A bridge operation state evaluation method based on multi-source heterogeneous data space-time fusion

Abstract

This invention relates to the field of condition assessment technology and discloses a bridge operation condition assessment method based on spatiotemporal fusion of multi-source heterogeneous data. The method includes: extracting vibration energy using low-frequency measured displacement and temperature and high-frequency acceleration to construct a multi-source spatiotemporal monitoring dataset; extracting the dominant temperature pattern to calculate the theoretical free temperature-induced displacement; constructing the distribution of unreleased energy and the distribution of instantaneous deformation release based on displacement difference and vibration energy; extracting the spatial offset between the accumulation center and the release center; locating the dominant imbalance measuring points by combining rearrangement costs; and solving the bridge operation condition index to classify levels and output abnormal locations. This invention can accurately identify spatially eccentric energy release phenomena and adaptively achieve graded condition assessment.

Description

Technical Field

[0001] This invention relates to the field of condition assessment technology, specifically to a bridge operation condition assessment method based on spatiotemporal fusion of multi-source heterogeneous data. Background Technology

[0002] The longitudinal boundary system of a bridge typically consists of core supporting devices such as continuous main girders, sliding bearings, and modular expansion joints. During long-term service, sliding bearings may exhibit stick-slip behavior due to wear, lubrication degradation, or localized corrosion, leading to localized stress concentration and causing their temperature-displacement relationship to exhibit hysteretic behavior with a loop. Simultaneously, under the combined effects of lateral and vertical vehicle wheel loads, the coordinated deformation of multiple internal components of modular expansion joints easily triggers multi-scale, nonlinear damage evolution. Recent in-depth research and field monitoring data indicate that the degradation of this boundary system not only alters the response of individual measuring points but also introduces a hidden and highly dangerous engineering problem: the free elongation of the main girder due to temperature changes is not necessarily released in situ from the accumulated constraint potential; instead, it is easily triggered and released at another bearing or joint by transient impacts caused by vehicles. However, existing bridge operational status assessment technologies often assume that low-frequency temperature or displacement monitoring sources and high-frequency dynamic response monitoring sources are located at the same point and coupled based on the same mechanism. This cognitive limitation makes it impossible for existing assessment methods to effectively characterize the spatial eccentricity and misalignment process in which temperature-induced longitudinal constraints accumulate in one place but are released in another. As a result, it is very easy to misjudge the local imbalance of the above boundary chain as ordinary temperature hysteresis, normal vehicle impact, or harmless sensor fluctuations. Summary of the Invention

[0003] The present invention aims to provide a bridge operation status assessment method based on spatiotemporal fusion of multi-source heterogeneous data, thereby addressing the shortcomings of the prior art mentioned in the background.

[0004] The bridge operation status assessment method based on spatiotemporal fusion of multi-source heterogeneous data, as described in this invention, includes: The measured displacement and temperature of the bridge boundary measuring points were collected using low-frequency monitoring equipment, and the acceleration of the bridge boundary measuring points was collected and the vibration energy was extracted using high-frequency monitoring equipment to construct a multi-source spatiotemporal monitoring dataset. Based on the dominant temperature pattern extracted from the multi-source spatiotemporal monitoring dataset, the theoretical free temperature-induced displacement of the bridge boundary measuring point is calculated. The distribution of unreleased temperature-induced displacement is constructed based on the difference between the theoretical free temperature-induced displacement and the measured displacement, and the instantaneous deformation release distribution is constructed by multiplying the vibration energy by the rate of change of the measured displacement. Calculate the cumulative center of the unreleased temperature-induced displacement distribution and the release center of the instantaneous deformation release distribution, and extract the spatial distance between the release center and the cumulative center as the cumulative-release spatial offset. The transformation from the temperature-induced displacement unreleased distribution to the instantaneous deformation release distribution is taken as a rearrangement process. The deformation rearrangement cost is calculated in combination with the cumulative-release spatial offset, and the dominant measurement point causing the imbalance of the cumulative-release spatial offset is located. The cumulative-release spatial offset and the deformation rearrangement cost are combined to form a comprehensive evaluation parameter set, and the bridge operation status index is solved. The comprehensive evaluation parameter set is classified into bridge operation status levels, and the abnormal locations are output according to the dominant imbalance measuring points.

[0005] The beneficial effects of this invention are as follows: by synchronously fusing quasi-static temperature-induced displacement data with high-frequency transient acceleration energy using a unified time grid, the spatial eccentricity between the thermal accumulation center and the actual energy release center is quantified; this invention abandons the traditional mode of relying on manual hard thresholds and isolated judgment of single measuring points, and adaptively delineates the state baseline and evaluation level based on the evolution law of the data itself, realizing the leap from passive monitoring to active and precise positioning of local imbalances and disease evolution locations in the boundary chain, greatly improving the accuracy and scientific nature of bridge long-term service safety assessment. Attached Figure Description

[0006] Figure 1 This is a flowchart of a bridge operation status assessment method based on spatiotemporal fusion of multi-source heterogeneous data according to the present invention. Detailed Implementation

[0007] like Figure 1 As shown, this invention provides a bridge operation status assessment method based on spatiotemporal fusion of multi-source heterogeneous data, the implementation steps of which include: The measured displacement and temperature of the bridge boundary measuring points were collected using low-frequency monitoring equipment, and the acceleration of the bridge boundary measuring points was collected and the vibration energy was extracted using high-frequency monitoring equipment to construct a multi-source spatiotemporal monitoring dataset. Based on the dominant temperature pattern extracted from the multi-source spatiotemporal monitoring dataset, the theoretical free temperature-induced displacement of the bridge boundary measuring point is calculated. The distribution of unreleased temperature-induced displacement is constructed based on the difference between the theoretical free temperature-induced displacement and the measured displacement, and the instantaneous deformation release distribution is constructed by multiplying the vibration energy by the rate of change of the measured displacement. Calculate the cumulative center of the unreleased temperature-induced displacement distribution and the release center of the instantaneous deformation release distribution, and extract the spatial distance between the release center and the cumulative center as the cumulative-release spatial offset. The transformation from the temperature-induced displacement unreleased distribution to the instantaneous deformation release distribution is taken as a rearrangement process. The deformation rearrangement cost is calculated in combination with the cumulative-release spatial offset, and the dominant measurement point causing the imbalance of the cumulative-release spatial offset is located. The cumulative-release spatial offset and the deformation rearrangement cost are combined to form a comprehensive evaluation parameter set, and the bridge operation status index is solved. The comprehensive evaluation parameter set is classified into bridge operation status levels, and the abnormal locations are output according to the dominant imbalance measuring points.

[0008] Preferably, low-frequency monitoring equipment is used to collect measured displacement and temperature at bridge boundary measuring points, and high-frequency monitoring equipment is used to collect acceleration at the bridge boundary measuring points and extract vibration energy to construct a multi-source spatiotemporal monitoring dataset, including: Set a unified synchronization time point Using spline basis functions The measured displacement With the temperature Synchronize to the unified synchronization time point The calculation formula is:

[0009]

[0010] in, The measured displacement after synchronization The temperature after synchronization. The total number of samples of the measured displacement. The total number of temperature samples. This refers to the original sampling time of the low-frequency monitoring device. This refers to the numbering of the measuring points at the bridge boundary. This refers to the numbering of the temperature measurement points; Calculate the acceleration acceleration autocorrelation value The acceleration autocorrelation value The time interval corresponding to the first drop to zero or less than zero is used as the adaptive smoothing scale for the measurement point. And construct an exponentially decaying kernel function. The calculation formula is:

[0011]

[0012] in, For time difference variables, This indicates the fimum operation; Calculate the acceleration The square of the acceleration energy density is obtained by using the exponential decay kernel function. The acceleration energy density is integrally smoothed to extract the time point synchronized to the unified synchronization point. The vibrational energy The calculation formula is:

[0013] in, This refers to the original sampling time of the high-frequency monitoring device; The spatial coordinates of the bridge boundary measuring points The measured displacement after synchronization The temperature and the vibrational energy By combining the data, the multi-source spatiotemporal monitoring dataset is constructed. The formula is as follows:

[0014] A unified synchronization time point refers to a common time node that aligns low-frequency displacement data, low-frequency temperature data, and high-frequency vibration energy data to the same analysis moment. Preferably, it is a time grid that is consistent with or an integer multiple thereof (10 to 60 seconds) of the low-frequency displacement sampling interval. This is because it is necessary to cover the actual sampling period of low-frequency monitoring while avoiding compressing high-frequency events to an indistinguishable degree.

[0015] Spline basis functions are a set of basis functions used to smoothly map discrete low-frequency samples to a unified synchronization time point.

