Welding stress prediction and point location optimization method for bridge assembly and electronic equipment

Abstract

The invention provides a welding stress prediction and point location optimization method for bridge assembly and electronic equipment, and relates to the technical field of bridge assembly. The method comprises the following steps: acquiring three-dimensional digital assembling data of the steel member; based on the data, welding characteristic parameters in the steel member assembling process are extracted, a parameterized command stream is adopted to establish a welding stress finite element model of multiple assembling point positions within a preset deviation range, and a welding assembling characteristic data set is constructed; on the basis of the welding and assembling feature data set, the welding residual stress of each assembling point position is predicted through a nonlinear regression prediction model; and analysis is carried out based on an interpretability analysis algorithm, welding stress influence factors are identified, and an optimal splicing scheme is determined by taking the minimum welding residual stress as a target. According to the method, accurate prediction and active optimization of the welding stress can be achieved, stress control is achieved before construction, welding stress concentration is effectively avoided, and the long-term stability, fatigue durability and overall construction precision of a bridge structure are improved.

Description

Technical Field

[0001] This invention relates to the field of beam bridge assembly technology, and in particular to a method and electronic equipment for predicting welding stress and optimizing the location of welding points in bridge assembly. Background Technology

[0002] In the construction of modern steel beam bridges and steel-concrete composite beam bridges, the control of assembly accuracy and welding stress is crucial to ensuring the overall safety and long-term stable operation of the bridge. With the widespread application of digital measurement methods such as 3D laser scanning, the ability to assemble, position, and control the geometric dimensions of bridge structures has been significantly improved. However, relying solely on improved geometric accuracy is insufficient to fully address the residual stress generated during welding. If this stress is not effectively managed, it will pose a potential threat to the fatigue life and structural performance of the bridge.

[0003] Currently, 3D laser scanning technology is widely used in bridge construction for measurement and 3D modeling, enabling precise acquisition of component dimensions and assembly point information, thereby optimizing assembly processes and improving the accuracy of structural assembly. Furthermore, in welding stress analysis, traditional finite element analysis (FEA) methods are frequently employed, but their analysis is often limited to simulating single welding points. Therefore, although existing technologies can effectively control assembly dimensional errors, they remain insufficient in the systematic management and multi-point coordinated analysis of welding stress.

[0004] Therefore, although existing technologies can improve assembly accuracy through digital means, they are difficult to effectively control the resulting welding stress, which affects the long-term performance of the bridge. Summary of the Invention

[0005] This invention provides a method and electronic device for predicting welding stress and optimizing the location of weld stress during bridge assembly, in order to solve the problem that while existing methods can improve assembly accuracy through digital means, it is difficult to effectively control the resulting welding stress, which affects the long-term performance of the bridge.

[0006] In a first aspect, embodiments of the present invention provide a method for predicting welding stress and optimizing welding points in bridge assembly, including: Obtain three-dimensional digital assembly data of steel components; Based on three-dimensional digital assembly data, welding characteristic parameters of steel components during assembly are extracted, and a finite element model of welding stress at multiple assembly points is established within a preset deviation range using a parametric command flow, thus constructing a welding assembly characteristic dataset. Based on the welding assembly feature dataset, a nonlinear regression prediction model is used to predict the welding residual stress at each assembly point; The residual welding stress at each assembly point is analyzed based on the interpretability analysis algorithm, the influencing factors of welding stress are identified, and the optimal assembly scheme is determined with the goal of minimizing the residual welding stress.

[0007] In one possible implementation, based on three-dimensional digital assembly data, welding characteristic parameters during the steel component assembly process are extracted, and a parametric command flow is used to establish a finite element model of welding stress at multiple assembly points within a preset deviation range, constructing a welding assembly characteristic dataset, including: Welding characteristic parameters are determined based on three-dimensional digital assembly data; The Cartesian product method is used to combine welding characteristic parameters to generate various assembly conditions; each assembly condition corresponds to a set of welding characteristic parameters. For any assembly condition, a finite element model of welding stress at each assembly point is established within a preset deviation range using a parameterized command flow. Based on the finite element model of welding stress at each assembly point, the welding stress at each assembly point is determined. By associating each set of welding characteristic parameters with the welding stress at each assembly point, a welding assembly characteristic dataset is formed.

[0008] In one possible implementation, the Cartesian product method is used to combine welding characteristic parameters to generate various assembly conditions, including: The Cartesian product method is used to combine welding feature parameters to form the initial welding feature; Interaction terms are added based on the degree of influence between each group of welding characteristic parameters to form interactive features; The initial welding features and interactive features are combined to form a variety of assembly conditions; interactive features refer to derived features constructed by combining two or more welding feature parameters.

[0009] In one possible implementation, a parameterized command flow is used to establish a finite element model of the welding stress at each assembly point under the assembly condition within a preset deviation range, including: The parametric command flow is used to draw the butt joint section at each assembly point; the first adjustable parameters of the butt joint section include the tensile length, weld width, and offset of the butt joint section; the offset of the butt joint section takes a value within a first preset deviation range; Parametric assembly is performed based on the docking sections of each assembly point to form a docking model for each assembly point; the second adjustable parameters of the docking model include the flange uplift and flange offset of the steel components, which are used to simulate the actual docking state of the docking model within a second preset deviation range. The material properties, weld area meshing, thermo-mechanical coupling finite element analysis steps, weld birth and death elements, thermal interaction conditions, loads and boundary conditions of the docking model are set respectively to form the welding stress finite element model of each assembly point under this assembly condition.

[0010] In one possible implementation, the nonlinear regression prediction model is the XGBoost model.

