stress field.

[0021] As a further description of the present invention, the initial value of the equivalent elastic modulus correction vector in S3 is taken as 100% of the nominal elastic modulus of steel, and its upper and lower limits of the search domain are set as 110% and 90% of the nominal value, respectively.

[0022] The dimension of the equivalent elastic modulus correction vector is equal to the number of strength grades of steel actually used in the real-type iron tower, thereby achieving independent decoupled correction of the elastic modulus of steel of different strength grades.

[0023] As a further description of the present invention, the specific process of S4 is as follows:

[0024] S41: Calculate the normalized residual vector between the “simulation-measured” response fields.

[0025] S42: Using the residual vector calculated in S41 as the network input and the equivalent elastic modulus correction vector in S3 as the network output, a deep learning neural network is constructed and trained to establish a nonlinear mapping relationship from the "response deviation domain" to the "parameter correction domain".

[0026] S43: Use the trained network to iteratively invert and solve the parameters to be corrected. If the deviation between the simulated response field and the measured response field meets the preset convergence criterion, proceed to S44; if the preset convergence criterion is not met, return to S41 to recalculate the residual vector.

[0027] S44: Output the optimal elastic modulus.

[0028] S45: Back-substitution model, correction complete.

[0029] As a further description of the present invention, the normalized residual vector in S41 is composed of a weighted combination of the normalized displacement residual components and stress residual components, and the weighting coefficients are allocated according to the measurement signal-to-noise ratio and engineering importance of each response quantity.

[0030] As a further description of the present invention, the deep learning neural network in S42 is any one of a fully connected feedforward neural network with multiple hidden layers, a convolutional neural network, or a radial basis function neural network.

[0031] As a further description of the present invention, the preset convergence criterion in S43 is:

[0032] The weighted root mean square error between the simulated response field and the measured response field of the corrected finite element model is less than 5%.

[0033] A smart correction system for a finite element simulation model of a transmission line tower is provided to implement the aforementioned correction method for a finite element model of a transmission line tower based on static coordination and residual driving. The system includes:

[0034] The hybrid model building module is used to generate a conformal, reduced-order hybrid initial model of the truss beam.

[0035] The multi-condition collaborative loading and synchronous acquisition module is used to apply differentiated load spectra to the initial finite element model, apply the same load spectrum to the real iron tower, and acquire the measured displacement field and stress field corresponding to the simulation measurement points.

[0036] The residual-driven deep learning inversion module is used to construct a residual-driven deep learning neural network inversion model and iteratively correct it, solve for the optimal equivalent elastic modulus correction vector, and complete the model correction.

[0037] The beneficial effects of this invention are:

[0038] This invention presents a method for correcting the finite element model of transmission towers based on static synergy and residual driving. This method establishes clear element type selection criteria based on the slenderness ratio and stress characteristics of the components. Beam elements are used for the main material to preserve shape and accuracy, while rod elements are used for the diagonal and auxiliary materials to eliminate redundant degrees of freedom. Under the premise of ensuring geometric and topological consistency, the mechanical degrees of freedom of the model are reduced and simplified, effectively balancing the contradiction between simulation accuracy and solution efficiency.

[0039] This invention relates to a method for correcting the finite element model of transmission towers based on static synergy and residual drive. This method defines the elastic modulus as an independent parameter to be corrected according to the differences in steel strength grades. Compared with the traditional matrix finite element model correction (which directly corrects the stiffness, mass, or damping matrix of the finite element model), this method comprehensively characterizes the equivalent reduction effect of multiple factors such as material discreteness and nodal semi-rigidity on the overall stiffness, and achieves decoupled correction of multiple material parameters. The correction results have clear physical meaning and are easy to interpret in engineering.

[0040] This invention is based on a method for correcting finite element models of transmission towers using static synergy and residual driving. This method introduces deep learning neural networks into the field of transmission tower model correction, establishes a nonlinear inversion mapping from the "response deviation domain" to the "parameter correction domain", does not rely on the sensitivity matrix, and can efficiently and accurately solve complex inverse problems. The corrected model also has high-precision prediction capabilities under verification conditions without training. Attached Figure Description

[0041] Figure 1 This is a flowchart of the method for correcting the finite element model of a power transmission tower based on static synergy and residual driving proposed in this invention.