[0016] Measured displacement refers to the longitudinal displacement response actually monitored at bridge boundary measuring points during operation. It can be collected using wire displacement gauges, magnetostrictive displacement gauges, fiber optic displacement gauges, GPS receivers, or total stations.

[0017] Temperature refers to the structural ambient temperature or component surface temperature obtained at the monitoring time at bridge boundary measuring points or adjacent measuring points. It can be collected through thermocouples, resistance temperature gauges, digital temperature sensors, or distributed fiber optic temperature measurement devices.

[0018] The measured displacement after synchronization refers to the displacement sequence after resampling according to the unified synchronization time point.

[0019] Synchronized temperature refers to the temperature sequence after resampling according to a unified synchronization time point.

[0020] The total number of measured displacement samples refers to the number of original displacement samples that participate in the synchronous displacement calculation.

[0021] The total number of temperature samples refers to the number of original temperature samples that participate in the temperature synchronization calculation.

[0022] The original sampling time of low-frequency monitoring equipment refers to the timestamp corresponding to each original sample from the low-frequency displacement and temperature sensors. This can be obtained through the system clock and timestamp recording function of the data acquisition instrument.

[0023] The bridge boundary measuring point number refers to the data identification sequence number used to distinguish each bridge boundary measuring point.

[0024] Temperature measurement point number refers to the data identification number used to distinguish each temperature measurement point.

[0025] Acceleration refers to the transient dynamic response of a bridge boundary measuring point under vehicle load and boundary vibration. It can be acquired using piezoelectric accelerometers, microelectromechanical accelerometers, or servo accelerometers.

[0026] Acceleration autocorrelation value refers to the degree of correlation of acceleration sequences at the same measuring point under different time lags.

[0027] The adaptive smoothing scale of the measuring point refers to the vibration energy smoothing time width automatically determined based on the autocorrelation decay characteristics of the acceleration at each measuring point.

[0028] The exponentially decaying kernel function is a smooth weighted function that decreases exponentially as the time difference increases.

[0029] The time difference variable refers to the time difference between the current synchronization time and the original high-frequency sample time when the kernel function is calculated.

[0030] The original sampling time of the high-frequency monitoring equipment refers to the timestamp corresponding to each sample from the high-frequency accelerometer. It can be obtained through a high-frequency data acquisition card and a unified time synchronization module.

[0031] Vibration energy refers to an energy-type index that reflects the intensity of local dynamic excitation, obtained by squared acceleration and smoothing it with a kernel function.

[0032] Acceleration energy density refers to the instantaneous energy representation formed by squaring the acceleration.

[0033] The spatial coordinates of bridge boundary measuring points refer to the positions of each boundary measuring point in the bridge's longitudinal coordinate system. These coordinates can be obtained through as-built survey results, total station re-measurement, global navigation satellite system measurements, or verification using bridge design drawings.

[0034] Multi-source spatiotemporal monitoring datasets refer to composite data sets that use spatial coordinates as indexes, unified synchronization time points as time indexes, and simultaneously include synchronous displacement, synchronous temperature, and synchronous vibration energy.

[0035] Specifically, because the sampling frequencies of low-frequency displacement data, low-frequency temperature data, and high-frequency acceleration data are different, the original sampling times usually do not coincide. Therefore, if they are directly matched point by point, the data at different times will be mismatched to the same physical state, which will cause the displacement-temperature relationship and vibration release relationship to be distorted. Therefore, it is necessary to first set a unified synchronization time point and then complete the fusion of multi-source data on a unified time grid.

[0036] Specifically, because low-frequency monitoring data often suffers from packet loss, clock drift, and non-constant sampling intervals at engineering sites, relying solely on nearest neighbor sampling introduces step errors and phase errors. Spline basis functions, on the other hand, can utilize multiple neighboring samples to jointly reconstruct continuous trends. Therefore, using spline basis functions for synchronization is more suitable for bridge quasi-static temperature displacement data.

[0037] Specifically, since the rate of dynamic decay at different bridge boundary measurement points is not consistent, a preset fixed smoothing window will cause some measurement points to be over-smoothed and some measurement points to be under-smoothed. Therefore, the time interval corresponding to the first drop of the acceleration autocorrelation value to zero or less than zero is used as the adaptive smoothing scale, which can make the energy extraction window of each measurement point conform to its own dynamic correlation length.

[0038] Specifically, since high-frequency acceleration signals are fast oscillators, direct downsampling would cancel out short-term impact events. Therefore, we first form the energy density by squared acceleration, and then use an exponentially decaying kernel function for integral smoothing. This can preserve the event intensity information and map it to a unified synchronization time point.

[0039] Specifically, since the spatial eccentric energy release phenomenon is not a single-point anomaly, but a misalignment of the thermal accumulation and dynamic release relationship between different spatial locations at the same time, it is necessary to combine the spatial coordinates with the synchronized displacement, temperature and vibration energy to provide a basis for subsequent spatial distribution comparison and center offset extraction.

[0040] Specifically, the method for setting the unified synchronization time point is as follows: First, read all low-frequency displacement sample timestamps, low-frequency temperature sample timestamps, and high-frequency acceleration sample timestamps, and take the overlapping time interval covered by the three types of data as the effective analysis interval; second, use the nominal sampling interval of low-frequency displacement as the basic step size, or an integer multiple of this step size as the common step size, and generate a unified synchronization time point sequence at equal intervals within the effective analysis interval; third, if the nominal intervals of low-frequency displacement and low-frequency temperature are different, the larger interval is preferred to avoid artificially generating pseudo-refined time points that exceed the true resolution of low-frequency monitoring; finally, reserve at least one low-frequency sampling interval buffer at the start and end points to reduce boundary interpolation errors.

[0041] Specifically, the spline basis functions are constructed as follows: cubic spline basis functions are preferred; the original low-frequency sampling times are sorted chronologically as spline nodes; spline functions are established for the displacement and temperature sequences of each measuring point; natural boundary conditions are used at the boundaries to prevent excessive amplification of the second-order changes at both ends; if there are missing samples, the missing times are deleted first, and the remaining valid nodes are retained before reconstructing the spline; the generated spline functions are sampled point by point at a unified synchronization time point to obtain the synchronized displacement and synchronized temperature; when the length of continuous missing data exceeds 3 low-frequency sampling intervals, this interval is not extrapolated by spline, but is directly marked as an invalid interval and removed from subsequent analysis.

[0042] Specifically, the discrete calculation method for the acceleration autocorrelation value is as follows: first, subtract the mean value from the high-frequency acceleration sequence of each measurement point to eliminate the influence of the DC component; then, calculate the mean value of the product of the acceleration sequence and its delay sequence one by one according to the discrete lag step size to form the autocorrelation sequence; starting from zero lag, search along the time lag direction to find the first lag point with an autocorrelation value less than or equal to zero; divide the number of sampling points corresponding to the lag point by the high-frequency sampling frequency to obtain the measurement point adaptive smoothing scale; if no non-positive value appears within the preset maximum search time, then the maximum search time is taken as the alternative smoothing scale.

[0043] Specifically, the integral smoothing method of the exponential decay kernel function is as follows: taking each unified synchronization time point as the center, select no less than 3 high-frequency samples within the adaptive smoothing scale range before and after it; calculate the time difference between each high-frequency sample and the synchronization time point; substitute the time difference into the exponential decay kernel function to obtain the weight; then multiply the square value of the high-frequency acceleration by the corresponding weight and perform discrete summation on all samples to obtain the vibration energy at the synchronization time point; finally, normalize the kernel weights to ensure that the vibration energy between different time lengths and different measurement points is comparable.

[0044] Specifically, the boundary and missing data handling methods during vibration energy synchronization are as follows: when the unified synchronization time point is close to the start or end of the analysis interval, causing the kernel function support interval to exceed the original data range, only the effective sample portion is used to calculate the weight and re-normalize; if the missing proportion of high-frequency samples in the kernel function support interval does not exceed one-tenth, linear filling with neighboring samples is allowed; if the missing proportion exceeds one-tenth, the vibration energy at that synchronization time point is marked as invalid and does not participate in subsequent distribution normalization.

[0045] Specifically, the unified method for spatial coordinates is as follows: the longitudinal axis of the bridge is used as the only analysis coordinate axis; all boundary measurement points are projected onto this longitudinal axis; the starting point of the boundary at one end of the bridge is used as the zero point; the cumulative distance along the bridge direction is used as the spatial coordinates of the measurement points; if the original measurement is a three-dimensional coordinate, the coordinate system and projection transformation are completed first before taking the longitudinal component; only one set of coordinate references is allowed to be used for the entire length of the same bridge, and the design mileage and temporary on-site coordinates are prohibited from being mixed.

