Bridge lattice column structure bearing capacity state monitoring method

Abstract

The present application relates to the technical field of bridge health monitoring, and discloses a bridge lattice column structure bearing capacity state monitoring method. The method comprises the following steps: constructing an original monitoring field according to a sensor time sequence signal, and generating a state evolution sequence representing the overall dynamic evolution of the bridge lattice column through multidimensional space reconstruction, so as to capture the nonlinear characteristics of the structure behavior and provide a basis for anomaly detection. The potential abnormal area is located based on the deviation degree from the benchmark model, and the time-varying characteristic spectrum thereof is extracted to drive the digital twin model to simulate and obtain a simulation response spectrum. The simulation response spectrum is spatiotemporally fused with the measured state evolution sequence to generate a hybrid state field combining real data and physical mechanism inference, which makes up for the monitoring blind area and improves the data reliability. Through multiscale decomposition of the hybrid state field, the specific state mode of the bridge lattice column bearing capacity is identified, and more accurate and complete evaluation of the bearing capacity anomaly is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bridge health monitoring, in particular to a bridge lattice column structure bearing capacity state monitoring method. BACKGROUND

[0002] At present, the monitoring of the bearing capacity state of the bridge lattice column generally relies on the time series data collected by the sensor network. The existing technical solutions mainly fall into two categories: one is based on statistical analysis or machine learning model, directly judging the abnormal threshold of the original time series signal features; the other is using a finite element model or a simplified mechanical model for offline simulation, and comparing the simulation results with the measured data statically. These methods constitute the mainstream technical path of the current bridge structure health monitoring.

[0003] The conventional technical solutions have defects. The method of directly analyzing the original time series signal can only capture single or local data anomalies, and is difficult to depict the nonlinear evolution process of the bridge structure as a whole under the coupling action of complex environment and load over time, resulting in insufficient recognition ability of the overall evolution and early weak anomalies of the bearing capacity state. At the same time, the method of relying solely on physical model simulation is decoupled from the dynamic changes of the actual state of the bridge lattice column, the model parameters are updated with lag, and the real-time response characteristics of the structure under a specific damage mode or abnormal state cannot be reflected, so that the matching degree of the simulation results with the actual state is limited, and it is difficult to accurately evaluate the bearing capacity degradation.

[0004] How to effectively reconstruct and track the overall state evolution trajectory of the bridge structure from the massive monitoring data, and how to deeply integrate the high-fidelity dynamic simulation with the real-time monitoring data to form a more accurate and reliable state evaluation field, have become key problems to be solved to improve the monitoring level of the bearing capacity state of the bridge lattice column. SUMMARY

[0005] The purpose of the present application is to provide a bridge lattice column structure bearing capacity state monitoring method to solve the problems raised in the background.

[0006] To achieve the above purpose, the present application provides a bridge lattice column structure bearing capacity state monitoring method, which comprises:

[0007] Based on the time series signal collected by the sensor array, an original monitoring field of the bridge lattice column structure is constructed, the original monitoring field is reconstructed in a multi-dimensional space, and a state evolution sequence of the bridge lattice column is generated;

[0008] According to the state evolution sequence, the deviation degree thereof from a preset reference model is calculated, and a potential abnormal area of the bridge lattice column is located according to the deviation degree;

[0009] extracting a time-varying feature spectrum of the potential abnormal area, driving a digital twin model by using the time-varying feature spectrum to simulate a bearing capacity response, and obtaining a simulation response spectrum;

[0010] spatiotemporally fusing the simulation response spectrum and the state evolution sequence to generate a hybrid state field of the bridge lattice column;

[0011] performing multi-scale decomposition on the hybrid state field, identifying a state mode of the bearing capacity of the bridge lattice column based on a decomposition result, and generating a bearing capacity state monitoring result according to the state mode.

[0012] Preferably, the original monitoring field is subjected to multi-dimensional space reconstruction to generate a state evolution sequence of the bridge lattice column, including:

[0013] spatial grid division is performed on the original monitoring field, and a sensing data vector is assigned to each spatial grid;

[0014] coupling strength between sensing data vectors of adjacent spatial grids is calculated, and a spatial correlation graph is constructed according to the coupling strength;

[0015] random walk is performed on the spatial correlation graph to generate multiple state paths;

[0016] node information of the multiple state paths is fused to generate the state evolution sequence of the bridge lattice column.

[0017] Preferably, the deviation degree of the state evolution sequence from a preset reference model is calculated, including:

[0018] a standard response mode of the bridge lattice column is loaded from the preset reference model;

[0019] the state evolution sequence and the standard response mode are convolved in a frequency domain to obtain a convolution response;

[0020] energy distribution of the convolution response is calculated, and a difference value between the energy distribution and a standard energy template is calculated according to the energy distribution, and the difference value is the deviation degree.

[0021] Preferably, the potential abnormal area of the bridge lattice column is located according to the deviation degree, including:

[0022] a thermal distribution map of a surface of the bridge lattice column is generated based on the deviation degree;

[0023] the thermal distribution map is subjected to adaptive threshold segmentation to obtain multiple candidate regions;

[0024] geometric moments and deviation degree moments of each candidate region are calculated, and the potential abnormal area is selected according to the geometric moments and the deviation degree moments; specifically including:

[0025] for each candidate region, geometric moments are calculated, including zero-order moments representing area of the region, first-order moments representing coordinates of a center of gravity of the region, and second-order moments representing moments of inertia of the region;

[0026] For each candidate region, calculate its moment of skewness, including the mean of skewness, the variance of skewness, the skewness of skewness;

[0027] Combine the moments of geometry and the moments of skewness of each candidate region into a feature vector, and input it into a pre-trained support vector machine classifier;

[0028] According to the output result of the support vector machine classifier, mark the candidate region whose feature vector meets the abnormal condition as a potential abnormal area;

[0029] Perform a morphological closing operation on the screened potential abnormal area to fill the internal cavities and connect adjacent regions to obtain the final potential abnormal area.