[0011] In one possible implementation, based on a welding assembly feature dataset, a nonlinear regression prediction model is used to predict the welding residual stress at each assembly point, including: Based on the welding assembly feature dataset, the XGBoost model is trained and its hyperparameters are optimized to obtain a welding residual stress prediction model. The welding feature parameters are used as input features, and the corresponding maximum welding residual stress value is used as the output label. The welding feature parameters include at least one of the following: the thickness of the steel component, the butt joint length deviation, the butt joint offset, the weld width, and the ambient temperature. The assembly characteristic parameters of the assembly point to be predicted are input into the welding residual stress prediction model, and the welding residual stress of the point under various assembly conditions is output.

[0012] In one possible implementation, the residual welding stress at each assembly point is analyzed based on an interpretable analysis algorithm to identify factors influencing welding stress, and the optimal assembly scheme is determined with the goal of minimizing residual welding stress, including: Calculate the contribution of each input feature to the welding residual stress output by the nonlinear regression prediction model; Based on the contribution degree, a global sensitivity analysis was conducted to identify and rank the characteristics affecting welding residual stress and their nonlinear influence laws. Based on the contribution degree, a local interpretation analysis is performed to quantify the positive or negative contribution of the characteristics affecting welding residual stress to the prediction of welding residual stress for specific assembly points or working conditions. Based on the characteristics of the residual welding stress, its nonlinear influence, and its positive or negative contribution, the welding characteristic parameters are adjusted to determine the optimal assembly scheme with the goal of minimizing the residual welding stress.

[0013] In one possible implementation, the optimal assembly scheme is determined with the goal of minimizing welding residual stress, including: determining the maximum residual stress, average residual stress, and variance of stress distribution based on the welding residual stress at each assembly point; The mixed residual stress index is determined based on the maximum residual stress, the average residual stress, and the variance of the stress distribution. The optimal assembly scheme is determined by minimizing the mixed residual stress index and ensuring that the change in the mixed residual stress index is less than a preset threshold.

[0014] In one possible implementation, acquiring the three-dimensional digital assembly data of the steel components includes: The actual geometric information of steel components is acquired using three-dimensional laser scanning technology; Based on the actual geometric information of the steel components, a digital assembly of the steel components is constructed; The assembly feature dimensions are extracted from the digitally assembled assembly to form three-dimensional digital assembly data of the steel components.

[0015] In a second aspect, embodiments of the present invention provide an electronic device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method for predicting welding stress and optimizing the location of bridge assembly as described in any of the first aspects.

[0016] This invention provides a method and electronic device for predicting welding stress and optimizing welding points in bridge assembly. By integrating parametric finite element simulation, nonlinear machine learning prediction, and interpretability analysis, it achieves accurate prediction and proactive optimization of welding stress. This method can systematically predict the distribution of residual welding stress under different assembly error conditions and clearly quantify the contribution of each assembly parameter to the corresponding stress using interpretability analysis algorithms, transforming the complex welding thermodynamic process into an understandable and operable decision-making basis. Based on this, with the goal of minimizing welding stress, it intelligently recommends the optimal assembly points and parameter combinations, thereby achieving stress control before construction, effectively avoiding welding stress concentration, and improving the long-term stability, fatigue durability, and overall construction accuracy of the bridge structure. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the implementation of a method for predicting welding stress and optimizing welding points in bridge assembly, as provided in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the implementation of step 102 provided in an embodiment of the present invention; Figure 3 This is a three-dimensional welding view of the I-beam provided in an embodiment of the present invention; Figure 4 It is a two-dimensional projection of the three-dimensional welding view of the I-beam provided in the embodiment of the present invention; Figure 5 This is a schematic diagram of automated modeling for upper flange welding provided in an embodiment of the present invention; Figure 6 This refers to the residual stress in the upper flange butt weld provided in this embodiment of the invention. Figure 7 This is a schematic diagram of automated modeling for web plate welding provided in an embodiment of the present invention; Figure 8 This refers to the residual stress in the web butt weld provided in the embodiments of the present invention. Figure 9 This is a schematic diagram of automated modeling for lower flange welding provided in an embodiment of the present invention; Figure 10 This refers to the residual stress in the lower flange butt weld provided in the embodiments of the present invention; Figure 11 This refers to the global contribution percentage of each welding stress influencing factor provided in the embodiments of the present invention; Figure 12 This is a basic error diagram of the component provided in the embodiment of the present invention; Figure 13 This is a schematic diagram of the beam micro-adjustment provided in an embodiment of the present invention; Figure 14 This is an enlarged schematic diagram of the beam micro-adjustment provided in an embodiment of the present invention; Figure 15 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention may be implemented in other embodiments without these specific details.

[0020] Considering the limitations of current digital methods in effectively controlling welding stress, as mentioned in the background section, further analysis reveals two main shortcomings: First, while 3D scanning-based digital assembly improves component dimensional accuracy, it fails to effectively address the uneven stress distribution caused by assembly deviations during welding. Second, welding stress concentration is particularly pronounced in complex structures with multi-point welding (such as multi-span steel beam bridges or steel-concrete composite beam bridges). Traditional finite element methods, failing to adequately consider the interactions and stress transfer effects between multiple welding points in complex bridge structures, lack the ability to comprehensively model and analyze multi-point welding stress, making it difficult to accurately assess and optimize the distribution of welding stress, thus affecting the long-term durability and structural safety of the bridge.

[0021] To this end, this invention combines finite element digital modeling, nonlinear regression prediction models, and interpretable analysis algorithms to provide a method that can accurately predict welding stress during the assembly process of steel beam bridges or steel-concrete composite beam bridges. By optimizing the assembly points, the method ensures that welding stress is minimized, thereby improving the stability, durability, and construction accuracy of the bridge.

