Method for monitoring and analyzing rapid installation of main girder of cable-stayed bridge based on AI technology

Abstract

The application belongs to the technical field of bridge installation monitoring, and specifically discloses a method for quickly installing and monitoring a main girder of a cable-stayed bridge based on AI technology.The method comprises the following steps: S1, real-time acquisition of displacement, stress and environmental load data of a current installation section and adjacent installed sections by means of a distributed sensor network, and construction of a global installation dataset; S2, time-space alignment processing of the dataset, extraction of a joint feature vector representing the instantaneous state of a single point and the cumulative deviation between sections, and dynamic calculation of the attitude prediction tolerance and the cumulative error trend tolerance under the current working condition based on the environmental load; S3, input of the joint feature vector into a pre-trained time series prediction model, and output of the attitude prediction result and the cumulative error trend evaluation result at a future key time point; and S4, error grading decision making.The application realizes the transformation of bridge installation from passive monitoring to active prediction control, and effectively improves the installation accuracy, efficiency and safety.

Description

Technical Field

[0001] This invention belongs to the field of bridge installation monitoring technology, specifically, it relates to a rapid installation monitoring and analysis method for the main girder of a cable-stayed bridge based on AI technology. Background Technology

[0002] In the construction of flexible cable-stayed bridges, the main beams are usually assembled using the cantilever method, extending symmetrically from the bridge towers to both sides, and the prefabricated beam segments are gradually hoisted and connected to form the bridge, which is affected by the dynamic load environment.

[0003] Currently, after each beam segment is hoisted into place, surveyors manually and intermittently collect data and evaluate the current installation posture through offline calculations. This method relies heavily on manual inspections and offline data analysis. On the one hand, it is prone to monitoring delays and high risks of missed detections. On the other hand, it may overlook the cumulative installation error, the development trend of installation error, and the impact between adjacent installed bridge segments, which may lead to structural safety hazards. Furthermore, current judgments are mostly based on experience thresholds, which do not adequately consider the impact of environmental loads, making it difficult to guarantee the accuracy of the judgment results and resulting in untimely and ineffective correction. Summary of the Invention

[0004] In view of this, in order to solve the above problems, a rapid installation monitoring and analysis method for the main girder of cable-stayed bridge based on AI technology is proposed.

[0005] The objective of this invention can be achieved through the following technical solution: This invention provides a rapid installation monitoring and analysis method for the main beam of a cable-stayed bridge based on AI technology: The method includes: S1, real-time acquisition of displacement, stress and environmental load data of the currently installed segment, and synchronous integration of historical monitoring data of adjacent installed segments to construct a global installation dataset.

[0006] S2. Perform spatiotemporal alignment processing on the global installation dataset, extract the joint feature vector representing the instantaneous state of a single point and the cumulative deviation between segments, and dynamically calculate the attitude prediction tolerance and cumulative error trend tolerance under the current working condition based on the real-time acquired environmental load data.

[0007] S3. Based on the joint feature vector, output the future attitude prediction result of the current installation segment and the cumulative error trend evaluation result of the overall installation alignment through time-series prediction.

[0008] S4. Error Classification Decision: First-level judgment: If any attitude prediction value exceeds the prediction tolerance, an installation deviation warning instruction is generated and fed back to the on-site installation personnel.

[0009] Second-level judgment: If the cumulative error trend assessment results all exceed the cumulative error trend tolerance, then adjustment instructions are generated, including the real-time adjustment plan for the hoisting parameters of the current segment and the pre-correction plan for the subsequent uninstalled segments.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention achieves real-time acquisition of multi-source monitoring data of the entire process of main beam installation by deploying a distributed sensor network and automatically and continuously collecting data, replacing the traditional manual intermittent measurement method, effectively avoiding the data loss and observation delay caused by manual inspection, and improving the timeliness and completeness of monitoring data.

[0011] (2) This invention constructs a joint feature vector that reflects the state of a single point and the cumulative deviation between segments, and accurately captures the evolution law and development trend of installation error through time series analysis, thus overcoming the shortcomings of traditional methods that ignore the cumulative effect of error and the mutual influence between adjacent segments.

[0012] (3) This invention automatically adjusts the judgment threshold based on real-time wind speed, temperature and other load conditions, so that the evaluation standard can adapt to complex environmental changes, solves the problem of insufficient adaptability of fixed experience threshold in variable environments, and improves the accuracy and reliability of state judgment.

[0013] (4) By establishing two processing methods, namely graded early warning and forward adjustment instructions, this invention realizes early identification and proactive intervention of installation deviations, significantly improves the timeliness and effectiveness of correction measures, and ensures precise control and structural safety of the main beam installation process. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the overall implementation process of the present invention.

[0016] Figure 2 This is a schematic diagram of the error classification decision-making process of the present invention.

[0017] Figure 3 This is a simplified schematic diagram illustrating the joint feature vector extraction process of the present invention.

[0018] Figure 4 This is a simplified diagram illustrating the calculation process of the attitude prediction tolerance and cumulative error trend tolerance of the present invention.