[0042] Figure 2 This is a schematic diagram of the finite element simulation model (truss-beam hybrid model) of the transmission line tower in Embodiment 3 of the method for correcting the finite element model of the transmission tower based on static coordination and residual driving proposed in this invention.

[0043] Figure 3 This is a flowchart of the BP neural network optimization algorithm in Embodiment 3 of the method for correcting the finite element model of a power transmission tower based on static coordination and residual driving proposed in this invention. Detailed Implementation

[0044] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0045] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0046] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0047] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0048] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0049] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0050] like Figures 1-3 As shown, it illustrates a specific embodiment of the present invention:

[0051] Example 1:

[0052] A method for correcting the finite element model of transmission towers based on static synergy and residual driving includes the following steps:

[0053] S1: Construct a form-preserving, reduced-order hybrid initial model of the truss and beam. Use beam elements to simulate the main tower body and crossarm main materials, and use rod elements to simulate the diagonal members, cross diaphragms, and auxiliary materials. Establish the initial finite element model of the transmission tower. Maintain consistency with the real tower at the geometric topology level, and achieve order reduction and simplification at the mechanical degree of freedom level.

[0054] Specifically, the construction principle of the form-preserving and order-reduced hybrid initial model of the truss beam in S1 is as follows:

[0055] For secondary members with a slenderness ratio greater than 80 or whose end-node constraints are hinged, rod elements are used for simulation.

[0056] For main members with a slenderness ratio not greater than 80 and subjected to significant bending moments, beam elements considering shear deformation are used for simulation.

[0057] The initial model undergoes mesh refinement at the tower slope variation nodes, crossarm roots, and tower foot plate areas to reduce the mesh distortion rate to below 5%, balancing computational accuracy and efficiency.

[0058] In this embodiment, using beam elements to simulate the main tower body and crossarm main material can retain the contribution of the transmission tower to bending stiffness, while using pole elements to simulate diagonal members, crossbars and auxiliary materials can eliminate redundant degrees of freedom.

[0059] S2: Establish a synchronous response field for multi-condition parametric simulation and real-world testing. Apply differentiated load spectra to the initial finite element model, extract the simulated displacement field and stress field of key measuring points, apply the same load spectrum to the real-world tower, and collect the measured displacement field and stress field corresponding to the simulated measuring points one by one through a high-precision sensor network to form a "simulation-measurement" response field.

[0060] Specifically, the selection of key measuring points in S2 is determined based on the results of sensitivity analysis. Priority is given to selecting the nodes that rank in the top 20% in terms of the sensitivity of displacement to changes in elastic modulus, as well as the surface areas of high-stress components with stress levels exceeding 50% of the material's yield limit.

[0061] Specifically, the high-precision sensing network in S2 includes an array of stress sensors and displacement sensors arranged in layers along the height of the tower. The data sampling of the stress sensors, displacement sensors and strain gauges is triggered by the same clock to ensure the time synchronization of the displacement field and the stress field.

[0062] S3: Define an equivalent elastic modulus correction vector with clear physical meaning. Based on the strength grade differences of the steel used in the tower components, define the elastic modulus of the members corresponding to different strength grades of steel as mutually independent parameters to be corrected, thus forming a multi-maintenance positive vector.

[0063] In this embodiment, the correction vector comprehensively represents the equivalent reduction of overall stiffness caused by factors such as material discreteness, nodal semi-rigidity effect, and residual stress.

[0064] Specifically, the initial value of the equivalent elastic modulus correction vector in S3 is taken as 100% of the nominal elastic modulus of steel, and its upper and lower limits of the search domain are set as 110% and 90% of the nominal value, respectively.

[0065] The dimension of the equivalent elastic modulus correction vector is equal to the number of strength grades of steel actually used in the real-type iron tower, thereby achieving independent decoupled correction of the elastic modulus of steel of different strength grades.

[0066] S4: Construct a residual-driven deep learning neural network inversion model and iteratively correct it.

[0067] Specifically, the process of S4 is as follows:

[0068] S41: Calculate the normalized residual vector between the “simulation-measured” response fields.

[0069] S42: Using the residual vector calculated in S41 as the network input and the equivalent elastic modulus correction vector in S3 as the network output, a deep learning neural network is constructed and trained to establish a nonlinear mapping relationship from the "response deviation domain" to the "parameter correction domain".