[0022] By combining the prediction of welding stress with the optimization of assembly points, this invention can assess welding stress before assembly, thereby selecting the optimal assembly points and avoiding structural problems caused by stress concentration during welding.

[0023] To make the objectives, technical solutions, and advantages of the present invention clearer, specific embodiments will be described below in conjunction with the accompanying drawings.

[0024] The technical solution of this invention is particularly applicable to the assembly process of steel beam bridges and steel-concrete composite beam bridges, improving the long-term stability and safety of bridge structures through welding stress optimization. It can be applied to the welding and assembly process of steel structure buildings, reducing the negative impact of welding stress on the structure and improving the stability of the building.

[0025] The technical solution of the present invention can also be applied to the welding and assembly process of complex mechanical equipment to ensure the strength and stability of the equipment structure, and can also be used in the assembly process of other large steel structures, tunnels, transportation facilities, etc. to ensure that the stress of the structure is minimized during the welding process.

[0026] Figure 1 The implementation flowchart of the bridge assembly welding stress prediction and point optimization method provided in the embodiment of the present invention is described in detail below: Step 101: Obtain the three-dimensional digital assembly data of the steel components.

[0027] For example, acquiring three-dimensional digital assembly data of steel components includes: The actual geometric information of steel components is collected using three-dimensional laser scanning technology; based on the actual geometric information of the steel components, a digital assembly of the steel components is constructed; and the assembly feature dimensions are extracted from the digital assembly to form three-dimensional digital assembly data of the steel components.

[0028] The actual geometric information of the steel components includes the dimensional parameters of each component. The assembly feature dimensions include the thickness of the spliced ​​portions, the butt joint length, the width of the butt weld, and the butt offset. When digitally assembling the steel components based on their actual geometric information, a margin needs to be added to the theoretical butt joint gap according to the welding process specifications, considering the actual welding scenario, to reflect the space required for the deposited metal to fill the gap during actual welding. Simultaneously, a butt offset exists when comparing the actual shape of the steel components with the theoretical assembly position. The digitally assembled assembly forms three-dimensional digital assembly data containing complete geometric deviations and process parameters. Therefore, to provide accurate input for subsequent parametric finite element modeling, it is necessary to extract the assembly feature dimensions of the digitally assembled assembly.

[0029] Step 102: Based on the three-dimensional digital assembly data, extract the welding characteristic parameters in the steel component assembly process, and use the parametric command flow to establish a welding stress finite element model of multiple assembly points within a preset deviation range to construct a welding assembly characteristic dataset.

[0030] This step aims to provide high-quality, large-scale training data for subsequent machine learning predictions. A parametric command-flow modeling method is employed to achieve efficient batch simulations. Based on 3D digital assembly data, key welding characteristic parameters affecting residual welding stress are extracted, such as component thickness, butt joint deviation, weld geometry, and ambient temperature. Subsequently, using parametric command-flow technology, these parameters are input as variables. Within a preset assembly deviation range, finite element models of welding stress corresponding to different "assembly conditions" are automatically and in batches generated. This replaces the traditional manual modeling approach, achieving efficient systematic simulation that covers various possible error combinations. Finally, the input parameters and maximum output welding stress values ​​of all finite element models are collected. After cleaning and standardization, a structured welding assembly feature dataset suitable for machine learning training is constructed, ensuring that the prediction model has a solid physical foundation and sufficient data support.

[0031] Step 103: Based on the welding assembly feature dataset, use a nonlinear regression prediction model to predict the welding residual stress at each assembly point.

[0032] This step uses the dataset constructed in step 102 to train a nonlinear regression prediction model. The nonlinear regression prediction model can be other machine learning algorithms such as XGBoost or LightGBM. The generation of welding residual stress is a complex nonlinear physical process involving strong thermo-mechanical coupling, which is difficult to accurately describe using traditional linear models.

[0033] Step 104: Analyze the welding residual stress at each assembly point based on the interpretability analysis algorithm, identify the factors affecting welding stress, and determine the optimal assembly scheme with the goal of minimizing welding residual stress.

[0034] Interpretive analysis algorithms are applied to analyze the model's predictions, quantifying the contribution of each input feature—such as assembly offset in a certain direction and weld width—to the final predicted stress. This clearly identifies the key factors affecting welding residual stress. Based on this, with minimizing welding residual stress as the core optimization objective, and combining the patterns revealed by interpretive analysis, controllable assembly parameters are systematically adjusted, such as fine-tuning component positions and optimizing weld dimensions. The model predicts and evaluates stress levels under different parameter combinations, thereby intelligently selecting and determining the final assembly scheme with minimum stress and optimal performance.

[0035] This invention achieves accurate prediction and proactive optimization of welding stress by integrating parametric finite element simulation, nonlinear machine learning prediction, and interpretability analysis. This method can systematically predict the distribution of residual welding stress under different assembly error conditions and clearly quantify the contribution of each assembly parameter to the corresponding stress using interpretability analysis algorithms, transforming the complex welding thermodynamic process into an understandable and operable decision-making basis. Based on this, with the goal of minimizing welding stress, it intelligently recommends the optimal assembly points and parameter combinations, thereby achieving stress control before construction, effectively avoiding welding stress concentration, and improving the long-term stability, fatigue durability, and overall construction accuracy of the bridge structure.

[0036] In one embodiment, see Figure 2 This section describes how to construct a structured welding assembly feature dataset that can be used for machine learning training. Step 102 includes: Step 1021: Determine welding characteristic parameters based on three-dimensional digital assembly data.

