A bridge preventive maintenance decision system based on digital twinning

Abstract

This invention relates to the field of bridge structural health monitoring technology, specifically disclosing a bridge preventive maintenance decision-making system based on digital twins. The system collects bridge deck strain, ambient temperature, and vehicle load response data; constructs a slow-time state vector composed of nodal temperature and local thermal expansion coefficient after time-domain averaging of the strain; outputs a temperature strain baseline value through statistical assimilation; obtains residual strain by subtracting the temperature strain baseline value from the original strain; tracks crack width and its rate of change using particle swarm sequential resampling; performs sign consistency judgment on crack width and load data to identify true cracks and output their evolution trends; when the crack width exceeds a threshold and its duration exceeds a preset window, a maintenance instruction including location, evolution trend, and repair timing is generated. This invention effectively separates temperature strain from crack strain, reduces misjudgments and omissions, and provides quantifiable preventive maintenance decision-making basis.

Description

Technical Field

[0001] This invention relates to the field of bridge structural health monitoring technology, specifically to a bridge preventive maintenance decision-making system based on digital twins. Background Technology

[0002] During operation, concrete bridge decks of steel-concrete composite beam bridges are prone to cracking. Accurate monitoring of crack width is a prerequisite for implementing preventative maintenance decisions. Existing technologies typically use fiber optic strain sensors to collect bridge deck strain data and employ digital twin models to simulate and predict the structural condition. To identify crack strain, some methods attempt to fuse measured strain with finite element models using data assimilation techniques such as Kalman filtering or particle filtering, in order to separate the strain components caused by temperature changes and those caused by crack opening.

[0003] In steel-concrete composite beam bridges, due to the difference in the thermal expansion coefficients of the two materials and the change in the local heat conduction path after the appearance of cracks, the temperature strain exhibits a non-uniform and non-linear multi-peak distribution. This makes it impossible for traditional data assimilation methods based on the Gaussian assumption to distinguish between the actual crack strain and non-Gaussian temperature interference, resulting in missed crack detection during high-temperature periods and false crack detection during low-temperature periods. Maintenance personnel are unable to accurately obtain the true evolution trend of early cracks. Summary of the Invention

[0004] The purpose of this invention is to provide a bridge preventive maintenance decision system based on digital twins to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions: A bridge preventive maintenance decision-making system based on digital twins includes: The data acquisition module collects strain data, ambient temperature data, and vehicle load response data of the steel-concrete composite beam bridge deck as the raw input set. The slow-time assimilation module averages the strain data over a fixed time window to obtain the time-averaged strain, constructs a slow-time state vector composed of the temperature of each finite element node and the local thermal expansion coefficient, and recursively matches the time-averaged strain with the ambient temperature data through slow-time statistical assimilation, outputting the temperature strain baseline value and the corrected thermal expansion coefficient. The residual tracking module subtracts the temperature strain baseline value from the strain data in the original input set time by time to obtain the residual strain. Based on the residual strain and vehicle load response data, it uses particle swarm sequential resampling to dynamically track the crack width and its rate of change, and outputs the time-series estimate of the crack width. The crack identification module performs synchronous correlation and consistency judgment between the estimated crack width time series value and the vehicle load response data. When the crack width exceeds the instantaneous threshold at multiple consecutive times and the correlation coefficient with the vehicle load is higher than the set value, it is determined to be a real crack, and the confirmed crack width and evolution trend are output. The maintenance decision module generates a preventive maintenance decision instruction when the confirmed crack width exceeds the maintenance intervention threshold and the duration exceeds the preset window. The preventive maintenance decision instruction includes the location of the crack, the width evolution trend, and the repair timing.

[0006] As a further aspect of the present invention: the output temperature strain baseline value and the corrected coefficient of thermal expansion specifically include: The initial temperature values ​​of each finite element node are derived from the average strain over the time period, and the initial temperature values ​​are combined with the initial values ​​of the local thermal expansion coefficients to form a slow-time state vector. Using ambient temperature data as the driving condition, heat conduction recursion is performed on the temperature of each node in the slow-time state vector to obtain the recursed node temperature. The theoretical strain is calculated using the recursive node temperature. The difference between the theoretical strain and the average strain over a time period is obtained. The local thermal expansion coefficient is recursively adjusted based on the difference until the difference converges. The temperature strain baseline value and the corrected thermal expansion coefficient are then output.

