Intensive single-column pier overload overturning risk assessment method based on convolutional neural network

Abstract

The application provides an intensive single-column pier overload overturning risk assessment method based on a convolutional neural network, belongs to the technical field of bridge engineering safety assessment, integrates vehicle load parameters, temperature effects, concrete creep and vehicle-bridge interactive dynamic amplification factors in a multi-dimensional tensor input, uses stage-by-stage training and gradient descent optimization of the convolutional neural network, realizes second-level reaction force prediction and dynamic amplification adjustment value output, expands the reaction force data through a synthetic overturning axis method, controls the error within a reasonable range, after integration with an existing bridge health monitoring system, can collect sensor data in real time, outputs a four-level risk grade, which is safety, criticality, danger and immediate intervention respectively, thereby reducing bridge maintenance costs, avoiding resource waste, improving structural stability in a high traffic volume environment, and being suitable for various bridge types such as straight bridges and curved bridges, realizing intensive assessment of single-column pier bridge overload overturning risk, and greatly improving the real-time performance and accuracy of the assessment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bridge engineering safety assessment, in particular to an intensive single-column pier overload overturning risk assessment method based on a convolutional neural network. BACKGROUND

[0002] Single-column pier bridges are one of the most common beam bridges in urban overpasses and ramps due to their small land occupation, convenient design, low construction cost, and easy adaptation to the surrounding environment. The increasing traffic volume and increasing heavy vehicles increase the overturning risk of single-column pier bridges, making it more likely to occur overall overturning accidents and causing very adverse social impact. Bridge health monitoring can effectively send early warning signals when the bridge is in special weather, traffic conditions, or when the bridge operation condition is abnormally severe, providing basis and guidance for bridge maintenance, repair and management decision-making. The current bridge health monitoring system can already provide daily diagnosis and post-accident cause analysis for bridges, but it cannot provide early and rapid warning to avoid accidents. Finite element software for data analysis, such as Midas, Abaqus or Ansys, can accurately calculate the stress and strain of the bridge structure, but the analysis time is long, and it cannot be quickly evaluated before an accident occurs, and the timeliness is not strong. At present, artificial intelligence has entered a new era of development, and neural networks can effectively solve the problem of the timeliness of the current bridge health monitoring system and slow calculation speed based on their advantages of cross-physical quantities, high accuracy and fast calculation speed.

[0003] The prior art has the following technical problems:

[0004] 1. The existing single-column pier bridge health monitoring relies on finite element software, which takes several hours to analyze, making it difficult to meet the real-time warning needs of overload scenarios, and is prone to overturning accidents. The traditional method only considers static loads and ignores the influence of other factors such as ring and vehicle interaction, resulting in large prediction errors and insufficient accuracy, especially in complex environments, lack of models integrating multi-source data, and inability to achieve second-level response, limiting the safety management of high-traffic bridges, and not conducive to actual use.

[0005] 2. The existing technology is limited to static analysis and time diagnosis, ignoring the influence of other factors such as temperature and concrete creep, resulting in inaccurate risk classification and ignoring potential failure paths, making maintenance passive, especially in complex structures such as curved bridges, the overturning process evaluation is not refined enough, the risk level is single and lacks effective reinforcement suggestions, resulting in insufficient risk of bearing disengagement, serious waste of resources, and difficulty in adapting to the high load demand of urban bridges. SUMMARY

[0006] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide a method for evaluating overload overturning risk of intensive single-column pier based on convolutional neural network, which comprises the following steps:

[0007] A full-bridge finite element model is established according to bridge geometric parameters, vehicle load parameters, temperature effects, concrete creep and vehicle-bridge interaction are simulated, support reaction data are calculated and normalized by layering, and key reaction parameters are extracted;

[0008] A convolutional neural network is designed based on the full-bridge finite element model, a multi-dimensional tensor is taken as input data, the output is the key reaction parameters and the mean square error is taken as the loss function, the training is carried out in stages and the gradient descent method is used for optimization until convergence, and a dynamic amplification adjustment value is output;

[0009] The key reaction parameters are input into a linear regression model, the key reaction parameters are expanded by using a synthetic overturning axis method based on bridge symmetry and temperature, concrete creep and vehicle-bridge interaction, and a support combined reaction force is generated by predicting core reaction parameters through the linear regression model;

[0010] In combination with bridge geometric parameters, the risk of voiding under temperature and concrete creep is checked, the anti-overturning stability coefficient is calculated, the dynamic amplification and concrete creep effect are taken into account, the overturning process is evaluated in four stages and divided into four risk levels, the convolutional neural network is integrated into a bridge health monitoring system, sensor data are collected in real time and dynamic early warning and reinforcement suggestions are output.

[0011] In another aspect, the embodiments of the present application also provide a method for evaluating overload overturning risk of intensive single-column pier based on convolutional neural network, the hardware of the method comprises a processor and a machine-readable storage medium, the machine-readable storage medium is connected with the processor, the machine-readable storage medium is used for storing programs, instructions or codes, and the processor is used for executing the programs, instructions or codes in the machine-readable storage medium to realize the above-mentioned method.

[0012] Based on the above aspects, the method for evaluating overload overturning risk of intensive single-column pier based on convolutional neural network has the following beneficial effects:

[0013] This invention integrates vehicle load parameters, temperature effects, concrete creep, and vehicle-bridge interactive dynamic amplification factors into a multidimensional tensor input. Utilizing staged training and gradient descent optimization of a convolutional neural network, it achieves second-level reaction force prediction and dynamically amplified adjustment value output. By synthesizing the overturning axis, the reaction force data is expanded, with errors controlled within a reasonable range. After integration with existing bridge health monitoring systems, it can collect sensor data in real time and output four risk levels: safe, critical, dangerous, and requiring immediate intervention. This effectively avoids the limitations of post-hoc diagnosis in existing systems, significantly reduces the probability of overturning accidents, thereby reducing bridge maintenance costs, preventing resource waste, and improving structural stability under high traffic conditions. It is applicable to various bridge types, including straight and curved bridges, enabling intensive assessment of overload overturning risks in single-column pier bridges, greatly improving the real-time performance and accuracy of the assessment.

[0014] This invention, based on highway bridge and culvert design specifications and combined with bridge geometric parameters, examines the risk of bearing delamination, calculates the overturning stability coefficient, incorporates dynamic amplification and concrete creep effects, assesses the overturning process in four stages and classifies it into four risk levels, automatically outputs reinforcement recommendations based on the risk level, and simulates the reinforcement effect through a digital twin platform. This provides a comprehensive prevention strategy for the overload overturning risk management of single-column pier bridges, achieving closed-loop management from assessment to action, improving the accuracy of bridge remaining life prediction, reducing human intervention errors, promoting the transformation of bridge maintenance from passive repair to proactive prevention, and enhancing the safety and sustainability of long-term bridge operation. Attached Figure Description

[0015] Figure 1 This is a flowchart of the intensive single-column pier overload overturning risk assessment method based on convolutional neural networks provided in this embodiment of the invention.