[0030] Preferably, the time-varying characteristic spectrum is used to drive the digital twin model to simulate the bearing capacity response, including:

[0031] The time-varying characteristic spectrum is input as a boundary condition into the digital twin model;

[0032] A dynamic load application sequence is constructed inside the digital twin model;

[0033] The digital twin model is run to perform iterative calculations under the action of the dynamic load application sequence, and a response data set during the simulation process is output;

[0034] The response data set is feature compressed to obtain a simulation response spectrum;

[0035] The construction steps of the digital twin model include:

[0036] Based on the design drawings of the bridge lattice column structure, the geometric structure data of the bridge is extracted, including the size of the limb, the arrangement of the patch, and the position of the node;

[0037] According to the bridge lattice column material test report, the constitutive relationship parameters of the material are set, including the elastic modulus, the Poisson's ratio, and the density;

[0038] Define the boundary conditions of the bridge lattice column to simulate the constraint conditions of the lattice column end, including the degree of freedom constraint of the fixed end and the hinged end;

[0039] Generate a finite element mesh on the geometric model, set the element type and mesh density, and ensure that the mesh is refined at key positions;

[0040] Establish a load mapping table to convert the actual load type into model load conditions, including self-weight load, vehicle load, and wind load;

[0041] Calibrate the model through modal test data, adjust the material parameters and boundary conditions, so that the model frequency error with the measured frequency is less than a preset threshold;

[0042] The calibrated model is saved as a digital twin model for carrying out the capacity response simulation.

[0043] Preferably, the simulation response spectrum is spatiotemporally fused with the state evolution sequence to generate a hybrid state field of the bridge lattice column, including:

[0044] The simulation response spectrum is mapped onto the physical space coordinates of the bridge lattice column to form a virtual response field;

[0045] The state evolution sequence is interpolated onto the same spatiotemporal grid as the virtual response field to form a measured state field;

[0046] The virtual response field and the measured state field are subjected to tensor product operation to obtain the hybrid state field.

[0047] Preferably, the hybrid state field is subjected to multi-scale decomposition, and a state mode of the capacity of the bridge lattice column is identified based on the decomposition result, including:

[0048] The hybrid state field is subjected to multi-layer decomposition by wavelet packet transform to obtain sub-band coefficient matrices at different scales;

[0049] The singular value vectors of each sub-band coefficient matrix are calculated, and the singular value vectors at all scales are combined to form a state feature vector;

[0050] The state feature vector is input into a pre-trained mode classifier, and the mode classifier outputs a corresponding state mode.

[0051] Preferably, the singular value vectors of each sub-band coefficient matrix are calculated, including:

[0052] Each sub-band coefficient matrix is subjected to singular value decomposition, and the first several singular values on the main diagonal line thereof are extracted;

[0053] The extracted singular values are arranged in descending order to form a singular value vector corresponding to the sub-band coefficient matrix.

[0054] Preferably, a capacity state monitoring result is generated according to the state mode, including:

[0055] A detailed state description template is configured for each state mode;

[0056] According to the state mode output by the mode classifier, a corresponding state description template is called;

[0057] The spatial information of the potential abnormal area, the numerical information of the deviation degree, and the summary information of the state evolution sequence are filled into the state description template to generate the capacity state monitoring result.

[0058] Preferably, the spatial information of the potential abnormal area, the numerical information of the deviation degree, and the summary information of the state evolution sequence are filled into the state description template, including:

[0059] The vertex coordinates of the circumscribed polygon of the potential abnormal area are extracted as spatial information;

[0060] The statistical mean and variance of the deviation degree are calculated as numerical information;

[0061] The state evolution sequence is down-sampled and normalized to obtain summary information;

[0062] The spatial information, numerical information and summary information are written into the corresponding fields of the state description template according to the predefined format.

[0063] Compared with the prior art, the beneficial effects of the present application are:

[0064] By multi-dimensional spatial reconstruction of the original monitoring field constructed by the time series signal of the sensor array, a state evolution sequence capable of depicting the overall dynamic behavior of the bridge lattice column is generated. The dependence on single signal features in traditional methods is broken, and the discrete time series data is mapped to a high-dimensional state space that can reveal the inherent evolution law of the system. The nonlinear dynamic characteristics and evolution trend of the structure behavior are captured from a global perspective, enhancing the detection sensitivity of the overall drift of the bearing capacity state and early potential weak anomalies, so that the abnormal positioning is no longer dependent on isolated over-standard signal points.

