Machine learning evaluation method and system for bridge response under mixed traffic of large transport vehicle

Abstract

The application discloses a large piece of transport vehicle mixed traffic under bridge response machine learning evaluation method and system, and belongs to the technical field of civil engineering structure health monitoring and machine learning. The method comprises the following steps: establishing a grillage finite element model of a typical span bridge and extracting a load influence surface; adopting a moving load loading method to calculate a dynamic response history when mixed formation passes through the bridge; constructing a comprehensive database covering various bridge types, multiple large piece of transport vehicles and various design loads; establishing multiple machine learning prediction models with vehicle load distribution parameters and bridge key parameters as inputs and load effect amplification coefficients as outputs; and screening an optimal model to predict bridge load effect results under mixed traffic conditions. The application combines finite element simulation and machine learning, constructs training samples through offline simulation, predicts online, rapidly outputs evaluation results, and provides an efficient and accurate technical tool for large piece of transport vehicle approval and bridge safety operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of civil engineering structural health monitoring and machine learning technology, specifically to a machine learning evaluation method and system for bridge response under mixed traffic conditions involving heavy transport vehicles. Background Technology

[0002] As the core carrier for transporting critical equipment in major infrastructure projects, the safety of heavy-duty transport vehicles crossing bridges is directly related to the operational safety of the road network. Currently, the traffic load on the road network is increasing day by day, and the contradiction between long-term traffic control and socio-economic efficiency is becoming more prominent. In actual transportation, there are many instances of open traffic conditions where heavy-duty vehicles and other vehicles share the road, resulting in a fundamental difference between the actual load conditions of bridges and the single-vehicle load conditions assumed by the specifications.

[0003] Existing bridge assessment methods suffer from irreconcilable technical contradictions. On the one hand, methods that simplify single-beam models and incorporate lateral distribution coefficients reduce spatial load problems to planar problems, failing to capture the spatial load coupling effect when heavy haulage vehicles and other vehicles travel together. This leads to an underestimation of the mixed-condition load effect by 15% to 30%, posing a serious safety hazard. On the other hand, while refined simulation methods such as beam grid finite element and solid finite element methods can accurately reflect load distribution, their modeling cycles are long (over 8 hours for a single bridge) and over 2 hours for a single mixed-condition simulation, failing to meet the daily demand for permit approvals for thousands of heavy haulage vehicles. The fundamental contradiction between "accuracy" and "efficiency" in traditional methods has become a technical bottleneck restricting the safety assessment of heavy haulage transportation.

[0004] Furthermore, current applications of machine learning in bridge assessment are mostly limited to single-vehicle scenarios, establishing predictive models only for a single bridge or vehicle type. They fail to cover comprehensive mixed scenarios involving multiple bridge types, vehicles, and load levels, and lack accurate predictive models for load effect amplification factors. Therefore, there is an urgent need to develop a bridge response prediction method that can capture the load coupling effects of mixed traffic spaces and achieve rapid assessment, providing an efficient and accurate technical tool for permitting heavy transport vehicles and ensuring safe bridge operation and maintenance. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a machine learning evaluation method and system for bridge response under mixed traffic conditions involving heavy transport vehicles. This method establishes a typical bridge beam finite element model and extracts the load influence surface, constructs a comprehensive database covering various bridge types, heavy transport vehicles, and standard design loads, and then establishes a machine learning prediction model with vehicle load distribution parameters and key bridge parameters as inputs and load effect amplification coefficient as output, thereby achieving rapid and accurate evaluation of bridge response under mixed traffic conditions.

[0006] This invention is achieved through the following technical solution: Firstly, this application provides a machine learning evaluation method for bridge response under mixed traffic conditions involving heavy transport vehicles, including: Step 1: Establish a finite element model of a typical span bridge and extract the load influence surface of the key control section; Step 2: Based on the load influence surface, calculate the dynamic response history of the mixed formation when passing through the bridge using the moving load loading method; Step 3: Based on the dynamic response history, construct a comprehensive database covering multiple bridge types, multiple heavy transport vehicles, and multiple design loads; Step 4: Based on the comprehensive database, establish multiple machine learning prediction models with vehicle load distribution parameters and key bridge parameters as input features and load effect amplification coefficient as output target. The load effect amplification coefficient is the ratio of the extreme value of bridge load effect under mixed traffic conditions to the extreme value of bridge load effect under individual traffic conditions. Step 5: Select the optimal model based on the machine learning prediction model, and use the optimal model to predict the bridge load effect results under mixed traffic conditions.