[0037] Step 1022: The welding characteristic parameters are combined using the Cartesian product method to generate various assembly conditions.

[0038] Each assembly condition corresponds to a set of welding characteristic parameters.

[0039] Step 1023: For any assembly condition, establish the welding stress finite element model of each assembly point under the assembly condition using the parameterized command flow within the preset deviation range, and determine the welding stress of each assembly point based on the welding stress finite element model of each assembly point.

[0040] Step 1024: Associate the welding characteristic parameters of each group with the welding stress at each assembly point to form a welding assembly characteristic dataset.

[0041] To more clearly illustrate the above steps, let's take two H-beams and a connecting diaphragm as an example to explain the specific implementation process of steps 1021-1024. The steel component includes two H-beams and a connecting diaphragm; this digitally assembled assembly forms an I-beam, such as... Figure 3 The three-dimensional welding view of the I-beam shown includes the actual geometric information of the steel component, including the dimensional parameters of the diaphragms and web of the I-beam; the assembly feature dimensions include flange thickness, web thickness, flange butt joint length, flange butt joint weld width, flange butt joint offset, web butt joint length, web butt joint weld width, and web butt joint offset.

[0042] The key parameters are determined from the assembly feature dimensions to form a dataset of welding feature parameters, and in step 1022, various assembly conditions are generated by combining these parameters. In step 1021, key welding characteristic parameters are extracted from the assembly feature dimensions, and based on engineering experience and process specifications, relevant parameters and their discrete values ​​for welding simulation are set for each parameter. Specifically, for the thermo-mechanical coupling simulation process of flange docking, the selected welding characteristic parameters and their set discrete values ​​are as follows: Flange thickness: 15.0mm, 16.0mm, 17.0mm (3 values); Flange butt weld widths: 15.0mm, 20.0mm, 25.0mm, 30.0mm (4 values); Flange mating offset: 0.0mm, 1.5mm, 3.0mm (3 values); Flange butt joint length error: -2.0mm, 0.0mm, 2.0mm (3 values); Ambient temperature: 8.0°C (autumn), 23.0°C (summer) (2 values); Flange upturn: 0.5mm, 1.0mm (2 values).

[0043] In step 1022, the Cartesian product method is used to combine all the discrete values ​​of the above six parameters to generate a complete parameter mesh for flange assembly. A total of 3×3×4×3×2×2=432 different flange assembly conditions can be formed. Each condition corresponds to a set of determined parameter combinations for subsequent batch finite element simulations.

[0044] The simulation formula for the Cartesian product method is as follows:

[0045] in, This represents the predicted maximum welding stress at the flange butt joint. For flange docking offset, For flange butt joint length error, For the flange thickness, Where T is the width of the flange butt weld, and T is the ambient temperature. The wing edges are raised; The noise term is expressed in MPa. The simulation formula of the Cartesian product method is derived from the principle of thermo-mechanical coupling and has a correlation coefficient >0.95 with the finite element results. It is used for preliminary label calculations and will be subsequently verified through true finite element analysis.

[0046] Specifically, for the thermo-mechanical coupling simulation process of flange butt welding, the selected welding characteristic parameters and their set discrete values ​​are as follows: Web butt joint length error: -2.0mm, 0.0mm, 2.0mm (3 values); Web plate offset: 0.0mm, 0.5mm, 1.0mm (3 values); Weld width: 15.0mm, 20.0mm, 25.0mm, 30.0mm (4 values); Ambient temperature: 8.0°C (autumn), 23.0°C (summer) (2 values).

[0047] Using the Cartesian product method, all discrete values ​​of the above four parameters are combined to generate a web parameter mesh, which can form a total of 3×3×4×2=72 different web assembly conditions. Each condition corresponds to a set of determined parameter combinations for subsequent batch finite element simulations.

[0048] The simulation formula for the Cartesian product method is as follows:

[0049] in, The predicted maximum welding stress for the web butt joint; For web plate misalignment, This is for the error in the length of the web butt joint. For web thickness, Where is the width of the web weld, and T is the ambient temperature. The noise term is expressed in MPa. The simulation formula of the Cartesian product method is derived from the principle of thermo-mechanical coupling and has a correlation coefficient >0.95 with the finite element results. It is used for preliminary label calculations and will be subsequently verified through true finite element analysis.

[0050] For example, datasets formed from different flange assembly conditions and web assembly conditions can be preprocessed. Preprocessing may include outlier removal (IQR method), feature normalization (Z-score), and addition of interactive terms. Therefore, step 1022 may include: The Cartesian product method is used to combine welding feature parameters to form the initial welding feature; Interaction terms are added based on the degree of influence between each group of welding characteristic parameters to form interactive features; The initial welding features and interactive features are combined to form a variety of assembly conditions; interactive features refer to derived features constructed by combining two or more welding feature parameters.

[0051] The interaction items can be limited to one or more. For example, adding the product of offset and ambient temperature as the interaction item will result in a 1500-row × 7-column matrix (6 initial welding features + 1 derived feature) for the flange assembly condition and a 1500-row × 5-column matrix (4 initial welding features + 1 derived feature) for the web assembly condition. The training and test sets for the assembly conditions are divided in an 80:20 ratio.

[0052] In step 1023, since the parametric modeling of the overall H-beam is too redundant, the flange butt joint and web butt joint are modeled separately, and constraints are added to the model to simulate the stress situation during the welding process of the H-beam.

