A multi-factor coupled freezing-thawing damage risk quantitative evaluation method for electric poles in cold regions

Abstract

The present application relates to the technical field of electric pole freeze-thaw damage risk quantitative evaluation, and particularly relates to a multi-factor coupling calculation method for distribution network electric pole freeze-thaw damage risk quantitative evaluation in severe cold regions, comprising the following steps: obtaining multi-source heterogeneous data of electric poles in severe cold regions, including air temperature time series data, soil frost heaving displacement data, layered soil volume water content data, electric pole concrete strength and reinforcement parameters, icing thickness and duration data, terrain and micro-geographical space attribute data, underground water level dynamic monitoring data and historical operation and maintenance fault label data; and extracting electric pole apparent crack feature parameters from the crack segmentation model in the electric pole full-circumferential visible light image collected by the unmanned aerial vehicle, and extracting vibration modal frequency offset from the vibration response signal collected by the acceleration sensor arranged on the electric pole through the operational modal analysis algorithm.

Description

Technical Field

[0001] This invention relates to the field of quantitative assessment of freeze-thaw damage risk of utility poles, and specifically to a multi-factor coupled calculation method for quantitative assessment of freeze-thaw damage risk of utility poles in cold regions of power distribution networks. Background Technology

[0002] Assessment of freeze-thaw damage to concrete power poles in frigid regions is a key technology for disaster prevention and mitigation in power distribution networks. With the development of sensor networks and digital data processing technology, both domestic and international approaches are gradually shifting from qualitative judgment based on manual inspections to quantitative assessment based on environmental monitoring data. Existing methods mostly rely on statistical analysis of the number of freeze-thaw cycles, threshold determination of soil frost heave, or empirical regression formulas. Some schemes introduce machine learning models to fit historical fault samples, constructing a preliminary risk classification system.

[0003] However, existing technologies still have room for improvement in terms of data acquisition dimensions, multi-factor coupled modeling capabilities, model adaptive update mechanisms, and evaluation closed-loop evolution, especially in the areas of integrating pole structure damage characterization information, constructing multi-physics nonlinear coupled quantitative models, and realizing regional damage propagation analysis. Summary of the Invention

[0004] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0005] This invention provides a multi-factor coupled method for quantitatively assessing the risk of freeze-thaw damage to utility poles in frigid regions, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides a multi-factor coupled method for quantitatively assessing the risk of freeze-thaw damage to utility poles in frigid regions, executed by a computer system, comprising the following steps:

[0007] Preferred steps: Step 1: Acquire multi-source heterogeneous data of power poles in frigid regions, including time-series air temperature data, soil frost heave displacement data, layered soil volumetric moisture content data, power pole concrete strength and reinforcement parameters, ice thickness and duration data, topographic and micro-geographical attribute data, groundwater level dynamic monitoring data, and historical operation and maintenance fault tag data; and extract apparent crack feature parameters of power poles from circumferential visible light images of power poles collected by UAVs using a crack segmentation model, and extract vibration mode frequency offsets from vibration response signals collected by acceleration sensors deployed on power poles using a working modal analysis algorithm.

[0008] Specifically: The following further explains the extraction methods of crack feature parameters and vibration modal frequency offsets involved in step one. The vibration modal frequency offsets are obtained as follows: a triaxial accelerometer is deployed at the top or upper middle part of the pole to collect the vibration response signal of the pole under environmental excitation. After detrending and bandpass filtering preprocessing of the vibration response signal, the first three natural frequencies and damping ratio of the pole are extracted using a random subspace modal recognition algorithm. The difference between the current measured natural frequency and the initial reference natural frequency of the pole is taken as the vibration modal frequency offset. The crack feature parameters are extracted as follows: the visible light image is segmented at the pixel level using a deep convolutional neural network based on the U-Net architecture to output a crack region mask. The maximum crack width, crack distribution density, and average crack orientation angle are extracted from the mask as crack feature parameters. The training data of the deep convolutional neural network is historical pole inspection images labeled with crack pixel-level tags. The loss function is a weighted combination of cross-entropy loss and Dice coefficient.