[0007] Preferably, the establishment of a typical span bridge finite element model and the extraction of the load influence surface of key control sections include: Beam elements are used to simulate the main beams and crossbeams, and the transverse connection method between the main beams is set according to the bridge type. Define key control sections according to bridge type; A dimensionless unit concentrated force is used to apply the unit load point by point with a lateral step length equal to the main beam spacing and a longitudinal step length smaller in the negative bending moment region than in the non-negative bending moment region of the continuous beam. The response values ​​of each main beam section are extracted to form a load influence surface data matrix.

[0008] Preferably, the calculation of the dynamic response history of the mixed formation passing over the bridge using the moving load method includes: The mixed formation of heavy transport vehicles and other vehicles is discretized into several axle groups, with each axle group corresponding to a concentrated load. Based on the load magnitude and wheel spacing of each axis group, determine the corresponding loading position of each axis group in the load influence surface data matrix; Based on the load values ​​of each axis group and the influence surface matrix at the loading position, the response values ​​of the key control sections at each moment are obtained, and the mixed formation moves longitudinally to form a dynamic response time history curve.

[0009] Preferably, the construction of the comprehensive database includes: For each type of bridge, the extreme values ​​of bridge load effects are calculated for the case of heavy transport vehicles traveling alone, and the extreme values ​​of bridge load effects are calculated for the case of heavy transport vehicles traveling together with other vehicles. The ratio of the extreme value of bridge load effect under the mixed traffic condition to the extreme value of bridge load effect under the single traffic condition is calculated and used as the load effect amplification factor. The load distribution parameters of heavy transport vehicles, key parameters of bridges, and corresponding load effect amplification factors are linked and stored to form a comprehensive database.

[0010] Preferably, in step 4, the vehicle load distribution parameters include axle load, number of axles, and wheelbase; the key bridge parameters include span, moment of inertia, and cross-sectional area.

[0011] Preferably, in step 4, the method for establishing multiple machine learning prediction models includes: establishing prediction models using decision tree-based ensemble learning algorithms, gradient-based boosting algorithms, or neural network-based deep learning algorithms, respectively. Cross-validation combined with Bayesian optimization was used to automatically search and optimize the hyperparameters of the model, resulting in multiple trained machine learning prediction models.

[0012] Preferably, in step 5, the method for selecting the optimal model is as follows: calculate the goodness of fit and prediction error of each prediction model, prioritize the model with the largest goodness of fit, and if the goodness of fit is the same, select the model with the smallest prediction error as the optimal model.

[0013] Preferably, in step 5, the method for predicting the bridge load effect under mixed traffic conditions using the optimal model is as follows: Input the load distribution parameters of the target heavy transport vehicle and the key parameters of the target bridge into the optimal model, and output the corresponding load effect amplification factor; Multiply the load effect amplification factor by the extreme value of the bridge load effect under the condition of heavy transport vehicles traveling alone to obtain the predicted value of the bridge load effect under mixed traffic conditions.

[0014] Secondly, this application provides a machine learning evaluation system for bridge response under mixed traffic conditions involving heavy transport vehicles, comprising: The finite element modeling module is used to build finite element models of beam grids for typical span bridges and extract the load influence surfaces of key control sections. The dynamic response calculation module, connected to the finite element modeling module, is used to calculate the dynamic response history of the mixed formation when passing through the bridge based on the load influence surface and using the moving load loading method. The database construction module, connected to the dynamic response calculation module, is used to construct a comprehensive database covering multiple bridge types, multiple heavy-duty transport vehicles, and multiple design loads based on the dynamic response history. The model training module, connected to the database construction module, is used to establish multiple machine learning prediction models based on the comprehensive database, with vehicle load distribution parameters and key bridge parameters as input features and load effect amplification coefficient as output target. The load effect amplification coefficient is the ratio of the extreme value of bridge load effect under mixed traffic conditions to the extreme value of bridge load effect under individual traffic conditions. The prediction output module, connected to the model training module, is used to select the optimal model based on the machine learning prediction model and use the optimal model to predict the bridge load effect results under mixed traffic conditions.

[0015] Thirdly, this application provides an electronic device, comprising: Memory, used to store computer programs; A processor, when executing the computer program, implements the steps of the machine learning evaluation method for bridge response under mixed traffic conditions involving heavy transport vehicles.