[0007] As a further aspect of the present invention: the process of constructing the slow-time state vector is as follows: Extract the difference sequence of average strain of three adjacent time periods for each finite element node, and locate the temperature fluctuation boundary node according to the sign rate of change of the difference sequence; The temperature disturbance amplitude is inferred from the strain amplitude of the boundary nodes, and the temperature of each node is spatially interpolated and corrected using the structural thermal conduction constraint to obtain the initial temperature value of each finite element node. The initial temperature value is vector-grouped with the initial local thermal expansion coefficient of the corresponding node to form a slow-time state vector arranged in node order.

[0008] As a further aspect of the present invention: the method of inferring the temperature disturbance amplitude from the strain amplitude of the boundary nodes and using structural thermal conduction constraints to perform spatial interpolation correction on the temperature of each node to obtain the initial temperature value of each finite element node specifically includes: For each boundary node, its strain amplitude is multiplied by the reciprocal of the pre-stored thermal expansion coefficient to obtain the temperature disturbance amplitude of the boundary node, and the temperature disturbance amplitude is superimposed on the ambient temperature reference value as the initial value of the boundary temperature. With the initial temperature values ​​of all boundary nodes as constraints, the temperature influence is transmitted layer by layer to the internal nodes along the structural heat conduction path. During the transmission at each layer, the temperature increment is allocated according to the thermal resistance weight between adjacent nodes. The temperature increment received by the internal node is added to the temperature of the previous iteration, and the consistency of the temperature gradient on both sides of the node is checked under the condition of heat flow continuity until the temperature gradient deviation of all nodes is less than the preset tolerance, and the initial temperature value of each finite element node is output.

[0009] As a further aspect of the present invention: the output crack width time-series estimate specifically includes: The instantaneous ratio of the residual strain at the current moment to the vehicle load response data is calculated, and the initial values ​​of the crack width and rate of change carried by each particle are initialized based on the instantaneous ratio. Using vehicle load response data as the driving condition, a first-order hold-behind recursion is performed on the crack width and rate of change of each particle to obtain the predicted crack width and rate of change. The likelihood of each predicted crack width is calculated using residual strain. The particle swarm is discarded or resampled according to the likelihood. The center crack width of the resampled particle swarm is output as the time series estimate of the crack width at the current moment.

[0010] As a further aspect of the present invention: obtaining the predicted crack width and rate of change specifically includes: Based on the instantaneous amplitude of the vehicle load response data at the current moment, calculate the load driving coefficient of each particle, and multiply the load driving coefficient by the time step to obtain the width recursive step. Add the product of the rate of change and the time step to the current crack width of each particle, and then add the width recursion step size to obtain the predicted crack width. The difference between the predicted crack width and the current crack width is taken, divided by the time step to obtain the updated rate of change. The predicted crack width and the updated rate of change are then combined to form the initial recursive value for the next time step.

[0011] As a further aspect of the present invention: the output confirmed crack width and evolution trend specifically include: Align the crack width time-series estimate with the vehicle load response data on the same time axis, calculate the crack width increment at each time step relative to the previous time step, and simultaneously calculate the load increment of the vehicle load response data. The width increment and load increment at multiple consecutive time points are compared for sign consistency. The proportion of time points with the same sign is used as the correlation coefficient. When the correlation coefficient is higher than the set value and the crack width at each time point exceeds the instantaneous threshold, it is determined to be a real crack. The crack width at the last moment in a series of consecutive moments is taken as the confirmed crack width, and the average width change rate per unit time is calculated based on the crack width sequence of consecutive moments as the evolution trend output.

[0012] As a further aspect of the present invention: the generation of preventive maintenance decision instructions specifically includes: The moment when the confirmed crack width exceeds the maintenance intervention threshold is recorded as the damage initiation moment. The duration is calculated with the current moment as the end point. When the duration exceeds the preset window, the command is triggered to generate. Based on the location of the crack, the pre-stored structural safety margin table is called, and the remaining time required for the crack width to reach the limit allowable width is calculated in combination with the width evolution trend. Half of the remaining time is taken as the repair time. The location of the crack, its width evolution trend, and the timing of repair are encoded into preventive maintenance decision instructions, and the repair timing in the instructions is converted into the corresponding calendar date for output.