[0046] Preferably, the theoretical free temperature-induced displacement of the bridge boundary measuring points is calculated based on the dominant temperature pattern extracted from the multi-source spatiotemporal monitoring dataset, including: Calculate the temperature in the multi-source spatiotemporal monitoring dataset. Temperature covariance matrix and the temperature covariance matrix Eigenvalues ​​are obtained by performing eigenvalue decomposition. With feature vectors Combined with the feature vector Extract the dominant temperature pattern The calculation formula is:

[0047]

[0048]

[0049] in, For the unified synchronization time point Total quantity Let be the mean vector of the temperatures, with superscript... Indicates matrix transpose; According to the characteristic value Calculation percentage Obtain the information entropy and adaptively determine the number of modes to retain. The calculation formula is:

[0050]

[0051] in, Index of the total number of eigenvalues This represents the floor function; The vibration energy of each bridge boundary measuring point in the multi-source spatiotemporal monitoring dataset is used. The summation constructs a dynamic weighting factor A spatial topological Laplace matrix is ​​constructed based on the spatial coordinates of the bridge boundary measurement points. The measured displacement in the multi-source spatiotemporal monitoring dataset To approach the target, the dominant temperature pattern is used. The dynamic weighting factor and the spatial topological Laplace matrix Perform graphical regularization to solve for the temperature-to-displacement mapping matrix. The calculation formula is:

[0052]

[0053] in, This represents the total number of measuring points at the bridge boundary. Let be the mapping matrix to be solved. To plot the regularity strength, Represents the trace operation of a matrix; Using the mapping matrix With the dominant temperature mode Multiply and extract the corresponding components as the theoretical free temperature-induced displacement of the bridge boundary measuring points. The calculation formula is:

[0054] Among them, subscript This indicates the corresponding element in the matrix operation result. The components of the bridge boundary measuring points.

[0055] The temperature covariance matrix is ​​a matrix used to describe the common changes in the synchronous temperatures of various temperature measuring points.

[0056] Eigenvalues ​​refer to the variance intensity corresponding to each principal direction of change in the temperature covariance matrix.

[0057] Eigenvectors are vectors in the temperature covariance matrix that correspond to the main direction of change.

[0058] The dominant temperature mode refers to the modal response obtained by projecting the main eigenvectors, which can represent the main trend of temperature field change.

[0059] The total number of unified synchronization time points refers to the number of unified synchronization time points that participate in subsequent modeling and calculations.

[0060] The temperature mean vector is a vector composed of the average temperature of all temperature measurement points over the entire analysis period.

[0061] The proportion of an eigenvalue refers to the relative proportion of a certain eigenvalue in the sum of all eigenvalues.

[0062] The number of retained modes refers to the number of dominant temperature modes that are retained after adaptive filtering.

[0063] The total number of eigenvalues ​​index refers to the ordinal variable used when traversing all eigenvalues.

[0064] The dynamic weighting factor refers to the regression weights automatically calculated based on the vibration energy levels of all measuring points at the same synchronous moment.

[0065] The spatial topological Laplace matrix is ​​a structural smoothing constraint matrix constructed based on the spatial adjacency relationship of bridge boundary measurement points.

[0066] The measured displacement vector is the vector composed of the synchronous displacements of all boundary measuring points at the same synchronous moment.

[0067] The dominant temperature mode vector is a vector composed of the coordinates of all retained temperature modes at the same synchronization moment.

[0068] The mapping matrix to be solved refers to the temperature-to-displacement mapping coefficient matrix that has not yet been determined in the regression optimization process.

[0069] The temperature-to-displacement mapping matrix refers to the coefficient matrix used to map the temperature-dominant mode to the theoretical free temperature-induced displacement after solving the graph regular regression.

[0070] The graph regularization strength refers to the adjustment parameter that controls the strength of the spatial smoothness constraint of the mapping matrix. It is preferably between 0.01 and 10. The value needs to strike a balance between fitting accuracy and spatial smoothness; a larger value is preferable when the data noise is high, while a value between 0.1 and 1 is preferred when the number of measurement points is small and the displacement response is relatively smooth.

[0071] The total number of bridge boundary measuring points refers to the total number of all boundary measuring points included in the calculation using this method.

[0072] Theoretical free temperature-induced displacement refers to the displacement that bridge boundary measuring points should have, derived from the dominant temperature model, without considering the effects of local hindrance and abnormal release.

[0073] Modal index refers to the sequence number variable that distinguishes different dominant temperature modes.

[0074] Specifically, since the temperature field at the bridge boundary is often composed of multiple related measuring points, a single temperature measuring point cannot fully represent the thermal driving state of the entire structure. Therefore, it is necessary to first construct the temperature covariance matrix and perform eigenvalue decomposition to compress the temperature information from multiple measuring points into a few dominant temperature patterns, thereby extracting the most representative thermal change direction.

[0075] Specifically, since the effective number of temperature-dominant modes is not fixed for different bridges and different seasons, directly presetting the number of modes can easily lead to information omissions or redundancies. Therefore, by forming information entropy through the proportion of feature values ​​and adaptively determining the number of modes to be retained, the number of modes to be retained can be automatically changed with the complexity of the data.

[0076] Specifically, since vehicle impacts and sudden vibrations can contaminate the quasi-static relationship between temperature and displacement, if these moments are treated equally with pure thermal drive moments, the dynamic disturbance will be mistakenly learned as a temperature-displacement mapping. Therefore, a dynamic weighting factor composed of the sum of vibration energy is introduced to automatically reduce the impact of dynamically active moments on the regression results.

[0077] Specifically, since the temperature-induced free displacements of adjacent boundary measurement points are usually continuous in space, if no spatial constraints are applied, the mapping matrix may exhibit drastic changes that do not conform to the continuity of the structure. Therefore, by constructing a spatial topological Laplace matrix based on spatial coordinates and introducing graph regularization, the structural adjacency relationship can be incorporated into the mapping solution process.

[0078] Specifically, since the goal is to reconstruct the theoretical free temperature-induced displacement under conditions of no local obstruction, rather than directly fitting the measured abnormal displacement, the dominant temperature model, dynamic weighting factor, and spatial topological Laplace matrix must be used together in the regression to ensure that the results reflect temperature dominance, suppress dynamic contamination, and maintain spatial continuity.

[0079] Specifically, the temperature vector and the temperature mean vector are formed as follows: at each unified synchronization time point, the synchronized temperature values ​​of all temperature measuring points are arranged into a column vector according to a fixed measuring point order; then, the synchronized temperature of each temperature measuring point at all unified synchronization time points is averaged to form the mean component at the corresponding position; finally, all mean components are arranged into a temperature mean vector according to the same measuring point order; if there is a missing temperature value at a certain moment, it is first linearly filled with the adjacent valid synchronization time of that measuring point; if more than 3 consecutive synchronization time intervals are missing, the corresponding moment is removed from the covariance statistics.

[0080] Specifically, the method for determining the number of retained modes is as follows: first, calculate the proportion of each feature value, then calculate the effective dimension corresponding to the information entropy according to the proportion; then, use the rounding up method to obtain the number of retained modes; when the rounding up result is greater than the total number of temperature measurement points, take the total number of temperature measurement points; when the result is less than 1, force it to be 1; in this way, at least one dominant mode is retained, and the maximum number of achievable modes is not exceeded.

[0081] Specifically, the dynamic weighting factor is calculated as follows: at each unified synchronization time point, the synchronous vibration energy of all boundary measurement points is summed, and the sum is substituted into the weight expression; if the sum is very close to zero, the weight of that synchronization time is directly set to 1; if the sum is significantly greater than zero, the weight is automatically reduced according to the inverse relationship; to avoid the instability of the value caused by extremely small weights, the lower limit of the weight can be set to 0.05.

[0082] Specifically, the spatial topological Laplace matrix is ​​constructed as follows: First, all boundary measurement points are arranged in ascending order of their longitudinal coordinates; adjacent measurement points are considered as directly adjacent nodes; the adjacency weight is preferentially taken as the reciprocal of the longitudinal distance between adjacent measurement points; then, a matrix is ​​constructed in which the diagonal elements are the sum of the adjacent weights and the off-diagonal elements are the corresponding negative adjacency weights; the first and last measurement points are only connected to their respective unique adjacent measurement points; if there are structurally strongly coupled support and joint measurement points on site, an additional adjacent edge is allowed and assigned a smaller weight.

[0083] Specifically, the solution method for graph regularization is as follows: First, the dominant temperature mode vectors at all unified and synchronized time points are arranged in chronological order to form a design matrix, and the measured displacement vectors at the corresponding time points are arranged to form a response matrix; then, the dynamic weighting factor is written as a diagonal weight matrix; then, the mapping matrix is ​​solved by a system of linear equations consisting of weighted least squares and regularization terms; when the condition number of the system of equations is too large, singular value decomposition or ridge stability solution is preferred to obtain a numerically stable solution.