[0019] Figure 5 This is a simplified diagram illustrating the real-time steps of the timing prediction in this invention. Detailed Implementation

[0020] 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.

[0021] Please see Figure 1 As shown, the present invention provides a rapid installation monitoring and analysis method for the main beam of a cable-stayed bridge based on AI technology. The method includes: S1, acquiring displacement, stress and environmental load data of the currently installed segment in real time, and simultaneously integrating historical monitoring data of adjacent installed segments to construct a global installation dataset.

[0022] Specifically, the environmental load data includes, but is not limited to, real-time environmental wind speed and real-time environmental temperature.

[0023] The displacement, stress, and environmental load data are acquired through a distributed sensor network deployed on the main beam segments and hoisting equipment. The sensor network includes multiple devices such as GNSS receivers, tilt sensors, strain gauges, and anemometers.

[0024] Understandably, a distributed sensor network can be deployed at the four corners, mid-span section, and near the lifting points of each precast main beam segment. All sensors transmit data to the on-site edge computing gateway via an industrial-grade ZigBee wireless network. Testing has shown that a 1Hz transmission frequency can operate stably under the ZigBee protocol, or the LoRaWAN protocol can be used as needed.

[0025] S2. Perform spatiotemporal alignment processing on the global installation dataset, extract the joint feature vector representing the instantaneous state of a single point and the cumulative deviation between segments, and dynamically calculate the attitude prediction tolerance and cumulative error trend tolerance under the current working condition based on the real-time acquired environmental load data.

[0026] In this context, it is understandable that a distributed sensor network can be deployed at the four corner points, mid-span section, and near the lifting points of each precast main beam segment. All sensors transmit data to the on-site edge computing gateway via an industrial-grade ZigBee wireless network. The gateway packages and timestamps the data at a frequency of 1Hz and transmits it to the cloud analytics server via a 5G network.

[0027] Further, please refer to Figure 3 , Figure 3 The overall extraction steps of the joint feature vector are shown, specifically including steps S21-S24.

[0028] S21. Read the displacement, stress, tilt angle, hoisting equipment status parameters, and cable force change rate of the currently installed segment in real time from the distributed sensor network, and use them as characteristic data to characterize the instantaneous state of a single point.

[0029] The status parameters of the hoisting equipment include, but are not limited to, the rotational speed fluctuation coefficient, the pressure stability coefficient, and the lifting point height deviation of the hoisting equipment.

[0030] It is understandable that the rotational speed fluctuation coefficient of the hoisting equipment can be calculated by the relative deviation between the current rotational speed and the average rotational speed over a set time period in the past. The set time period can be 30 minutes. The pressure stability coefficient is calculated by the coefficient of variation of the pressure data. The hoisting point height deviation is represented by the average of the relative deviations between the actual elevation and the theoretical elevation. The coefficient of variation uses the existing calculation method and will not be shown here.

[0031] S22. Subtract the displacement, stress, and tilt values ​​of the currently installed segment from the corresponding reference data recorded by one or more adjacent installed segments at the moment of installation and locking to obtain the displacement difference sequence, stress difference sequence, and tilt difference sequence.

[0032] S23. For each difference sequence, calculate its arithmetic mean and standard deviation over the complete time period from the start of hoisting the current segment to the current moment, and fit its trend slope based on the least squares method.

[0033] S24. The feature data representing the instantaneous state of a single point are concatenated with the arithmetic mean, standard deviation and trend slope to form the joint feature vector.

[0034] Understandably, before performing vector concatenation, all data must be normalized, specifically using a min-max normalization method.

[0035] It should be noted that by calculating the difference between the current segment and the installed segments to highlight the local impact, subsequent calculation errors can be further eliminated. Then, the statistical characteristics such as the average level, volatility and trend of the difference sequence can be extracted. Finally, these multi-dimensional information are stitched together into a fused state profile, which can provide high-quality and understandable input data for accurate and rapid safety assessment and early warning in time series forecasting.

[0036] In one specific embodiment, taking the currently installed Nth segment as an example, the process of extracting the joint feature vector is explained, and the hoisting equipment status parameters are only exemplified by the hoisting point elevation: The following data of the Nth segment at time point T are read in real time: displacement value [ , , = [10.2, -5.3, 0.8] , and These represent the displacement components in the x, y, and z directions, respectively.

[0037] Stress value [ , = [25.6, 18.7], and These represent the horizontal stress component and the vertical stress component, respectively.

[0038] Inclination value [ , = [0.12, -0.05], and These represent the horizontal tilt angle and the vertical tilt angle, respectively.

[0039] The elevation of the suspension point is H=55.32.

[0040] Rate of change of cable force =-0.15, Indicates storage, Indicates time.

[0041] Extract the reference data of the adjacent segment N-1 that has been installed and locked at the moment of installation and locking: Reference displacement value: [ , , ] = [10.0, -5.5, 0.9], , and These represent the reference displacement components in the x, y, and z directions, respectively.