[0013] The beneficial effects of this invention are: (1) Through the hierarchical processing of slow-time assimilation and particle group sequential resampling, temperature strain and crack strain can be effectively distinguished. Combined with vehicle load response data, sign consistency judgment is performed, which significantly reduces the misjudgment and omission of crack width caused by temperature gradient changes. This enables maintenance personnel to accurately identify early cracks that actually need intervention and avoid unnecessary on-site verification or delayed repair.

[0014] (2) Based on the confirmed trend of crack width evolution and structural safety margin, the remaining time required for the crack to reach the limit allowable width is automatically calculated, and the repair timing is converted into a specific calendar date output, providing clear location, time and trend information for bridge maintenance management, making preventive maintenance decisions quantifiable and executable, and reducing the uncertainty caused by relying on experience judgment. Attached Figure Description

[0015] The invention will now be further described with reference to the accompanying drawings.

[0016] Figure 1 This is a system block diagram of the present invention; Figure 2 This is a flowchart illustrating the construction process of the slow-time state vector in this invention. Detailed Implementation

[0017] 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.

[0018] Please see Figure 1 As shown, this invention is a bridge preventive maintenance decision-making system based on digital twins, comprising: The data acquisition module collects strain data, ambient temperature data, and vehicle load response data of the steel-concrete composite beam bridge deck as the raw input set. The slow-time assimilation module averages the strain data over a fixed time window to obtain the time-averaged strain, constructs a slow-time state vector composed of the temperature of each finite element node and the local thermal expansion coefficient, and recursively matches the time-averaged strain with the ambient temperature data through slow-time statistical assimilation, outputting the temperature strain baseline value and the corrected thermal expansion coefficient. The residual tracking module subtracts the temperature strain baseline value from the strain data in the original input set time by time to obtain the residual strain. Based on the residual strain and vehicle load response data, it uses particle swarm sequential resampling to dynamically track the crack width and its rate of change, and outputs the time-series estimate of the crack width. The crack identification module performs synchronous correlation and consistency judgment between the estimated crack width time series value and the vehicle load response data. When the crack width exceeds the instantaneous threshold at multiple consecutive times and the correlation coefficient with the vehicle load is higher than the set value, it is determined to be a real crack, and the confirmed crack width and evolution trend are output. The maintenance decision module generates a preventive maintenance decision instruction when the confirmed crack width exceeds the maintenance intervention threshold and the duration exceeds the preset window. The preventive maintenance decision instruction includes the location of the crack, the width evolution trend, and the repair timing.

[0019] The data acquisition module collects strain data, ambient temperature data, and vehicle load response data of the steel-concrete composite beam bridge deck as the raw input set, specifically including: Multiple fiber Bragg grating strain sensors were pre-deployed on the concrete bridge deck of the steel-concrete composite beam bridge to be monitored. One fiber Bragg grating strain sensor was placed every 5 meters along the longitudinal direction of the bridge, and three were evenly distributed transversely at each monitoring section, each affixed to the underside of the bridge deck. All fiber Bragg grating strain sensors continuously acquired strain data at a sampling frequency of 10 Hz, which was then converted into digital signals by a fiber optic demodulator and transmitted to the data aggregation terminal.

[0020] A platinum resistance temperature sensor is installed adjacent to each fiber Bragg grating strain sensor, close to the surface of the bridge deck, to synchronously acquire ambient temperature data at that measurement point. The platinum resistance temperature sensor also operates at a sampling frequency of 10 Hz, with a measurement accuracy of 0.1 degrees Celsius. Temperature and strain data are time-synchronized using the same data acquisition instrument, ensuring that the strain data at each moment has a corresponding temperature data for comparison.

[0021] A dynamic weighing system is installed beneath the lanes on the bridge's approach side to collect vehicle load response data. This system includes piezoelectric sensors and inductive loops embedded in the road surface, enabling real-time detection of axle load, number of axles, vehicle speed, and passage time. The system outputs an instantaneous axle load sequence for each vehicle and displays the instantaneous load pressure waveform of the lane at a frequency of 20 Hz. This vehicle load response data, along with strain and temperature data, is timestamped and combined to form the raw input set, which is stored in a real-time database on a data server.