[0070] S43: Use the trained network to iteratively invert and solve the parameters to be corrected. If the deviation between the simulated response field and the measured response field meets the preset convergence criterion, proceed to S44; if the preset convergence criterion is not met, return to S41 to recalculate the residual vector.

[0071] S44: Output the optimal elastic modulus.

[0072] S45: Back-substitution model, correction complete.

[0073] Specifically, in S41, the normalized residual vector is composed of a weighted combination of the normalized displacement residual components and the stress residual components, and the weighting coefficients are allocated according to the measurement signal-to-noise ratio and engineering importance of each response quantity.

[0074] Specifically, the deep learning neural network in S42 can be any one of a fully connected feedforward neural network with multiple hidden layers, a convolutional neural network, or a radial basis function neural network.

[0075] Specifically, the preset convergence criterion in S43 is:

[0076] The weighted root mean square error between the simulated response field and the measured response field of the corrected finite element model is less than 5%.

[0077] In this embodiment, as Figure 1As shown, this correction method establishes clear element type selection criteria based on the slenderness ratio and stress characteristics of the components. Beam elements are used for the main material to preserve form and accuracy, while rod elements are used for the diagonal and auxiliary materials to eliminate redundant degrees of freedom. This achieves a reduction in the mechanical degrees of freedom of the model while maintaining geometric topological consistency, effectively balancing the contradiction between simulation accuracy and solution efficiency. Furthermore, this correction method defines the elastic modulus as an independent parameter to be corrected based on the differences in steel strength grades. Compared to traditional matrix-type finite element model correction (which directly corrects the stiffness, mass, or damping matrix of the finite element model), it comprehensively characterizes the equivalent reduction effect of multiple factors such as material discreteness and nodal semi-rigidity on the overall stiffness, achieving decoupled correction of multiple material parameters. The correction results have clear physical meaning and are easy to interpret in engineering. Finally, this correction method introduces deep learning neural networks into the field of transmission tower model correction, establishing a nonlinear inversion mapping from the "response deviation domain" to the "parameter correction domain." It does not rely on the sensitivity matrix and can efficiently and accurately solve complex inverse problems. The corrected model also has high-precision prediction capabilities under verification conditions without training.

[0078] Example 2:

[0079] Based on the above embodiment one, this embodiment provides an intelligent correction system for the finite element simulation model of transmission line towers, used to implement the aforementioned correction method for the finite element model of transmission line towers based on static coordination and residual driving. The system includes:

[0080] The hybrid model building module is used to generate a conformal, reduced-order hybrid initial model of the truss beam.

[0081] The multi-condition collaborative loading and synchronous acquisition module is used to apply differentiated load spectra to the initial finite element model, apply the same load spectrum to the real iron tower, and acquire the measured displacement field and stress field corresponding to the simulation measurement points.

[0082] The residual-driven deep learning inversion module is used to construct a residual-driven deep learning neural network inversion model and iteratively correct it, solve for the optimal equivalent elastic modulus correction vector, and complete the model correction.

[0083] In this embodiment, the system performs adaptive correction of the finite element model for a 220kV transmission line straight-line tower (tower height 42.5m, nominal height 30m) that is actually in operation. The system mainly includes three core modules: a hybrid model construction module, a multi-condition collaborative loading and synchronous acquisition module, and a residual-driven deep learning inversion module.

[0084] The hybrid model building module shown is responsible for generating a conformal, reduced-order hybrid initial model of the truss structure. The system first reads the tower's design drawings and BIM model files, automatically extracting the geometric parameters (length, cross-sectional shape and dimensions, spatial coordinates) and connection node information of all members. For the straight tower in this embodiment, a total of 64 main members, 216 diagonal members, and 180 auxiliary members are extracted, totaling 460 member elements and 158 key nodes. The system automatically determines the simulation type of each element based on its slenderness ratio, end connection method, and its stress role in the overall structure.

[0085] Beam element determination: The main members (slenderness ratio <80 and rigid connection at both ends) and the main crossbars in the load-bearing diaphragm are selected using three-dimensional elastic beam elements (B31 elements, 6 degrees of freedom per node, capable of withstanding axial force, shear force, bending moment and torque).