[0084] Specifically, the determination of the graph regularization strength is as follows: cross-validation or leave-one-out-of-cycle validation on historical temperature cycle data is preferred; the selection criterion is the optimal combination of the fitting error of theoretical free temperature-induced displacement to synchronous measured displacement and the spatial smoothness of the mapping matrix; in engineering implementation, a coarse search can be performed first at the four orders of magnitude of 0.01, 0.1, 1, and 10, and then a fine search can be performed near the optimal order of magnitude.

[0085] Specifically, the extraction method for theoretical free temperature-induced displacement components is as follows: when the number of temperature measurement points is inconsistent with the number of displacement measurement points, the number of rows in the mapping matrix corresponds to the number of displacement measurement points, and the number of columns corresponds to the number of retained modes; the length of the resulting vector obtained by multiplying the matrix by the dominant temperature mode vector is consistent with the number of displacement measurement points; each component directly corresponds to the theoretical free temperature-induced displacement of the displacement measurement point with the same sequence number; the order of measurement point numbering must be consistent with the order in the measured displacement vector.

[0086] Preferably, a distribution of unreleased temperature-induced displacement is constructed based on the difference between the theoretical free temperature-induced displacement and the measured displacement, and an instantaneous deformation release distribution is constructed by multiplying the vibration energy by the rate of change of the measured displacement, including: Calculate the theoretical free temperature-induced displacement The measured displacement after synchronization The difference is used to obtain the constraint difference of the measuring point. The calculation formula is:

[0087] in, To unify and synchronize time points, The numbering of the bridge boundary measuring points; Calculate the constraint difference of the measuring points. The square of the value yields the pseudo-thermal potential surrogate quantity at the measuring point. The pseudo-thermal potential surrogate quantity at each of the bridge boundary measuring points is then divided by the sum of the pseudo-thermal potential surrogate quantities at all the bridge boundary measuring points to construct the distribution of the unreleased temperature-induced displacement. The calculation formula is:

[0088] in, This represents the total number of measuring points at the bridge boundary. The cumulative index number of the bridge boundary measuring points; Calculate the measured displacement after synchronization The absolute value of the first-order time derivative is used to obtain the measured rate of displacement change. The synchronized vibration energy Compared with the measured displacement change rate Multiplying these values ​​yields the instantaneous release intensity at each measuring point. Dividing the instantaneous release intensity at each of the bridge boundary measuring points by the sum of the instantaneous release intensities at all the bridge boundary measuring points then constructs the instantaneous deformation release distribution. The calculation formula is:

[0089] in, This represents the operation of taking the first derivative with respect to time. For the first The bridge boundary measuring points were at the unified synchronous time point. The vibrational energy after synchronization For the first The bridge boundary measuring points were at the unified synchronous time point. The measured displacement after synchronization.

[0090] The measurement point constraint difference refers to the difference between the theoretical free temperature-induced displacement and the synchronously measured displacement at the same boundary measurement point.

[0091] The pseudo-thermal potential energy proxy quantity at the measuring point refers to the proxy quantity formed by the square of the constraint difference at the measuring point, which is used to characterize the degree of constraint accumulation caused by unreleased heat.

[0092] The distribution of unreleased temperature-induced displacement refers to the spatial distribution of the pseudo-thermal potential energy proxy at each boundary measuring point after normalization.

[0093] The measured displacement change rate refers to the absolute value index of how fast the synchronously measured displacement changes over time.

[0094] The instantaneous release strength at the measuring point refers to the local instantaneous release strength index obtained by multiplying the synchronous vibration energy by the measured displacement change rate.

[0095] Instantaneous deformation release distribution refers to the spatial distribution of instantaneous release intensity at each boundary measuring point after normalization.

[0096] The cumulative index number refers to the sequence number variable used when performing summation on all boundary measurement points.

[0097] Specifically, since the theoretical free temperature-induced displacement represents the displacement that the structure should undergo under ideal free expansion and contraction conditions, while the measured displacement includes the effects of local hindrance and release delay, the difference between the two can directly reflect the degree of thermal constraint that has not yet been released at the current measuring point. Therefore, it is defined as the measuring point constraint difference having directionality.

[0098] Specifically, since the cumulative strength of unreleased thermal constraints needs to be expressed by a quantity that is always non-negative and more sensitive to large deviations, the square of the constraint difference at the measurement point is used as a pseudo-thermal potential energy proxy, which can highlight the high-constraint measurement points without introducing unknown stiffness parameters.

[0099] Specifically, since high-frequency vibration does not necessarily mean that a real release has occurred at the boundary, it is more likely to correspond to support slippage or gap release events only when dynamic excitation and displacement change occur simultaneously. Therefore, multiplying the synchronous vibration energy with the measured displacement change rate to construct the instantaneous release intensity at the measuring point can more accurately distinguish between real release and ordinary background vibration.

[0100] Specifically, the first-order time derivative of the measured displacement is calculated as follows: the central difference method is preferred to calculate the rate of change of displacement at the unified synchronous time point; forward difference and backward difference are used to supplement the sequence at the start and end points; in order to reduce the amplification effect of low-frequency displacement noise on the derivative, a short window smoothing can be performed on the synchronous displacement sequence before the derivative is calculated, and the smoothing window is preferably taken as 3 to 5 synchronous time points.

[0101] Specifically, the pseudo-thermal potential energy proxy is implemented in a squared form as follows: at each unified synchronization time point, the difference between the theoretical free temperature-induced displacement and the synchronous measured displacement is first calculated, and then the difference is squared to ensure that the proxy is non-negative and to enhance the sensitivity to large deviation measurement points; unknown equivalent stiffness of measurement points is no longer introduced, thereby avoiding the model being unfeasible due to the difficulty in identifying the stiffness in the field in real time.

[0102] Specifically, the normalization method for the distribution of unreleased temperature-induced displacement is as follows: first, sum the pseudo-thermal potential energy surrogate quantities of all boundary measuring points; if the sum is greater than the preset minimum value, then divide the surrogate quantity of each measuring point by the sum; if the sum is less than or equal to the preset minimum value, then it is considered that there is no significant unreleased constraint at the current moment, and the distribution of unreleased quantity of all measuring points is directly set to 0.

[0103] Specifically, the instantaneous deformation release distribution is normalized as follows: first, the product of the synchronous vibration energy and the rate of change of displacement of all boundary measuring points is summed; if the sum is greater than the preset minimum value, the product of each measuring point is divided by the sum; if the sum is less than or equal to the preset minimum value, it is considered that there is no significant release event at the current moment, and the release distribution of all measuring points is directly set to 0.

[0104] Specifically, the preprocessing method for vibration energy and displacement change rate is as follows: before calculating the product, it is confirmed that the vibration energy has been normalized according to the kernel function weight and the displacement change rate has been calculated according to the unified synchronous time interval; if there are differences in the dimensional scale between different bridges or different time periods, robust standardization can be performed according to the median and quartile scales respectively in the same analysis window before entering the product operation, so as to maintain the comparability between different measurement points.

[0105] Preferably, calculating the cumulative center of the unreleased temperature-induced displacement distribution and the release center of the instantaneous deformation release distribution, and extracting the spatial distance between the release center and the cumulative center as the cumulative-release spatial offset, includes: Calculate the theoretical free temperature-induced displacement at all the bridge boundary measuring points. The first-order time derivative is used to calculate the average thermal driving velocity. The calculation formula is:

[0106] in, This represents the total number of measuring points at the bridge boundary. This refers to the numbering of the measuring points at the bridge boundary. This represents the operation of taking the first derivative with respect to time. According to the average thermal drive speed The temperature cycle is obtained by dividing the time interval at the zero crossing point. Determine each of the temperature cycles. start and end times and and duration ,in ; At the start and end times and Between, the distribution of the amount of unreleased temperature-induced displacement Perform time integration and divide by the duration. Obtain the distribution of average unreleased amount in the cycle The instantaneous deformation release distribution Perform time integration and divide by the duration. Obtain the cyclic average release distribution The calculation formula is:

[0107]

[0108] The spatial coordinates of each of the bridge boundary measurement points The distribution of the average unreleased amount in the cycle Multiply and sum to obtain the cumulative center. The spatial coordinates of each of the bridge boundary measurement points are... With the average release distribution of the cycle Multiply and sum to obtain the release center. The calculation formula is:

[0109]

[0110] Calculate the release center With the accumulation center The difference is calculated and then divided by the total longitudinal length of the bridge boundary. The accumulated-release space offset is obtained. The calculation formula is:

[0111] The average thermally driven velocity refers to the average rate of change of theoretically free temperature-induced displacement at all measuring points on the bridge boundary.