[0022] Please see Figure 2 As shown, in the slow-time assimilation module, the strain data is averaged over a fixed time window to obtain the time-averaged strain. A slow-time state vector is constructed, consisting of the temperature of each finite element node and the local thermal expansion coefficient. The time-averaged strain is recursively matched with the ambient temperature data through slow-time statistical assimilation, and the temperature strain baseline value and the corrected thermal expansion coefficient are output. Specifically, this includes: The strain data is averaged over a fixed time window to obtain the time-domain average strain. The fixed time window is 900 seconds; that is, for each fiber Bragg grating strain sensor, the strain data is continuously acquired, and the sum of all strain data points within the window is divided by the number of points to obtain the time-domain average strain for that window. Adjacent windows do not overlap, and the time-domain average strain for each window corresponds to a specific time point. Simultaneously, the ambient temperature data is averaged over the same 900-second window to obtain the time-domain average ambient temperature.

[0023] A slow-time state vector is constructed, consisting of the temperature and local thermal expansion coefficient of each finite element node. The time-interval average strain difference sequence of three adjacent windows for each finite element node is extracted. Let the... The node at the th The average strain over the time period of each window is Calculate the difference between adjacent windows Take three consecutive differences. , , The number of times the product of two adjacent differences is negative is counted, and this number is divided by 2 to obtain the sign rate of change. When the sign rate of change is greater than 0.5, the node is marked as a temperature fluctuation boundary node.

[0024] For each boundary node, the difference between the maximum and minimum average strain over the next three windows is taken as the strain amplitude. Pre-store the initial value of the thermal expansion coefficient of the material at this node. Calculate the amplitude of temperature disturbance The ambient temperature reference value (The arithmetic mean of all platinum resistance temperature sensor readings at 0:00 on the same day) is added to the temperature disturbance amplitude to obtain the initial value of the boundary temperature. .

[0025] Spatial interpolation correction of the temperature at each node is performed using structural heat conduction constraints. Using the initial temperature values ​​of all boundary nodes as constraints, the temperature influence is transmitted layer by layer along the structural heat conduction path to the internal nodes. The thermal resistance weighting coefficient between adjacent nodes is calculated using the following formula: ;in, The straight-line distance between two adjacent finite element nodes is expressed in meters. The reference thermal diffusion length is taken as 5 meters; This is the distribution coefficient for transferring temperature from the boundary node to the internal nodes. The value is between 0 and 1. For each internal node, collect the temperature increments from all its neighboring nodes with known temperatures. The temperature increment from each neighboring node is equal to the initial temperature value of that neighboring node multiplied by itself. Then divide by the number of adjacent nodes of that internal node. Sum all the incoming temperature increments to obtain the initial temperature value of that internal node.

[0026] After adding the temperature increment received by the internal node to the temperature of the previous iteration, the consistency of the temperature gradient on both sides of the node is checked. The specific check method is as follows: For each element edge connecting adjacent nodes, the temperature difference between the two nodes is calculated and divided by the distance to obtain the gradient. Then, the temperature gradient difference between the elements on both sides of that edge is calculated. If the absolute value of the gradient difference for all edges is less than 0.01 degrees Celsius per meter, the gradients are considered consistent. If not, the gradient difference is distributed in reverse to the temperatures of the nodes on both sides, the temperature increment is recalculated, and the above layer-by-layer propagation and check steps are repeated until the temperature gradient deviation of all nodes is less than the preset tolerance of 0.01 degrees Celsius per meter. Finally, the initial temperature value of each finite element node is output.

[0027] The initial temperature value and the initial local thermal expansion coefficient of each finite element node are vectorized and arranged in ascending order of node number to form a slow-time state vector. The initial local thermal expansion coefficient is taken from the factory calibration value of the node material; for concrete nodes, it is taken from... For each degree Celsius, the steel joint is taken as follows: Every degree Celsius.

[0028] Using the average ambient temperature over a time period as the driving condition, heat conduction recursion is performed on the temperature of each node in the slow-time state vector. The recursion method employs an explicit finite difference method, with a time step of 900 seconds and a spatial step equal to the distance between adjacent nodes. According to the heat conduction equation, the new temperature of each node is equal to its old temperature plus the product of the temperature difference between adjacent nodes, the thermal diffusivity, and the time step, divided by the square of the spatial step. The thermal diffusivity is taken as [value missing] for steel. square meters per second, concrete is Square meters per second. The node temperature is obtained after the recursion is completed.