[0065] By extracting the time-varying feature spectrum of the potential abnormal area to drive the digital twin model for simulation, and spatiotemporally fusing the simulation response spectrum with the real state evolution sequence, a hybrid state field is generated. The mode of static operation or simple data comparison of the digital twin model is changed, and the dynamic adjustment and response of the model parameters and boundary conditions according to the measured abnormal features are realized. The simulation process can focus on abnormal patterns, and the obtained response has high physical relevance and scenario consistency. Through subsequent spatiotemporal fusion, the hybrid state field not only contains the actual measured information, but also incorporates simulation inference based on physical mechanisms, filling the data gap in the sensor deployment blind area and correcting the noise or errors that may exist in a single data source, ultimately improving the integrity and accuracy of the bearing capacity state pattern recognition. BRIEF DESCRIPTION OF DRAWINGS

[0066] Figure 1 is a working principle diagram of the bridge lattice column structure bearing capacity state monitoring method of the present application;

[0067] Figure 2 is a flowchart of multi-dimensional spatial reconstruction;

[0068] Figure 3 is a flowchart of spatiotemporal fusion processing;

[0069] Figure 4 For the bridge lattice column surface deviation degree thermal distribution map;

[0070] Figure 5 For the bridge lattice column simulation response spectrum and the measured state field of the moderate damage stage are compared with the space-time fusion response diagram. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0072] Please refer to Figure 1 The present application provides a bridge lattice column structure bearing capacity state monitoring method, which comprises the following steps: constructing an original monitoring field based on the time sequence signals of the bridge lattice column structure collected by a sensor array, wherein the original monitoring field contains time and space dimension sensing information; performing a multi-dimensional space reconstruction operation on the original monitoring field to generate a state evolution sequence capable of representing the state change of the bridge structure; calculating the deviation degree between the state evolution sequence and a preset reference model according to the state evolution sequence, wherein the deviation degree is used to quantify the difference between the current state and the standard state; locating the potential abnormal area of the bridge lattice column according to the spatial distribution of the deviation degree; extracting the time-varying characteristic spectrum of the potential abnormal area, wherein the characteristic spectrum reflects the dynamic characteristics of the abnormal area; driving a digital twin model using the time-varying characteristic spectrum to perform bearing capacity response simulation and obtain a simulation response spectrum; performing space-time fusion processing on the simulation response spectrum and the state evolution sequence to generate a mixed state field that combines simulation and measured information; performing multi-scale decomposition on the mixed state field, identifying the state mode of the bearing capacity of the bridge lattice column based on the decomposition results, and generating the final bearing capacity state monitoring result according to the identified state mode.

[0073] Example 1: refer to Figure 2The original monitoring field is spatially gridded, and a sensing data vector composed of sensor data in the grid is assigned to each spatial grid. The coupling strength between the sensing data vectors of adjacent spatial grids is calculated, and a spatial correlation graph representing the correlation between spatial units is constructed according to the coupling strength. Random walk is performed on the constructed spatial correlation graph to generate multiple state paths covering different spatial paths. The node information of the multiple state paths is fused to generate a state evolution sequence that comprehensively reflects the evolution of the structural space state. The deviation degree of the state evolution sequence from the preset reference model is calculated, which is the standard response mode of the bridge lattice column in the normal state loaded from the preset reference model. The state evolution sequence and the standard response mode are convolved in the frequency domain to obtain a convolution response. The energy distribution of the convolution response is calculated, and the difference value between the energy distribution and the standard energy template is calculated, which is the deviation degree.

[0074] In specific implementation, the original monitoring field is reconstructed in multiple dimensions to generate a state evolution sequence, and the deviation degree of the state evolution sequence from the preset reference model is calculated. The sensor array is arranged in a uniform grid form on the bridge deck and the key section of the bridge lattice column, and the original signal collected by the sensor array is acceleration time series data. The original monitoring field collected by the sensor array is a data set containing time dimension and spatial position information of all sensors. In specific implementation, the spatial grid division of the original monitoring field is based on the physical coordinates of the sensors to divide the bridge surface into rectangular spatial grid units. An acceleration sensor is arranged at the center point of each spatial grid, and a sensing data vector is assigned to each spatial grid, which is directly composed of acceleration time series data collected by the sensor in the grid in the latest time window. The sensing data vector is an array containing amplitude information. The coupling strength between the sensing data vectors of adjacent spatial grids is calculated by calculating the similarity of the joint distribution of the two vectors in the time-frequency domain. Specifically, the calculation formula of the coupling strength is:

[0075] ;

[0076] Wherein: represents the coupling strength between the spatial grid and the adjacent spatial grid , represents the total number of preset frequency bands, represents the power spectral density value of the sensing data vector of the spatial grid at the frequency point , represents the sensing data vector of the spatial grid at the same frequency point The coupling strength between each pair of spatial grids is calculated according to the power spectral density values of the acceleration response signals. A spatial correlation graph is constructed according to the calculated coupling strength, in which each node corresponds to a spatial grid, and each edge connects adjacent spatial grids. The weight of each edge in the spatial correlation graph is assigned by the coupling strength value between the two spatial grids corresponding to the edge. In some embodiments, a transition probability matrix is set before random walk is performed on the spatial correlation graph, and the elements in the transition probability matrix are obtained by normalizing the coupling strength between the connected nodes. The random walk on the spatial correlation graph is performed by starting from each spatial grid node and selecting the next node according to the rules defined by the transition probability matrix, thereby generating multiple state paths with fixed length. The node information of the multiple state paths is fused by extracting the node sequence passed by each state path and performing weighted average on the sensor data vectors of the nodes at the same position in all sequences, thereby generating a state evolution sequence representing the evolution of the overall structural state in time sequence.