[0112] Temperature cycling refers to a continuous thermally driven stage unit obtained by dividing the average thermally driven velocity at the zero-crossing point.

[0113] The start time of a temperature cycle refers to the unified and synchronized point in time at which a certain temperature cycle begins.

[0114] The temperature cycle termination time refers to the unified synchronous point in time when a certain temperature cycle ends.

[0115] The duration of a temperature cycle refers to the length of time between the end of the temperature cycle and the start of the cycle.

[0116] The current temperature cycle refers to the latest temperature cycle that is currently being evaluated.

[0117] Historical temperature cycles refer to all completed temperature cycles preceding the current temperature cycle.

[0118] The cyclic mean unreleased quantity distribution refers to the average spatial distribution of the temperature-induced displacement unreleased quantity distribution within a certain temperature cycle after integral over time and normalization by duration.

[0119] Cyclic average release distribution refers to the average spatial distribution of instantaneous deformation release distribution within a certain temperature cycle after integral over time and normalization by duration.

[0120] The accumulation center refers to the location of the thermally induced accumulation center obtained by weighted summation of the spatial coordinates of each measuring point using the cyclic average unreleased amount distribution.

[0121] The release center refers to the actual location of the release center obtained by weighted summation of the spatial coordinates of each measuring point using the cyclic average release distribution.

[0122] The total longitudinal length of the bridge boundary refers to the total length of the bridge boundary chain included in the analysis of this method along the bridge direction. It can be obtained through bridge design drawings, as-built survey results, total station measurements, or bridge mileage verification.

[0123] Cumulative release space offset refers to the normalized vertical offset index of the release center relative to the cumulative center.

[0124] The time integral variable refers to the continuous time variable used when performing integral averaging over a temperature cycle interval.

[0125] Specifically, because this invention corresponds to the spatial correspondence between thermal accumulation and power release within a complete thermal drive phase, rather than the instantaneous fluctuation at a single moment, it is necessary to first obtain the average thermal drive velocity using the average value of the first-order time derivative of the theoretical free temperature-induced displacement, and then use its zero-crossing point as the thermal cycle boundary, so that each temperature cycle corresponds to a relatively complete thermal drive process.

[0126] Specifically, since both the instantaneous non-released amount distribution and the instantaneous release distribution may be affected by single-vehicle impact and local noise, by integrating the two types of distributions over time and normalizing them according to the duration within the temperature cycle range, the occasional spikes can be transformed into a stable cyclic average spatial distribution.

[0127] Specifically, since the distribution center is the most direct statistical measure for judging spatial offset, the accumulation center reflects where thermal constraints mainly accumulate, and the release center reflects where actual energy release mainly occurs, the difference between the two and then divided by the total longitudinal length of the bridge boundary can yield a dimensionless offset index that can be compared between different bridges.

[0128] Specifically, the calculation method for the first time derivative of the theoretical free temperature-induced displacement is as follows: the theoretical free temperature-induced displacement sequence of each boundary measuring point is centrally differentiald at a uniform synchronization time interval; unilateral differential is used at the first and last time points; when there is a short-term spike in the theoretical free temperature-induced displacement sequence, the isolated anomaly is first removed by smoothing with the median of the three synchronization time points, and then the derivative is calculated.

[0129] Specifically, the zero-crossing identification method is as follows: first, the average thermally driven velocity sequence is slightly smoothed; then, it is determined whether the velocity sign changes between two adjacent synchronization time points; if a change occurs from positive to negative or from negative to positive, the zero-crossing time is determined by linear interpolation between the two points; to avoid noise repeatedly crossing near zero, a zero threshold less than 1 / 10 of the sequence standard deviation can be set, and zero crossing is confirmed only when the absolute value continuously crosses the threshold.

[0130] Specifically, the current temperature cycle and historical temperature cycles are divided as follows: the latest closed zero-crossing interval is taken as the current temperature cycle; all previously closed intervals are taken as historical temperature cycles; if the latest interval has not yet closed, it is allowed to generate a cycle under construction with the current time as a temporary endpoint, but this cycle should be marked as an unclosed state separately during statistical modeling.

[0131] Specifically, the time integral averaging is implemented as follows: the trapezoidal integral method is used to integrate the unreleased quantity distribution and the released quantity distribution at discrete synchronization time points; then the integration result is divided by the duration of the temperature cycle; if the synchronization time points within the cycle are not equidistant, the actual time difference is used as the weight for each small segment of integration.

[0132] Specifically, the coordinates of the accumulation center and release center are used as follows: This method prioritizes the weighted summation of the one-dimensional longitudinal projection coordinates of the bridge; if the original measurement point coordinates are three-dimensional, they must first be projected onto the longitudinal principal axis of the bridge; calculations should not be performed directly on unprojected three-dimensional coordinates, otherwise transverse and vertical deviations will be introduced.

[0133] Specifically, the total longitudinal length of the bridge boundary is determined by subtracting the longitudinal projection coordinates of the starting point and ending point of the analysis chain; if the analysis object only covers a local boundary chain, the total length is taken as the length of that local chain; the total length must use the same coordinate reference as the spatial coordinates of the measurement points.

[0134] Preferably, the transformation from the distribution of unreleased temperature-induced displacement to the distribution of released instantaneous deformation is taken as a rearrangement process. The deformation rearrangement cost is calculated in conjunction with the cumulative-release spatial offset, and the dominant imbalance measuring point causing the offset is located, including: combining the spatial coordinates of the bridge boundary measuring points. and The total longitudinal length of the bridge boundary The accumulated-release space offset and the accumulation center Calculate the cost of rearranging between measurement points The calculation formula is:

[0135] in, and The numbering of the bridge boundary measuring points; based on the cyclic average unreleased amount distribution. and the cycle average release distribution Given the constraints, solve for the optimal rearrangement allocation share. The calculation formula is:

[0136] The constraints are satisfied:

[0137] in, To rearrange the allocation of shares, For the set of optimal allocation schemes, The total number of the bridge boundary measuring points; the cost of rearranging the measuring points. With the optimal rearrangement allocation share Multiply and sum to obtain the cost of the deformed rearrangement. The calculation formula is:

[0138] For any of the bridge boundary measuring points, calculate the corresponding measuring point rearrangement participation intensity. The calculation formula is:

[0139] in, Allocate shares for the corresponding reverse optimal rearrangement. To correspond to the reverse rearrangement cost between the measurement points; the participation intensity of the measurement point rearrangement. The largest bridge boundary measuring point is located as the dominant imbalance measuring point.

[0140] The cost of rearrangement between measuring points refers to the unit transfer cost corresponding to the allocation of the unreleased amount at one measuring point to another measuring point for release.

[0141] The geometric midpoint refers to the midpoint between the spatial coordinates of two bridge boundary measurement points.

[0142] The rearrangement allocation share variable refers to the decision variable that represents the proportion of unreleased quantity at one measurement point to released quantity at another measurement point.

[0143] The optimal rearrangement allocation share refers to the optimal transfer ratio obtained under the condition of minimizing the total cost.

[0144] The optimal allocation scheme set refers to the overall scheme composed of the optimal rearrangement allocation share among all measurement point pairs.

[0145] The cost of rearrangement is the global sum of the product of the rearrangement cost and the rearrangement share of all measurement points under the optimal allocation scheme.

[0146] The cost of reordering refers to the total cost incurred when a certain measuring point outputs unreleased quantities to other measuring points.

[0147] The cost of transitioning to a rearrangement refers to the total cost incurred when other measuring points input a release amount to a certain measuring point.

[0148] The intensity of measurement point rearrangement participation refers to the sum of the cost of moving out of the same measurement point and the cost of moving in.

[0149] The imbalance-dominant measuring point refers to the bridge boundary measuring point that contributes the most to the distribution rearrangement within the same temperature cycle.

[0150] Specifically, because this invention does not simply regard the distribution of unreleased energy and the distribution of released energy as two independent statistical graphs, but rather regards them as the entire process of thermally induced constraints transforming from an accumulation state to a release state, the relationship between the two is modeled as a rearrangement process, which can directly reflect the essence of spatial migration and eccentric energy release.

[0151] Specifically, since the release of unreleased amount is affected by the geometric distance between the two measurement points, as well as by whether the transfer is far from the accumulation center and the magnitude of the current offset, writing the distance from the geometric midpoint to the accumulation center, the absolute value of the offset, and the distance between measurement points into the rearrangement cost between measurement points can more sensitively characterize abnormal eccentric rearrangements.