[0009] Preferred step: Step 2, perform adaptive spatiotemporal fusion filtering on the multi-source heterogeneous data, use spatial neighborhood reference station weighted fusion and time decay window to remove outliers, interpolate missing values ​​and perform spatiotemporal alignment, and extract micro-topography correction coefficients to locally correct the temperature cycle factor and soil moisture content factor, thereby generating a standardized freeze-thaw metadata dataset.

[0010] Specifically: The adaptive spatiotemporal fusion filtering involved in step two is further explained below. The adaptive spatiotemporal fusion filtering uses the spatial location of the target pole as an index, selecting multiple reference stations within a preset spatial radius as spatial neighbor reference stations. A fusion weight is assigned to the observation values ​​of each reference station. The fusion weight is determined by normalizing the product of the spatial distance attenuation factor and the time attenuation factor. The spatial distance attenuation factor decays exponentially with the square of the spatial distance, and the attenuation rate is controlled by the spatial attenuation bandwidth; the time attenuation factor decays exponentially with the absolute value of the observation time difference, and the attenuation rate is controlled by the time attenuation constant. Fusion weights The calculation formula is:

[0011]

[0012] in Target pole Reference Station Spatial distance, For spatial attenuation bandwidth, The time decay constant, For the observation time, For reference station For missing data segments, interpolation is performed using the spatiotemporal weighted average of multiple reference stations. The micro-topography correction coefficients are extracted as follows: using a high-resolution digital elevation model acquired by airborne lidar, the elevation, slope, aspect, sky visibility factor, and topographic humidity index at each pole are extracted. These topographic attributes are input into a pre-trained lightweight gradient boosting tree model, which outputs the local correction coefficients for the temperature cycle factor and the soil moisture content factor. The training labels of the lightweight gradient boosting tree model are derived from the deviation between the measured local temperature of the poles equipped with tower-level meteorological sensors and the temperature of the macro-meteorological station in the region. During training, this deviation value is used as the regression target, and the topographic attributes are used as input features.

[0013] Preferred Step 3: Based on the standardized freeze-thaw metadata, construct a multi-factor coupled freeze-thaw damage quantification mathematical model. This mathematical model consists of a main damage term formed by summing the cumulative freeze-thaw energy dissipation term, the additional damage term from icing, and the material degradation attenuation term. This is then multiplied sequentially by a crack correction factor, a modal stiffness correction factor, and a Bayesian posterior probability correction factor to output a single pole freeze-thaw damage risk quantification value. Based on the value range of the risk quantification value, four risk levels are defined. This mathematical model serves as the baseline model for physical information enhancement. Its output intermediate features will be used as part of the input features of the neural network constructed in Step 4. The feature library is used to uniformly store and manage the multimodal features of each pole, including measured features and predicted features generated by simulation. Before each risk assessment, the latest feature vector of the corresponding pole is extracted from the feature library and used as the input to the mathematical model and neural network. When the feature library is incrementally updated, the updated features will be automatically referenced in the next assessment cycle.

[0014] Specifically: Regarding the mathematical model for step three, the detailed composition of its internal terms is as follows: the cumulative term of freeze-thaw energy dissipation. The calculation method is as follows:

[0015]

[0016] In the formula, The weighted value for time-varying decay of stratified moisture content. As an effectiveness discrimination factor, it is set to 1 when the temperature drop is greater than or equal to the minimum effective freeze-thaw cycle temperature difference threshold and the freezing duration is greater than or equal to the minimum effective freeze-thaw cycle freezing duration threshold; otherwise, it is set to 0. The total number of effective freeze-thaw cycles, For the first The temperature drop in the next cycle, For the duration of the freeze, For the first Soil volumetric moisture content, For the first Soil layer thickness, The total soil integral depth, This is the displacement due to frost heave. and These are the minimum temperature difference threshold for effective freeze-thaw cycles and the minimum freezing duration threshold, respectively. to For hyperparameters, , This is a normalized reference value.

[0017] Additional damage from icing for:

[0018]

[0019] in To assess the maximum icing thickness during the time period, This is a reference value for icing thickness. , For hyperparameters,

[0020] Material degradation and degradation term for:

[0021]

[0022] in , These are the measured reinforcement ratio and the measured compressive strength, respectively. , For the corresponding nominal value, to This is a hyperparameter.