[0016] Figure 2 This is a line graph showing the real-time model iteration count and loss value of the intensive single-column pier overload overturning risk assessment method based on convolutional neural networks provided in this embodiment of the invention.

[0017] Figure 3 This is a scatter plot of the target value and predicted value of the intensive single-column pier overload overturning risk assessment method based on convolutional neural network provided in this embodiment of the invention.

[0018] Figure 4 This is a data processing relationship diagram of the intensive single-column pier overload overturning risk assessment method based on convolutional neural networks provided in this embodiment of the invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Specific Implementation Example 1:

[0021] like Figures 1 to 4 As shown, the intensive method for assessing the overload overturning risk of single-column piers based on convolutional neural networks includes:

[0022] Step S100: Establish a full-bridge finite element model based on the bridge's geometric parameters. Set vehicle load conditions according to the "General Specifications for Highway Bridge and Culvert Design" (JTG D60-2015). Simulate data on vehicle load parameters, temperature effects, concrete creep, and vehicle-bridge interaction. Calculate bearing reaction force data and perform hierarchical normalization processing. Extract key reaction force parameters to ensure the assessment considers the dynamic impact of actual operation, improving data representativeness and model generalization ability. Bearing reaction force data refers to the reaction force generated by bridge bearings when subjected to superstructure and external loads, a key indicator reflecting the bridge's stress state. The dynamic amplification factor is the load amplification effect caused by vibration and road surface roughness when vehicles pass over the bridge, typically ranging from 1.1 to 1.5. Data acquisition can be achieved using a drone equipped with a LiDAR to scan the bridge and generate a 3D point cloud model, ensuring geometric parameter accuracy down to the millimeter level. Simulating the bridge's dynamic response under real operating conditions using the finite element model allows for the identification of bearing reaction force variation patterns, providing high-quality input data for subsequent convolutional neural networks while reducing the cost and risk of on-site testing.

[0023] The vehicle-bridge interaction refers to the change in the stress state of the bridge due to the interaction between the movement (speed, acceleration, etc.) of the vehicle, the vehicle load (axle load, wheelbase, off-center load position, etc.), and the dynamic response (vibration, road roughness, etc.) of the bridge. The vehicle-bridge interaction is reflected by a dynamic amplification factor, which is usually 1.1 to 1.5, reflecting the load amplification effect due to vibration and road roughness when the vehicle passes through the bridge, as well as the dynamic influence of the vehicle on the bridge in actual operation. It is a key input for the stress state analysis of the bridge. The parameters of the vehicle-bridge interaction are collected based on the General Code for Design of Highway Bridges and Culverts (JTG D60-2015), finite element model simulation, field data collection, random traffic flow model, and environmental factors. The related data of the vehicle-bridge interaction (such as support reaction force and dynamic amplification factor) are generated by the finite element model simulation, then normalized by layer, stored in CSV format, uploaded to the local server or cloud database as multi-dimensional tensor input, and provided to the convolutional neural network of step S200 for training to predict key reaction force parameters. Subsequently, the data of the vehicle-bridge interaction are expanded in step S300 by the combined overturning axis method and linear regression model to generate support concurrent reaction force combinations, which are finally used in step S400 to calculate the anti-overturning stability coefficient and risk classification. The data of the vehicle-bridge interaction are collected in real time by sensors of the bridge health monitoring system and uploaded to the cloud database to support real-time risk assessment and dynamic early warning. By simulating the dynamic load of the vehicle and the vibration response of the bridge, the stress change of the bridge in actual operation can be accurately captured.

[0024] The vehicle load parameters mainly include axle load, wheelbase, lateral off-center load position, vehicle speed, road roughness level, and dynamic amplification factor generated thereby, which are used to reflect the influence of the vehicle on the bridge in actual operation and are derived from the simulation of bridge design specifications and actual operation data. According to the General Code for Design of Highway Bridges and Culverts (JTG D60-2015), the vehicle load working conditions are set, and the dynamic response of the bridge under real operation conditions is simulated by a full-bridge finite element model, such as axle load range 10 to 50 tons, wheelbase 2 to 6 meters, lateral off-center load position change range -1.5 to 1.5 meters, vehicle speed 20 to 80 kilometers / hour, road roughness level A to C, etc. The dynamic amplification factor is derived from the load amplification effect due to vibration and road roughness when the vehicle passes through the bridge, and the value is usually 1.1 to 1.5, which is generated by at least 200 simulations (each simulation lasting 5 minutes).

[0025] In the process of vehicle load parameter processing, firstly, the segmented modeling strategy is adopted in the full-bridge finite element model, and the key components of the bridge such as the beam body, pier column and support are modeled respectively, and then integrated into the full-bridge model through coupling analysis to improve the calculation efficiency; the multi-physical field coupling analysis is introduced, combined with wind load and seismic load, to further simulate the bridge response under extreme environment; then, the generated support reaction data is normalized by layer, and the numerical value is mapped to the interval of 0 to 1 using the NumPy library of Python, and the data distribution is monitored to avoid overflow. Abnormal values such as data exceeding 3 times the standard deviation are removed by data cleaning tools to ensure the quality of the data set. The extraction of key reaction parameters includes the maximum value of the double-support reaction of the side pier (unit: kN, precision: 0.1 kN), the distribution standard deviation of the single-support reaction of the middle pier (precision: 0.01), etc. The processed vehicle load parameters and related support reaction data are extracted as key reaction parameters to generate a data set covering straight bridges, curved bridges and at least 20 types of bridges, containing at least 1000 samples, stored in CSV format, and uploaded to a local server or cloud database, providing input data for the convolutional neural network training in subsequent step S200, and the extension and risk assessment in step S300, improving the representativeness of the data and the generalization ability of the model, identifying the change rule of the support reaction by simulating the influence of vehicle load parameters on vehicle-bridge interaction, providing high-quality input data for the convolutional neural network, reducing the cost and risk of field tests, supporting real-time risk assessment, avoiding bridge overturning accidents, and improving the efficiency and safety of bridge risk management under complex working conditions.

[0026] Concrete creep specifically refers to the sustained deformation of concrete under long-term load, including short-term concrete creep and long-term concrete creep, which is included in the bridge simulation. In the simulation, the different concrete creep effects in the construction phase and the operation phase are distinguished by adopting staged loading. The concrete creep coefficient is taken as a key parameter and is included in the multi-dimensional tensor input, combined with temperature effects and vehicle-bridge interaction. When grouping, it is grouped according to the concrete creep stage, which is divided into short-term 28 days and long-term 1 year to 10 years, with at least 300 samples in each group. Concrete creep can affect the long-term stability of the bridge, increase the deflection of the beam body, redistribute the support reaction, and may cause the risk of support voiding, especially under the action of temperature gradient and heavy load vehicles, which amplifies the overturning risk. For example, in a curved bridge, concrete creep may cause an increase in beam rotation angle and a decrease in stability coefficient. The concrete creep data is derived from finite element model simulation, which is updated once a month according to the JTG 3362-2018 specification. The data is collected by measuring the bridge geometric parameters and importing and simulating them using finite element software such as Midas Civil or Ansys. The data acquisition combines total station and laser scanner to obtain pier inclination angle and beam deflection data with an accuracy of less than 0.005 meters.