[0042] Reference stress value: [ , = [24.8, 18.5], and These represent the reference horizontal stress component and the reference vertical stress component, respectively.

[0043] Reference tilt angle value: [ , = [0.10, -0.06] and These represent the horizontal reference tilt angle and the vertical reference tilt angle, respectively.

[0044] Calculate the difference between the output displacement values ​​respectively. , , Stress difference , ] and the difference in tilt angle [ , ].

[0045] Extract all displacement differences from the start of hoisting segment N to the current time T. Calculate the arithmetic mean of the sequence. ), standard deviation std( ), and fit its trend slope (slope) based on the least squares method. ).

[0046] right , , , , , Repeat the above calculations to concatenate all the feature data into a joint feature vector V: V=[ , , , , , , H, mean( std( ), slope ), ..., slope( )).

[0047] In this embodiment, the vector is a 35-dimensional real vector.

[0048] Further, please refer to Figure 4 , Figure 4 The overall calculation steps for attitude prediction tolerance and cumulative error trend tolerance are shown, specifically including steps A1 to A7.

[0049] A1. Obtain the attitude prediction baseline tolerance and cumulative error trend baseline tolerance under the reference environmental conditions according to the bridge design specifications.

[0050] A2. Real-time reading of current ambient wind speed, statistics of the duration of wind speed exceeding the benchmark ambient wind speed, and the standard deviation of the current ambient wind speed.

[0051] Understandably, instantaneous wind speeds and sustained strong winds have drastically different effects on structures. Instantaneous wind speeds may cause swaying, while sustained strong winds can lead to cumulative displacement and fatigue effects on structures, threatening long-term safety; therefore, the duration of the wind must be considered.

[0052] Understandably, the response of bridge structures to wind is not linear. Highly volatile winds can amplify dynamics, causing structural vibrations and stress changes far greater than those caused by steady winds at the same average speed. This directly threatens the stability and safety of the bridge's instantaneous posture during hoisting. Therefore, the volatility of wind speed needs to be considered, and thus the standard deviation—a quantitative value characterizing wind speed fluctuations—is introduced to comprehensively assess the specific impact and extent of wind speed.

[0053] A3. Based on the current environmental wind speed, duration, and standard deviation, output the dynamic adjustment amount of wind load on attitude prediction tolerance through preset wind load adjustment statistical rules.

[0054] Preferably, the specific content of the preset wind load adjustment statistics rule is as follows: A31, compare the current ambient wind speed with the preset wind speed intensity level corresponding to each wind speed range threshold to determine the wind speed intensity level to which it belongs.

[0055] Understandably, wind speed intensity levels are set with reference to the wind load classification in the design code for cable-stayed bridges, and based on the specific structural characteristics of the bridge, local wind environment characteristics, and construction control objectives. For example, 0-10 m / s is defined as the normal level; 10-15 m / s is the stronger level; and above 15 m / s is the stronger level.

[0056] A32. Obtain the baseline attitude prediction tolerance adjustment amount corresponding to the wind speed intensity level according to the preset mapping relationship.

[0057] Understandably, the preset mapping relationship is established based on wind tunnel tests, structural finite element simulations, and historical monitoring data, using a combination of theoretical analysis and statistical regression. The specific process is as follows: first, intensity levels are divided according to wind speed ranges; then, the baseline adjustment range corresponding to each level is determined through mechanical calculations and data back-analysis; finally, a deterministic correspondence between wind speed intensity levels and baseline adjustment amounts is formed, with a queryable data table at its core.

[0058] For example, when the wind speed is at the normal level, the baseline adjustment is 0%-5% of the baseline value. When the wind speed is at the stronger level, the baseline adjustment is 5%-15% of the baseline value. When the wind speed is at the stronger level, the baseline adjustment is 15%-30% of the baseline value. Furthermore, the threshold values ​​and adjustment ranges for each interval are calibrated through wind tunnel tests and historical monitoring data, forming a deterministic mapping rule from wind speed parameters to tolerance adjustment amounts.

[0059] A33. After performing min-max normalization on the duration and standard deviation, linear weighted fusion calculation is performed, and the calculation result is input into the Sigmoid function to obtain the attitude prediction compensation factor.

[0060] Understandably, the Sigmoid function can transform linear changes in parameters into risk perception changes that conform to engineering realities, preventing over-adjustment due to continuous parameter growth, thereby ensuring the smoothness and robustness of control decisions.

[0061] It should be added that the duration reflects the cumulative effect of wind load exceeding the limit, while the standard deviation reflects the severity of wind speed fluctuations. Generally, the duration has a higher weight than the standard deviation. The weight of the normalized duration is usually 0.6-0.7, and the weight of the normalized standard deviation is usually 0.3-0.4. The specific weight values ​​can be verified through historical data or calibrated through engineering experience.

[0062] A34. Based on the attitude prediction compensation factor, the adjustment amount of the baseline attitude prediction tolerance is proportionally adjusted, and the final dynamic adjustment amount of wind load on attitude prediction tolerance is output.