[0077] In some embodiments, the deviation degree of the state evolution sequence from a preset reference model is calculated according to the state evolution sequence. The preset reference model is stored in a database. The standard response mode of the bridge lattice column is loaded from the preset reference model, and the standard response mode is a reference acceleration response data template collected and preprocessed under the healthy state of the bridge. The state evolution sequence and the standard response mode are convolved in the frequency domain. Before convolution, the state evolution sequence and the standard response mode are converted to the frequency domain by fast Fourier transform, and the convolution operation is completed in the frequency domain in the form of complex multiplication. The energy distribution of the convolution response is calculated by taking the power spectrum of the frequency domain representation of the convolution result, which represents the distribution of energy at each frequency component. The difference value between the energy distribution and a standard energy template is calculated according to the energy distribution, and the standard energy template is a standard power spectrum corresponding to the standard response mode loaded from the preset reference model. The difference value is calculated by integrating and summing the absolute differences of the corresponding frequency point values of the two power spectra. The integral sum value obtained is the deviation degree. It can be understood that the deviation degree is a scalar value, and the size of the deviation degree directly reflects the overall inconsistency between the structural response represented by the state evolution sequence and the energy distribution characteristics of the healthy reference state.

[0078] Example 2: Locating potential anomaly regions of bridge lattice columns based on deviation begins with generating a thermal distribution map of the column surface based on deviation. Adaptive threshold segmentation is applied to the thermal distribution map to obtain multiple candidate regions with high deviation. The geometric moments and deviation moments of each candidate region are calculated. Geometric moments include the zeroth-order moment representing the region area, the first-order moment representing the region's centroid coordinates, and the second-order moment representing the region's moment of inertia. Deviation moments include the average deviation, variance of deviation, and skewness of deviation. The geometric moments and deviation moments of each candidate region are combined into a feature vector, which is then input into a pre-trained support vector machine (SVM) classifier. Based on the SVM classifier's output, candidate regions whose feature vectors satisfy the anomaly criteria are marked as potential anomaly regions. Morphological closure operations are performed on the selected potential anomaly regions to fill internal voids and connect adjacent regions, resulting in the final potential anomaly regions.

[0079] In practical implementation, potential abnormal areas of the bridge lattice columns are located based on the deviation degree. The deviation degree is a scalar value representing the difference between the state evolution sequence calculated in Example 1 and the preset benchmark model. A thermal distribution map of the bridge lattice column surface is generated based on the deviation degree. The thermal distribution map maps the deviation degree value corresponding to each spatial grid on the surface of the bridge lattice column as a color intensity, thus forming a two-dimensional image. The darker the color in the image, the higher the deviation degree value. Adaptive threshold segmentation is performed on the thermal distribution map. The adaptive threshold segmentation uses the Otsu algorithm to automatically determine a global threshold, segmenting the regions in the thermal distribution map with pixel values ​​higher than the threshold, obtaining multiple connected candidate regions. The candidate regions are the sets of pixels with high deviation degree values ​​in the thermal distribution map. The geometric moments and deviation moments of each candidate region are calculated. The geometric moments include the zeroth moment representing the region area, the first moment representing the region's centroid coordinates, and the second moment representing the region's moment of inertia. The deviation moments include the mean deviation, the variance of the deviation, and the skewness of the deviation. In some embodiments, the zeroth moment in the geometric moments is obtained by counting the total number of pixels within the candidate region, the first moment is obtained by calculating the mean of the coordinates of all pixels within the candidate region, and the second moment is obtained by calculating the inertia tensor of the candidate region relative to its centroid. The average deviation is obtained by calculating the arithmetic mean of the deviation values ​​of all pixels within the candidate region, the variance of the deviation describes the dispersion of the deviation values ​​within the candidate region, and the skewness of the deviation describes the asymmetry of the distribution of the deviation values ​​within the candidate region. Optionally, the variance of the deviation... The calculation formula is:

[0080] ;

[0081] in: Indicates candidate region Deviation variance Indicates candidate region The total number of pixels in the middle. Indicates candidate region The Middle The deviation value of each pixel. Indicates candidate region The average deviation. It can be understood that the variance of the deviation... The larger the value, the stronger the candidate region. The more drastic the fluctuation in the internal deviation value, the more likely it is to indicate local damage or anomaly. The deviation skewness is calculated using the ratio of the third-order central moment to the cube of the standard deviation. The geometric moments and deviation moments of each candidate region are combined to form a feature vector, which is a numerical array containing multiple dimensions such as region area, centroid coordinates, moment of inertia, mean deviation, variance of deviation, and deviation skewness. In some embodiments, the feature vector is normalized before being input into the classifier to eliminate the influence of different dimensions. The feature vector is input to a pre-trained support vector machine (SVM) classifier, a binary classification model trained on historical data containing feature vectors and labels of normal and abnormal region samples. Based on the output of the SVM classifier, it outputs a classification label for each feature vector, marking candidate regions whose feature vectors meet anomaly conditions as potential anomaly regions. Anomaly conditions refer to the classification label output by the SVM classifier corresponding to the "abnormal" category. Optionally, the SVM classifier uses a radial basis function as its kernel function, and the classification decision function is based on the position of the feature vector in the mapping space. Morphological closing is performed on the selected potential anomaly regions. This involves first dilating the binarized potential anomaly region image, then performing erosion to fill in internal holes and connect adjacent regions, resulting in the final potential anomaly region. Essentially, morphological closing smooths the boundaries of potential anomaly regions and merges adjacent small regions that are broken due to noise or incomplete segmentation, making the final potential anomaly region more spatially complete and continuous.