[0023] Crack semantic correction factor for:

[0024]

[0025] in (Maximum crack width) (Crack density) (Sine value of the average strike angle of the crack) For the corresponding critical value, , These are the weighting coefficients and the power exponent.

[0026] Modal stiffness correction factor for:

[0027]

[0028] in This is the measured natural frequency offset. The initial reference natural frequency, This is the gain coefficient.

[0029] Bayesian posterior probability correction factor for:

[0030]

[0031] in , This is a priori pseudo-counting. This represents the number of faulty pole samples within the neighborhood. This represents the total number of pole samples within the neighborhood.

[0032] Final risk quantification value The combination of the above items yields:

[0033]

[0034] The hyperparameters of the above coupling models to , , The solution strategy is as follows: using historical fault samples, a step-by-step estimation strategy is employed to determine each hyperparameter; whereby... to The expectation-maximization algorithm is used to perform joint maximum likelihood estimation of the freeze-thaw energy dissipation accumulation term, the icing-induced damage term, and the material degradation attenuation term. , and The estimation is based independently on the statistical distribution of crack features and the distribution of modal migration features. The step-by-step estimation strategy decomposes the high-dimensional parameter space into multiple low-dimensional subspaces by fixing other types of hyperparameters in each optimization step, thereby avoiding the curse of dimensionality and ensuring convergence stability. The value range of the four risk levels is divided as follows: a risk quantification value of less than 0.25 indicates mild freeze-thaw damage, greater than or equal to 0.25 and less than 0.50 indicates moderate freeze-thaw damage, greater than or equal to 0.50 and less than 0.75 indicates severe freeze-thaw damage, and greater than or equal to 0.75 indicates failure.

[0035] Preferred Step 4: Construct a four-branch multilayer feedforward neural network. The four branches include a freeze-thaw energy product branch, a soil frost heave spatial field branch, a crack visual feature branch, and a modal stiffness offset branch. The input of the freeze-thaw energy product branch includes the freeze-thaw energy dissipation accumulation term output in Step 3 as a physical prior feature. The input of the soil frost heave spatial field branch includes the time-varying decay weighted value of layered moisture content output in Step 3 as a physical prior feature. The backpropagation algorithm with Bayesian prior constraints is used to adaptively adjust the neuron connection weights of each branch. Based on the proportion of fault samples in the region and the presence or absence of fault labels, the working condition type is determined as a high-sample balanced working condition, a locally sparse working condition, or a newly added fault-free area working condition. The corresponding layered cyclic weight adjustment strategy is called according to the working condition type. The enhancement of the four-branch multilayer feedforward neural network by the physical prior features includes, in addition to using the output of the mathematical model as the branch input feature, using the intermediate quantity output by the mathematical model as the physical constraint regularization term in the loss function of the neural network to guide the network training process.

[0036] Specifically: The following further explains the network structure and weight adjustment strategy involved in step four. In the four-branch multilayer feedforward neural network: the freeze-thaw energy product branch receives an input feature subset consisting of the freeze-thaw cycle temperature difference sequence, the freezing duration sequence, and the freeze-thaw energy dissipation accumulation term; the soil frost heave spatial field branch receives an input feature subset consisting of the time-varying sequence of layered soil moisture content, the frost heave displacement sequence, and the time-varying attenuation weighted value of the layered moisture content; the crack visual feature branch receives an input feature subset consisting of crack feature parameters; and the modal stiffness offset branch receives an input feature subset consisting of the natural frequency offset and... The input feature subset is composed of the damping ratio change. After each branch independently completes the nonlinear mapping of the input feature subset to the high-dimensional latent space, the outputs of the four branches are concatenated by the fully connected fusion layer and input to the shared latent layer. The gating fusion coefficient between the four branches is a trainable scalar parameter that is multiplied by each branch output before it is input to the shared latent layer, which is used to adjust the contribution ratio of each branch output during fusion. The backpropagation algorithm with Bayesian prior constraints introduces a penalty term in the loss function to shrink the weights to the prior mean. The prior distribution adopts a Gaussian distribution form, and its hyperparameters are estimated by the empirical Bayesian method through joint training of historical data from multiple regions.