[0086] Truss element determination: diagonal members (slenderness ratio ≥ 80), auxiliary members and secondary cross supports are selected using three-dimensional truss elements (T3D2 elements, each node has 3 degrees of freedom and only bears axial force).

[0087] Hybrid connection treatment: At the beam-truss hybrid connection node, the system automatically sets the rotational degree of freedom constraint equation to ensure that the bending moment is transmitted only by the beam element, while the truss element only participates in the axial force transmission.

[0088] In this embodiment, to avoid computational redundancy due to excessive model size, the system implements a reduction strategy: dense diagonal members within a 3m range in the height direction are merged into a "super truss element" based on the principle of equivalent stiffness, retaining their axial stiffness contribution but reducing the number of elements; non-critical nodes are treated as slave nodes, and static condensation is achieved through master-slave degree-of-freedom condensation. After reduction, the total number of model elements is compressed from the original 460 to 188, the number of nodes is reduced from 158 to 67, and the computational degrees of freedom are reduced by approximately 56%, while ensuring that the displacement accuracy deviation of critical nodes (tower legs, crossarm hanging points, and tower slope changes) does not exceed 3%.

[0089] In this embodiment, the system assigns an initial equivalent elastic modulus to each unit according to the design specifications: the main beam unit takes... diagonal truss unit Poisson's ratio for all materials The initial finite element model generated at this time is denoted as... .

[0090] The multi-condition collaborative loading and synchronous acquisition module enables the application of the same differentiated load spectrum to the initial finite element model and the real iron tower, and synchronously acquires the corresponding displacement field and stress field.

[0091] In this embodiment, the system automatically generates a load spectrum containing five representative working conditions, covering typical stress scenarios in the operation of transmission towers:

[0092] Operating condition W1: Design vertical load (conductor self-weight + 2 times the weight of icing, total vertical force 180kN).

[0093] Condition W2: 90° lateral wind load (basic wind speed 30m / s).

[0094] Condition W3: 45° oblique wind load (synthetic wind load, used to test the bending-torsional coupling effect of the main material).

[0095] Condition W4: Wire breakage condition (sudden release of tension on one side of the conductor, simulating an accident).

[0096] Condition W5: Non-uniform icing load (ratio of icing thickness on the windward side to the leeward side is 2:1).

[0097] The load for each working condition is applied in 10 progressively increasing increments (each increment increasing the target load by 10%) to obtain the nonlinear response process of the structure.

[0098] The following sensors were installed on a full-scale transmission tower of the same model at the test site: High-precision reflective targets were attached to the corresponding positions of 67 reserved nodes in the simulation model, and three-dimensional displacement was collected using a Leica TS60 total station (measurement accuracy 0.5mm). Key measuring points were the tower top, crossarm ends, and tower leg slope changes. Resistance strain gauges were attached axially to the middle sections of beam and truss elements, with a total of 124 strain measuring points. Each measuring point simultaneously measured strain on both sides to eliminate the influence of bending. The system synchronized finite element simulation calculations with the physical loading of the full-scale tower through a distributed control system: For the finite element model, the Abaqus solver was called, and a static implicit algorithm was used to calculate the nodal displacements and element stresses under each load level. For the full-scale tower, a hydraulic servo actuator system (maximum loading force 500kN) was used to load the tower step-by-step according to the same load spectrum, and the data acquisition chassis synchronously recorded strain and displacement signals at a sampling rate of 100Hz. After each load level is applied, the system automatically stores the simulation and measured results into the time-series database according to a unified data structure (node ​​ID-load case number-load level-displacement component / stress component). The collected raw data is filtered by median filtering to remove impulse noise, and linear interpolation is used to fill in any missing points caused by sensor malfunctions. Finally, a "simulation-measurement paired dataset" is generated for inversion correction.

[0099] The residual-driven deep learning inversion module constructs a residual-driven deep learning neural network inversion model and completes the model correction by iteratively optimizing the solution of the optimal equivalent elastic modulus correction vector.