[0016] Compared with the prior art, the present invention has the following beneficial technical effects: This application provides a machine learning evaluation method for bridge response under mixed traffic conditions involving heavy transport vehicles. First, a finite element model of a typical span bridge is established, and the load influence surface of key control sections is extracted, transforming the complex spatial load response problem into an influence surface data matrix. Based on this, a moving load method is used to calculate the dynamic response history when heavy transport vehicles and other vehicles pass through the bridge in mixed formations. The principle of linear superposition is used to transform the simulation of numerous mixed conditions into efficient matrix operations. Furthermore, a comprehensive database covering various bridge types, multiple heavy transport vehicles, and various design loads is constructed. Using vehicle load distribution parameters and key bridge parameters as input features, and the load effect amplification factor as the output target, multiple machine learning prediction models are established. Finally, the optimal model is selected for rapid prediction of bridge load effects under mixed traffic conditions. The principle of this technical solution lies in combining finite element simulation with machine learning. High-precision simulation in the offline stage constructs training samples, allowing the online prediction stage to avoid time-consuming finite element calculations. This method accurately captures the spatial load coupling effect of mixed traffic through beam-grid model and influence surface method; it reduces the single assessment time from hours to seconds, meeting the time requirements for permit approval of heavy-duty transportation; the introduction of load effect amplification factor realizes the effective correlation between mixed working conditions and single-vehicle working conditions; the comparison and screening of multiple machine learning algorithms ensures prediction accuracy, and the goodness of fit of the test set can reach above 0.98.

[0017] This application also proposes a machine learning evaluation system for bridge response under mixed traffic conditions involving heavy-duty transport vehicles, an electronic device, and a computer storage medium, which possess all the advantages of the aforementioned machine learning evaluation method for bridge response under mixed traffic conditions involving heavy-duty transport vehicles. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the machine learning evaluation method for bridge response under mixed traffic conditions for heavy transport vehicles according to the present invention. Figure 2 This is a finite element model diagram of a small-span beam bridge in this invention; Figure 3 The graph shows the performance indicators of various machine learning models under different mixed loads and internal force responses of the present invention. Figure 4 This is a scatter plot showing the predicted distribution of each subset of data under the optimal model of this invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0022] Example 1 The machine learning evaluation method for bridge response under mixed traffic conditions provided in this embodiment establishes a finite element model of a typical bridge beam grid and extracts the load influence surface. It constructs a comprehensive database covering various bridge types, heavy transport vehicles, and standard design loads. Then, it establishes a machine learning prediction model with vehicle load distribution parameters and key bridge parameters as inputs and load effect amplification coefficient as output. This enables rapid and accurate evaluation of bridge response under mixed traffic conditions. The specific steps include: Step 1: Select a representative prefabricated concrete beam bridge as the analysis object, and obtain its geometric dimensions, material properties, and cross-sectional parameters as input data. Use finite element analysis software to establish a beam grid model of the bridge structure, simulating the main beams and crossbeams with beam elements. Based on the characteristics of different bridge types, set appropriate connection methods to obtain a high-fidelity beam grid finite element model that can be used for load influence surface extraction.

[0023] Specifically, it includes the following processes: S1.1 Four typical types of prefabricated concrete beam bridges are selected as the analysis objects: reinforced concrete simply supported hollow slab bridges, prestressed concrete simply supported hollow slab bridges, prestressed concrete simply supported T-beam bridges, and prestressed concrete continuous small box girder bridges.

[0024] S1.2. A beam grid model of the bridge structure was established using the finite element analysis software ANSYS, and beam elements were used to simulate the main beams and crossbeams. Different connection methods were set according to the characteristics of the bridge type: for hollow slab bridges, the hinge joints between the main beams were simulated by coupling the translational displacement of the nodes and releasing the rotational displacement; for simply supported T-beam bridges and continuous small box girder bridges, the cast-in-place wet joints between the main beams were treated as the flange plates of the main beams.

[0025] S1.3. The concrete pavement layer is included in the top slab of the section, while the influence of the asphalt pavement layer is ignored, in order to simplify the model while ensuring calculation accuracy. A typical bridge finite element model is as follows: Figure 2 As shown.