[0082] Example 3: See Figure 3The time-varying characteristic spectrum is input as a boundary condition to the digital twin model. A dynamic load application sequence is constructed inside the digital twin model. The digital twin model is run to perform iterative calculations under the action of the dynamic load application sequence, and a response data set during the simulation process is output. The response data set is feature compressed to obtain a simulation response spectrum. The construction of the digital twin model is based on the geometric structure data extracted from the design drawings of the bridge lattice column, including limb size, patch arrangement, and node position. The material constitutive relationship parameters are set according to the bridge material test report, including elastic modulus, Poisson's ratio, and density. The boundary conditions of the bridge lattice column are defined to simulate the end constraint conditions of the lattice column, including the degree of freedom constraints of fixed ends and hinged ends. Finite element meshes are generated on the geometric model, and the element types and mesh densities are set to ensure that the key parts are refined. A load mapping table is established to convert the actual load types into model load conditions, including self-weight load, vehicle load, and wind load. The model is calibrated through modal test data to adjust the material parameters and boundary conditions, so that the model frequency error is less than the preset threshold. The calibrated model is saved as a digital twin model for bearing capacity response simulation. The simulation response spectrum and the state evolution sequence are spatiotemporally fused to generate a mixed state field of the bridge lattice column, which maps the simulation response spectrum to the physical space coordinates of the bridge lattice column to form a virtual response field. The state evolution sequence is interpolated into the same spatiotemporal grid as the virtual response field to form a measured state field. The virtual response field and the measured state field are subjected to tensor product operation to obtain a mixed state field.

[0083] In a specific implementation, the bearing capacity response simulation driven by the time-varying feature spectrum utilizes the time-varying feature spectrum as a boundary condition input to the digital twin model. The boundary condition is applied in the form of time-varying load or displacement constraint on the nodes or elements corresponding to the final potential abnormal area in the digital twin model. A dynamic load application sequence is constructed inside the digital twin model. The dynamic load application sequence is interpolated according to the feature parameters in the time-varying feature spectrum within the simulation time step, generating a continuously changing load input. The digital twin model, which is a calibrated finite element model, is run under the action of the dynamic load application sequence and iteratively calculated according to the preset time increment step. The response data set in the simulation process is output, including stress, strain, displacement, and acceleration data of each node of the model at each simulation time step. Feature compression is performed on the response data set using principal component analysis, which extracts the first several principal components of the covariance matrix in the response data set, and uses the principal component coefficients as the simulation response spectrum. The construction steps of the digital twin model are based on the design drawings of the bridge lattice column to extract the geometric structure data of the bridge, including limb size, patch arrangement, and node position. The material constitutive relationship parameters are set according to the bridge material test report, including elastic modulus, Poisson's ratio, and density. The boundary conditions of the bridge are defined to simulate the end constraint conditions of the lattice column, including the degree of freedom constraints of fixed and hinged ends. Finite element grids are generated on the geometric model, with unit types and grid densities set to ensure grid refinement in key parts. A load mapping table is established to convert actual load types into model load conditions, including self-weight load, vehicle load, and wind load. The model is calibrated through modal test data. The process of calibrating the model is to adjust the material parameters and boundary conditions so that the error between the natural frequency calculated by the model and the natural frequency measured on site is less than a preset threshold. The calibrated model is saved as a digital twin model, which is used for bearing capacity response simulation.

[0084] In some embodiments, generating a hybrid state field for the bridge lattice column by spatiotemporally fusing the simulated response spectrum and the state evolution sequence involves mapping the simulated response spectrum onto the physical spatial coordinates of the bridge lattice column. The principal component coefficients of the simulated response spectrum correspond to the node positions of the digital twin model. These coefficients are mapped onto a grid with the same spatial resolution as the actual sensor array using spatial interpolation methods, forming a virtual response field, which is a spatiotemporal data field. It can be understood that the virtual response field reflects the structural response prediction of the digital twin model under simulated abnormal load conditions. The state evolution sequence is interpolated onto the same spatiotemporal grid as the virtual response field. The state evolution sequence originates from actual sensor data. Through linear interpolation or spline interpolation methods, the state evolution sequence is aligned with the virtual response field at both time and spatial points, forming a measured state field, which is another spatiotemporal data field. Optionally, the measured state field and the virtual response field have the same number of time points and spatial grid division. Tensor multiplication is performed on the virtual response field and the measured state field. Tensor multiplication multiplies the data values ​​of the two fields at corresponding time and space points. In specific implementations, a hybrid state field is used. The calculation formula is:

[0085] ;

[0086] in: Represents spatial coordinates and time The numerical value of the mixed state field at that location. This represents the virtual response field values ​​at the same coordinates. Represents the measured state field values ​​under the same coordinates, with the sign... This represents tensor product operations, which in practice involve multiplying corresponding elements. In some embodiments, tensor product operations can amplify the consistency or inconsistency between simulated and measured responses at specific spatiotemporal locations. It can be understood that the hybrid state field integrates physical simulation information from the digital twin model and measured evolution information from sensors, providing a richer feature base for subsequent state pattern recognition.

[0087] Example 4: Multi-scale decomposition of the mixed state field Based on the decomposition results, the state mode of the bearing capacity of the bridge lattice column is identified, which is to use wavelet packet transform to perform multi-layer decomposition on the mixed state field to obtain sub-band coefficient matrices at different scales. The singular value vector of each sub-band coefficient matrix is calculated, and the singular value vectors of all scales are combined to form a state feature vector. The state feature vector is input into a pre-trained mode classifier, and the mode classifier outputs the corresponding state mode. The singular value vector of each sub-band coefficient matrix is calculated by singular value decomposition of each sub-band coefficient matrix to extract the first several singular values on the main diagonal line. The extracted singular values are arranged in descending order to form the singular value vector corresponding to the sub-band coefficient matrix.