[0053] For example, a finite element model of welding stress at each assembly point under the assembly condition is established using a parameterized command flow within a preset deviation range, including: First, the docking cross-section of each assembly point is drawn using a parameterized command flow.

[0054] For example, draw a sketch of the butt joint section of the upper flange, web, and lower flange, such as... Figure 4 The two-dimensional projection of the three-dimensional welded view of the I-beam shown is used for contour point extraction in finite element modeling. The first adjustable parameters of the butt joint section include the tensile length, weld width, and offset of the butt joint section; the offset of the butt joint section takes values ​​within a first preset deviation range.

[0055] Secondly, parametric assembly is performed based on the docking sections of each assembly point to form a docking model for each assembly point.

[0056] The second adjustable parameters of the docking model include the flange uplift and flange offset of the steel components, which are used to simulate the actual docking state of the docking model within a second preset deviation range.

[0057] Finally, the material properties, weld area mesh generation, thermo-mechanical coupling finite element analysis steps, weld birth and death elements, thermal interaction conditions, loads and boundary conditions of the docking model are set respectively to form the welding stress finite element model of each assembly point under this assembly condition.

[0058] For example, material properties may include, but are not limited to, the thermal conductivity, density, elastic modulus, coefficient of thermal expansion, plasticity parameters, and specific heat capacity of Q235 steel.

[0059] For example, the meshing of the weld area includes meshing the structure, using a uniform mesh for the weld structure, and using a fixed number of eccentric meshes for the flange. The purpose is to ensure that the mesh in the weld area is dense while optimizing the consumption of computational resources.

[0060] For example, the thermo-mechanical coupled finite element analysis step is the thermo-mechanical coupled step for weld simulation. For instance, Step-0 is initialized with a small time step (1×10^{-10}s); the analysis step size is calculated from the weld element length; and the cooling step duration is 3000s. Adjustable parameters include the Step-0 time step, the analysis step size, and the cooling step duration.

[0061] For example, setting weld life and death units includes simulating the gradual filling and shaping of weld material as it moves with the heat source by controlling the state changes of "activation" and "death" of the control unit, thereby activating weld elements layer by layer.

[0062] For example, setting thermal interaction conditions can include convection conditions for Film Condition and radiation conditions for Radiation to Ambient. The load is a double ellipsoidal heat source model, and the boundary conditions constrain the docking flange and docking web.

[0063] Command streams enable efficient batch processing of models, rapid adjustment of welding deviations, and the generation of large-scale simulation results. This provides robust data support for welding stress prediction, ultimately forming finite element models of welding stress at each assembly point, such as... Figure 5-10 As shown, Figure 5 A schematic diagram of automated modeling for upper flange welding; Figure 6 This refers to the residual stress in the butt weld of the upper flange; Figure 7 A schematic diagram of automated web welding modeling; Figure 8 Residual stress in the butt weld of the web plate; Figure 9 This is a schematic diagram of automated modeling for lower flange welding. Figure 10 The residual stress in the butt weld of the lower flange is shown in Table 1. In step 1024, the flange welding assembly feature dataset obtained by the finite element model is shown in Table 1, and the web welding assembly feature dataset is shown in Table 2.

[0064] Table 1. Examples of statistical results based on the finite element simulation flange docking dataset.

[0065] Table 2. Examples of statistical results based on the finite element simulation flange docking dataset.

[0066] In this embodiment, by combining parameterized command flow with the Cartesian product method, efficient, systematic, and large-scale generation of welding stress simulation data is achieved. This method utilizes finite element simulation to provide high-quality data labels with clear physical meaning for machine learning, which not only significantly improves modeling efficiency and overcomes the limitations of traditional single-point simulation, but also ensures the comprehensiveness and representativeness of the dataset through full parameter combination coverage, thus laying a solid and reliable data foundation for subsequent high-precision and interpretable intelligent prediction and optimization of welding stress.

[0067] In one embodiment, the welding stress prediction process is described using the XGBoost model as an example of a nonlinear regression prediction model. Step 103 includes: Based on the welding assembly feature dataset, the XGBoost model was trained and its hyperparameters were optimized to obtain a welding residual stress prediction model.

[0068] The welding characteristic parameters are used as input features; the corresponding maximum welding residual stress value is used as the output label; the welding characteristic parameters include at least one of the following: the thickness of the steel component, the butt joint length deviation, the butt joint offset, the weld width, and the ambient temperature.

[0069] The XGBoost model learns the relationship between different assembly points and welding stresses, enabling it to accurately predict the welding stress at each point. As a gradient boosting decision tree algorithm, XGBoost is suitable for handling multi-feature regression problems, and is particularly well-suited for complex nonlinearities in welding thermo-mechanical coupling (such as offset-dominated stress peak amplification).

[0070] For example, flange welding feature parameters and web welding feature parameters are used as input features, and the corresponding maximum welding stress is used as the regression label. One interaction feature is introduced for each feature, resulting in a total dimension of 14. This ensures that the XGBoost model captures both independent and cross-effects of the flange / web (such as the joint contribution of flange warping and web offset, consisting of 10 initial welding features + 2 derived features + 2 welding stresses). For instance, the maximum welding stress of 387.00 MPa is corresponding to the baseline combination of flange welding feature parameters "flange thickness 16.00 mm, flange butt weld width 15.00 mm, flange butt offset 0.00 mm, ambient temperature 23.00°C, flange warping 0.50 mm", and the maximum welding stress of 399.00 MPa is corresponding to the baseline combination of web welding feature parameters "web butt offset 0.00 mm, weld width 30.00 mm, ambient temperature 23.00°C", as part of the training samples.