[0026] A beam-grid model was adopted to discretize the complex spatial bridge structure into a system of crisscrossing beam elements, significantly reducing the model's degrees of freedom while maintaining computational accuracy. Compared to solid finite element models, the beam-grid model can significantly improve computational efficiency while capturing the spatial stress characteristics of the bridge. A high-fidelity bridge finite element model was obtained, which can be used for subsequent influence surface extraction, laying the foundation for accurately capturing the spatial load coupling effects under mixed traffic conditions.

[0027] In some embodiments, the finite element analysis software is not limited to ANSYS, but may also include other finite element software such as ABAQUS, MIDASCivil, or SAP2000.

[0028] In some embodiments, the beam element is not limited to BEAM4, but may also be BEAM188, BEAM189 or other beam elements that take into account shear deformation, in order to improve the accuracy of the model.

[0029] In some embodiments, for special bridge structures, a solid finite element model can be established for local verification to calibrate the calculation accuracy of the beam grid model.

[0030] Table 1 Span and Section Parameters of Representative Prefabricated Beam Bridges

[0031] Step 2: Based on the finite element model of the beam grid, determine the location of the key control sections of the target bridge. Using dimensionless unit concentrated force, apply unit loads point-by-point across the entire bridge deck according to the set longitudinal and transverse step sizes, extracting the section response values ​​of all main beams under each working condition to form load influence surface data. Using the extracted influence surface data, calculate the dynamic response history of the key sections when a mixed convoy of heavy transport vehicles and ordinary vehicles passes over the bridge through matrix operations.

[0032] Specifically, it includes the following processes: S2.1 Define key control sections according to bridge type: For simply supported beams, key control sections include the mid-span positive bending moment section and the support maximum shear force section; For continuous beams, the key control sections include the positive bending moment section of the secondary span and the negative bending moment section of the support.

[0033] S2.2 Set load movement parameters. For example, use dimensionless unit concentrated force (P=1), set the longitudinal step size to 0.25 meters (to 0.1 meters in the negative bending moment zone of continuous beams), and the lateral step size to be equal to the spacing between the main beams. The load movement range covers the entire bridge deck.

[0034] S2.3 Apply unit load point by point, and after static analysis of each working condition, extract the section response values ​​of all main beams in batches to form a load influence surface data matrix covering the entire bridge deck.

[0035] S2.4 In the matrix laboratory environment, using the extracted influence surface data matrix, the dynamic response history of the key control section is calculated through matrix operations when a mixed formation of heavy transport vehicles and ordinary vehicles passes over a bridge.

[0036] Based on the principle of linear superposition, the bridge response under moving loads is represented as a convolution operation between the load and the influence surface. By pre-extracting influence surface data, the complex moving load analysis is transformed into efficient matrix operations, avoiding the inefficient process of re-performing finite element analysis every time vehicle parameters are changed. This enables efficient batch simulation of numerous mixed traffic conditions, reducing single calculation time from hours to seconds, and providing a feasible technical path for building large-scale comprehensive databases.

[0037] In some embodiments, the longitudinal step size can be adjusted according to the bridge span and accuracy requirements. For long-span bridges, a step size of 0.5 meters can be used, and for fine analysis areas, a step size of 0.1 meters can be used.

[0038] In some embodiments, the matrix laboratory environment is not limited to MATLAB, but may also employ Python (NumPy / SciPy), Julia, or other scientific computing platforms.

[0039] In some embodiments, the calculation of the dynamic response history can be performed using a parallel computing method, utilizing multi-core processors or graphics processors to accelerate matrix operations.

[0040] Step 3: Collect key parameters from multiple typical small-to-medium span bridges, actual load distribution parameters from multiple heavy transport vehicles, and various standard design load levels as data sources. Classify and organize the data according to internal force response types (positive bending moment, negative bending moment, and support shear force) to obtain multiple subset datasets covering various bridge types, vehicles, and load levels, forming a comprehensive database.

[0041] Specifically, it includes the following processes: S3.1 Collect key parameters of 17 typical small and medium span prefabricated beam bridges, including span, total length, longitudinal moment of inertia of the main beam, lateral moment of inertia, torsional moment of inertia and cross-sectional area.

[0042] The number of bridges mentioned is not limited to 17; it is only an example and can be expanded to a greater number according to actual engineering needs.

[0043] S3.2 Collect real load distribution parameters of more than 2,000 heavy transport vehicles, including tractor axle load, trailer axle load, number of trailer axles (load limit of 18 tons), five wheelbases and total wheelbase.

[0044] Similarly, the number of heavy-duty transport vehicles is not limited to more than 2,000 and can be continuously expanded as data accumulates.