[0027] Step S101: When establishing the full-bridge finite element model, first measure and collect the bridge geometric parameters, including the support spacing accurate to the centimeter level, the self-weight distribution of the beam body per meter, the pier column height, and the material properties such as the concrete strength grade C30 to C60 and the reinforcement yield strength HRB40, use measuring instruments such as laser scanners to obtain data, and ensure the accuracy within 0.01 meters;

[0028] Then use the finite element software to import the bridge geometric parameters and set the vehicle load working condition, the finite element software such as Midas Civil or Ansys, configure the axle load range of 10 tons to 50 tons, the axle distance of 2 meters to 6 meters, the lateral load position change range of -1.5 meters to 1.5 meters, and simulate the temperature effect, set the uniform temperature change from -30°C to 60°C, collect once every 5°C, update the temperature data once every 12 hours, and the gradient temperature positive and negative direction is 5°C per meter;

[0029] Finally, calculate the concrete creep, adopt the long-term self-weight deformation model based on the JTG 3362-2018 specification to update once a time, analyze the vehicle-bridge interaction, set the vehicle speed of 20 to 80 kilometers / hour, the step of 10 kilometers / hour, the road roughness grade A to C, run at least 200 times of simulation, each simulation time length of 5 minutes, calculate and export the dynamic amplification factor, the range [1.1-1.5], generate the support reaction force data, and store to the local server.

[0030] wherein the long-term self-weight deformation model is a concrete creep simulation model based on the JTG 3362-2018 specification for calculating the sustained deformation of concrete under long-term loads, including concrete creep coefficients and staged loading, covering short-term and long-term stages, combined with temperature effects and vehicle interaction effects. Based on the finite element software, after importing the bridge geometric parameters, the long-term self-weight deformation simulation is set. First, update the concrete creep data at a fixed time (for example, update the temperature data every 12 hours, but the concrete creep is updated at a fixed time for a long time); second, simulate the different concrete creep effects in the construction and operation stages by staged loading; then, through the vehicle-bridge interaction analysis and integration, run at least 200 simulations, each for 5 minutes, to generate support reaction force data. The concrete creep time is 28 days to 10 years, and the composite overturning axis method is used to process the curved bridge reaction force expansion. When processing, first group the support reaction force data according to the concrete creep stage (short-term 28 days and long-term 1 year to 10 years), with at least 300 samples in each group. Then, use the NumPy library of Python for normalization processing, mapping the numerical value to the 0 to 1 interval. Abnormal value detection combined with machine learning algorithms (such as Isolation Forest) removes abnormal values (such as data more than 3 times the standard deviation). Store the data to the local server to generate a CSV format dataset. By calculating the concrete creep, ensure the dynamic response of the bridge under real operating conditions, provide high-quality input data to the convolutional neural network, reduce the cost and risk of field tests, and be used for training in step S200 and expansion in step S300.

[0031] Step S102: When processing the support reaction force data to extract key reaction force parameters, first group the support reaction force data according to the temperature interval, with groups of -30°C to 0°C, 0°C to 30°C, 30°C to 60°C, and concrete creep stage groups, respectively, short-term 28 days and long-term 1 year to 10 years, with at least 300 samples in each group; then use the NumPy library of Python to normalize each group of data, mapping the numerical value to the 0 to 1 interval, monitoring the data distribution during processing to avoid overflow, extracting key reaction force parameters, including the maximum value of the double support reaction force of the side pier (unit: kilo Newton, precision: 0.1 kilo Newton), the distribution standard deviation of the single support reaction force of the middle pier (precision: 0.01), the beam rotation angle (precision: 0.001 radian), and the long-term variation coefficient (precision: 0.001), generating a dataset covering straight bridges, curved bridges, and at least 20 types of bridges, containing at least 1000 samples, stored as a CSV format, and removing abnormal values such as data more than 3 times the standard deviation through a data cleaning tool, ensuring the quality of the dataset, providing standardized input for step S200 through the complete link from simulation to extraction, and realizing the conversion of physical data to machine learning data.

[0032] In this embodiment, the laser scanner can generate high-precision three-dimensional point cloud data by using a device that measures the distance of the object surface with a laser beam, which is used for bridge geometric modeling; the sustained deformation phenomenon of concrete under long-term load is called concrete creep, which affects the long-term stability of the bridge; the index of road surface flatness is divided into A (flat), B (general), C (poor) levels, which affects the dynamic effect of vehicle-bridge interaction; when measuring, the total station and the laser scanner can be combined to obtain the inclination angle of the pier column and the deflection data of the beam body, with the precision controlled within 0.005 meters; when simulating vehicle load, a random traffic flow model is introduced to simulate the scene of multiple vehicles passing through at the same time during peak hours; the temperature effect simulation adds a sunshine radiation model to consider the non-uniform temperature distribution of the bridge surface caused by solar radiation; the concrete creep simulation adopts a phased loading to simulate the different concrete creep effects in the construction and operation stages; through accurate geometric parameters and multi-working condition simulation, the model can reflect the real stress state of the bridge under complex environment, and provide diversified data samples for subsequent machine learning training.

[0033] The key reaction force parameters reflect the core indicators of the bridge stress state, including the maximum value of the reaction force, the distribution standard deviation, etc., which are used to evaluate the overturning risk; when grouping data, time series analysis can be introduced to dynamically group the reaction force data combined with the daily traffic flow changes; the normalization processing adopts Min-Max standardization or Z-score standardization, and the optimal method is selected according to the data distribution; the abnormal value detection can further improve the data cleaning efficiency combined with machine learning algorithms (such as isolated forest); the development of visualization tools (such as Matplotlib to generate reaction force distribution graph) can assist engineers to verify the data quality; through grouping and normalization processing, the data noise is reduced, and the representativeness of the key reaction force parameters is improved, providing standardized feature input for subsequent neural network training.

[0034] Step S200: Design a convolutional neural network based on the full-bridge finite element model, use a multi-dimensional tensor as input data, output the key reaction force parameters, and use mean square error as the loss function; through stage-by-stage training and gradient descent method optimization to convergence, output the dynamic amplification adjustment value, learn the nonlinear relationship between vehicle load parameters and key reaction force parameters from multi-source data, realize efficient prediction, replace traditional finite element iterative calculation, and improve the calculation speed and adaptability of evaluation;

[0035] The specific way of convolution operation in neural network is:

[0036] Let the input data be a one-dimensional sequence , use a convolution kernel with length k, The convolution operation at the jth position is:

[0037] ;

[0038] where, is the last value of the vehicle load parameter sequence, and the input sequence x contains a multi-dimensional tensor; is the last parameter of the convolution kernel, which is learned by training the convolutional neural network and used to weight and sum the local window of the input sequence; is the jth output value after convolution calculation, representing an element in the result sequence after convolution calculation, which is the output after local feature extraction in the convolutional neural network for subsequent layer processing; is the local window data in the input sequence, specifically the ith input value in the window starting from position j; is the ith weight value of the convolution kernel parameter, which is a parameter learned during the training process of the convolutional neural network and is used to extract features from input data; is the bias term, used to adjust the offset of the convolution output and improve the fitting ability of the model.