[0100] In this embodiment, the entire tower structure is divided into six correction zones based on the stress characteristics of its members: R1, main members of the tower legs (elevation 0-8m); R2, main members of the lower part of the tower body (8-20m); R3, main members of the upper part of the tower body (20-42.5m); R4, all diagonal members and cross braces (uniform correction factor); R5, main members of the crossarms; R6, diagonal members of the crossarms. The vector to be corrected is... each This represents the multiplicative correction factor for the equivalent elastic modulus of this region, with an initial value of 1.0. The elastic modulus of each element after correction is...

[0101] A residual-driven neural network is constructed. The system builds a deep neural network with physical constraints, whose input is a correction vector. The output is the corresponding full-field residual vector. The network structure is as follows:

[0102] Input layer: 6 nodes, corresponding to .

[0103] Hidden layers: 4 fully connected layers, with 32, 64, 64, and 32 neurons in each layer, respectively, using LeakyReLU as the activation function. Each hidden layer is followed by a batch normalization layer and a Dropout layer to prevent overfitting.

[0104] Output layer: Output residual vector ,in The displacement residual is (dimension 67×3=201). The stress residual (dimension 124×1=124) is output with a total dimension of 325.

[0105] Residual connection: Introducing a bypass path in the middle of the network that connects directly from the input layer to the output layer (through only a linear transformation). This means that what the network actually learns is "the residual within the residual": .

[0106] Define a loss function, which takes the form of a weighted combination. The core driving term is the difference between the measured residuals and the model predicted residuals.

[0107] .

[0108] in:

[0109] , which is the data fitting term.

[0110] That is, the physical residual term. The overall stiffness matrix under the current corrected parameters is denoted by F, which is the external load vector. This term forces the model to satisfy the static equilibrium equation, meaning that the residual drive comes not only from data comparison but also from the degree of violation of mechanical equilibrium.

[0111] This is an L2 regularization term to prevent the correction coefficient from deviating excessively from the physically permissible range.

[0112] Pick

[0113] Iterative inversion optimization first uses Latin hypercube sampling in... Generate 2000 sets of samples in a 6-dimensional space, and for each set... Call the finite element solver to calculate the corresponding and Obtain the training set. Train the RDN network for 200 epochs using the Adam optimizer (learning rate 0.001, batch size 64) to enable the network to quickly predict any... The corresponding residual R.

[0114] To minimize To achieve the goal, use Bayesian optimization (20 initial points, 80 iterations) in... The optimal correction vector is searched within the space. In each iteration, RDN provides residual predictions in milliseconds, eliminating the need to repeatedly call time-consuming finite element method calculations.

[0115] Local fine-tuning, with optimal results obtained through Bayesian optimization. For initial values, the L-BFGS-B algorithm (boundary constraints) is used. Further gradient descent optimization is performed. At this point, the loss function... The weight is increased to 0.9 to finely fit the local measurement point response.

[0116] Convergence determination and output: Two metrics are calculated in real time during the iteration process.

[0117] The iteration terminates when the root mean square error of displacement residuals and the relative error of stress residuals are satisfied or the number of iterations exceeds 150, and the optimal correction vector is output.

[0118] In this embodiment, the final optimal correction vector is:

[0119] .

[0120] The physical meaning of the optimal correction vector is as follows: The measured stiffness of the main members of the tower legs is lower than the design value (correction factor 0.928), which may be due to minor corrosion during long-term operation; the stiffness of the main members in the upper part of the tower is higher (1.112), reflecting the actual material strength margin; while the overall stiffness of the diagonal members is lower (0.873), indicating that the semi-rigid assumption of the diagonal member connections in the original design is too ideal and there is actual connection slip.

[0121] For the final model correction and verification, the system writes the optimal correction vector into the initial model M0 to generate a corrected high-fidelity finite element model. . Under two additional verification conditions (Condition W6: 75° oblique wind load; Condition W7: 2 times the vertical load + 80% of the design wind load), the comparison between, M0 and the measured results are shown in the following table:

[0122]

[0123] The displacement and stress errors of the corrected model are both reduced to within 5%, verifying the effectiveness of this system. At the same time, the corrected model retains the same calculation efficiency as the initial model (the number of elements does not increase), meeting the requirements of real-time engineering analysis.