[0045] S3.3 Select three standard design load levels as reference benchmarks: Automobile-Super 20, Highway 04-Class I, and Highway 15-Class I.

[0046] In some embodiments, the standard design load may include load levels from the latest specifications or local standard loads to account for regional differences.

[0047] S3.4. Based on the three types of internal force response—positive bending moment, negative bending moment, and support shear force—and combined with three standard design load levels, establish nine subset datasets (positive bending moment-04, positive bending moment-15, positive bending moment-20, negative bending moment-04, negative bending moment-15, negative bending moment-20, support shear force-04, support shear force-15, and support shear force-20).

[0048] By systematically collecting and organizing multi-source data on bridges, vehicles, and loads, a comprehensive database covering a wide range of working conditions is constructed, providing ample sample data for subsequent machine learning model training. This database encompasses various combinations of different bridge types, spans, vehicle types, and load levels, exhibiting good representativeness and generalization ability. Consequently, it lays a data foundation for establishing high-precision, highly generalizable predictive models.

[0049] Step 4: Using the vehicle load distribution parameters and key bridge parameters from the constructed integrated database as input features, and the load effect amplification factor as the output target, the data undergoes standardized preprocessing and is divided into training and test sets according to a fixed ratio. Finally, various machine learning algorithms are used to build prediction models, and parameters are fine-tuned through cross-validation and Bayesian optimization.

[0050] S4.1. The nine-dimensional load distribution parameters of heavy transport vehicles (including tractor axle load, trailer axle load, number of trailer axles, five wheelbases and the sum of wheelbases) and the six-dimensional bridge key parameters (including span, total length, longitudinal moment of inertia of the main beam, lateral moment of inertia, torsional moment of inertia and cross-sectional area) are used as input parameters of the prediction model; the load effect amplification factor is used as the output parameter of the prediction model.

[0051] The load effect amplification factor is defined as the ratio of the extreme value of the bridge load effect under mixed traffic conditions to the extreme value of the bridge load effect under individual traffic conditions.

[0052] S4.1. Adopt a consistent partitioning scheme, divide the samples into training set and test set in a 7:3 ratio using a fixed random seed, and standardize all input features so that their mean is 0 and their standard deviation is 1.

[0053] S4.1 Select the following seven machine learning algorithms to build prediction models respectively: Random Forest, Gradient Boosting Algorithm, Adaptive Boosting, Multilayer Perceptron, K-Nearest Neighbors Algorithm, Extreme Gradient Boosting, and Lightweight Gradient Boosting Machine.

[0054] S4.1. The hyperparameters of each model are automatically searched and optimized using a five-fold cross-validation combined with the Optuna Bayesian optimization framework (100 iterations).

[0055] S4.1. Based on three evaluation metrics—goodness of fit, root mean square error, and mean absolute error—the model performance is evaluated from multiple dimensions. The performance metrics of each machine learning model under different mixed loads and internal force responses are as follows: Figure 3 As shown.

[0056] The bridge response assessment problem is transformed into a regression prediction problem. Machine learning algorithms are used to learn the nonlinear mapping relationship between input features (vehicle parameters, bridge parameters) and the output target (load effect amplification factor). Compared to traditional empirical formulas or mechanical derivations, machine learning methods can automatically discover the complex interactions between high-dimensional features without requiring pre-assumed functional forms. A high-precision prediction model is established, capable of quickly outputting the load effect amplification factor under mixed traffic conditions, providing core algorithmic support for subsequent rapid assessment.

[0057] In some embodiments, the machine learning algorithm is not limited to the seven types mentioned above, and may also employ support vector regression, decision tree, extreme learning machine, or deep learning algorithm (such as deep neural network, convolutional neural network).

[0058] In some embodiments, the data division ratio is not limited to 7:3, but can be adjusted to 8:2 or 6:4 depending on the total sample size.

[0059] In some embodiments, the optimization framework is not limited to Optuna, and may also employ Hyperopt, Scikit-Optimize, grid search, random search, or other methods.

[0060] Step 5: Based on the model performance evaluation results, select the optimal prediction model for each subset of data from various machine learning models according to preset screening rules. Next, use the selected optimal model to predict the load effects on each subset of data. Finally, output the prediction results for rapid assessment of bridge response under mixed traffic conditions involving heavy transport vehicles.