[0088] In specific implementation, the mixed state field obtained by the tensor product operation of the virtual response field and the measured state field in Example 3 is decomposed by wavelet packet transform. The wavelet packet transform decomposes the time series dimension of the mixed state field layer by layer. The wavelet packet transform uses the selected mother wavelet function and the number of decomposition layers. Each layer of decomposition decomposes the input signal into low-frequency approximation coefficients and high-frequency detail coefficients, and further decomposes all sub-bands. The mixed state field is decomposed layer by layer to obtain sub-band coefficient matrices at different scales. Each sub-band after decomposition corresponds to a frequency band range and a time resolution. Each sub-band coefficient matrix is a two-dimensional array, and the rows of the two-dimensional array correspond to spatial positions, and the columns of the two-dimensional array correspond to time points of the sub-band. In specific implementation, the wavelet packet transform uses the "db4" wavelet basis function, and the number of decomposition layers is set to 3. The singular value vector of each sub-band coefficient matrix is calculated, and the singular value decomposition of each sub-band coefficient matrix is performed. Singular value decomposition decomposes the sub-band coefficient matrix into the product of three matrices. The singular value decomposition formula is represented as:

[0089] ;

[0090] wherein: represents the sub-band coefficient matrix at a specific scale, is the left singular vector matrix, is the right singular vector matrix, represents the transpose matrix of , is a diagonal matrix, The main diagonal elements of the matrix are singular values. The first several singular values on the main diagonal are extracted, and in a specific implementation, five singular values are extracted. The extracted singular values are arranged in descending order to form a singular value vector corresponding to the subband coefficient matrix. Singular value vectors of all scales are combined to form a state feature vector, which is a one-dimensional array obtained by sequentially concatenating singular value vectors of all subbands. Wavelet packet decomposition produces multiple subbands, and the dimension of the state feature vector is equal to the number of subbands multiplied by the number of singular values extracted from each subband. Referring to Table 1, the state feature vector contains information.

[0091] Table 1: State feature vector information table

[0092] Subband number Band description Singular value 1 Singular value 2 Singular value 3 Singular value 4 Singular value 5 1 Low frequency approximation 11 ]]> ​ 12 ]]> ​ 13 ]]> ​ 14 ]]> ​ 15 ]]> ​ 2 High frequency detail 1 21 ]]> ​ 22 ]]> ​ 23 ]]> ​ 24 ]]> ​ 25 ]]> ​ 3 High frequency detail 2 31 ]]> ​ 32 ]]> ​ 33 ]]> ​ 34 ]]> ​ 35 ]]> ​ 4 High frequency detail 3 41 ]]> ​ 42 ]]> ​ 43 ]]> ​ 44 ]]> ​ 45 ]]> ​

[0093] In Table 1, each row corresponds to a subband, the subband number indicates the order of the decomposed subband, the frequency band description qualitatively describes the approximate frequency range of the subband, and the singular value 11 to singular value 45 The symbols represent the singular value values extracted from the corresponding subband coefficient matrix, and singular value 1 to singular value 5 represent the first to fifth singular values arranged in descending order. Singular values reflect the energy concentration characteristics and main modes of the corresponding subband coefficient matrix. In some embodiments, the state feature vector contains a total of 20 singular values extracted from multiple subbands. It can be understood that the state feature vector encodes the main structural features of the mixed state field in different time scales and frequency scales in a compact mathematical form. The state feature vector is input to a pre-trained pattern classifier, and the pattern classifier uses a deep neural network model. The pattern classifier is trained on a historical data set containing state feature vector samples and their labels corresponding to the bridge in various known bearing capacity states. The pattern classifier receives the state feature vector as input, and the pattern classifier calculates the probability of the state feature vector belonging to each preset state mode category through multiple layers of nonlinear transformation inside the pattern classifier. The pattern classifier outputs a state mode corresponding to the input state feature vector, which is a classification label indicating the most likely category to which the current bridge lattice column bearing capacity state belongs. Optionally, the output layer of the pattern classifier uses a Softmax activation function to obtain the probability distribution of each category. It can be understood that by performing wavelet packet multi-scale decomposition on the mixed state field and extracting singular value vectors, subtle difference patterns of structural responses in the time-frequency domain under different states can be effectively captured, and these patterns are classified and recognized by a deep neural network, thereby realizing automatic discrimination of the bearing capacity state.

[0094] Referring to Figure 4With the bridge lattice column transverse and longitudinal grid coordinates as two-dimensional space dimensions, the deviation degree distribution of each space grid on the surface of the bridge lattice column is quantitatively displayed through color gradient (corresponding to the right color scale): the numerical interval of 0.00 to 2.00 in the color scale represents the change of deviation degree from low to high, and the color transitions from light color system (low deviation degree) to dark color system (high deviation degree). The "potential abnormal area" is clearly marked in the figure. Combined with the thermal distribution of the deviation degree, it can be known that: the area (corresponding to the grid coordinates about 8-14 transversely and 8-14 longitudinally) presents the characteristics of deep color aggregation, and its deviation degree is significantly higher than that of the surrounding area, which is consistent with the technical logic of locating the potential abnormal area based on the deviation degree, that is, the spatial area with significantly abnormal deviation degree is identified by visualizing the spatial distribution of the deviation degree through the thermal map and combining the characteristics of the area geometric moment and the deviation degree moment, which provides accurate abnormal area positioning basis for subsequent extraction of time-varying feature spectrum and bearing capacity simulation.