[0037] The hierarchical cyclic weight adjustment strategy for the three types of working conditions is as follows: Under the high sample equilibrium working condition, the parameters of the bottom convolutional layer and attention layer of the crack visual feature branch and the modal stiffness offset branch are frozen, the weights and biases of the high-level fully connected layers of the two branches are updated, and the adaptive learning rate update based on the weight Fisher information matrix is ​​performed on all hidden layer weights of the freeze-thaw energy product branch and the soil frost heave spatial field branch. L1 sparsity constraints are applied to the gated fusion coefficients to achieve automatic branch pruning and reactivation. Under the local sample sparsity working condition, the hierarchical Bayesian method is used to bind the weight parameters of each branch to a shared super-prior distribution. Elastic weight consolidation is implemented for the crack visual feature branch and the modal stiffness offset branch. The virtual sample generation cycle is started. The concrete freeze-thaw damage constitutive model is extrapolated along the dimensions of freeze-thaw cycle number and frost heave displacement amplitude to generate synthetic severe samples and synthetic failure samples. The weight of the virtual samples in the loss function is gradually reduced with the accumulation of real samples according to the exponential decay scheduling. Under the working condition of new deployment area with no faults in the whole domain, the meta-model trained by multiple regions is loaded. Using a finite element forward modeling solver as the teacher network and the four-branch multilayer feedforward neural network as the student network, physical-guided self-supervised adaptation is performed using local measured temperature sequences and soil parameters. A virtual sample set covering four levels of damage is generated in batches as initial training data. After the first real fault label is generated, the process gradually transitions from the cold start stage to the semi-supervised incremental update stage. The specific method for extrapolating and generating synthetic severe samples and synthetic failure samples in the virtual sample generation loop is as follows: taking the environmental parameters and material parameters of poles with light and moderate freeze-thaw damage in the area as the starting point, the parameters are increased by a preset incremental step size along the freeze-thaw cycle number dimension and by a linear proportional relationship between the freeze-thaw displacement amplitude and the freeze-thaw cycle number along the frost heave displacement amplitude dimension. Each set of incremented parameters is input into the finite element forward modeling solver, and the corresponding theoretical modal frequency offset value and crack width expansion are calculated. When the modal frequency offset value obtained by the forward modeling exceeds 10% or the crack width exceeds the critical crack width, the feature vector corresponding to the set of parameters is marked as a synthetic severe sample or a synthetic failure sample.

[0038] Preferably, in step five, a three-dimensional virtual equipment model of the pole is constructed using multi-source inspection data, and the multi-modal freeze-thaw damage features are mapped to the corresponding spatial location of the model. The finite element forward modeling solver embedded in the three-dimensional virtual equipment model of the pole is called to perform virtual simulation forward modeling of freeze-thaw damage with thermal-water-mechanical three-field coupling. Based on the comparison between the simulation results and the existing feature library, the feature library is incrementally updated at the geometric level, damage level, or modal level.

[0039] Specifically: The following further explains the 3D model construction, simulation solution, and incremental update involved in step five. The construction process of the 3D virtual equipment model of the utility pole is as follows: Data from UAV visible light inspection images, airborne LiDAR point clouds, and handheld terminal ledger metadata are collected. After spatiotemporal consistency verification, statistical outlier removal filtering and moving least squares upsampling are performed on the LiDAR point clouds. A region growing segmentation algorithm based on the geometric prior of the utility pole is used to extract the main point clusters of the utility pole. The central axis of the utility pole is fitted, and orthogonal cross-sections are cut along the central axis. Each cross-section is fitted with a closed B-spline curve contour, and then skinned and lofted to generate a parametric 3D surface model of the main body of the utility pole. Visible light images are used to generate auxiliary component models through motion recovery structure multi-view reconstruction and Poisson surface reconstruction. After iterative nearest point precise registration and fusion, a complete 3D virtual equipment model of the utility pole is formed. The thermal-hydraulic-mechanical three-field coupled finite element forward model solver solves the transient heat conduction equation containing the latent heat of phase change in the temperature field, the moisture field solves the unsaturated Richards equation, and the mechanical field solves the linear elastic stress equilibrium equation considering the strain of the frost-swelled body. The freeze-thaw damage evolution of concrete materials is described by the Mazars damage model. Within each freeze-thaw cycle step, the three fields are sequentially coupled and solved to update the mesh node displacement, stress, and isotropic damage scalar, and the elastic modulus of the damage element is reduced proportionally according to the damage scalar. The Mazars damage model drives the evolution of damage variables with equivalent strain, and characterizes the material stiffness degradation through two independent damage scalars: tensile damage scalar and compressive damage scalar. The larger value of the tensile damage scalar and the compressive damage scalar is taken as the damage scalar used for subsequent incremental update judgment of the three-level feature library.