[0071] In the model building and optimization phase, an XGBoost regressor was used for training, with the squared error as the objective function. The initial learning rate was set to 0.05, the maximum tree depth to 4, the subsampling rate to 0.8, and the regularization parameters L1 and L2 to 0.8. Hyperparameters were further optimized using grid search combined with 5-fold cross-validation, with n_estimators tuned from 200 to 400. The model performance was evaluated using R² score and root mean square error (RMSE). After optimization, the model achieved excellent performance on the test set with an R² score of 0.97 and an RMSE of 11.2, indicating that its prediction error was less than 5% of the stress peak and it possessed high prediction accuracy.

[0072] The model was iteratively trained using an early stopping mechanism, with 10 early stopping rounds set, and the validation loss was continuously monitored until convergence during training. The trained model revealed the core principles of welding stress influence: flange butt joint offset and web butt joint offset contributed approximately 60% to residual stress, making them the dominant factors leading to stress amplification; flange warping and component thickness contributed approximately 25% combined, showing a negative correlation, meaning increasing thickness helps reduce stress; weld width and ambient temperature contributed approximately 15%, with increased weld width leading to a 15-25% increase in heat input, while low-temperature environments such as 8°C increased stress by approximately 30 MPa. For example, in a simulated autumn 8°C subset, under the combined condition of a 30 mm flange butt weld width and a 1.0 mm web butt joint offset, the peak welding stress increased by 28% compared to the baseline.

[0073] After obtaining the welding residual stress prediction model, the assembly characteristic parameters of the assembly point to be predicted are input into the welding residual stress prediction model, and the welding residual stress of the point under various assembly conditions is output.

[0074] For example, the model supports real-time prediction. For instance, inputs include: flange thickness 16.00 mm, flange butt weld width 15.00 mm, flange butt offset 0.00 mm, ambient temperature 23.00°C, flange warpage 0.50 mm, web butt offset 0.00 mm, weld width 30.00 mm, and ambient temperature 23.00°C; the output will be: predicted stress of 387.00 MPa for the flange and 399.00 MPa for the web. The confidence interval for the prediction is ±11.2 MPa, which can be further expanded and refined using quantile regression to assess the range of uncertainty in the prediction.

[0075] This embodiment utilizes the XGBoost model for welding stress prediction, achieving a significant efficiency leap from time-consuming simulation to millisecond-level real-time prediction. It ensures prediction accuracy with an R² score exceeding 0.97 and reveals the quantitative impact of each assembly parameter on the corresponding forces through interpretable output, thus integrating prediction, interpretation, and optimization. This method is easily integrated into actual construction systems, driving welding quality control from experience-based to data-driven approaches, and significantly improving the scientific rigor, reliability, and overall engineering efficiency of steel structure assembly.

[0076] In one embodiment, see Figure 11 This describes the detailed process of using the Interpretable Analysis (SHAP) algorithm to analyze the welding residual stress at each assembly point, identify the factors affecting welding stress, and determine the optimal assembly scheme with the goal of minimizing welding residual stress. Step 104 includes: First, the contribution of each input feature to the welding residual stress output by the nonlinear regression prediction model is calculated.

[0077] Secondly, a global sensitivity analysis was conducted based on the contribution level to identify and rank the characteristics affecting welding residual stress and their nonlinear influence laws.

[0078] Then, based on the contribution degree, a local interpretation analysis is performed to quantify the positive or negative contribution of the characteristics affecting welding residual stress to the prediction of welding residual stress for specific assembly points or working conditions.

[0079] Finally, based on the characteristics of the residual welding stress, its nonlinear influence, and its positive or negative contribution, the welding characteristic parameters are adjusted to determine the optimal assembly scheme with the goal of minimizing the residual welding stress.

[0080] For example, consider a typical multi-diaphragm connection scenario on the side of a steel girder bridge: with the central beam as a fixed spatial reference, H-shaped steel side beams are arranged on both sides, and each side is connected to the central beam through several diaphragms. The butt joint formed between each diaphragm and the beam body includes three welds: the upper flange, the lower flange, and the web, for a total of 16 butt joints × 3 welds = 48 welds.

[0081] The optimization process begins with a quantitative interpretation of the factors influencing welding residual stress. First, the SHAP algorithm is applied to calculate the contribution of each input feature (such as butt joint deviation, weld width, and ambient temperature) to the stress predicted by the XGBoost model, yielding SHAP values. Based on these SHAP values, a global sensitivity analysis is performed to identify and rank the key features affecting welding residual stress and their nonlinear effects. Simultaneously, through local interpretation analysis, the positive or negative contribution of each parameter to the force at specific assembly points is quantified.

[0082] Based on the interpretability analysis, an optimization model for assembly parameter adjustment and stress response is established. Figure 12 A schematic diagram of the basic errors of the component; Figure 13 This is a schematic diagram of the beam's micro-adjustment. Figure 14 This is an enlarged schematic diagram of the micro-motion adjustment of the beam. Inherent dimensional and orientation errors exist between each component, and every two components are constrained by their connection points. Errors between components propagate and accumulate through the kinematic chain. To control the overall error propagation and amplification of the bridge, a micro-motion adjustment mechanism needs to be introduced at key nodes. This mechanism actively and in a small range corrects the orientation of components to effectively suppress the chain propagation of errors. The overall configuration of the edge beam is described by a three-dimensional micro-motion vector, and the micro-motion adjustment of the edge beam is achieved through three-dimensional translation. Represented as:

[0083] in, ; , , These represent the micro-adjustments of the edge beam in the longitudinal, transverse, and vertical directions, respectively. , , These represent the maximum fine-tuning adjustments of the edge beam in the longitudinal, transverse, and vertical directions, respectively. This adjustment ensures that the edge beam's position meets dimensional requirements during assembly and does not exceed the maximum error range.