[0039] By inputting a one-dimensional sequence such as a multi-dimensional tensor, a convolution kernel of length k is selected, and for each position j in the output sequence, a local window of length k is extracted from the input sequence (from to , but the index is ), each element in the window is multiplied by the corresponding convolution kernel weight and summed, plus the bias b, to get the output , repeat this process, slide the window through the entire input sequence, generate the output sequence. The step size s determines the sliding interval (if s > 1, skip the calculation), the padding p determines whether to pad zeros at both ends of the input (SAME mode keeps the output length the same, VALID mode does not pad zeros, resulting in output shortening), which is used to extract key features from multi-dimensional tensors, quickly predict key reaction force parameters, build a prediction model of vehicle load parameters and key reaction force parameters, improve calculation efficiency, achieve second-level evaluation, replace time-consuming finite element simulation, overall, support real-time risk assessment, avoid bridge overturning accidents.

[0040] The output length is calculated as:

[0041] ;

[0042] where, is the length of the output sequence; is the input sequence length; k is the length of the convolution kernel; p is the padding length, the same padding is zero to make the output length the same as the input length, and no padding is shorter than the input; s is the step length, which represents the distance that the convolution kernel slides each time, and the default is 1. If the step length s > 1, the convolution kernel will jump calculation by s units. If s = 1 and p = 0, it is the VALID mode, and the output is shortened. If p = (k-1) / 2, it is the SAME mode, and the output length is close to the input. Through the calculation of the output length, the hyperparameters of the convolutional neural network can be optimized, such as the size of the convolution kernel, the pooling strategy, etc., to improve the model training efficiency and prediction accuracy, and ultimately serve the rapid assessment of bridge overload overturning risk.

[0043] The data samples generated by finite element analysis are used for training, and the network hyperparameters including the size of the convolution kernel, the pooling strategy, and the activation function are optimized. The model accuracy is evaluated by the mean square error loss function until convergence. Figure 2 Figure for the number of training times-loss function of the convolutional network in the embodiment.

[0044] Step S201: Construct a convolutional neural network. Set the input layer to contain a multi-dimensional tensor of axis redistribution vectors, load position vectors, temperature gradient distribution, concrete creep coefficient, and vehicle-bridge interactive dynamic amplification factor using the TensorFlow framework. The dimensions are as follows: [sample number, feature number, time step]. Set 4 convolution layers with a convolution kernel length of 3 to 7 and a step length of 1. The padding mode is zero padding to keep the output consistent and avoid information loss. Configure 3 fully connected layers using the ReLU activation function. Output the key reaction force parameters, including the maximum value and distribution characteristics of the side pier and middle pier reaction forces. Use the mean square error as the loss function and use the gradient descent method for optimization. Monitor the gradient explosion by gradient clipping (threshold 1.0) control.

[0045] Step S202: Based on the basic load data, only the axle load, axle distance, and eccentric load position, at least 600 groups of samples are divided into 80% training set and 20% validation set using Python script for pre-training. The model is preliminarily trained on the basic data set to obtain general weights, which provides a basis for subsequent fine-tuning. The batch size is set to 32, the training rounds are 100 to 200 times, the validation loss is evaluated every 10 rounds, and the intermediate model is saved for easy breakpoint training. The data containing temperature effect, concrete creep, and vehicle-bridge interaction are used to fine-tune the intermediate model. The pre-trained model is optimized based on the specific data set to improve the prediction accuracy of the specific scene. The batch size is set to 16, the training rounds are 50 to 100 times, and the actual data of at least 20 bridge cases (including counterforce measurement of straight and curved bridges) are used to verify the hyperparameters, including convolution kernel size and pooling strategy. When the loss function converges to less than 0.005, the training is stopped, and the counterforce parameters containing dynamic amplification adjustment values are output and uploaded to the cloud server. The data in step S100 is converted into prediction output to ensure complete iteration from model design to optimization, providing an extension basis for step S300, and realizing real-time prediction engineering application.

[0046] In this embodiment, learning rate decay strategy (such as from 0.01 to 0.001) is used during training to improve late-stage convergence accuracy. Early stopping mechanism is introduced to stop training when the validation loss does not decrease for 10 consecutive rounds. Customized optimization is performed for specific bridge types (such as curved bridges with a radius of curvature less than 50 meters) during fine-tuning. An automated pipeline is developed to update training data from the cloud database regularly. Through phased training and fine-tuning, the model can adapt to different bridge types and working conditions, improving the generalization ability and engineering applicability of the prediction.

[0047] Step S300: Key counterforce parameters are input into a linear regression model. Based on bridge symmetry, temperature, concrete creep, and vehicle-bridge interaction, the key counterforce parameters are expanded using the synthetic overturning axis method. The core counterforce parameters are predicted by the linear regression model to generate support and concurrent counterforce combinations. The key counterforce parameters are expanded using symmetry and regression models to generate multi-scenario concurrent combinations, making the evaluation conform to the specification requirements and improving the intensification and robustness of the method.

[0048] The expression of the linear regression model is as follows:

[0049] ;

[0050] wherein, The predicted value represents the core reaction force parameter output by the model, which is used for the expansion of subsequent key reaction force parameters; x is the input value, which identifies the key reaction force parameter; w is the weight, which represents the degree of influence of the input x on y; b is the bias, and all are scalars. In linear regression prediction, the smaller the difference between the predicted value and the true value, the higher the accuracy of the model, and the error follows a normal distribution with a mean of 0. Figure 3 Scatter plot for linear regression model prediction in the example.

[0051] The synthetic overturning axis method is a technique based on bridge symmetry and multi-source influence to expand key reaction force parameters, used to generate support concurrent reaction force combinations, making the evaluation meet the specification requirements, and improving the intensification and robustness of the method. The synthetic overturning axis method is an algorithm for handling curved bridge reaction force expansion, based on the symmetric distribution of double support piers and the axial symmetry of single support piers, considering temperature gradient, concrete creep and vehicle-bridge interaction, simulating reaction force distribution under different scenarios by synthesizing virtual overturning axes, setting the error control range to -38.8% to 33.9%, and expanding single key reaction force parameters to multi-scenario concurrent combinations. When used, the key reaction force parameters output in step S200 are input into the linear regression model, and Python scripts are used to divide symmetric units (such as dividing the side piers into left and right symmetric parts), analyze the influence of temperature gradient and concrete creep, and collect data every 4 hours with a temperature of 5°C per meter in both positive and negative directions; the concrete creep time is 28 days to 10 years, and the abnormal expansion is filtered by pre-support, running at least 300 iterations, optimizing the limited position each iteration, generating expanded key reaction force data, covering at least 30 concurrent support reaction force combinations.