[0063] Understandably, multiplying the baseline attitude prediction tolerance adjustment by the attitude prediction compensation factor yields the final output dynamic adjustment.

[0064] A4. Subtract the corresponding dynamic adjustment amount from the attitude prediction baseline tolerance to generate the attitude prediction tolerance under the current operating condition.

[0065] A5. Real-time reading of the current ambient temperature, and calculation of the current analysis temperature value based on historical temperature data within the time window using a time-weighted algorithm.

[0066] Understandably, wind speed, as a dynamic load, has a direct and highly fluctuating impact on the instantaneous attitude of the main girder, requiring real-time monitoring and careful consideration of its fluctuation values ​​and duration. Temperature changes, on the other hand, have a slow and continuous impact on the long-term deformation and cumulative error trend of the main girder. This requires weighted calculations using historical data to reflect its lag effect and trend, thus necessitating greater attention to its trend average to predict and control the long-term error accumulation trend. This allows for precise adaptation to both short-term attitude control and long-term error prediction.

[0067] Furthermore, the specific calculation process of the time-weighted algorithm includes: A51, continuously collecting the ambient temperature within a preset time window and calculating its arithmetic mean as the initial analysis temperature value.

[0068] A52. Based on the interval between the preset time window and the current time point, the time decay weight under the corresponding time window is calculated by the exponential decay function.

[0069] The specific formula for the exponential function is as follows: , For time decay weight, It is a natural constant. The normalized value is the interval between the preset time window and the current time point, after minimum-maximum normalization processing. The decay rate constant determines the rate at which the weight decays over time. The larger the value, the faster the decay, and the more rapidly the importance of historical data declines. Typically, the specific value needs to be calibrated based on engineering experience or historical data analysis.

[0070] A53. Correct the initial analysis temperature value using the time decay weight, and output the corrected initial analysis temperature value.

[0071] Understandably, directly using the arithmetic mean as the initial value may not reflect the time decay characteristics of the data. Since the initial value is calculated based on data from a past period, but these data points are at different distances from the current time point, a time decay weighting can be used to correct for this, giving more weight to more recent data points and making the initial value closer to the current state.

[0072] It is also understandable that temperature data exhibits time correlation and volatility. Directly using the arithmetic mean as the initial value ignores the time decay characteristics of the data points, and the subsequent EWMA algorithm formula itself requires a reasonable initial value to avoid recursion bias. Calculating the time decay weight using the exponential decay function and correcting the initial value allows the time dimension of the data to be taken into account, making the initial value more consistent with the engineering practice where recent data is more important.

[0073] A54. Based on the initial preset sampling period, perform the following iterative calculations: In the first calculation period, use the corrected initial analysis temperature value as the historical input value, denoted as... .

[0074] Real-time reading of current ambient temperature value .

[0075] The new analytical temperature value is calculated using the EWMA algorithm formula. , , This is a preset smoothing factor, between 0 and 1, with a default value of 0.3. Used to control the weights of current measurements and historical values.

[0076] Understandably, the EWMA algorithm assigns higher weights to recent data, enabling it to respond quickly to changes while smoothing out random fluctuations. This is important for temperature data, as temperatures may fluctuate, but we need a smooth, representative value.

[0077] A55. Output the new analysis temperature value obtained in each calculation cycle as the current analysis temperature value, and assign this value to... Iterative calculations for the next calculation cycle.

[0078] It should be noted that EWMA iterative calculations enable recursive updates, dynamically balancing current measurements with historical trends. This filters out random noise in temperature data while simultaneously tracking continuous temperature changes. Consequently, it provides reliable temperature input for subsequent dynamic tolerance adjustments, enhances the ability to perceive changes in environmental loads, and supports precise decision-making during the installation of the cable-stayed bridge's main girder.

[0079] A6. Calculate the relative deviations between the current ambient wind speed and the current analyzed temperature value and their reference ambient values. After performing minimum-maximum standardization on the deviation values, obtain the adjustment ratio of the environmental load on the cumulative error trend tolerance by linear weighted summation.

[0080] Understandably, the weighting is primarily based on the different mechanisms and sensitivities of wind load and temperature load on the cumulative error trend of the main girder installation in cable-stayed bridges. Generally, the thermal expansion and contraction effect caused by temperature is the dominant factor leading to long-term, systematic drift in the main girder's alignment, and therefore it is given a higher weight, such as 0.6-0.7. Wind speed, due to its greater volatility, has a greater impact on instantaneous attitude, but its contribution to the long-term cumulative trend is relatively minor, resulting in a relatively lower weight, such as 0.3-0.4. Furthermore, the specific weight ratio can be determined jointly through finite element model simulation analysis and regression analysis of historical installation data.

[0081] A7. Multiply the adjustment ratio by the cumulative error trend baseline tolerance to output the cumulative error trend tolerance.