[0124] Example 3:

[0125] In this example, a certain double-circuit 220 kV transmission line tower is taken as the research object. The tower is 36 m high, with a root opening of 8 m. The main members are made of Q345 steel, and the cross braces and diagonal members are made of Q235 steel, all of which are common specifications in engineering. The test site is selected as a certain power engineering test base (with a flat terrain and no external vibration and other interference factors). The finite element simulation model is corrected by using the method provided by this invention. The specific steps are as follows:

[0126] (1) Construction of the initial finite element simulation model.

[0127] Referring to the complete design drawings of the double-circuit 220 kV transmission line tower, including component dimensions, joint connection details, and material property parameters), a transmission tower is established using a truss-beam hybrid model. The Beam188 beam element in the simulation software is used to simulate the main members of the tower body, cross braces, and the main members of the tower head. Its cross-section is determined according to the actual materials used. The Link10 rod element is used to simulate the diagonal members of the tower. The cross-section of its rod is equivalently calculated according to the actual cross-sectional area size of the angle steel. Since the auxiliary members of the tower in the pole and tower structure are less stressed, the modeling of this part of the auxiliary members is ignored during the modeling process, and their mass is added to the tower in the form of gravitational acceleration.

[0128] The initial elastic modulus of the material is set according to the standard value of GB / T 1591-2018. The initial elastic modulus of Q345 steel is 2.06×10 5MPa, Poisson's ratio 0.3; the initial elastic modulus of Q235 steel is 2.03×10⁻⁶. 5 MPa, Poisson's ratio 0.3; the model mesh was generated using free meshing, with mesh size controlled between 50 and 100 mm, mesh distortion rate of 2.2%, and mesh quality compliance rate of over 98%. The initial finite element simulation model schematic is shown below. Figure 2 As shown.

[0129] (2) Establishment of parametric finite element model.

[0130] Based on the initial finite element simulation model described above, the static analysis module of the simulation software was used to conduct finite element static analyses under four different load conditions, covering typical stress scenarios in actual service of the tower, namely:

[0131] A: No-load condition (no external load, only the self-weight of the tower is considered).

[0132] B: Single vertical load condition (load size is 50kN, corresponding to the equivalent load of conductor self-weight).

[0133] C: Single horizontal load condition (load size is 30kN, corresponding to the equivalent value of light wind load).

[0134] D: Single tension load condition (load size is 40kN, corresponding to the equivalent value of conductor tension).

[0135] The input and output parameters of the model under each working condition are extracted. The input parameters are the load magnitude, load direction, and position of each load point. The vertical load is applied to the conductor suspension point in the middle of the crossarm, with the direction vertically downward; the horizontal load is applied to the top of the tower, with the direction perpendicular to the tower axis; and the tension load is applied to the conductor suspension point, with the direction along the conductor erection direction. The output parameters are the displacement, stress, and strain values ​​of several preset measuring points on the tower. Three characteristic monitoring points are selected for each measuring point, and the average value is taken as the output data for that measuring point to reduce simulation errors.

[0136] Using the elastic modulus of the rod as the core variable, and the elastic modulus of Q345 steel and Q235 steel as adjustable variables, the input parameters and output parameters are correlated to construct a variable-response mapping relationship, and a parametric finite element model is established to provide a foundation for subsequent parameter iterative optimization.

[0137] (3) Static test of real-type transmission tower.

[0138] To conduct static testing on a full-scale transmission tower, the testing equipment was first deployed. The load application system employed a series of wire rope tension sensors for load application and real-time load monitoring, ensuring load application accuracy. Displacement sensors were installed at pre-set measuring points on the tower (consistent with the initial simulation model's measuring point locations) for real-time displacement detection. The sensors were vertically aligned with the measuring points to avoid detection deviations. Strain gauges were adhered to the surface of the components at each measuring point. Before adhesion, the component surfaces were ground, degreased, and dried. After adhesion, the surfaces were left to stand for 24 hours to ensure firm bonding. Stress and strain values ​​were simultaneously acquired using a series data acquisition instrument.

[0139] Table 1 Test Equipment Model Parameter Table

[0140]

[0141] Loads were applied according to four different load conditions in the simulation model: no load (no external load), single vertical load (load magnitude 50kN), single horizontal load (load magnitude 30kN), and single tensile load (load magnitude 40kN). After each load application, the load was allowed to stabilize for 4 minutes before data acquisition to avoid the impact of instantaneous load fluctuations and tower structure creep on the accuracy of the test data. Data was collected three times for each load gradient, with each acquisition lasting 10 seconds, ultimately forming a complete database of static test data for the real-world tower.