[0052] When expanding key reaction force parameters, symmetry and regression models are used to generate multi-scenario combinations from a single input, ensuring comprehensive coverage of specification requirements (such as JTG 3362-2018), using the key reaction force parameters (including the maximum values and distribution characteristics of side pier and central pier reaction forces) output by the convolutional neural network in step S200 as input, combining bridge geometric parameters, and dividing symmetric units based on bridge symmetry using Python scripts, such as dividing side piers into left and right symmetric parts and central piers based on axial symmetry. Analyze the influence of temperature gradient and concrete creep, and for curved bridges, synthesize virtual overturning axes to simulate beam rotation and reaction force redistribution. Set the error control range to -38.8% to 33.9%, filter abnormal expansion values by pre-support, run at least 300 iterations, optimize the limited position each iteration, generate expanded key reaction force data, such as expanding the maximum value of double support pier reaction forces into multiple combinations, cover at least 30 concurrent support reaction force combinations, and store them as an index table using a MySQL database, including timestamp and expansion type fields for easy retrieval, and provide the expansion data to the linear regression model for further prediction of core reaction force parameters.

[0053] The core counterforce parameters predicted by the linear regression model mainly include the support counterforce values in the failure state, such as the counterforce value of the support on the ith pier under the action of the dead load when the support is in the failure state , the counterforce value of the support on the ith pier under the action of the live load when the support is in the failure state , and the support counterforce in the normal working state , which is used for the calculation of the anti-overturning stability coefficient. The key counterforce parameters derived from the full-bridge finite element model simulation and the data output by the convolutional neural network in step S200 are input as a feature vector containing the concrete creep coefficient and the dynamic amplification factor of the vehicle-bridge interaction, based on the expanded key counterforce data. The training sample is at least 900 groups, the test sample is 300 groups, and it meets the JTG 3362-2018 specification. The predicted core counterforce parameters generate support concurrent counterforce combinations and are stored in CSV format, supporting subsequent step S400 calling, which is used for calculating the anti-overturning stability coefficient and risk classification, uploading to the cloud database, and facilitating real-time calling by the health monitoring system. The role of the core counterforce parameters is to expand the key counterforce parameters to generate multi-scenario concurrent combinations for predicting overturning risk, supporting four-stage evaluation and generating four-level risk grades.

[0054] Step S301: Input the key counterforce parameters output in step S200 into the linear regression model, and based on the symmetry distribution of the side pier double supports and the axial symmetry characteristics of the middle pier single support, use Python script to divide symmetric units according to the “Highway Reinforced Concrete and Prestressed Concrete Bridge and Culvert Design Specification” (JTG3362-2018), for example, divide the side pier into left and right symmetric parts, analyze the influence of temperature gradient and concrete creep, temperature adopts 5℃ per meter in positive and negative directions, collect data every 4 hours, concrete creep time is 28 days to 10 years, use synthetic overturning axis method to process curved bridge counterforce expansion, set error control range to-38.8% to 33.9%, filter abnormal expansion through pre-support, run at least 300 iterations, and optimize the limited position each time;

[0055] Generate expanded key counterforce data covering at least 30 support concurrent counterforce combinations, and store and expand the key counterforce data to generate an index table, use MySQL database, contain timestamp and expansion type field, facilitate fast retrieval and update.

[0056] Step S302: Extract the simulated key counterforce parameters from the full-bridge finite element model and the core counterforce parameters predicted by the linear regression model. Set the input as a feature vector containing the concrete creep coefficient and the vehicle-bridge interaction dynamic amplification factor. Train the linear regression model, assign at least 900 training samples and 300 test samples to adjust the weights and biases, minimize the difference between the predicted value and the true value, generate a combination of support and concurrent counterforces, meet the requirements of JTG 3362-2018 specifications, and store it in CSV format to support subsequent step S400 calls. Extend the prediction of step S200 to specification data to ensure efficient conversion from single input to multiple combination outputs, achieve comprehensive coverage of evaluation and data optimization.

[0057] Step S400: Combine the bridge geometric parameters, which are loaded from the local server, such as support spacing and beam self-weight distribution. According to article 3.6.8 of the General Specification for Highway Bridge and Culvert Design (JTG D60-2015), the setting of bridge transverse and vertical supports should consider the problem of support voiding. The upper and lower transmission surfaces of the support should remain horizontal. Check the risk of voiding under temperature and concrete creep, calculate the overturning stability coefficient, include dynamic amplification and concrete creep effects, evaluate the overturning process in four stages and divide it into four risk levels. Integrate the convolutional neural network into the bridge health monitoring system, real-time collect sensor data and output dynamic early warning and practical reinforcement suggestions. Integrate the aforementioned data for final calculation and output, provide risk classification and prevention guidance, realize the closed loop of the evaluation process and practical application, and ensure the initiative and sustainability of bridge management.

[0058] According to article 4.1.8 of the Highway Reinforced Concrete and Prestressed Concrete Bridge and Culvert Design Specification (JTG 3362-2018), under sustained conditions, the structural system of the beam bridge should not change and should meet the following requirements: ① Under the action of the basic combination, the single compression support always remains in compression state; ② When the standard value is combined, the action effect of the integral section simply supported beam and continuous beam should meet the following requirements:

[0059] ;

[0060] ;

[0061] ;

[0062] In the formula: is the transverse overturning stability coefficient, taken as ; is the effect design value of the superstructure stability; is the effect design value of the superstructure instability; denotes the reaction force value when the i-th pier support is in failure state under dead load; denotes the reaction force value when the i-th pier support is in failure state under live load; denotes the distance between the support center of the i-th pier in normal working state and the support center of the i-th pier in failure state. According to the threshold value of the anti-overturning coefficient, such as k≥2.5 is safe, the stability level of the bridge is divided into safe, critical, dangerous, and immediate intervention.

[0063] Step S401: input the support and combined reaction force, evaluate the overturning process by combining the dynamic amplification factor and the concrete creep effect, the four-stage evaluation of the overturning process includes the stable stage of all supports under compression, i.e. all supports are in normal working state and the reaction force is uniform, the transition stage of the first support failure, i.e. the first support is detached, the reaction force is redistributed, the risk stage of three support, i.e. the remaining three supports bear the load, the stability decreases, the overturning stage of two support, i.e. only two supports are in working state, the structure loses stability, calculate the anti-overturning stability coefficient, according to JTG D60-2015 3.6.8 and JTG 3362-2018 4.1.8, check the risk of voiding under temperature and concrete creep (simulate the threshold value of beam rotation angle 0.02 radian), according to the results of four-stage evaluation of overturning process, divide four risk levels, four risk levels include safe (stability coefficient≥2.5, normal operation), critical (coefficient 1.5 to 2.5, need to monitor), dangerous (coefficient 1.0 to 1.5, immediate repair) and immediate intervention (coefficient <1.0, stop traffic), generate evaluation report in PDF format and update daily, notify the manager through email.