[0061] S5.1 Set the screening rules as follows: prioritize the model with the highest goodness of fit; when the goodness of fit is the same, select the model with the smallest root mean square error; if the root mean square error is still the same, select the model with the smallest mean absolute error.

[0062] S5.1. Based on the selection rules, determine the optimal model for each of the nine subsets: the optimal model for seven datasets (positive bending moment -04, positive bending moment -15, positive bending moment -20, negative bending moment -04, negative bending moment -15, negative bending moment -20, and fulcrum shear force -15) is the extreme gradient boosting model; the optimal model for two datasets (fulcrum shear force -04 and fulcrum shear force -20) is the gradient boosting algorithm model.

[0063] S5.1 Predicting Load Effects. Using the selected optimal model, predict the load effect amplification factor for the test set of each subset of data, and output the predicted value.

[0064] S5.1 Compare the predicted values ​​with the actual values ​​to verify the prediction accuracy of the optimal model.

[0065] Based on the selection criteria, the optimal models and performance indices under different mixed loads and internal force responses are shown in Table 2. It can be seen that the optimal model for the SSF-04 and SSF-20 datasets is GradientBoost, while the optimal model for the other seven datasets is XGBoost. Furthermore, the prediction R² for all datasets exceeds 0.98, and the absolute error is controlled within 0.02. Using the obtained optimal models, predictions were made on nine datasets. The scatter distribution of prediction points for each subset of datasets under the optimal model is shown below. Figure 4 As shown.

[0066] Table 2. Optimal Model and Performance Indicators under Different Mixed Loads and Internal Force Responses Table 2. Optimal Model and Performance Indicators under Different Mixed Loads and Internal Force Responses

[0067] Specific application examples are as follows: This embodiment takes the mixed working condition assessment of an 8m reinforced concrete hollow slab bridge (RCS-8m) as an example and applies the method described in this invention.

[0068] An ANSYS beam grid model was established using the parameters in Table 1. The section parameters for beam4 elements were: beam height 0.42m, area 0.40m². 2 I x =0.024m 4 I y =0.011m 4 I z =0.030m 4 ; The key control sections are the mid-span positive bending moment section and the support shear section. The unit load is P=1, the longitudinal step is 0.25m, the transverse step is 1.6m, and the influence surface data matrix is ​​extracted. Select a 199t heavy transport vehicle (axle load distribution is 10+10+10+5×17.8t, wheelbase is 3.5-1.5-8-4×1.25m), the standard load is 15 highway level 1, the design lane number is 3, and the internal force response is "positive bending moment", input 15-dimensional characteristic parameters; By calling the XGBoost optimal model, the load effect amplification factor is predicted. ξ= 1.418; The positive bending moment at mid-span of a single heavy transport vehicle is 150kN. The mid-span positive bending moment under mixed working conditions is 150 × 1.418 = 212.7 kN. m, compared with the actual beam grid finite element simulation result (213kN) The error (m) is only 0.14%.

[0069] For different types of internal force responses and different load levels, the optimal machine learning model was selected separately to achieve site-specific model selection. Since the mapping relationship between input features and output targets varies under different working conditions, using a single model uniformly is unlikely to achieve optimal results on all subsets of datasets. By selecting the optimal model from different datasets, prediction accuracy can be maximized. High-precision prediction models for various working conditions were obtained, enabling rapid and accurate assessment of bridge responses under mixed traffic conditions, with single prediction times reaching the second level, meeting the timeliness requirements for permit approval of heavy-duty transport.

[0070] In some embodiments, the optimal model selection rule can be adjusted according to actual application needs, for example, in scenarios with high security requirements, a conservatively estimated model can be selected first.

[0071] In some embodiments, the prediction model may employ an ensemble learning strategy to weighted average the prediction results of multiple models to obtain a more robust prediction performance.

[0072] In some embodiments, the optimal model can be deployed on a cloud server or edge computing device to enable real-time and rapid assessment of heavy transport vehicles before they cross the bridge.

[0073] This invention uses a high-fidelity beam grid finite element model and a moving load analysis method based on load influence surfaces to accurately capture the spatial load coupling effect when heavy transport vehicles and other vehicles are mixed, while significantly reducing computational costs and enabling efficient batch simulation of a large number of working conditions. By constructing a comprehensive database covering typical small and medium span bridges, heavy transport vehicles, and standard design load combinations, and based on multi-dimensional input features and various machine learning algorithms, the optimal model trained has high prediction accuracy (R²>0.98) and can quickly and accurately output the load effect amplification factor.