[0142] (4) Construct a residual-driven deep learning neural network inversion model and iteratively correct it.

[0143] Based on the static analysis data of the initial finite element simulation model and the static test data of the real iron tower, the least squares method is used to calculate the difference between the two (displacement difference and stress difference) to ensure the accuracy of the difference calculation.

[0144] The elastic modulus of the tower members was selected as the sole target parameter for correction, and the correction range was strictly set to 90%~110% of the standard value to avoid the correction parameter exceeding the actual performance range of the material.

[0145] This embodiment employs a backpropagation (BP) neural network within a fully connected feedforward neural network. The BP neural network is constructed using MATLAB software. The input layer has 10 nodes (5 displacement and 5 stress differences from 5 measurement points), the hidden layer has 12 nodes, and the output layer has 2 nodes (representing the elastic modulus corrections for Q345 and Q235 steel, respectively). The activation function is the Sigmoid function, the number of iterations is set to 50, and the convergence accuracy is set to [value missing]. With a learning rate of 0.01, the momentum gradient descent method is used to optimize the iterative process, avoiding getting trapped in local optima and ensuring that the iterative results are stable and reliable.

[0146] The difference between the initial simulation data and the actual experimental data is used as the input to the BP neural network, and the elastic modulus correction is used as the output to start parameter iterative optimization. In each iteration, the elastic modulus parameters in the parametric finite element model are automatically updated, and the ANSYS software is called to re-perform the finite element static analysis to calculate the difference between the new simulation data and the experimental data. The difference is fed back to the BP neural network for the next round of iteration. During the iteration, the convergence curve is monitored in real time. If the difference does not change significantly after 5 consecutive iterations, it is determined to be converged. The iteration ends when the difference meets the termination condition (the relative error between the output parameters of the corrected simulation model and the actual experimental data is ≤5%).

[0147] After the iteration terminated, the optimized elastic modulus parameters were obtained. These optimized parameters were then substituted into the initial finite element simulation model to complete the model correction. Further model verification was performed using a single vertical load condition without correction. Under the same test environment and at the same measuring point locations, output data from the simulation model before and after correction were collected, along with actual test data, and the deviations were compared. Before correction, the relative errors between the displacement and stress data at the bottom, middle, and top of the tower, as well as at both ends of the crossarm, and the test data were 8.2%–11.5%, with an average error of 9.8%. After correction, the relative errors at each measuring point decreased to 2.1%–4.3%, with an average error of 3.2%, all meeting the termination condition of ≤5%. This indicates that the corrected finite element simulation model accurately fits the actual working conditions of the real-world tower, truly reflecting the static response characteristics of the tower. The correction effect is excellent and fully meets the accuracy requirements of actual engineering simulation.

[0148] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

[0149] Many other changes and modifications can be made without departing from the concept and scope of this invention. It should be understood that this invention is not limited to the specific embodiments, and the scope of this invention is defined by the appended claims.

Claims

1. A method for correcting the finite element model of transmission towers based on static synergy and residual driving, characterized in that, Includes the following steps: S1: Construct a form-preserving and order-reduced initial truss-beam hybrid model. Use beam elements to simulate the main tower body and crossarm main materials, and use rod elements to simulate diagonal members, cross diaphragms and auxiliary materials. Establish the initial finite element model of the transmission tower. Maintain consistency with the real tower at the geometric topology level, and achieve order reduction and simplification at the mechanical degree of freedom level. S2: Establish a synchronous response field for multi-condition parametric simulation and real-world test. Apply differentiated load spectra to the initial finite element model, extract the simulated displacement field and stress field of key measuring points, apply the same load spectrum to the real-world tower, and collect the measured displacement field and stress field corresponding to the simulated measuring points one by one through a high-precision sensor network to form a "simulation-measurement" response field. S3: Define an equivalent elastic modulus correction vector with clear physical meaning. Based on the strength grade differences of the steel used in the tower components, define the elastic modulus of the members corresponding to different strength grades of steel as mutually independent parameters to be corrected, thus forming a multi-maintenance positive vector. S4: Construct a residual-driven deep learning neural network inversion model and iteratively correct it.