[0095] In example 5, according to the state pattern output by the pattern classifier, a corresponding state description template is called. The spatial information of the potential abnormal area, the numerical information of the deviation degree, and the summary information of the state evolution sequence are filled into the state description template to generate the bearing capacity state monitoring result. The spatial information of the potential abnormal area, the numerical information of the deviation degree, and the summary information of the state evolution sequence are filled into the state description template, which takes the vertex coordinates of the circumscribed polygon of the potential abnormal area as the spatial information. The statistical mean and variance of the deviation degree are calculated as the numerical information. The state evolution sequence is down-sampled and normalized to obtain the summary information. The spatial information, numerical information, and summary information are written into the corresponding fields of the state description template according to the predefined format.

[0096] For each state pattern, a detailed state description template is configured, and the state pattern includes "normal", "slight damage", "moderate damage", and "early warning". The state description template is a text framework with predefined structure fields, and the state description template includes monitoring conclusions, abnormal positions, quantitative indicators, and historical change summaries. According to the state pattern output by the pattern classifier, the corresponding state description template is called, and the pattern classifier outputs the "moderate damage" label to call the state description template configured for the "moderate damage" state pattern. The vertex coordinates of the circumscribed polygon of the potential abnormal area are extracted as spatial information, and the circumscribed polygon is the smallest convex polygon enclosing the potential abnormal area. The vertex coordinates are the two-dimensional coordinate sequence of each corner point of the polygon in the global coordinate system of the bridge, and the vertex coordinates can be described as The statistical mean and variance of the deviation degree are calculated as numerical information, the statistical mean is the arithmetic mean of all the deviation degree values of the spatial grid cells, and the variance is the second-order central moment describing the dispersion degree of all the deviation degree values. The state evolution sequence is down-sampled and normalized to obtain the summary information, the down-sampling is uniformly interval sampling of the state evolution sequence in the time dimension to reduce the data length, and the normalization is linear transformation of the down-sampled data to the interval [0, 1]. It can be understood that the summary information reflects the main change trend of the state evolution sequence without containing all the details. Optionally, the formula of the normalization processing is:

[0097] ;

[0098] wherein: represents the state evolution sequence data vector after down-sampling, represents the minimum value in the vector represents the maximum value in the vector represents the normalized summary information vector, and all the element values in the vector are between 0 and 1. In specific implementation, the spatial information, the numerical information and the summary information are written into the corresponding fields of the state description template according to the predefined format. The spatial information is written into the field marked as “abnormal area vertex coordinates” in the form of a coordinate list, the numerical information is written into the field marked as “deviation degree statistical indicators” in the text format of “mean: X, variance: Y”, and the summary information is written into the field marked as “evolution trend summary” in the form of the simplified data vector. In some embodiments, the state description template also contains fixed information fields such as time stamp and bridge identifier, which are directly obtained from the system configuration and filled. The bearing capacity state monitoring result is generated, which is a structured text report generated after filling all the fields. It can be understood that the bearing capacity state monitoring result integrates all the analysis conclusions from anomaly detection, positioning, simulation to pattern recognition, and presents in a standardized form, which is convenient for engineering and technical personnel to directly check and make decisions. Optionally, the bearing capacity state monitoring result can also be converted into JSON structured data format for calling by other systems in addition to the text report format.

[0099] Referring to

[0100] Figure 5 ​​In the spatiotemporal fusion analysis of the moderate damage stage, the coupling of the simulated response spectrum and the measured state field relies on the tensor product fusion technology. In the specific operation, the virtual response field is composed of the discrete field structure of the simulation response spectrum mapped to the physical space coordinates of the bridge lattice column, and the measured state field is interpolated to the continuous state distribution field through the state evolution sequence in the same space-time grid. The fusion of the two field structures is realized by tensor product operation: the tensor product of the virtual response field (simulation response spectrum) and the measured state field (after interpolation) is performed grid by grid, and the mixed state field is generated to represent the comprehensive response state of the bridge lattice column. In the quantitative comparison of field data, the spatial grid number corresponds to the spatial discrete unit of the bridge structure, and the response value reflects the mechanical response strength of each unit in the moderate damage stage. In the parameter configuration process, the spatial grid resolution of the virtual response field and the measured state field is consistent to ensure the spatial matching of the tensor product operation.