[0082] Understandably, wind loads are highly impactful and random. Their dynamic adjustments are directly superimposed on the attitude prediction baseline tolerance in absolute value form, allowing for rapid response to instantaneous changes and quantification of safety margins. In contrast, the impact of temperature loads is relatively gradual. By standardizing deviation values ​​and adjusting the cumulative error trend tolerance using weighted proportions, the proportional impact of temperature fluctuations on long-term errors can be more reasonably reflected, avoiding abrupt changes in absolute values ​​and ensuring the smoothness and stability of trend analysis. This approach not only guarantees the real-time reliability of attitude prediction under wind loads but also optimizes the long-term predictive capability of cumulative error trends under temperature changes, further enhancing the overall safety and efficiency of construction control and reducing the risk of installation deviations caused by environmental fluctuations.

[0083] It should be noted that by constructing a two-level correction architecture of benchmark adjustment and compensation factor, and decoupling and re-integrating the steady-state influence of wind load determined by wind speed intensity with the dynamic time-varying influence determined by duration and volatility, it is possible not only to quickly respond to the main intensity effect of wind load through benchmark adjustment, but also to finely characterize the amplification effect of the time-varying characteristics of wind load on cumulative risk through compensation factor. This enables a hierarchical and refined assessment of structural safety risks in complex wind field environments, making the tolerance boundary more consistent with the actual physical process.

[0084] S3. Based on the joint feature vector, output the future attitude prediction result of the current installation segment and the cumulative error trend evaluation result of the overall installation alignment through time-series prediction.

[0085] Specifically, please refer to Figure 5 , Figure 5 The overall process of time series prediction is shown, and the prediction process using a pre-trained machine learning model is presented in general terms, including steps S31-S34.

[0086] S31. Import the displacement, stress, and tilt angle values ​​from the joint feature vector into the pre-trained random forest regression model, and output the predicted displacement, stress, and tilt angle values ​​for each preset key time point in the future.

[0087] S32. Based on the state parameters of the hoisting equipment and the rate of change of cable force, the above predicted values ​​are corrected, and the corrected results are used as the attitude prediction results.

[0088] S33. Perform first-order difference calculation on the attitude prediction result to obtain the instantaneous rate of change of the cumulative error, and obtain the cumulative error rate of change curve by curve fitting.

[0089] S34. Extract the slope and intercept from the curve, and calculate the probability that the cumulative error will exceed the threshold in the future preset time period based on the slope and intercept. At the same time, use the slope as the error growth rate, and combine the probability and the error growth rate to obtain the cumulative error trend evaluation result.

[0090] It should be noted that, in one specific embodiment, the training data for the random forest regression model comes from historical bridge installation projects, such as monitoring data including 150 segments, and the model parameters are optimized through grid search. The grid search optimization is an existing optimization algorithm, which will not be described in detail here.

[0091] For example, the training process of the random forest regression model includes: evaluating the contribution of each feature to the prediction result using the Gini index, selecting the top 60% of features by importance as model input, and using a grid search method to determine the optimal number of trees, maximum depth, and minimum number of leaf node samples. It also ensures that the average absolute error of the displacement prediction is less than 2 mm and the average relative error of the stress prediction is less than 5%.

[0092] The training process includes collecting joint feature vector data from the installation process of 150 main beam segments of three similar cable-stayed bridges in the past as a training set, with approximately 3,600 time steps for each segment.

[0093] Using RandomForestRegressor from the Scikit-learn library, set n_estimators=200 and max_depth=15.

[0094] The model input is the joint feature vector at time point T, and the output is the displacement, stress, and tilt angle values ​​at three key future time points. The output is the model's mean absolute error on the test set.

[0095] The joint feature vector V is input into a pre-trained random forest model, and the model outputs predicted values ​​for three key future time points.

[0096] Preferably, the key time points include the expected completion time of the current process, the start time of tensioning the next pair of stay cables, and the time when the structure tends to stabilize.

[0097] Understandably, the first preset critical time point is determined based on the standard operating time of the current hoisting process, such as precise positioning and temporary locking, to ensure that the prediction is synchronized with the construction progress. The third critical time point is determined based on the stabilization time of time-varying effects such as concrete shrinkage and creep, and prestress loss, and is usually 2-6 hours after installation to ensure that the prediction takes into account long-term effects.

[0098] Furthermore, the process of correcting the above-mentioned predicted value includes: 1) real-time acquisition of the status parameters of the hoisting equipment, including the rotational speed fluctuation coefficient and pressure stability coefficient of the hoisting equipment, as well as the height deviation of the hoisting point.

[0099] 2) Classify the state parameters of the hoisting equipment and the rate of change of cable force according to their physical nature and their different influence mechanisms on displacement, stress, and tilt angle. Specifically, classify the hoisting point height deviation and the rate of change of cable force as the first type of parameters that have a dominant influence on the predicted values ​​of displacement and tilt angle, and classify the rotational speed fluctuation coefficient and the pressure stability coefficient as the second type of parameters that have a dominant influence on the predicted values ​​of stress.