[0074] Example 2 This application provides a machine learning evaluation system for bridge response under mixed traffic conditions involving heavy transport vehicles, including: The finite element modeling module is used to build finite element models of beam grids for typical span bridges and extract the load influence surfaces of key control sections. The dynamic response calculation module, connected to the finite element modeling module, is used to calculate the dynamic response history of the mixed formation when passing through the bridge based on the load influence surface and using the moving load loading method. The database construction module, connected to the dynamic response calculation module, is used to construct a comprehensive database covering multiple bridge types, multiple heavy-duty transport vehicles, and multiple design loads based on the dynamic response history. The model training module, connected to the database construction module, is used to establish multiple machine learning prediction models based on the comprehensive database, with vehicle load distribution parameters and key bridge parameters as input features and load effect amplification coefficient as output target. The load effect amplification coefficient is the ratio of the extreme value of bridge load effect under mixed traffic conditions to the extreme value of bridge load effect under individual traffic conditions. The prediction output module, connected to the model training module, is used to select the optimal model based on the machine learning prediction model and use the optimal model to predict the bridge load effect results under mixed traffic conditions.

[0075] It should be noted that, in the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another device, or some features may be ignored or not executed. The modules described as separate components may or may not be physically separated. The components shown as modules may be one or more physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs.

[0076] Furthermore, in the various embodiments of the present invention, the modules can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The integrated unit described above can be implemented in hardware or as a software functional unit.

[0077] An electronic device provided in this application includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the machine learning evaluation method for bridge response under mixed traffic of heavy transport vehicles as described in any of the above embodiments.

[0078] Another electronic device provided in this application embodiment may further include: an input port connected to a processor for transmitting multimodal data collected by an external acquisition device to the processor; a display unit connected to the processor for displaying the processor's processing results to the outside world; and a communication module connected to the processor for enabling communication between the electronic device and the outside world. The display unit may be a display panel, a laser scanning display, etc.; the communication method adopted by the communication module includes, but is not limited to, Mobile High Definition Link (HML), Universal Serial Bus (USB), High Definition Multimedia Interface (HDMI), and wireless connection (including Wi-Fi, Bluetooth, Bluetooth Low Energy, and IEEE 802.11s-based communication technology).

[0079] This application provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the steps of the machine learning evaluation method for bridge response under mixed traffic conditions involving heavy transport vehicles as described in any of the above embodiments.

[0080] For descriptions of relevant parts of the bridge response machine learning evaluation system, electronic device, and computer-readable storage medium under mixed traffic conditions for heavy transport vehicles provided in this application's embodiments, please refer to the detailed descriptions of the corresponding parts in the bridge response machine learning evaluation method under mixed traffic conditions for heavy transport vehicles provided in this application's embodiments; they will not be repeated here. Furthermore, parts of the technical solutions provided in this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0081] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A machine learning evaluation method for bridge response under mixed traffic conditions involving heavy transport vehicles, characterized in that, include: Step 1: Establish a finite element model of a typical span bridge and extract the load influence surface of the key control section; Step 2: Based on the load influence surface, calculate the dynamic response history of the mixed formation when passing through the bridge using the moving load loading method; Step 3: Based on the dynamic response history, construct a comprehensive database covering multiple bridge types, multiple heavy transport vehicles, and multiple design loads; Step 4: Based on the comprehensive database, establish multiple machine learning prediction models with vehicle load distribution parameters and key bridge parameters as input features and load effect amplification coefficient as output target. The load effect amplification coefficient is the ratio of the extreme value of bridge load effect under mixed traffic conditions to the extreme value of bridge load effect under individual traffic conditions. Step 5: Select the optimal model based on the machine learning prediction model, and use the optimal model to predict the bridge load effect results under mixed traffic conditions.

2. The machine learning evaluation method for bridge response under mixed traffic conditions involving heavy transport vehicles as described in claim 1, characterized in that, The establishment of a typical span bridge finite element model and the extraction of the load influence surface of key control sections include: Beam elements are used to simulate the main beams and crossbeams, and the transverse connection method between the main beams is set according to the bridge type. Define key control sections according to bridge type; A dimensionless unit concentrated force is used to apply the unit load point by point with a lateral step length equal to the main beam spacing and a longitudinal step length smaller in the negative bending moment region than in the non-negative bending moment region of the continuous beam. The response values ​​of each main beam section are extracted to form a load influence surface data matrix.