[0152] Specifically, since the actual release distribution is often not a one-to-one correspondence between a single cumulative measurement point and a multiple cumulative measurement point jointly distributing to multiple release measurement points, it is necessary to introduce a rearrangement allocation share variable and find the globally optimal solution under conservation constraints. Only in this way can the spatial rearrangement path with the lowest cost be determined as a whole.

[0153] Specifically, since a measuring point may be both a major output and a major input, looking only at unilateral transfer cannot fully reflect its role in eccentric energy release. Therefore, the sum of the cost of out-of-transfer rearrangement and the cost of in-transfer rearrangement is used to define the strength of the measuring point's rearrangement participation, thereby locating the dominant measuring point of imbalance.

[0154] Specifically, the solution to the rearrangement problem is as follows: Form a non-negative matrix by representing the rearrangement allocation share variables among all boundary measurement points; assign the row sum constraints of each output measurement point to the corresponding cyclic average unreleased quantity distribution, and assign the column sum constraints of each input measurement point to the corresponding cyclic average release distribution; then establish a linear programming model with the objective of minimizing the sum of the products of the rearrangement cost and the allocation share for all measurement point pairs; preferentially use the network simplex method, interior point method, or entropy regularization iterative method to solve the problem; the solution result is the optimal rearrangement allocation share.

[0155] Specifically, the geometric midpoint distance term in the cost of rearrangement between measurement points is set as follows: first, find the geometric midpoint of the longitudinal coordinates of the two measurement points, and then calculate the absolute distance from the geometric midpoint to the cumulative center. This distance is used to amplify abnormal rearrangements in areas far from the normal cumulative center. If the geometric midpoint is close to the cumulative center, the impact of this term is small. If the geometric midpoint is significantly deviated from the cumulative center, this term will increase the cost weight of the corresponding rearrangement path.

[0156] Specifically, when the feasible region is not unique, the optimal allocation share is selected as follows: if the minimum objective value corresponding to multiple solutions is exactly the same, the solution with the shorter total transition distance is preferred; if the total transition distance is still the same, the solution with the sparser allocation matrix is ​​preferred; if the software supports entropy regularization, a unique approximate solution can be obtained by minimizing the regularization parameter and maintaining numerical stability.

[0157] Specifically, the discrete summation method for the deformation rearrangement cost is as follows: read the rearrangement cost and the optimal rearrangement allocation share between all measurement point pairs one by one; multiply the two and then sum them; the total number of measurement points determines how many measurement point pairs are traversed; finally, a scalar is obtained as the deformation rearrangement cost for this temperature cycle.

[0158] Specifically, the statistical methods for the out-of-row rearrangement cost and the in-row rearrangement cost are as follows: For any given test point, first multiply all the optimal rearrangement allocation shares of the row where the test point is located by the corresponding rearrangement cost item by item and sum them up to obtain the out-of-row rearrangement cost; then multiply all the optimal rearrangement allocation shares of the column where the test point is located by the corresponding reverse rearrangement cost item by item and sum them up to obtain the in-row rearrangement cost; finally, add the two together to obtain the test point's rearrangement participation intensity.

[0159] Specifically, the method for determining when imbalance-dominant measurement points are in parallel is as follows: if multiple measurement points have the same participation intensity when rearranged, the measurement point farther from the accumulation center is selected first; if they are still in parallel, the measurement point with a larger cyclic average release distribution is selected first; if they still cannot be distinguished, multiple parallel imbalance-dominant measurement points are allowed to be output and simultaneously marked in the abnormal location results.

[0160] Preferably, the cumulative-release spatial offset and the deformation rearrangement cost are combined to form a comprehensive evaluation parameter set, and the bridge operation status index is solved, including: Compare the distribution of the cycle-averaged unreleased amount at the same bridge boundary measuring point. With the average release distribution of the cycle The smaller of the two values ​​is extracted and summed over all the bridge boundary measurement points to obtain the cyclic distribution overlap. The calculation formula is:

[0161] in, This represents the total number of measuring points at the bridge boundary. This refers to the numbering of the measuring points at the bridge boundary. This indicates the minimum value operation; The constant 1 overlaps with the cyclic distribution. The difference is defined as the cyclic distribution overlap loss, and the accumulated-release spatial offset is... The absolute value of the rearrangement cost Combined with the cyclic distribution overlap loss, the comprehensive evaluation parameter set is constructed. The formula is as follows:

[0162] Extract historical temperature cycles to the current temperature cycle The robust central parameter is calculated using the minimum covariance determinant estimation method for all the aforementioned comprehensive evaluation parameter sets. With robust covariance matrix The calculation formula is:

[0163] in, This represents the robust estimation operation of the minimum covariance determinant; Based on the comprehensive evaluation parameter set The robust central parameter and the robust covariance matrix The robust state distance is obtained by calculating the Mahalanobis distance from the inverse matrix. And divide constant one by constant one and the robust state distance. The sum of these factors is used to solve for the bridge's operational status index. The calculation formula is:

[0164]

[0165] Among them, superscript Indicates matrix transpose operation, superscript This represents the matrix inversion operation.

[0166] Cyclic distribution overlap refers to the total degree of overlap between the cyclic average unreleased amount distribution and the cyclic average released amount distribution at each measuring point within the same temperature cycle.

[0167] Cyclic distribution overlap loss refers to the degree of non-overlapping distribution obtained by subtracting the cyclic distribution overlap from a constant.

[0168] The comprehensive evaluation parameter set refers to a 3-dimensional evaluation vector consisting of the absolute value of the cumulative release spatial offset, the cost of deformation rearrangement, and the overlapping loss of cyclic distribution.

[0169] The robust center parameter refers to the center position estimated using robust statistical methods based on a comprehensive set of evaluation parameters from historical temperature cycles to the current temperature cycle.

[0170] The robust covariance matrix is ​​a matrix of dispersion estimated using robust statistical methods based on a comprehensive set of evaluation parameters from historical temperature cycles to the current temperature cycle.

[0171] Robust state distance refers to the degree of multivariate deviation of the current comprehensive evaluation parameter set from the robust central parameter.

[0172] The bridge operational status index is a status evaluation index between 0 and 1 obtained by robust state distance transformation.

[0173] Specifically, since only looking at the spatial offset can reflect the change in the center position, and only looking at the rearrangement cost can reflect the overall migration cost, but cannot reflect whether the two distributions still have local overlap, the cyclic distribution overlap degree is introduced as a third complementary index. Then, the cyclic distribution overlap loss is formed by subtracting the overlap degree, which can make up for the information on the difference in distribution shape.

[0174] Specifically, because a small number of abnormal cycles may exist during the long-term operation of a bridge, if the baseline is established directly using the ordinary mean and ordinary covariance, the abnormal cycles will skew the baseline. Therefore, the minimum covariance determinant estimation method is used to estimate the center and dispersion in a more robust way.

[0175] Specifically, since the comprehensive evaluation parameter set is a multivariate vector, the size of a single component cannot fully represent the overall degree of deviation. Therefore, the Mahalanobis distance is used to measure the comprehensive deviation of the current cycle relative to the robust baseline, and then converted into a bridge operation status index to obtain a more stable comprehensive status characterization.

[0176] Specifically, the preprocessing method for the comprehensive evaluation parameter set is as follows: before forming the historical sequence, check whether the absolute value of the cumulative release spatial offset, the cost of deformation rearrangement, and the overlapping loss of the cyclic distribution are all non-negative finite numbers; if there are missing values ​​or infinite values, the cycle will not enter the robust statistical sample; if the magnitude difference of the three components is too large, robust standardization can be performed first according to the historical median and quartile scale, and then it will enter the subsequent robust center and robust covariance estimation.

[0177] Specifically, the sample usage method for minimum covariance determinant estimation is as follows: robust statistical modeling is started only after accumulating at least 10 complete temperature cycles in history; if the sample size is insufficient, the empirical mean and empirical covariance of the existing cycles are used as transitional estimates, and the minimum covariance determinant estimation is switched to after the sample size meets the requirements; each update includes all valid complete cycles before the current cycle and the current cycle in the candidate samples.

[0178] Specifically, the handling method when the robust covariance matrix is ​​not invertible is as follows: first check whether the sample size is significantly smaller than the parameter dimension; if the sample size is too small, delay the update; if the sample size is sufficient but the covariance is still ill-conditioned, add a very small positive value to the diagonal of the covariance matrix to stabilize it, and then perform the inverse matrix calculation; the positive value is preferably taken as 1 / 100,000 to 1 / 10,000 of the mean of the covariance diagonal.