[0100] 3) Based on the parameter classification results, query the displacement correction factor, stress correction factor and tilt correction factor corresponding to each type of parameter from the preset influence factor mapping table.

[0101] Understandably, the mapping table is pre-calibrated through parametric finite element simulation and historical data regression analysis, wherein the correction factor includes positive and negative values, with a range of [-0.3, 0.3], to reflect the increase or decrease effect that may be caused by changes in different parameters.

[0102] For example, the construction process of the mapping table includes: First, by establishing a parameterized finite element model of the main girder installation of a cable-stayed bridge, the theoretical response values ​​of the main girder displacement, stress, and inclination angle under different combinations of hoisting control parameters and cable force change rates are simulated, obtaining the theoretical relationship matrix of the influence of each parameter on the predicted value. Then, a large amount of actual monitoring data from historical construction is collected, and the theoretical relationship matrix is ​​corrected and calibrated using a multiple linear regression method to eliminate errors caused by model simplification and make the mapping relationship more consistent with engineering reality. Finally, the parameter value range is divided into several continuous intervals, and calibrated influence factor values ​​are assigned to each interval, forming a directly queryable mapping table. This mapping table is optimized and updated monthly based on newly added construction data to ensure its continuous adaptability.

[0103] 4) The displacement prediction value, stress prediction value and tilt angle prediction value are corrected and calculated respectively using the correction factor.

[0104] Understandably, the correction factor is applied to correct the predicted value through a linear weighted model, and the specific application formula is as follows: ,in, This represents the corrected predicted value. This represents the original predicted value. For the first Correction factors corresponding to each parameter For parameter numbering, specifically The value is 2. For displacement prediction, the correction factors corresponding to the lifting point height deviation and cable force change rate are mainly applied. For stress prediction, the correction factors corresponding to the rotational speed fluctuation coefficient and pressure stability coefficient are mainly applied. For tilt angle prediction, the correction factor corresponding to the lifting point height deviation is mainly applied. For the first The normalized weight coefficients corresponding to each parameter are dynamically adjusted according to the deviation between the measured value of the parameter and the threshold range; the greater the deviation, the higher the weight.

[0105] Meanwhile, the correction process considers the directionality of influencing factors: when equipment condition parameters deteriorate or cable tension changes exceed safe limits, the correction factor takes a positive value, and the predicted value is increased accordingly to reflect the risk amplification effect. When parameters improve, the correction factor takes a negative value, and the predicted value is decreased accordingly to reflect the condition improvement. This correction model is recalibrated monthly using the least squares method to ensure its continued accuracy.

[0106] Please see Figure 2 As shown, S4, Error Classification Decision: First-level judgment: If any attitude prediction value exceeds the prediction tolerance, an installation deviation warning instruction is generated and fed back to the on-site installation personnel.

[0107] Specifically, generating and feeding back installation deviation warning instructions includes: comparing the predicted longitudinal displacement, predicted lateral displacement, and predicted tilt angle with the generated dynamic attitude prediction tolerance in real time.

[0108] When the predicted value of any key parameter exceeds its corresponding tolerance, a graded early warning message is immediately generated. The early warning message is divided into three levels according to the degree of exceedance: attention, warning, and danger, and is pushed to the smart terminal of the on-site installation personnel in real time via wireless communication network.

[0109] The degree of exceedance is quantified by the relative deviation between the predicted value and the corresponding tolerance. For example, a warning level alert is triggered when the degree of exceedance is between 5% and 10%. A warning level alert is triggered when the degree of exceedance is between 10% and 20%. A danger level alert is triggered when the degree of exceedance exceeds 20%.

[0110] Second-level judgment: If the cumulative error trend assessment results all exceed the cumulative error trend tolerance, then adjustment instructions are generated, including the real-time adjustment plan for the hoisting parameters of the current segment and the pre-correction plan for the subsequent uninstalled segments.

[0111] Specifically, the real-time adjustment scheme for the hoisting parameters of the current segment includes coordinating the real-time elevation of each hoisting point of the current segment based on the adjustment command, and redistributing the cable force of the connected stay cables.

[0112] The pre-correction scheme for subsequent uninstalled segments specifically includes correcting the design installation positioning coordinates of one or more subsequent uninstalled segments based on the cumulative error trend assessment results, or adjusting their pre-camber setting value during the manufacturing stage.

[0113] It should be noted that the real-time adjustment scheme for the hoisting parameters of the current segment is an instantaneous closed-loop control based on predictive feedback. It is used to coordinately fine-tune the elevation of each hoisting point of the segment being installed, and at the same time redistribute or reset the cable force of the connected stay cables. By directly intervening in the mechanical state applied to the segment, it can quickly correct its installation posture deviation.

[0114] In one specific embodiment, taking the example that the elevation of the north front end of the currently installed P5 segment is continuously lower than the predicted tolerance, the specific adjustment scheme is as follows: Instruction 1: Increase the lifting speed of the No. 1 (north side) lifting point of the P5 segment by 3% on the existing basis, and at the same time reduce the lifting speed of the No. 2 (south side) lifting point by 2% to work together to correct the lateral tilt angle of the segment.