[0029] Theoretical strain is calculated using the recursive nodal temperature. The theoretical strain equals the recursive nodal temperature minus the reference temperature (20 degrees Celsius), multiplied by the initial value of the nodal's coefficient of thermal expansion. The absolute value δ of the difference between the theoretical strain and the time-averaged strain is calculated. The convergence factor is defined as follows: Where δ is in units of microstrain; is a dimensionless convergence factor. When If the difference is greater than 0.99, the difference is considered convergent; otherwise, depending on the sign of the difference, the local thermal expansion coefficient of the current node is increased or decreased by 1% of its current value, and the heat conduction recursion and theoretical strain calculation are performed again. This process is repeated until... Greater than 0.99. At this point, the recursive node temperature and thermal expansion coefficient at the final convergence are used as outputs, where the theoretical strain calculated from the recursive node temperature is the temperature strain baseline value, and the final thermal expansion coefficient is the corrected thermal expansion coefficient.

[0030] In the residual tracking module, the temperature strain baseline value is subtracted from the strain data in the original input set time-by-time to obtain the residual strain. Based on the residual strain and vehicle load response data, particle swarm sequential resampling is used to dynamically track the crack width and its rate of change, and outputs a time-series estimate of the crack width, specifically including: The residual strain is obtained by subtracting the temperature strain baseline value from the strain data in the original input set time-by-time. The strain data in the original input set is sampled at a frequency of 10 Hz, with one strain value corresponding to each sampling time. The temperature strain baseline value is output from the aforementioned slow-time statistical assimilation step, and this baseline value is given at 1-second intervals. For each sampling time, the original strain value at that time is subtracted from the temperature strain baseline value at the same time, and the difference is the residual strain. The residual strain reflects the strain component caused by crack opening and closing and vehicle load after removing the influence of temperature.

[0031] Based on residual strain and vehicle load response data, a particle swarm sequential resampling method is used to dynamically track crack width and its rate of change. The particle swarm consists of multiple particles carrying state information; each particle contains an initial value for the crack width and an initial value for the rate of change. The number of particles is fixed at 50. First, the particles are initialized: the residual strain at the current moment is divided by the instantaneous load pressure value in the vehicle load response data to obtain an instantaneous ratio. The instantaneous load pressure value is taken from the lane instantaneous load pressure waveform output by the dynamic weighing system at that moment. This instantaneous ratio is used as the benchmark for the initial crack width value, and an initial crack width value is assigned to each particle. Specifically, 50 initial crack width values ​​are randomly generated using a uniform distribution, with the mean of this instantaneous ratio and a variance of 10%. The initial rate of change value for each particle is set to zero.

[0032] Using vehicle load response data as the driving condition, a first-order hold-and-forth recursion is performed on the crack width and rate of change of each particle to obtain the predicted crack width and rate of change. The recursion time step is set to 0.1 seconds, consistent with the data sampling interval. First, based on the instantaneous amplitude of the vehicle load response data at the current moment, the load driving coefficient of each particle is calculated. The instantaneous amplitude is taken as the instantaneous lane load pressure waveform value output by the dynamic weighing system. This value is divided by the reference load value (the reference load value is taken as the standard vehicle axle load of 50 kN) to obtain the dimensionless load ratio. The load driving coefficient of each particle is equal to this load ratio multiplied by 0.01. The load driving coefficient is multiplied by the time step of 0.1 seconds to obtain the width recursion step.

[0033] The predicted crack width is obtained by adding the product of the crack width of each particle at the current moment and the time step of 0.1 seconds, and then summing this product with the aforementioned width recursion step size. The difference between the predicted crack width and the current crack width is then taken and divided by the time step of 0.1 seconds to obtain the updated rate of change. The predicted crack width and the updated rate of change are combined as the initial value for the recursion at the next moment.

[0034] The likelihood of the predicted crack width for each particle is calculated using residual strain. The likelihood measures the degree of match between the predicted crack width and the current observed value (i.e., residual strain). Specifically, for each particle, its predicted crack width is multiplied by a preset gauge length (0.1 meters, taken from the gauge length of a fiber optic grating sensor) to obtain the theoretical crack strain. The absolute difference between this theoretical crack strain and the residual strain is calculated. The reciprocal of this absolute difference is used as the preliminary likelihood. The preliminary likelihoods of all particles are then summed, and the preliminary likelihood of each particle is divided by this sum to obtain the normalized likelihood.

[0035] The particle swarm is discarded and resampled based on likelihood. The resampling method uses a conventional roulette wheel selection algorithm: a random number between 0 and 1 is generated, and the selected particle is determined according to the cumulative likelihood sequence of each particle. Particles with higher likelihood have a higher probability of being replicated, while particles with lower likelihood are discarded. This selection is repeated 50 times to obtain a new particle swarm, where the crack width and rate of change of each particle are derived from the selected particles. The arithmetic mean of the crack widths of all particles in the resampled swarm is calculated to obtain the central crack width, which is then output as the time-series estimate of the crack width at the current time step. This process is repeated sequentially at each time step to obtain the crack width estimate over a continuous time series.