3. The machine learning evaluation method for bridge response under mixed traffic conditions involving heavy transport vehicles as described in claim 1, characterized in that, The calculation of the dynamic response history of the mixed formation passing over the bridge using the moving load method includes: The mixed formation of heavy transport vehicles and other vehicles is discretized into several axle groups, with each axle group corresponding to a concentrated load. Based on the load magnitude and wheel spacing of each axis group, determine the corresponding loading position of each axis group in the load influence surface data matrix; Based on the load values ​​of each axis group and the influence surface matrix at the loading position, the response values ​​of the key control sections at each moment are obtained, and the mixed formation moves longitudinally to form a dynamic response time history curve.

4. The machine learning evaluation method for bridge response under mixed traffic conditions involving heavy transport vehicles as described in claim 1, characterized in that, The construction of the comprehensive database includes: For each type of bridge, the extreme values ​​of bridge load effects are calculated for the case of heavy transport vehicles traveling alone, and the extreme values ​​of bridge load effects are calculated for the case of heavy transport vehicles traveling together with other vehicles. The ratio of the extreme value of bridge load effect under the mixed traffic condition to the extreme value of bridge load effect under the single traffic condition is calculated and used as the load effect amplification factor. The load distribution parameters of heavy transport vehicles, key parameters of bridges, and corresponding load effect amplification factors are linked and stored to form a comprehensive database.

5. The machine learning evaluation method for bridge response under mixed traffic conditions involving heavy transport vehicles as described in claim 4, characterized in that, In step 4, the vehicle load distribution parameters include axle load, number of axles, and wheelbase; the key bridge parameters include span, moment of inertia, and cross-sectional area.

6. The machine learning evaluation method for bridge response under mixed traffic conditions involving heavy transport vehicles as described in claim 1, characterized in that, In step 4, the methods for establishing multiple machine learning prediction models include: using decision tree-based ensemble learning algorithms, gradient-based boosting algorithms, or neural network-based deep learning algorithms to establish prediction models respectively. Cross-validation combined with Bayesian optimization was used to automatically search and optimize the hyperparameters of the model, resulting in multiple trained machine learning prediction models.

7. The machine learning evaluation method for bridge response under mixed traffic conditions involving heavy transport vehicles as described in claim 6, characterized in that, In step 5, the method for selecting the optimal model is as follows: calculate the goodness of fit and prediction error of each prediction model, prioritize the model with the largest goodness of fit, and if the goodness of fit is the same, select the model with the smallest prediction error as the optimal model.

8. The machine learning evaluation method for bridge response under mixed traffic conditions involving heavy transport vehicles as described in claim 1, characterized in that, In step 5, the method for predicting the bridge load effect under mixed traffic conditions using the optimal model is as follows: Input the load distribution parameters of the target heavy transport vehicle and the key parameters of the target bridge into the optimal model, and output the corresponding load effect amplification factor; Multiply the load effect amplification factor by the extreme value of the bridge load effect under the condition of heavy transport vehicles traveling alone to obtain the predicted value of the bridge load effect under mixed traffic conditions.

9. A machine learning evaluation system for bridge response under mixed traffic conditions involving heavy transport vehicles, characterized in that, include: The finite element modeling module is used to build finite element models of beam grids for typical span bridges and extract the load influence surfaces of key control sections. The dynamic response calculation module, connected to the finite element modeling module, is used to calculate the dynamic response history of the mixed formation when passing through the bridge based on the load influence surface and using the moving load loading method. The database construction module, connected to the dynamic response calculation module, is used to construct a comprehensive database covering multiple bridge types, multiple heavy-duty transport vehicles, and multiple design loads based on the dynamic response history. The model training module, connected to the database construction module, is used to establish multiple machine learning prediction models based on the comprehensive database, with vehicle load distribution parameters and key bridge parameters as input features and load effect amplification coefficient as output target. The load effect amplification coefficient is the ratio of the extreme value of bridge load effect under mixed traffic conditions to the extreme value of bridge load effect under individual traffic conditions. The prediction output module, connected to the model training module, is used to select the optimal model based on the machine learning prediction model and use the optimal model to predict the bridge load effect results under mixed traffic conditions.

10. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, when executing the computer program, implements the steps of the machine learning evaluation method for bridge response under mixed traffic conditions for heavy transport vehicles as described in any one of claims 1 to 8.

Images