[0115] Instruction 2: Temporarily increase the target tension value of the C12 cable, which is already connected to the north side of this segment, from the design value of 2500kN to 2550kN, and immediately execute a tensioning stroke to provide additional elastic support.

[0116] It should be noted that the pre-correction scheme for subsequent uninstalled segments is a forward-looking open-loop compensation based on trend prediction. It mainly corrects the design installation positioning coordinates of one or more subsequent uninstalled segments based on the cumulative error trend assessment results, or adjusts their pre-camber setting value during the manufacturing stage, thereby combating systematic deviations from the root and ensuring that the final bridge alignment meets the design requirements.

[0117] In one specific embodiment, taking the discovery that the overall alignment of the bridge main beam has a cumulative offset of 12mm in the southwest direction as an example, the generated pre-correction scheme is as follows: Instruction 1: Starting from P6, the design installation positioning coordinates of all subsequent uninstalled segments shall be uniformly pre-offset by 5mm in the northeast direction based on the original design value in their plane position.

[0118] Instruction 2: During the prefabrication of segments P7 to P10, the pre-camber should be increased by 10mm from the theoretical value to compensate for the observed overall deflection trend.

[0119] It is important to note that the specific adjustment values ​​in the above instructions are determined based on a comprehensive analysis of bridge structural mechanics model simulations, historical installation data statistical analysis, and expert experience databases. Specifically, a finite element model of the bridge is established to simulate the structural response under different adjustment amounts to determine the effective adjustment range. Furthermore, regression analysis of historical successful cases is used to quantify the relationship between the adjustment amount and the corrective effect. The final specific values ​​are then output based on this quantified relationship, ensuring that they remain within the controllable range of the equipment and the structural safety tolerance.

[0120] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A rapid installation monitoring and analysis method for the main girder of a cable-stayed bridge based on AI technology, characterized in that, The method includes: S1. Real-time acquisition of displacement, stress and environmental load data of the currently installed segment, and synchronous integration of historical monitoring data of adjacent installed segments to construct a global installation dataset. The environmental load data includes environmental wind speed and environmental temperature. S2. Perform spatiotemporal alignment processing on the global installation dataset, extract the joint feature vector representing the instantaneous state of a single point and the cumulative deviation between segments, and dynamically calculate the attitude prediction tolerance and cumulative error trend tolerance under the current working condition based on the real-time acquired environmental load data. The specific steps for extracting the joint feature vector include: The displacement, stress, tilt angle, hoisting equipment status parameters, and cable force change rate of the currently installed segment are read in real time from the distributed sensor network as characteristic data representing the instantaneous state of a single point. The displacement, stress, and tilt values ​​of the currently installed segment are subtracted from the corresponding reference data recorded by one or more adjacent installed segments at the moment of installation and locking to obtain the displacement difference sequence, stress difference sequence, and tilt difference sequence. For each difference sequence, calculate its arithmetic mean and standard deviation over the complete time period from the start of hoisting the current segment to the current moment, and fit its trend slope based on the least squares method; The feature data representing the instantaneous state of a single point are concatenated with the arithmetic mean, standard deviation and trend slope to form the joint feature vector. S3. Based on the joint feature vector, output the future attitude prediction result of the current installation segment and the cumulative error trend evaluation result of the overall installation alignment through time-series prediction; S4. Error grading decision-making: Level 1 Decision: If any attitude prediction value exceeds the prediction tolerance, an installation deviation warning command is generated and fed back to the on-site installation personnel. Second-level judgment: If the cumulative error trend assessment results all exceed the cumulative error trend tolerance, then adjustment instructions are generated, including the real-time adjustment plan for the hoisting parameters of the current segment and the pre-correction plan for the subsequent uninstalled segments.

2. The rapid installation monitoring and analysis method for the main girder of a cable-stayed bridge based on AI technology as described in claim 1, characterized in that: The displacement, stress, and environmental load data are acquired through a distributed sensor network deployed on the main beam segments and hoisting equipment. The sensor network includes multiple types of GNSS receivers, tilt sensors, strain gauges, and anemometers.

3. The rapid installation monitoring and analysis method for the main girder of a cable-stayed bridge based on AI technology as described in claim 1, characterized in that: The specific calculation process for the attitude prediction tolerance and the cumulative error trend tolerance includes: Based on bridge design specifications, obtain the attitude prediction baseline tolerance and cumulative error trend baseline tolerance under baseline environmental conditions; Real-time reading of current ambient wind speed, statistics on the duration of wind speed exceeding the benchmark ambient wind speed, and the standard deviation of the current ambient wind speed; Based on the current environmental wind speed, duration, and standard deviation, the dynamic adjustment amount of wind load on attitude prediction tolerance is output through preset wind load adjustment statistical rules. Subtract the corresponding dynamic adjustment amount from the attitude prediction baseline tolerance to generate the attitude prediction tolerance under the current operating condition. The system reads the current ambient temperature in real time and calculates the current analysis temperature value based on historical temperature data within a time window using a time-weighted algorithm. Calculate the relative deviations of the current ambient wind speed and the current analyzed temperature value from their reference ambient values. After performing minimum-maximum standardization on the deviation values, obtain the adjustment ratio of the environmental load on the cumulative error trend tolerance by weighted summation. The cumulative error trend tolerance is output by multiplying the adjustment ratio by the cumulative error trend baseline tolerance.