[0036] In the crack identification module, the temporal estimate of crack width is synchronously correlated with vehicle load response data for consistency determination. When the crack width exceeds the instantaneous threshold for multiple consecutive time periods and the correlation coefficient with vehicle load is higher than a set value, it is determined to be a real crack, and the confirmed crack width and evolution trend are output, specifically including: The crack width time-series estimate is aligned with the vehicle load response data along the same time axis. The crack width time-series estimate is output at a frequency of 10 Hz, with one crack width value corresponding to each moment. The instantaneous load pressure waveform in the vehicle load response data is also output at a frequency of 10 Hz. The two are matched one-to-one according to the sampling time to ensure that the crack width and load pressure value appear in pairs at the same moment.

[0037] Calculate the crack width increment at each moment relative to the previous moment. Let the current moment be a multiple of 0.1 seconds from the t-th moment, and the previous moment be t minus 0.1 seconds. Subtract the crack width from the previous moment's crack width at the current moment to obtain the width increment, in millimeters. Simultaneously calculate the load increment of the vehicle load response data: subtract the instantaneous load pressure value from the previous moment's instantaneous load pressure value at the current moment (in kilonewtons) to obtain the load increment.

[0038] Five consecutive time intervals (0.5 seconds each) are used to obtain the width increment sequence and load increment sequence. For each of these five time intervals, the signs of the width increment and load increment are compared. If both are positive or both are negative, they are marked as having the same sign; if one is positive and the other negative, or if either is zero, they are marked as having different signs. The number of times the signs are consistent is counted and divided by 5 to obtain the correlation coefficient. A correlation coefficient threshold of 0.8 is set. Simultaneously, an instantaneous threshold of 0.05 mm is set; that is, a crack width exceeding 0.05 mm instantaneously is considered to have exceeded the threshold.

[0039] When the crack width exceeds 0.05 mm at each of the five consecutive time points, and the correlation coefficient of these five time points is greater than or equal to 0.8, the current crack is determined to be a real crack, rather than a false alarm caused by temperature disturbance or measurement noise.

[0040] After a crack is identified as a real crack, the crack width at the last of the five consecutive time points is taken as the confirmed crack width, in millimeters. The evolution trend is calculated based on this sequence of crack widths over five consecutive time points: the difference between the crack widths at the first and last time points is divided by 0.5 seconds (the total duration of the five time points) to obtain the average width change rate, in millimeters per second. This average width change rate is output as the evolution trend. A positive trend indicates that the crack width is increasing; a negative trend indicates that the crack width is decreasing. The confirmed crack width and evolution trend are output for subsequent maintenance decisions.

[0041] In the maintenance decision-making module, when the confirmed crack width exceeds the maintenance intervention threshold and the duration exceeds a preset window, a preventative maintenance decision instruction is generated. This instruction includes the crack's location, width evolution trend, and repair timing, specifically: When the confirmed crack width exceeds the maintenance intervention threshold and the duration exceeds the preset window, a preventative maintenance decision instruction is generated. The maintenance intervention threshold is set to 0.08 mm, meaning a crack width greater than 0.08 mm is considered to have exceeded the threshold. The preset window is set to 30 seconds, meaning the instruction will only be triggered if the crack width continues to exceed the threshold for more than 30 seconds.

[0042] The moment when the confirmed crack width first exceeds 0.08 mm is recorded as the damage initiation moment. From this moment, the confirmed crack width is continuously monitored at each subsequent moment. If the crack width drops back to 0.08 mm or below, timing is interrupted; timing resumes the next time it exceeds 0.08 mm. The duration from the damage initiation moment to the current moment is calculated, in seconds. When the duration exceeds 30 seconds, a command is triggered.