[0064] Step S402: integrate the trained convolutional neural network model into the bridge health monitoring system, configure sensors to collect vehicle load, temperature and vibration data in real time, such as 15 sensors, including 8 temperature sensors and 7 vibration sensors, collect vehicle load, temperature (-30℃ to 60℃) and vibration data in real time, the collection frequency is every 3 seconds, support wireless transmission; set the beam rotation angle alarm, the threshold is 0.02 radian, display in real time through the Web interface of the monitoring system, generate reinforcement suggestions according to the risk level, including adding steel support to form steel cap beam and using super high performance concrete post-poured strip connection, pier beam consolidation, widening pier column or setting auxiliary anti-pulling constraint, preferentially aiming at the concrete creep risk of curved bridge, considering the case of curvature radius <50 meters, simulate the reinforcement effect through the digital twin platform, run at least 100 simulations (each time 10 minutes) to verify the force transmission path and bending moment performance, generate simulation report containing 3D visualization, store in cloud database, convert the data of the foregoing steps into action output, ensure the complete closed loop from input to prevention, realize dynamic monitoring and optimization management of bridge risk.

[0065] According to the above technical solution, when applied to single-column pier highway bridges, it faces the challenges of frequent passage of heavy vehicles (6000 vehicles per day, with a maximum axle load of 50 tons) and high temperature and humidity (-10°C to 50°C). By applying this technical solution, the team first measures the bridge geometric parameters (pier height 9 meters, support spacing 2.8 meters) using laser radar and total station, establishes a full-bridge finite element model through Midas Civil, simulates the effects of heavy vehicles and temperature gradient, generates support reaction force data, with a peak value of 1600 kN for the side pier and a dynamic amplification factor of 1.35; after processing the data using Python, the convolutional neural network is trained to predict the reaction force, with an error of less than 1%, replacing the traditional finite element iteration and saving computing time. Based on the symmetry of the bridge, 40 combinations of reaction forces are extended, and the linear regression model predicts and generates reaction forces, which meet the JTG 3362-2018 specification. The 15 sensors integrated into the health monitoring system collect data in real time, and it is found that during a typhoon, the beam rotation angle reaches 0.019 radians, and the stability coefficient is 2.6 (safe level). The system recommends pier beam consolidation and reinforcement, and the digital twin simulation verifies that the coefficient improves to 3.0 after reinforcement. The management side implements reinforcement and optimizes traffic control accordingly, successfully avoids the risk of voiding, and extends the service life of the bridge by more than 5 years, demonstrating that the technical solution significantly improves the efficiency and safety of bridge risk management under complex conditions through precise modeling, rapid prediction, and real-time monitoring.

[0066] When applied to single-column pier bridges in mountainous areas with a radius of curvature of 40 meters, the span is 45 meters, and it needs to cope with the influence of heavy mine cars (axle load 45 tons) and long-term concrete creep (10 years). By applying this technical solution, the team uses unmanned aerial vehicle laser scanning to obtain geometric parameters, with a pier height of 7.5 meters and a support spacing of 2.5 meters. A finite element model is established in Ansys, simulating vehicle unloading and temperature changes from -20°C to 45°C, generating a reaction force dataset with a mid-pier reaction force of 850 kN and a dynamic amplification factor of 1.28. After normalization processing through Python, the convolutional neural network is trained to predict the reaction force with an error of 0.8% and a time consumption of only 2 seconds. Based on the symmetry of the curved bridge, 35 combinations of reaction forces are generated, and the linear regression model optimizes and generates reaction forces with an error of 3.5%, meeting the JTG D60-2015 specification. The health monitoring system is equipped with 12 sensors (6 temperature, 6 vibration) for real-time monitoring, which found that during a heavy load vehicle fleet, the beam rotation angle was 0.021 radians, and the stability coefficient was 1.8 (critical level), triggering an alarm. The system recommends widening the pier and adding uplift resistance, and digital twin simulation shows that the coefficient increases to 2.7 after reinforcement. The management side implements reinforcement and restricts heavy load traffic at night, and after 3 months of re-measurement, it is confirmed that the risk has been eliminated. The adaptability of the curved bridge under complex stress scenarios effectively prevents overturning risks through efficient prediction and dynamic management, ensuring the safe operation of mountainous bridges.

[0067] In summary, by finite element modeling and convolutional neural network, the bridge reaction force is efficiently predicted, saving calculation time, and the reaction force parameter expansion and linear regression ensure comprehensive evaluation and standard compliance. Real-time monitoring and digital twin technology realize dynamic risk early warning and reinforcement scheme optimization, suitable for different bridge types and complex working conditions, and exhibit the robustness and engineering practicability of the scheme.

[0068] In this embodiment, an intensive single-column pier overload overturning risk assessment method based on convolutional neural network is proposed. By establishing a full-bridge finite element model, designing a convolutional neural network, expanding the reaction force parameters, and integrating a health monitoring system, a complete closed-loop application from data collection to risk assessment can be provided for highway 50-meter span single-column pier bridges, urban road viaducts, etc. The anti-overturning stability coefficient is generated, the beam rotation angle of 0.015 radian is output in real time, and reinforcement suggestions are provided. Through PDF reports and Web interfaces, managers are updated daily to ensure the safe operation of bridges under heavy vehicles, extreme temperatures, and long-term concrete creep, significantly improving management efficiency and structural safety. The entire process can be deployed on a cloud platform (such as AWS or Alibaba Cloud) to integrate data storage, model training, and real-time monitoring. An artificial intelligence assisted decision system is introduced to automatically generate maintenance plans. Combined with Internet of Things technology, it is extended to the collaborative management of regional bridge groups. Regular expert audits optimize models and reinforcement schemes. The technical solution combines physical modeling, machine learning, and real-time monitoring to achieve precise prediction and dynamic management of bridge overturning risk, suitable for various bridge types, and reduces operational risks.

[0069] To address the overturning risk of single-column pier bridges under overload, environmental changes, and dynamic interactions, this project provides a complete chain from data simulation to real-time early warning. First, data is generated through finite element simulation. Then, a convolutional neural network is built and trained for prediction. Next, the reaction force data is expanded and optimized, ultimately achieving risk assessment and prevention output. Based on physical simulation, machine learning and standardized calculations are integrated to combine offline data preparation with online real-time application. An iterative optimization mechanism is employed, processing the dataset and training the model on a local server before deployment to a bridge health monitoring system, supporting cloud data synchronization to ensure the accuracy, real-time performance, and scalability of the assessment. Vehicle load parameters, temperature effects, concrete creep, and vehicle-bridge interaction dynamic amplification factors are integrated into a multi-dimensional tensor input. Phased training and gradient descent optimization of the convolutional neural network achieve second-level reaction force prediction and dynamic amplification adjustment value output. The reaction force data is expanded using a synthetic overturning axis method, with errors controlled within a reasonable range. After integration with existing bridge health monitoring systems, sensor data can be collected in real time. This system outputs four risk levels: safe, critical, dangerous, and immediate intervention. It effectively avoids the limitations of post-event diagnosis in existing systems, significantly reduces the probability of overturning accidents, thereby reducing bridge maintenance costs and avoiding resource waste. It also improves structural stability in high-traffic environments and is applicable to various bridge types, including straight and curved bridges. It achieves a centralized assessment of overload overturning risks for single-column pier bridges, greatly improving the real-time nature and accuracy of the assessment. Based on highway bridge and culvert design specifications and bridge geometric parameters, it checks the risk of bearing dislodgement, calculates the overturning stability coefficient, incorporates dynamic amplification and concrete creep effects, and assesses the overturning process in four stages, classifying it into four risk levels. Based on the risk level, it automatically outputs reinforcement recommendations and simulates the reinforcement effect through a digital twin platform. This provides a comprehensive prevention strategy for overload overturning risk management of single-column pier bridges, achieving closed-loop management from assessment to action. It improves the accuracy of bridge remaining life prediction, reduces human intervention errors, promotes a shift in bridge maintenance from passive repair to proactive prevention, and enhances the long-term safety and sustainability of bridge operations. Specific Implementation Example 2:

[0071] like Figures 1 to 4 As shown, based on the content of the above specific embodiments, the following content is further disclosed:

[0072] The hardware of the intensive single-column pier overload overturning risk assessment method includes a processor and a machine-readable storage medium. The machine-readable storage medium and the processor are connected. The machine-readable storage medium is used to store programs, instructions or code. The processor is used to execute the programs, instructions or code in the machine-readable storage medium to implement the above method.

[0073] To achieve the above technical solution, at least the following hardware structure is required in practical use:

[0074] The hardware structure adopts a layered architecture of "sensor - edge device - cloud server - user terminal", ensuring real-time flow from bridge site data collection to cloud analysis, supporting offline training and online deployment, with high reliability, low delay and high computing power, suitable for urban bridge environment, including:

[0075] Bridge end sensor system (data collection layer): responsible for real-time collection of bridge load and environmental data, supporting analog input and real-time monitoring in steps S100 and S400. Hardware includes:

[0076] Sensor array: deploy at least 15 sensors, including 8 temperature sensors such as PT100 platinum resistance, measuring range -50℃ to 100℃, accuracy ±0.1℃, 7 vibration sensors such as IEPE accelerometer, frequency response 10Hz to 10kHz, sensitivity 100mV / g, and vehicle load sensors, strain gauge load cells, range 10 tons to 50 tons, accuracy ±0.5%, installed at key locations on piers, bearings and beam bodies, support wireless transmission based on LoRa or NB-IoT protocol, collection frequency every 3 seconds, ensure capture dynamic changes such as temperature gradient 5℃ per meter and concrete creep effect;

[0077] Data collector: use embedded microcontroller such as STM32 series, ARM Cortex-M core, 512KB memory, integrated ADC conversion module, responsible for preliminary filtering of noise data (threshold filtering of outliers), and aggregate data through RS485 interface, this layer outputs CSV format raw data stream, transmitted to edge device, supporting load simulation input in step S100 and real-time collection in step S400, ensuring data accuracy and anti-interference capability, IP67 protection level.

[0078] Edge computing device (local processing layer): located at or near the bridge site, responsible for preliminary data processing and model inference, supporting fast prediction in steps S200 and S300, reducing cloud load, achieving low delay response, hardware includes:

[0079] Edge server: use industrial-grade embedded computer such as NVIDIA Jetson Xavier NX, equipped with 6-core ARM CPU and 384-core Volta GPU, 16GB LPDDR4 memory, 256GB NVMe SSD storage), run simplified convolutional neural network, process multi-dimensional tensor input, execute inference part of phased training and linear regression prediction, device supports GPIO interface to connect sensors, integrated Wi-Fi / 5G module to transmit data to cloud, ensure local data storage when network interruption;

[0080] Power and protection system: Equipped with UPS uninterruptible power supply (capacity 500Wh, supporting solar recharge) and protective enclosure (IP65 level, dust and water proof), working temperature -20℃ to 60℃. The hierarchical normalization of step S100 (using NumPy library) and the counterforce expansion of step S300 (synthetic overturning axis method, 300 iterations) output the preprocessed key counterforce parameters (JSON format) for efficient input to the cloud, realizing edge intelligent computing.

[0081] Cloud server (computing and storage layer): Responsible for high-intensity computing tasks such as finite element simulation, complete model training and data expansion, supporting offline processing and global optimization of steps S100, S200, S300. Hardware includes:

[0082] Cloud computing nodes: Use high-performance server clusters such as AWS EC2 or Alibaba Cloud ECS instances, with Intel Xeon Platinum CPU 64 cores, NVIDIA A100 GPU 4 blocks, 512GB memory, 10TB SSD storage, servers running finite element software (Midas Civil) for 200 vehicle-bridge interaction simulations (axle load 10-50 tons, temperature -30℃ to 60℃), and CNN training, pre-training 150 rounds, fine-tuning 80 rounds, using Adam optimizer, integrating database management system (such as MySQL), storing support counterforce data and index table, supporting big data query, at least 1000 groups of samples;

[0083] Cloud storage and backup: Use distributed storage such as Amazon S3 compatible object storage, with unlimited capacity expansion, backup CSV and JSON data, daily automatic synchronization. Servers support containerized deployment, running Python scripts to handle concrete creep model updates and linear regression training (900 training samples), ensuring error control in step S300, receiving edge data through API interface, outputting expanded counterforce combinations, and realizing computationally intensive tasks of the scheme.

[0084] User terminal (output and interaction layer): Responsible for risk visualization and early warning output, supporting step S400 evaluation report and reinforcement suggestion display, realizing human-computer interaction. Hardware includes:

[0085] Monitoring terminal: Use a notebook computer or tablet (such as Dell XPS series, Intel i7 CPU, 16GB memory, 512GB SSD) or mobile APP device (Android / iOS smartphone with high-resolution screen), terminal running Web interface (based on HTML5 and JavaScript), real-time display of beam rotation angle alarm and four-level risk level.

[0086] Digital twin platform: integrate dedicated workstations such as desktop PCs equipped with RTX 3080 GPU, run Bentley Systems software, perform 100 reinforcement simulations, verify steel cap beam connection, pier beam consolidation, etc., generate PDF reports and 3D visualization, receive cloud data through 5G or wired network, support daily update evaluation report, ensure users get reinforcement recommendations in time.

[0087] The overall hardware structure realizes end-to-end operation of the scheme through the processes of sensor collection, edge preprocessing, cloud computing and terminal output, supports real-time monitoring and preventive maintenance of the bridge, and controls the total power consumption within 500W, with high system availability.

[0088] The above hardware structure and its corresponding hierarchical structure are only one way to realize the intensive overload overturn risk assessment method of single-column pier, including but not limited to the above hardware structure and its corresponding parameters, and further setting of the hardware structure can be made according to specific scenes and use requirements in actual use.

[0089] It should be noted that in this paper, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. Without more limitations, the element defined by the statement "including a reference structure" does not exclude the presence of another identical element in the process, method, article or equipment including the element.