[0084] The posture of the diaphragm Defined by both the rotation matrix and the translation vector, it can be expressed as:

[0085] in The angle of rotation of the diaphragm. This represents the translation amount of the diaphragm.

[0086] Then, the width adjustment amount for each weld is determined based on the butt joint deviation. The butt joint deviation for each weld point is calculated based on the connection relationship between the diaphragm and the side beams and center beams. :

[0087] in, This refers to the actual position of the end point of the diaphragm. Ideal docking endpoint position; k Indicates the first k One welding point.

[0088] The relationship between weld width and butt joint deviation is as follows:

[0089] in, , , These are the butt joint deviations of the upper flange, lower flange, and web welds, respectively. This is the adjustment amount for the weld width of the upper flange; This is the adjustment amount for the weld width of the lower flange; The width of the weld seam in the web plate; This is the reference weld width for the upper flange; This is the reference weld width for the lower flange. Indicates the width of the reference weld in the web; These are the amplification factors for the butt joint deviations of the upper flange, lower flange, and web welds, respectively, indicating how the weld width increases with changes in the butt joint deviation.

[0090] The above parameters, such as docking deviation, weld width, and ambient temperature, are input into the trained XGBoost model. Predict the residual stress in each weld:

[0091] in, Input features for the predicted residual stress Factors include butt joint deviation, weld width, and ambient temperature. Simultaneously, the SHAP algorithm is used to quantify the contribution of each feature to stress prediction. :

[0092] The SHAP value can clearly identify factors affecting welding stress. m These are the main factors that provide an explainable basis for subsequent optimization.

[0093] The optimal assembly scheme is determined with the goal of minimizing welding residual stress, including: determining the maximum residual stress, average residual stress, and variance of stress distribution based on the welding residual stress at each assembly point; and determining the mixed residual stress index based on the maximum residual stress, average residual stress, and variance of stress distribution.

[0094] To achieve optimal assembly point selection, this invention introduces a simplified hybrid residual stress index (SHRSI) to simultaneously minimize the maximum residual stress, average residual stress, and variance of stress distribution. X :

[0095] in, The maximum residual stress in all welds; The average residual stress of all welds; The variance of the residual stress; , and These are the coefficients for the maximum residual stress, the average residual stress, and the variance of the residual stress, respectively.

[0096] The optimal assembly scheme is determined by minimizing the mixed residual stress index and ensuring that the change in the mixed residual stress index is less than a preset threshold.

[0097] For example, the optimization process includes two main iterations: The first round is the micro-motion adjustment of the side beams: under the premise of meeting the dimensional constraints, a set of candidate side beam micro-motion vectors are selected for adjustment. The second round is the attitude update of the diaphragm: after each side beam micro-motion, the attitude of the diaphragm is adjusted by constraining the positions of the feature points at both ends of the diaphragm, updating its position and orientation, and iteratively optimizing the welding stress.

[0098] In each iteration, convergence is determined by the following conditions:

[0099] When the change in the objective function is less than a preset threshold And the residual stress in all welds All are below the safety limit. If the optimization process has converged, the final output is the optimal side beam configuration, diaphragm orientation, and assembly parameters of each weld that minimize the overall welding residual stress, forming the optimal assembly scheme that can guide construction.

[0100] Based on the accurate prediction results of the XGBoost model, this embodiment proposes an innovative global welding control prediction method. Through interpretable quantitative analysis and iterative optimization, a closed loop from prediction to control of welding stress is achieved, which significantly improves the fatigue performance and long-term reliability of the assembled structure and realizes the systematic optimization of the overall welding residual stress on the bridge side.

[0101] This invention utilizes command-flow modeling to generate finite element models of multiple assembly points in batches and perform rapid calculations. This significantly improves modeling efficiency and reduces human error in traditional methods. By combining the XGBoost algorithm with command-flow modeling results, the welding stress at multiple assembly points can be accurately predicted, enabling systematic optimization. Based on the principle of minimizing welding stress, this invention selects optimal assembly points, avoiding structural instability caused by excessive stress during welding. By pre-optimizing assembly points, corrections and adjustments during construction are reduced, construction efficiency is improved, and the possibility of errors is lowered. This technology is not only applicable to the assembly process of steel beam bridges and steel-concrete composite beam bridges but can also be extended to other steel structures, mechanical equipment assembly, and other fields.

[0102] The widely used virtual assembly technology has not only been validated and supported in terms of assembly accuracy, but also proposed a scientific decision-making method for managing welding residual stress. Through the welding stress prediction and optimization scheme provided by this invention, virtual assembly technology can effectively control welding stress while improving accuracy, thereby making the design and construction of bridge structures more precise and reliable, and providing more rigorous decision support for future engineering projects.

[0103] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0104] See Figure 15 This invention also provides an electronic device 8, including a memory 81 and a processor 80. The memory 81 stores a computer program, and the processor 80 executes the computer program to implement the methods described in the above method embodiments. Exemplarily, the electronic device 8 can be an industrial computer, a server, an embedded control system, a mobile terminal, or a dedicated welding process analysis and optimization device, etc., and is not limited thereto.