[0043] The system retrieves a pre-stored structural safety margin table based on the crack's location. This table is a pre-stored data table where each row corresponds to a bridge deck grid position and contains the maximum allowable width value for that position, in millimeters. The maximum allowable width is determined according to the crack limit in the bridge design code, and is set to 0.2 millimeters. The remaining time required for the crack width to reach the maximum allowable width is calculated based on the width evolution trend (in millimeters per second). Specifically, the calculation method is as follows: subtract the currently confirmed crack width from the maximum allowable width of 0.2 millimeters to obtain the width margin, in millimeters. Divide the width margin by the absolute value of the width evolution trend to obtain the remaining time, in seconds. If the width evolution trend is negative (crack shrinkage), no calculation is performed, and the remaining time is directly determined to be infinite. Half of the remaining time is rounded down (to the nearest integer second) as the suggested repair timing, in seconds.

[0044] The location of the crack (represented by bridge station number and lateral offset distance), its width evolution trend (retaining three decimal places, in millimeters per second), and the timing of repair (in seconds) are encoded into a preventative maintenance decision instruction. The repair timing is then converted to a corresponding calendar date: based on the current system time, the number of seconds for repair is added to obtain the expected repair calendar date, in the format of year-month-day-hour-minute-second. This instruction is then output for maintenance personnel to execute.

[0045] The working principle of this invention is as follows: Strain data, ambient temperature data, and vehicle load response data of the steel-concrete composite beam bridge deck are collected as the original input set. The strain data is averaged over time to obtain the time-interval average strain. A slow-time state vector composed of finite element node temperatures and local thermal expansion coefficients is constructed. The time-interval average strain and ambient temperature data are recursively matched through slow-time statistical assimilation to output the temperature strain baseline value and the corrected thermal expansion coefficient. The temperature strain baseline value is subtracted from the original strain time-by-time to obtain the residual strain. Based on the residual strain and vehicle load response data, particle swarm sequential resampling is used to dynamically track the crack width and its rate of change, outputting a time-series estimate of the crack width. The time-series estimate of the crack width is then compared with the vehicle load response data for synchronous correlation consistency judgment. When the crack width exceeds the instantaneous threshold for multiple consecutive times and the correlation coefficient with the vehicle load is higher than a set value, it is determined to be a real crack, and the confirmed crack width and evolution trend are output. When the confirmed crack width exceeds the maintenance intervention threshold and the duration exceeds a preset window, a preventive maintenance decision instruction containing the crack location, width evolution trend, and repair timing is generated.

[0046] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A bridge preventive maintenance decision-making system based on digital twins, characterized in that, include: The data acquisition module collects strain data, ambient temperature data, and vehicle load response data of the steel-concrete composite beam bridge deck as the raw input set. The slow-time assimilation module averages the strain data over a fixed time window to obtain the time-averaged strain, constructs a slow-time state vector composed of the temperature of each finite element node and the local thermal expansion coefficient, and recursively matches the time-averaged strain with the ambient temperature data through slow-time statistical assimilation, outputting the temperature strain baseline value and the corrected thermal expansion coefficient. The residual tracking module subtracts the temperature strain baseline value from the strain data in the original input set time by time to obtain the residual strain. Based on the residual strain and vehicle load response data, it uses particle swarm sequential resampling to dynamically track the crack width and its rate of change, and outputs the time-series estimate of the crack width. The crack identification module performs synchronous correlation and consistency judgment between the estimated crack width time series value and the vehicle load response data. When the crack width exceeds the instantaneous threshold at multiple consecutive times and the correlation coefficient with the vehicle load is higher than the set value, it is determined to be a real crack, and the confirmed crack width and evolution trend are output. The maintenance decision module generates a preventive maintenance decision instruction when the confirmed crack width exceeds the maintenance intervention threshold and the duration exceeds the preset window. The preventive maintenance decision instruction includes the location of the crack, the width evolution trend, and the repair timing.

2. The bridge preventive maintenance decision-making system based on digital twins according to claim 1, characterized in that, The output temperature strain baseline value and the corrected coefficient of thermal expansion specifically include: The initial temperature values ​​of each finite element node are derived from the average strain over the time period, and the initial temperature values ​​are combined with the initial values ​​of the local thermal expansion coefficients to form a slow-time state vector. Using ambient temperature data as the driving condition, heat conduction recursion is performed on the temperature of each node in the slow-time state vector to obtain the recursed node temperature. The theoretical strain is calculated using the recursive node temperature. The difference between the theoretical strain and the average strain over a time period is obtained. The local thermal expansion coefficient is recursively adjusted based on the difference until the difference converges. The temperature strain baseline value and the corrected thermal expansion coefficient are then output.