[0101] It should be noted that, in this document, the terms such as first and second are used merely to distinguish one entity or operation from another, and do not necessarily require or imply that these entities or operations exist in any actual relationship or order. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0102] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring the bearing capacity state of a bridge lattice column structure, characterized in that, The method includes: constructing an original monitoring field of the bridge lattice column structure based on time-series signals collected by a sensor array, reconstructing the original monitoring field in a multi-dimensional space, and generating a state evolution sequence of the bridge lattice column; The deviation from the preset benchmark model is calculated based on the state evolution sequence, and the potential anomaly zone of the bridge lattice column is located based on the deviation. Extract the time-varying feature spectrum of the potential anomaly region, and use the time-varying feature spectrum to drive the digital twin model to perform load-bearing capacity response simulation to obtain the simulation response spectrum; The simulated response spectrum and the state evolution sequence are spatiotemporally fused to generate a hybrid state field for the bridge lattice column. The mixed state field is decomposed at multiple scales, and the state patterns of the bridge lattice column bearing capacity are identified based on the decomposition results. Bearing capacity state monitoring results are generated based on these state patterns. The original monitoring field is then reconstructed in a multi-dimensional space to generate a state evolution sequence for the bridge lattice column. This process includes: dividing the original monitoring field into spatial grids and assigning a sensing data vector to each spatial grid; calculating the coupling strength between the sensing data vectors of adjacent spatial grids and constructing a spatial correlation graph based on the coupling strength; performing a random walk on the spatial correlation graph to generate multiple state paths; and fusing the node information of the multiple state paths to generate a state evolution sequence for the bridge lattice column. Locating potential anomaly zones in bridge lattice columns based on the deviation degree includes: generating a thermal distribution map of the bridge lattice column surface based on the deviation degree; performing adaptive threshold segmentation on the thermal distribution map to obtain multiple candidate regions; calculating the geometric moments and deviation moments of each candidate region, and selecting potential anomaly zones based on the geometric moments and deviation moments; specifically including: for each candidate region, calculating its geometric moments, including the zero-order moment representing the region area, the first-order moment representing the region's centroid coordinates, and the second-order moment representing the region's moment of inertia; for each candidate region, calculating its deviation moments, including the average deviation, the variance of the deviation, and the skewness of the deviation; combining the geometric moments and deviation moments of each candidate region into a feature vector, and inputting it into a pre-trained support vector machine classifier; based on the output of the support vector machine classifier, marking candidate regions whose feature vectors satisfy the anomaly conditions as potential anomaly zones; performing morphological closure operations on the selected potential anomaly zones to fill the internal voids and connect adjacent regions to obtain the final potential anomaly zones; The simulation response spectrum and the state evolution sequence are spatiotemporally fused to generate a hybrid state field for the bridge lattice column. This process includes: mapping the simulation response spectrum onto the physical space coordinates of the bridge lattice column to form a virtual response field; interpolating the state evolution sequence onto the same spatiotemporal grid as the virtual response field to form a measured state field; and performing a tensor product operation on the virtual response field and the measured state field to obtain the hybrid state field. The mixed state field is decomposed into multiple scales, and the state patterns of the bridge lattice column bearing capacity are identified based on the decomposition results. This includes: using wavelet packet transform to decompose the mixed state field into multiple layers to obtain sub-band coefficient matrices at different scales; calculating the singular value vector of each sub-band coefficient matrix, and combining the singular value vectors of all scales to form a state feature vector; inputting the state feature vector into a pre-trained pattern classifier, and the pattern classifier outputs the corresponding state pattern.

2. The method for monitoring the bearing capacity state of bridge lattice column structures as described in claim 1, characterized in that, The deviation from the preset benchmark model is calculated based on the state evolution sequence, including: loading the standard response mode of the bridge lattice column from the preset benchmark model; convolving the state evolution sequence and the standard response mode in the frequency domain to obtain the convolution response; calculating the energy distribution of the convolution response; and calculating the difference value between the energy distribution and the standard energy template, wherein the difference value is the deviation.

3. The method for monitoring the bearing capacity state of bridge lattice column structures as described in claim 1, characterized in that, The method of using the time-varying feature spectrum to drive a digital twin model for load-bearing capacity response simulation includes: inputting the time-varying feature spectrum as a boundary condition into the digital twin model; constructing a dynamic load application sequence within the digital twin model; running the digital twin model to perform iterative calculations under the action of the dynamic load application sequence, and outputting a set of response data during the simulation process; and performing feature compression on the response data set to obtain the simulation response spectrum. The construction steps of the digital twin model include: extracting the geometric structural data of the bridge lattice columns based on the design drawings, including member dimensions, sprue arrangement, and node positions; and based on the bridge lattice column material... The process involves: compiling a material test report; setting constitutive parameters for the material, including elastic modulus, Poisson's ratio, and density; defining boundary conditions for the bridge lattice columns; simulating end constraints of the lattice columns, including degrees of freedom constraints at fixed and hinged ends; generating a finite element mesh on the geometric model, setting element types and mesh density, and ensuring mesh refinement in key areas; establishing a load mapping table to convert actual load types into model load conditions, including self-weight load, vehicle load, and wind load; calibrating the model using modal test data, adjusting material parameters and boundary conditions to ensure that the error between the model frequency and the measured frequency is less than a preset threshold; and saving the calibrated model as a digital twin for load-bearing capacity response simulation.

4. The method for monitoring the bearing capacity state of bridge lattice column structures as described in claim 1, characterized in that, Calculating the singular value vector of each sub-band coefficient matrix includes: performing singular value decomposition on each sub-band coefficient matrix and extracting the first few singular values ​​on its main diagonal; arranging the extracted singular values ​​in descending order to form the singular value vector corresponding to the sub-band coefficient matrix.

5. The method for monitoring the bearing capacity state of a bridge lattice column structure as described in claim 1, characterized in that, The process of generating bearing capacity status monitoring results based on the state patterns includes: configuring a detailed state description template for each state pattern; calling the corresponding state description template according to the state pattern output by the pattern classifier; filling the state description template with the spatial information of the potential anomaly zone, the numerical information of the deviation, and the summary information of the state evolution sequence to generate bearing capacity status monitoring results.

6. The method for monitoring the bearing capacity state of a bridge lattice column structure as described in claim 5, characterized in that, Filling the state description template with the spatial information of the potential anomaly region, the numerical information of the deviation, and the summary information of the state evolution sequence includes: extracting the vertex coordinates of the circumscribed polygon of the potential anomaly region as spatial information; calculating the statistical mean and variance of the deviation as numerical information; performing downsampling and normalization on the state evolution sequence to obtain summary information; and writing the spatial information, the numerical information, and the summary information into the corresponding fields of the state description template according to a predefined format.

Images