[0105] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0106] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for predicting welding stress and optimizing welding points in bridge assembly, characterized in that, include: Obtain three-dimensional digital assembly data of steel components; Based on the three-dimensional digital assembly data, welding characteristic parameters of the steel component assembly process are extracted, and a finite element model of welding stress at multiple assembly points is established within a preset deviation range using a parameterized command flow to construct a welding assembly characteristic dataset. Based on the aforementioned welding assembly feature dataset, a nonlinear regression prediction model is used to predict the welding residual stress at each assembly point. The residual welding stress at each assembly point is analyzed based on the interpretability analysis algorithm, the influencing factors of welding stress are identified, and the optimal assembly scheme is determined with the goal of minimizing the residual welding stress.

2. The method for predicting welding stress and optimizing welding points in bridge assembly according to claim 1, characterized in that, Based on the three-dimensional digital assembly data, welding characteristic parameters during the steel component assembly process are extracted, and a parametric command flow is used to establish a finite element model of welding stress at multiple assembly points within a preset deviation range, constructing a welding assembly characteristic dataset, including: Based on the three-dimensional digital assembly data, the welding characteristic parameters are determined; The welding characteristic parameters are combined using the Cartesian product method to generate multiple assembly conditions; each assembly condition corresponds to a set of welding characteristic parameters. For any assembly condition, a finite element model of welding stress at each assembly point is established within a preset deviation range using a parameterized command flow. Based on the finite element model of welding stress at each assembly point, the welding stress at each assembly point is determined. By associating each set of welding characteristic parameters with the welding stress at each assembly point, a welding assembly characteristic dataset is formed.

3. The method for predicting welding stress and optimizing welding points in bridge assembly according to claim 2, characterized in that, The Cartesian product method is used to combine the welding characteristic parameters to generate various assembly conditions, including: The welding feature parameters are combined using the Cartesian product method to form the initial welding feature; Interaction terms are added based on the degree of influence between each group of welding characteristic parameters to form interactive features; The initial welding features and interactive features are combined to form the various assembly conditions; the interactive features refer to the derived features constructed by combining two or more of the welding feature parameters.

4. The method for predicting welding stress and optimizing welding points in bridge assembly according to claim 2, characterized in that, The step of establishing a finite element model of welding stress at each assembly point under the assembly condition using a parameterized command flow within a preset deviation range includes: The butt joint sections at each assembly point are drawn using a parameterized command flow; the first adjustable parameters of the butt joint section include the tensile length, weld width, and offset of the butt joint section; the offset of the butt joint section is taken within the first preset deviation range. Parametric assembly is performed based on the docking cross-sections of each assembly point to form a docking model for each assembly point; the second adjustable parameters of the docking model include the flange uplift and flange offset of the steel components, which are used to simulate the actual docking state of the docking model within a second preset deviation range. The material properties, weld area meshing, thermo-mechanical coupling finite element analysis steps, weld birth and death elements, thermal interaction conditions, loads and boundary conditions of the docking model are set respectively to form the welding stress finite element model of each assembly point under the assembly condition.

5. The method for predicting welding stress and optimizing welding points in bridge assembly according to claim 1, characterized in that, The nonlinear regression prediction model is the XGBoost model.

6. The method for predicting welding stress and optimizing welding points in bridge assembly according to claim 5, characterized in that, The step of predicting the residual welding stress at each assembly point using a nonlinear regression prediction model based on the welding assembly feature dataset includes: Based on the aforementioned welding assembly feature dataset, the XGBoost model is trained and its hyperparameters are optimized to obtain a welding residual stress prediction model. The welding feature parameters serve as input features, and the corresponding maximum welding residual stress value serves as the output label. The welding feature parameters include at least one of the following: steel component thickness, butt joint length deviation, butt joint offset, weld width, and ambient temperature. The assembly characteristic parameters of the assembly point to be predicted are input into the welding residual stress prediction model, and the welding residual stress of the point under various assembly conditions is output.

7. The method for predicting welding stress and optimizing welding points in bridge assembly according to claim 1, characterized in that, The method of analyzing the welding residual stress at each assembly point based on the interpretability analysis algorithm, identifying the factors affecting welding stress, and determining the optimal assembly scheme with the goal of minimizing welding residual stress includes: Calculate the contribution of each input feature to the welding residual stress output by the nonlinear regression prediction model; Based on the aforementioned contribution, a global sensitivity analysis is performed to identify and rank the characteristics affecting welding residual stress and their nonlinear influence laws. Based on the contribution degree, a local interpretation analysis is performed to quantify the positive or negative contribution of the features affecting welding residual stress to the prediction of welding residual stress for specific assembly points or working conditions. Based on the characteristics affecting welding residual stress, their nonlinear influence, and the positive or negative contributions, the welding characteristic parameters are adjusted to determine the optimal assembly scheme with the goal of minimizing welding residual stress.

8. The method for predicting welding stress and optimizing welding points in bridge assembly according to any one of claims 1-7, characterized in that, The process of determining the optimal assembly scheme with the goal of minimizing welding residual stress includes: The maximum residual stress, average residual stress, and variance of stress distribution are determined based on the welding residual stress at each assembly point. The mixed residual stress index is determined based on the maximum residual stress, the average residual stress, and the variance of the stress distribution. The optimal assembly scheme is determined by minimizing the mixed residual stress index and ensuring that the change in the mixed residual stress index is less than a preset threshold.

9. The method for predicting welding stress and optimizing welding points in bridge assembly according to claim 1, characterized in that, The acquisition of the three-dimensional digital assembly data of the steel components includes: The actual geometric information of steel components is acquired using three-dimensional laser scanning technology; Based on the actual geometric information of the steel components, a digital assembly of the steel components is constructed; The assembly feature dimensions are extracted from the digital assembly to form the three-dimensional digital assembly data of the steel component.

10. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method for predicting welding stress and optimizing the location of bridge assembly as described in any one of claims 1 to 9.

Images