[0179] Specifically, the bridge operation status index is updated as follows: it is updated once after each complete temperature cycle is identified; for ongoing unclosed cycles, only temporary indices are calculated and not directly incorporated into the formal historical baseline; when the cycle is closed, the formal results are used to overwrite the temporary results.

[0180] Preferably, the comprehensive evaluation parameter set is classified into bridge operational status levels, and the abnormal locations are output based on the dominant imbalance measuring points, including: Gaussian mixture modeling is performed on all the aforementioned comprehensive evaluation parameter sets, and the optimal number of state categories is automatically calculated using the Bayesian information criterion. The calculation formula is:

[0181] in, The number of state categories is set. Represents the Bayesian information criterion function; Calculate the current temperature cycle The comprehensive evaluation parameter set Posterior probability of belonging to each state category The calculation formula is:

[0182] in, and These are all indexes of the state category. , and The first The weights, mean, and covariance of each Gaussian component. Represents the Gaussian normal distribution function. , and The first The weights, mean, and covariance of each Gaussian component; The state category corresponding to the highest posterior probability is taken as the bridge operation state level. The calculation formula is:

[0183] Extract the accumulated-release space offset The positive and negative signs serve as indicators of the offset direction. The calculation formula is:

[0184] in, Indicates symbolic function operations; Map the dominant imbalance measuring points to the corresponding bridge physical components, and combine this with the offset direction markers. Output the location of the anomaly.

[0185] The set number of state categories refers to the number of categories that are tried one by one as candidate models during Gaussian mixture modeling. It is preferably 2 to 6. Bridge operational status classification typically distinguishes between at least normal and abnormal conditions, and too many categories can lead to unstable fitting when the number of temperature cycle samples is limited.

[0186] The optimal number of state categories refers to the number of best categories selected from all candidate categories based on the Bayesian information criterion.

[0187] Posterior probability refers to the conditional probability that the comprehensive evaluation parameter set of the current temperature cycle belongs to a certain state category.

[0188] The state category index is a sequence number variable used to identify a candidate state category.

[0189] The state category index is a sequence number variable used when traversing all state categories.

[0190] The weight of the kth Gaussian component refers to the prior proportion of the kth state category in the Gaussian mixture model.

[0191] The mean of the kth Gaussian component refers to the central position of the kth state category in the comprehensive evaluation parameter space.

[0192] The covariance of the kth Gaussian component refers to the dispersion matrix of the kth state category in the comprehensive evaluation parameter space.

[0193] The weight of the r-th Gaussian component refers to the prior proportion of the r-th state category in the traversal calculation.

[0194] The mean of the r-th Gaussian component refers to the central position of the r-th state category in the traversal calculation.

[0195] The covariance of the r-th Gaussian component refers to the matrix representing the degree of dispersion of the r-th state category during traversal calculation.

[0196] The bridge's operational status level refers to the operational status category to which the current temperature cycle belongs after probability classification.

[0197] The offset direction mark refers to the bridge offset direction identifier determined by the positive or negative sign of the cumulative released space offset.

[0198] Abnormal location refers to the defect indication result formed after mapping the dominant imbalance measuring point to a specific bridge physical component.

[0199] Specifically, since the classification of bridge operation status is not necessarily limited to only 2 or 3 categories, if the number of categories is pre-defined, it is easy to compress or confuse the differences in status under different bridges, different seasons, and different operating conditions. Therefore, Gaussian mixture modeling of the comprehensive evaluation parameter set and automatic determination of the optimal number of status categories through Bayesian information criterion can make the classification boundary conform to the clustering structure of the data itself.

[0200] Specifically, since the same set of comprehensive evaluation parameters may have a certain degree of membership to multiple state categories at the same time, it is not possible to simply divide them by the nearest distance. Instead, the posterior probability of each state category should be calculated, and the category corresponding to the highest posterior probability should be used as the bridge operation state level. Therefore, this step is essentially an adaptive state discrimination in a probabilistic sense.

[0201] Specifically, since the cumulative release space offset not only has magnitude but also positive and negative directions, and the positive and negative directions directly correspond to the offset direction of the thermal accumulation center and the actual release center along the bridge direction, extracting its positive and negative signs as offset direction markers can add directional information to the abnormal position output.

[0202] Specifically, since the measuring points are merely data carriers, and the maintenance and handling ultimately target physical components such as supports, expansion joints, and beam ends, it is necessary to map the main imbalance measuring points to specific bridge components and then combine them with the offset direction markers to output the abnormal locations in order to form an executable maintenance positioning result.

[0203] Specifically, the implementation of Gaussian mixture modeling is as follows: the comprehensive evaluation parameter set corresponding to all complete temperature cycles is summarized into a sample matrix in chronological order; a Gaussian mixture model is established for each candidate state category; the full covariance form is preferred to preserve the correlation between the three evaluation components; the model training adopts expectation maximization iteration; the initial mean is preferably given by multiple random initializations or cluster center initializations; each candidate model is initialized at least 5 times and the result with the maximum log-likelihood is retained.

[0204] Specifically, the search range for the number of state categories is set as follows: the lower limit of the number of candidate categories is fixed at 2; the upper limit is preferably 6; when the number of historical temperature cycle samples is less than 30, it is not recommended that the upper limit exceed 5; when the number of samples is greater than or equal to 30, it can be relaxed to 8; calculate the Bayesian information criterion value for each candidate category, and take the smallest value as the optimal number of state categories.

[0205] Specifically, the risk ranking method for status levels is as follows: after determining the optimal number of status categories, first calculate the average robust status distance or average bridge operation status index of samples in each category; then arrange them in ascending order of average robust status distance, or in descending order of average bridge operation status index; the best ranking is defined as low risk level, the worst ranking is defined as high risk level, and the remaining categories are arranged in order.

[0206] Specifically, the method for mapping the dominant imbalance measuring points to bridge physical components is as follows: First, establish a one-to-one correspondence list between measuring point numbers and physical components; the list should at least specify the bridge span, abutment or pier location, bearing type, expansion joint number, and beam end direction of the measuring point; after the dominant imbalance measuring point is output, directly search for its corresponding component in the list; if a measuring point covers both the bearing and the adjacent joint, the attribution should be determined based on the on-site installation location and the nearest component.

[0207] Specifically, the output of the offset direction flag and the abnormal location is as follows: if the offset direction flag is positive, the output is an abnormality of a component that is offset in the positive direction of the bridge; if the offset direction flag is negative, the output is an abnormality of a component that is offset in the negative direction of the bridge; if the offset direction flag is zero or close to zero, the output is an abnormality of a component that is not significantly offset but has a dominant imbalance feature; the output result should also give the component name, its location in the bridge direction and the corresponding temperature cycle number.

[0208] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0209] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A bridge operation status assessment method based on spatiotemporal fusion of multi-source heterogeneous data, characterized in that, include: The measured displacement and temperature of the bridge boundary measuring points were collected using low-frequency monitoring equipment, and the acceleration of the bridge boundary measuring points was collected and the vibration energy was extracted using high-frequency monitoring equipment to construct a multi-source spatiotemporal monitoring dataset. Based on the dominant temperature pattern extracted from the multi-source spatiotemporal monitoring dataset, the theoretical free temperature-induced displacement of the bridge boundary measuring point is calculated. The distribution of unreleased temperature-induced displacement is constructed based on the difference between the theoretical free temperature-induced displacement and the measured displacement, and the instantaneous deformation release distribution is constructed by multiplying the vibration energy by the rate of change of the measured displacement. Calculate the cumulative center of the unreleased temperature-induced displacement distribution and the release center of the instantaneous deformation release distribution, and extract the spatial distance between the release center and the cumulative center as the cumulative-release spatial offset. The transformation from the temperature-induced displacement unreleased distribution to the instantaneous deformation release distribution is taken as a rearrangement process. The deformation rearrangement cost is calculated in combination with the cumulative-release spatial offset, and the dominant measuring point causing the imbalance of the cumulative-release spatial offset is located. The cumulative-release spatial offset and the deformation rearrangement cost are combined to form a comprehensive evaluation parameter set, and the bridge operation status index is solved. The comprehensive evaluation parameter set is classified into bridge operation status levels, and the abnormal locations are output according to the dominant imbalance measuring points.

2. The bridge operation status assessment method based on spatiotemporal fusion of multi-source heterogeneous data as described in claim 1, characterized in that, Measured displacement and temperature at bridge boundary measuring points were collected using low-frequency monitoring equipment, and acceleration and vibration energy were extracted from these points using high-frequency monitoring equipment to construct a multi-source spatiotemporal monitoring dataset, including: A unified synchronization time point is set, and the measured displacement and temperature are synchronized to the unified synchronization time point using spline basis functions; Calculate the acceleration autocorrelation value of the acceleration, and use the time interval corresponding to when the acceleration autocorrelation value first drops to zero or less than zero as the adaptive smoothing scale of the measurement point; An exponential decay kernel function is constructed based on the adaptive smoothing scale of the measurement points; The acceleration energy density is obtained by calculating the square of the acceleration. The acceleration energy density is then smoothed by integral processing using the exponential decay kernel function to extract the vibration energy synchronized to the unified synchronization time point. The spatial coordinates of the bridge boundary measuring points are combined with the synchronized measured displacement, temperature, and vibration energy to construct the multi-source spatiotemporal monitoring dataset.