[0090] Although the embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for assessing the risk of overload overturning of an intensive single-column pier based on a convolutional neural network, characterized in that, The intensive single-column pier overload overturning risk assessment method comprises: According to the bridge geometric parameters, a full-bridge finite element model is established, the vehicle load parameters, temperature effect, concrete creep and the data of vehicle-bridge interaction are simulated, the support reaction force data is calculated, the support reaction force data is grouped according to the temperature interval, and the grouping is-30℃ to 0℃, 0℃ to 30℃, 30℃ to 60℃, and the concrete creep stage grouping is short-term 28 days and long-term 1 year to 10 years, and each group has at least 300 samples; the data of each group is normalized, the value is mapped to the interval of 0 to 1, the key reaction force parameters are extracted, the key reaction force parameters include the maximum value of the double support reaction force of the side pier, the distribution standard deviation of the single support reaction force of the middle pier, the beam rotation angle and the long-term variation coefficient, the data set covering straight bridges, curved bridges and at least 20 types of bridges is generated, and at least 1000 groups of samples are included; Based on the full-bridge finite element model, a convolutional neural network is designed, a multi-dimensional tensor is used as input data, the multi-dimensional tensor includes an axle load distribution vector, a load position vector, a temperature gradient distribution, a concrete creep coefficient and a vehicle-bridge interaction dynamic amplification factor, a convolution kernel length of 3 to 7 is configured, a step length of 1 is configured, a zero padding mode is configured, 3 fully connected layers are configured, and the output is the key reaction force parameter, including the maximum value and the distribution characteristics of the side pier and the middle pier reaction force, the mean square error is used as the loss function, the training is optimized by the gradient descent method in stages to convergence, and a dynamic amplification adjustment value is output; The key reaction force parameters are input into a linear regression model, based on the symmetry of the bridge and the influence of temperature, concrete creep and vehicle-bridge interaction, the key reaction force parameters are expanded by using a synthetic overturning axis method, the core reaction force parameters are predicted by the linear regression model, and a support concurrent reaction force combination is generated; Combined with the bridge geometric parameters, the risk of voiding under the temperature and concrete creep is checked, the anti-overturning stability coefficient is calculated, the dynamic amplification and the concrete creep effect are included, the overturning process is evaluated in four stages and divided into four risk levels, the convolutional neural network is integrated into the bridge health monitoring system, the sensor data is collected in real time, and dynamic early warning and reinforcement suggestions are output. 2.The method of claim 1, wherein, The full-bridge finite element model is established according to the bridge geometric parameters, comprising: The bridge geometric parameters are measured and collected, including the support spacing accurate to the centimeter level, the beam body per meter self-weight distribution, the pier column height and the material properties; The finite element software is used to import the bridge geometric parameters and set the vehicle load working condition, configure the axle load range of 10 tons to 50 tons, the wheelbase of 2 meters to 6 meters, the transverse load position change range of-1.5 meters to 1.5 meters, and the temperature effect simulation, set the uniform temperature change from-30℃ to 60℃; The concrete creep is calculated, the long-term self-weight deformation model is updated once, the vehicle-bridge interaction analysis is performed, the vehicle speed is set to 20 to 80 kilometers / hour, the road roughness level is A to C, at least 200 simulations are run, the dynamic amplification factor is calculated and exported, the support reaction force data is generated, and stored in the local server. 3.The method of claim 1, wherein, The dynamic amplification adjustment value is output by optimizing the training in stages and the gradient descent method to convergence, comprising: Based on the basic load data, only containing axle load, axle distance, eccentric load position, at least 600 groups of samples are divided into 80% training set and 20% validation set for pre-training, the training round is 100-200 times, and the intermediate model is saved; The data containing temperature effect, concrete creep, and vehicle-bridge interaction are used to fine-tune the intermediate model, the training round is 50-100 times, the actual data of at least 20 bridges are used to verify the hyperparameters, including convolution kernel size and pooling strategy, the reaction force parameters containing dynamic amplification adjustment value are output, and uploaded to the cloud server. 4.The method of claim 1, wherein, The key reaction force parameters are input into the linear regression model, based on the symmetry of the bridge and the influence of temperature, concrete creep, and vehicle-bridge interaction, the synthetic overturning axis method is used to expand the key reaction force parameters, including: The extracted key reaction force parameters are input into the linear regression model, based on the symmetric distribution of the side pier double support and the axial symmetry of the single support of the middle pier, the influence of temperature gradient and concrete creep is analyzed, the synthetic overturning axis method is used to process the reaction force expansion of curved bridge, the error control range is set to-38.8% to 33.9%, and at least 300 iterations are run; The expanded key reaction force data is generated, covering at least 30 concurrent support reaction force combinations, and the key reaction force data is stored and expanded to generate an index table.

5. The method according to claim 1, wherein, The linear regression model is used to predict the core reaction force parameters to generate the support concurrent reaction force combination, including: The key reaction force parameters simulated in the full-bridge finite element model and the core reaction force parameters predicted by the linear regression model are extracted, and the input is set as a feature vector containing the concrete creep coefficient and the vehicle-bridge interaction dynamic amplification factor; The linear regression model is trained, at least 900 training samples and 300 test samples are allocated to adjust the weight and bias, so that the difference between the predicted value and the true value is minimized, and the support concurrent reaction force combination is generated. 6.The method of claim 1, wherein, The bridge geometric parameters are combined to check the risk of voiding under temperature and concrete creep, and the anti-overturning stability coefficient is calculated, taking into account the dynamic amplification and concrete creep effect, including: The support concurrent reaction force combination is input, combined with the dynamic amplification factor and the concrete creep effect to evaluate the overturning process, the four-stage evaluation of the overturning process includes the stable stage of full support compression, the transition stage of the first support failure, the risk stage of three support support, and the overturning stage of two support support, according to the results of the four-stage evaluation of the overturning process, the four-level risk grade is divided, including safe, critical, dangerous, and immediate intervention, an evaluation report is generated and updated daily.

7. The convolutional neural network-based intensification single-column pier overload overturning risk assessment method according to claim 6, characterized in that, The convolutional neural network is integrated into the bridge health monitoring system, real-time sensor data is collected, and dynamic early warning and reinforcement suggestions are output, including: The trained convolutional neural network model is integrated into the bridge health monitoring system, the sensors are configured to collect vehicle load, temperature, and vibration data in real time, the beam rotation angle alarm is set, and the monitoring system Web interface is used to display the data in real time; According to the risk level, reinforcement suggestions are generated, including adding steel support to form a steel cap beam and using concrete post-cast strip connection, pier beam consolidation, widening pier column or setting auxiliary uplift restraint, which is preferentially aimed at the risk of concrete creep of curved bridge, the reinforcement effect is simulated through the digital twin platform, and the force transmission path and bending moment resistance performance are verified. 8.The method of claim 1, wherein, The hardware of the intensive single-column pier overload overturning risk assessment method includes a processor and a machine-readable storage medium, the machine-readable storage medium is connected with the processor, the machine-readable storage medium is used for storing programs, instructions or codes, and the processor is used for executing the programs, instructions or codes in the machine-readable storage medium to realize the above method.

Images