3. The bridge preventive maintenance decision-making system based on digital twins according to claim 2, characterized in that, The process of constructing the slow-time state vector is as follows: Extract the difference sequence of average strain of three adjacent time periods for each finite element node, and locate the temperature fluctuation boundary node according to the sign rate of change of the difference sequence; The temperature disturbance amplitude is inferred from the strain amplitude of the boundary nodes, and the temperature of each node is spatially interpolated and corrected using the structural thermal conduction constraint to obtain the initial temperature value of each finite element node. The initial temperature value is vector-grouped with the initial local thermal expansion coefficient of the corresponding node to form a slow-time state vector arranged in node order.

4. The bridge preventive maintenance decision-making system based on digital twins according to claim 3, characterized in that, The method of inferring the temperature disturbance amplitude from the strain amplitude at the boundary nodes and using structural thermal conduction constraints to perform spatial interpolation correction on the temperature of each node to obtain the initial temperature values ​​of each finite element node specifically includes: For each boundary node, its strain amplitude is multiplied by the reciprocal of the pre-stored thermal expansion coefficient to obtain the temperature disturbance amplitude of the boundary node, and the temperature disturbance amplitude is superimposed on the ambient temperature reference value as the initial value of the boundary temperature. With the initial temperature values ​​of all boundary nodes as constraints, the temperature influence is transmitted layer by layer to the internal nodes along the structural heat conduction path. During the transmission at each layer, the temperature increment is allocated according to the thermal resistance weight between adjacent nodes. The temperature increment received by the internal node is added to the temperature of the previous iteration, and the consistency of the temperature gradient on both sides of the node is checked under the condition of heat flow continuity until the temperature gradient deviation of all nodes is less than the preset tolerance, and the initial temperature value of each finite element node is output.

5. A bridge preventive maintenance decision-making system based on digital twins according to claim 1, characterized in that, The output crack width time-series estimate specifically includes: The instantaneous ratio of the residual strain at the current moment to the vehicle load response data is calculated, and the initial values ​​of the crack width and rate of change carried by each particle are initialized based on the instantaneous ratio. Using vehicle load response data as the driving condition, a first-order hold-behind recursion is performed on the crack width and rate of change of each particle to obtain the predicted crack width and rate of change. The likelihood of each predicted crack width is calculated using residual strain. The particle swarm is discarded or resampled according to the likelihood. The center crack width of the resampled particle swarm is output as the time series estimate of the crack width at the current moment.

6. A bridge preventive maintenance decision-making system based on digital twins according to claim 5, characterized in that, The predicted crack width and rate of change specifically include: Based on the instantaneous amplitude of the vehicle load response data at the current moment, calculate the load driving coefficient of each particle, and multiply the load driving coefficient by the time step to obtain the width recursive step. Add the product of the rate of change and the time step to the current crack width of each particle, and then add the width recursion step size to obtain the predicted crack width. The difference between the predicted crack width and the current crack width is taken, divided by the time step to obtain the updated rate of change. The predicted crack width and the updated rate of change are then combined to form the initial recursive value for the next time step.

7. A bridge preventive maintenance decision-making system based on digital twins according to claim 1, characterized in that, The output confirms the crack width and evolution trend, specifically including: Align the crack width time-series estimate with the vehicle load response data on the same time axis, calculate the crack width increment at each time step relative to the previous time step, and simultaneously calculate the load increment of the vehicle load response data. The width increment and load increment at multiple consecutive time points are compared for sign consistency. The proportion of time points with the same sign is used as the correlation coefficient. When the correlation coefficient is higher than the set value and the crack width at each time point exceeds the instantaneous threshold, it is determined to be a real crack. The crack width at the last moment in a series of consecutive moments is taken as the confirmed crack width, and the average width change rate per unit time is calculated based on the crack width sequence of consecutive moments as the evolution trend output.

8. A bridge preventive maintenance decision-making system based on digital twins according to claim 1, characterized in that, The generation of preventative maintenance decision instructions specifically includes: The moment when the confirmed crack width exceeds the maintenance intervention threshold is recorded as the damage initiation moment. The duration is calculated with the current moment as the end point. When the duration exceeds the preset window, the command is triggered to generate. Based on the location of the crack, the pre-stored structural safety margin table is called, and the remaining time required for the crack width to reach the limit allowable width is calculated in combination with the width evolution trend. Half of the remaining time is taken as the repair time. The location of the crack, its width evolution trend, and the timing of repair are encoded into preventive maintenance decision instructions, and the repair timing in the instructions is converted into the corresponding calendar date for output.

Images