2. The method for correcting the finite element model of transmission towers based on static synergy and residual driving as described in claim 1, characterized in that, The construction principle of the form-preserving and order-reduced truss hybrid initial model in S1 is as follows: For secondary members with a slenderness ratio greater than 80 or whose end-node constraints are hinged, rod elements are used for simulation. For main members with a slenderness ratio not greater than 80 and subjected to significant bending moments, beam elements considering shear deformation are used for simulation. The initial model refines the mesh at the tower slope change nodes, crossarm roots, and tower foot plate areas to reduce the mesh distortion rate to less than 5%.

3. The method for correcting the finite element model of transmission towers based on static synergy and residual driving as described in claim 1, characterized in that, The selection of key measuring points in S2 is determined based on the results of sensitivity analysis. Priority is given to selecting nodes that rank in the top 20% of the sensitivity of displacement to changes in elastic modulus, as well as surface areas of high-stress components with stress levels exceeding 50% of the material's yield limit.

4. The method for correcting the finite element model of transmission towers based on static synergy and residual driving as described in claim 1, characterized in that, The high-precision sensing network in S2 includes an array of stress sensors and displacement sensors arranged in layers along the height of the tower. The data sampling of the stress sensors, displacement sensors and strain gauges is triggered by the same clock to ensure the time synchronization of the displacement field and the stress field.

5. The method for correcting the finite element model of transmission towers based on static synergy and residual driving as described in claim 1, characterized in that, The initial value of the equivalent elastic modulus correction vector in S3 is taken as 100% of the nominal elastic modulus of steel, and its upper and lower limits of search domain are set as 110% and 90% of the nominal value, respectively. The dimension of the equivalent elastic modulus correction vector is equal to the number of strength grades of steel actually used in the real-type iron tower, thereby achieving independent decoupled correction of the elastic modulus of steel of different strength grades.

6. The method for correcting the finite element model of transmission towers based on static synergy and residual driving as described in claim 1, characterized in that, The specific process of S4 is as follows: S41: Calculate the normalized residual vector between the "simulation-measurement" response fields; S42: Using the residual vector calculated in S41 as the network input and the equivalent elastic modulus correction vector in S3 as the network output, a deep learning neural network is constructed and trained to establish a nonlinear mapping relationship from the "response deviation domain" to the "parameter correction domain". S43: Use the trained network to iteratively invert and solve the parameters to be corrected. If the deviation between the simulated response field and the measured response field meets the preset convergence criterion, proceed to S44. If the preset convergence criterion is not met, return to S41 to recalculate the residual vector; S44: Outputs the optimal elastic modulus; S45: Back-substitution model, correction complete.

7. The method for correcting the finite element model of transmission towers based on static synergy and residual driving as described in claim 6, characterized in that, The normalized residual vector in S41 is composed of a weighted combination of the normalized displacement residual components and stress residual components, and the weighting coefficients are allocated according to the measurement signal-to-noise ratio and engineering importance of each response quantity.

8. The method for correcting the finite element model of transmission towers based on static synergy and residual driving as described in claim 6, characterized in that, The deep learning neural network in S42 can be any one of a fully connected feedforward neural network with multiple hidden layers, a convolutional neural network, or a radial basis function neural network.

9. The method for correcting the finite element model of transmission towers based on static synergy and residual driving according to claim 6, characterized in that, The preset convergence criterion in S43 is: The weighted root mean square error between the simulated response field and the measured response field of the corrected finite element model is less than 5%.

10. An intelligent correction system for a finite element simulation model of a transmission line tower, characterized in that, The system is used to implement the method for correcting the finite element model of a transmission tower based on static synergy and residual driving as described in any one of claims 1-9, the system comprising: The hybrid model building module is used to generate a form-preserving, reduced-order hybrid initial model of trusses; The multi-condition collaborative loading and synchronous acquisition module is used to apply differentiated load spectra to the initial finite element model, apply the same load spectra to the real iron tower, and acquire the measured displacement field and stress field corresponding to the simulation measurement points one by one. The residual-driven deep learning inversion module is used to construct a residual-driven deep learning neural network inversion model and iteratively correct it, solve for the optimal equivalent elastic modulus correction vector, and complete the model correction.

Images