3. The bridge operation status assessment method based on spatiotemporal fusion of multi-source heterogeneous data according to claim 2, characterized in that, Based on the dominant temperature pattern extracted from the multi-source spatiotemporal monitoring dataset, the theoretical free temperature-induced displacement of the bridge boundary measuring points is calculated, including: Calculate the temperature covariance matrix of the temperature in the multi-source spatiotemporal monitoring dataset, and perform eigenvalue decomposition on the temperature covariance matrix to obtain eigenvalues ​​and eigenvectors; The information entropy is obtained by calculating the proportion based on the feature value, the number of retained modes is adaptively determined, and the dominant temperature mode is extracted by combining the number of retained modes and the feature vector. A dynamic weighting factor is constructed by using the sum of the vibration energy at each of the bridge boundary measuring points in the multi-source spatiotemporal monitoring dataset. Construct a spatial topological Laplace matrix based on the spatial coordinates of the bridge boundary measurement points; Using the measured displacement in the multi-source spatiotemporal monitoring dataset as the approximation target, and combining the dominant temperature model, the dynamic weighting factor, and the spatial topological Laplace matrix, a graph regular regression is performed to solve the temperature-to-displacement mapping matrix. By multiplying the mapping matrix with the dominant temperature pattern, the corresponding components are extracted as the theoretical free temperature-induced displacement of the bridge boundary measuring points.

4. The bridge operation status assessment method based on spatiotemporal fusion of multi-source heterogeneous data according to claim 3, characterized in that, Based on the difference between the theoretical free temperature-induced displacement and the measured displacement, a distribution of unreleased temperature-induced displacement is constructed, and the instantaneous deformation release distribution is constructed by multiplying the vibration energy by the rate of change of the measured displacement, including: The difference between the theoretical free temperature-induced displacement and the measured displacement after synchronization is calculated to obtain the constraint difference at the measuring point; The square of the constraint difference at the measuring point is calculated to obtain the pseudo thermal potential energy proxy quantity at the measuring point. The pseudo thermal potential energy proxy quantity at each of the bridge boundary measuring points is divided by the sum of the pseudo thermal potential energy proxy quantities at all the bridge boundary measuring points to construct the distribution of the unreleased temperature-induced displacement. Calculate the absolute value of the first time derivative of the measured displacement after synchronization to obtain the rate of change of the measured displacement; Multiply the synchronized vibration energy by the rate of change of the measured displacement to obtain the instantaneous release intensity at the measuring point; The instantaneous deformation release distribution is constructed by dividing the instantaneous release intensity of each of the bridge boundary measuring points by the sum of the instantaneous release intensities of all the bridge boundary measuring points.

5. The bridge operation status assessment method based on spatiotemporal fusion of multi-source heterogeneous data according to claim 4, characterized in that, Calculate the cumulative center of the unreleased temperature-induced displacement distribution and the release center of the instantaneous deformation release distribution, and extract the spatial distance between the release center and the cumulative center as the cumulative-release spatial offset, including: Calculate the first-order time derivative of the theoretical free temperature-induced displacement at all the bridge boundary measuring points, and obtain the average thermal driving velocity by taking the average value. Based on the zero-crossing time interval of the average thermal driving velocity, a series of continuous temperature cycles are obtained, and the start and end times and duration of each temperature cycle are determined. The temperature cycle being evaluated is defined as the current temperature cycle, and the temperature cycles preceding the current temperature cycle are defined as historical temperature cycles. Between the start and end times, the distribution of unreleased temperature-induced displacement is integrated over time and divided by the duration to obtain the cyclic average distribution of unreleased displacement. Between the start and end times, the instantaneous deformation release distribution is integrated over time and divided by the duration to obtain the cyclic average release distribution; The cumulative center is obtained by multiplying the spatial coordinates of each bridge boundary measuring point with the cyclic average unreleased amount distribution and summing the results. The release center is obtained by multiplying the spatial coordinates of each bridge boundary measuring point by the cyclic average release distribution and summing the results. Calculate the difference between the release center and the accumulation center, and divide the difference by the total longitudinal length of the bridge boundary to obtain the accumulation-release spatial offset.

6. The bridge operation status assessment method based on spatiotemporal fusion of multi-source heterogeneous data according to claim 5, characterized in that, The transformation from the temperature-induced unreleased displacement distribution to the instantaneous deformation release distribution is taken as a rearrangement process. The deformation rearrangement cost is calculated in conjunction with the cumulative-release spatial offset, and the dominant measurement point causing the imbalance of the cumulative-release spatial offset is located, including: Calculate the distance between the geometric midpoint of the spatial coordinates of any two bridge boundary measuring points and the cumulative center. Combine the distance, the absolute value of the cumulative-release spatial offset, the absolute difference between the two spatial coordinates, and the total longitudinal length of the bridge boundary to calculate the rearrangement cost between measuring points. A rearrangement allocation share variable is introduced to characterize the distribution transfer amount between any two bridge boundary measuring points; with the cyclic average unreleased amount distribution and the cyclic average released amount distribution as constraints, and with the objective of minimizing the global sum of the product of the rearrangement cost between the measuring points and the corresponding rearrangement allocation share variable, the optimal rearrangement allocation share is obtained. The rearrangement cost between each measurement point is multiplied by the optimal rearrangement allocation share and then summed globally to obtain the deformed rearrangement cost. For any of the bridge boundary measuring points, the sum of the products of the optimal rearrangement allocation share transferred from the bridge boundary measuring point to other bridge boundary measuring points and the corresponding inter-measuring point rearrangement cost is defined as the out-of-boundary rearrangement cost. The sum of the products of the optimal rearrangement allocation share transferred from other bridge boundary measuring points to the bridge boundary measuring point and the corresponding inter-measuring point rearrangement cost is defined as the in-boundary rearrangement cost. The sum of the out-of-boundary rearrangement cost and the in-boundary rearrangement cost is calculated to obtain the measuring point rearrangement participation intensity. The bridge boundary measuring point with the highest participating intensity is positioned as the dominant unbalance measuring point.

7. The bridge operation status assessment method based on spatiotemporal fusion of multi-source heterogeneous data according to claim 6, characterized in that, The cumulative-release spatial offset and the deformation rearrangement cost are combined to form a comprehensive evaluation parameter set, and the bridge operational status index is solved, including: Compare the cyclic average unreleased amount distribution with the cyclic average released amount distribution at the same bridge boundary measuring point, extract the smaller value of the two and sum them over all the bridge boundary measuring points to obtain the cyclic distribution overlap. The difference between the constant 1 and the cyclic distribution overlap is defined as the cyclic distribution overlap loss; The absolute value of the cumulative-release spatial offset, the deformation rearrangement cost, and the cyclic distribution overlap loss are combined to construct a comprehensive evaluation parameter set corresponding to the current temperature cycle and the historical temperature cycle; Extract the comprehensive evaluation parameter set corresponding to all temperature cycles, including the historical temperature cycle to the current temperature cycle, and calculate the robust central parameter and robust covariance matrix using the minimum covariance determinant estimation method. The Mahalanobis distance is calculated based on the comprehensive evaluation parameter set corresponding to the current temperature cycle, the robust center parameter, and the inverse matrix of the robust covariance matrix to obtain the robust state distance; The bridge operation status index is obtained by dividing constant 1 by the sum of constant 1 and the distance to the robust state.

8. The bridge operation status assessment method based on spatiotemporal fusion of multi-source heterogeneous data according to claim 7, characterized in that, The comprehensive evaluation parameter set is classified into bridge operational status levels, and the abnormal locations are output based on the dominant imbalance measuring points, including: Gaussian mixture modeling is performed on all the extracted comprehensive evaluation parameter sets, and the optimal number of state categories is automatically calculated using the Bayesian information criterion; Calculate the posterior probability of the comprehensive evaluation parameter set corresponding to the current temperature cycle belonging to each state category, and take the state category corresponding to the largest posterior probability as the bridge operation state level; Extract the positive or negative sign of the accumulated-release spatial offset as the offset direction indicator; The imbalance-dominant measuring point is mapped to the physical components of the bridge, and the abnormal location is output by combining the offset direction marker.

Images