4. The rapid installation monitoring and analysis method for the main girder of a cable-stayed bridge based on AI technology as described in claim 3, characterized in that: The specific details of the preset wind load adjustment statistics rules are as follows: The current ambient wind speed is compared with the preset wind speed intensity level thresholds corresponding to each wind speed intensity level to determine its wind speed intensity level. The baseline attitude prediction tolerance adjustment amount corresponding to the wind speed intensity level is obtained according to the preset mapping relationship; After normalizing the duration and standard deviation, a linear weighted fusion calculation is performed, and the calculation result is input into the Sigmoid function to obtain the attitude prediction compensation factor. Based on the attitude prediction compensation factor, the adjustment amount of the baseline attitude prediction tolerance is proportionally adjusted, and the final dynamic adjustment amount of wind load on the attitude prediction tolerance is output.

5. The rapid installation monitoring and analysis method for the main girder of a cable-stayed bridge based on AI technology as described in claim 3, characterized in that: The specific calculation process of the time-weighted algorithm includes: The ambient temperature is continuously collected within a preset time window, and its arithmetic mean is calculated as the initial analysis temperature value. Based on the interval between the preset time window and the current time point, the time decay weight under the corresponding time window is calculated by the exponential decay function; The initial analysis temperature value is corrected by the time decay weight, and the corrected initial analysis temperature value is output. The following iterative calculations are performed based on the initial preset sampling period: In the first calculation cycle, the corrected initial analysis temperature value is used as the historical input value, denoted as... ; Real-time reading of current ambient temperature value ; The new analytical temperature value is calculated using the EWMA algorithm formula. , , Set as the preset smoothing factor; The new analytical temperature value obtained in each calculation cycle is output as the current analytical temperature value, and this value is assigned to... Iterative calculations for the next calculation cycle.

6. The rapid installation monitoring and analysis method for the main girder of a cable-stayed bridge based on AI technology as described in claim 1, characterized in that: The time series prediction is achieved through a pre-trained machine learning model, and the specific process includes: The displacement, stress, and tilt angle values ​​from the joint feature vector are imported into a pre-trained random forest regression model, which outputs the predicted displacement, stress, and tilt angle values ​​at each preset key time point in the future. The above predicted values ​​are corrected based on the state parameters of the hoisting equipment and the rate of change of cable force, and the corrected results are used as the attitude prediction results. The attitude prediction results are subjected to first-order difference calculation to obtain the instantaneous rate of change of the cumulative error, and the cumulative error rate of change curve is obtained by curve fitting. The slope and intercept are extracted from the curve, and the probability of the cumulative error exceeding the threshold in the future preset time period is calculated based on the slope and intercept. At the same time, the slope is used as the error growth rate. The cumulative error trend evaluation result is obtained by combining the probability and the error growth rate.

7. The rapid installation monitoring and analysis method for the main girder of a cable-stayed bridge based on AI technology as described in claim 6, characterized in that: The key time points include the expected completion time of the current process, the start time of tensioning the next pair of stay cables, and the time when the structure tends to stabilize.

8. The rapid installation monitoring and analysis method for the main girder of a cable-stayed bridge based on AI technology as described in claim 6, characterized in that: The process of correcting the above-mentioned predicted values ​​includes: Real-time acquisition of hoisting equipment status parameters, including the hoisting equipment's rotational speed fluctuation coefficient, pressure stability coefficient, and hoisting point height deviation; The hoisting point height deviation and cable force change rate are classified as first-class parameters that have a dominant influence on displacement prediction and tilt angle prediction, and the rotation speed fluctuation coefficient and pressure stability coefficient are classified as second-class parameters that have a dominant influence on stress prediction. Based on the parameter classification results, the displacement correction factor, stress correction factor and tilt correction factor corresponding to each type of parameter are queried from the preset influence factor mapping table; The displacement prediction value, stress prediction value, and tilt angle prediction value are corrected and calculated using the correction factors respectively.

9. The rapid installation monitoring and analysis method for the main girder of a cable-stayed bridge based on AI technology as described in claim 1, characterized in that: The real-time adjustment scheme for the hoisting parameters of the current segment specifically includes coordinating the real-time elevation of each hoisting point of the current segment based on the adjustment command, and redistributing the cable force of the connected stay cables; The pre-correction scheme for subsequent uninstalled segments specifically includes correcting the design installation positioning coordinates of one or more subsequent uninstalled segments based on the cumulative error trend assessment results, or adjusting their pre-camber setting value during the manufacturing stage.

Images