[0040] The three-level feature library incremental update judgment logic is as follows: The first level is geometric deformation detection, which compares the horizontal displacement of the top of the pole obtained from simulation with the previously recorded displacement in the feature library. When the displacement increment exceeds the preset deformation increment threshold, geometric incremental update is triggered, and the full model version is upgraded and the corresponding node coordinate transformation matrix and skin surface shape are updated. The second level is damage propagation detection, which extracts the set of elements in the simulation cloud map whose damage scalar exceeds the preset damage propagation threshold. It performs a union-intersection operation with the previously recorded damage element set in the feature library. When the ratio of the area of ​​the newly added damage element to the total surface area of ​​the pole exceeds the preset area ratio threshold, damage-level incremental update is triggered. The crack texture is hot-applied to the crack feature layer and the generation source is recorded as simulation. The third level is modal migration detection, which compares the natural frequency change rate obtained from simulation with the most recent measured modal migration in the feature library. When the difference exceeds the preset modal difference tolerance, modal incremental update is triggered. The modal attribute field is written in real time and the update time and simulation confidence mark are attached.

[0041] Preferred step: Step 6, construct a freeze-thaw damage topology model and adjacency matrix based on the risk quantification value and material stiffness characteristics of all poles in the region, search for weak paths on the freeze-thaw damage topology model with high-risk poles as the source point and the risk values ​​of all nodes along the path exceeding a preset threshold, and identify key weak paths in the region.

[0042] Specifically: The following further explains the topology graph construction and path search involved in step six. The edge weights of the adjacency matrix are determined as follows: For each pair of adjacent poles, the product of their respective risk quantification values ​​is calculated. The formula is: Power the result, multiply the above three factors, and then normalize the result by using the sum of the products of the target pole and all its neighboring poles as the denominator. The specific formula is as follows:

[0043]

[0044] in For utility poles stiffness eigenvalues The connection distance threshold, Given a set of neighboring nodes, the weak path search algorithm is as follows: taking the confirmed severely damaged or failed poles as the source point and the healthy poles on the power supply side or main line as the target nodes, enumerate all feasible paths, select paths along the route where the risk quantification value of all nodes is not lower than the severe threshold to form a high-risk path subset, calculate the risk propagation resistance and comprehensive risk propagation cost of each path, take the one with the largest comprehensive risk propagation cost as the critical weak path, and output the path node sequence.

[0045] Specifically: Step 7: Optimize resources for the adaptive iteration of the neural network, 3D virtual simulation and deduction, and regional topology risk analysis through a three-priority computing power dynamic scheduling mechanism. Utilize bidirectional feedback verification between the 3D virtual device model parameters and the evaluation model parameters, as well as multi-dimensional evolutionary effect evaluation of simulation-actual consistency, computing power input-output ratio, and feature library quality to achieve full-domain closed-loop iteration.

[0046] Preferably, the following further explains the computing power scheduling and closed-loop evolution mechanism involved in step seven. The three-priority computing power dynamic scheduling mechanism maintains the following three priority queues: The first priority is the measured data write-back trigger queue, which allocates reserved computing resources for tasks that need to be written into the feature library after feature extraction of the new round of multi-source inspection data and processes them in real time; the second priority is the high-risk tower emergency incremental update queue, which preferentially allocates idle resources for the emergency incremental update tasks of the feature library of poles whose risk quantification value jumps across the risk level and their first-order neighbor poles; the third priority is the periodic batch simulation update queue, which performs simulations on all poles in the area sequentially according to a preset period. The simulation and hierarchical incremental update tasks dynamically adjust the number of parallel poles and the granularity of the simulation step size based on resource reserves. Before the start of each scheduling cycle, a rolling resource estimate is performed. If the estimated demand exceeds the preset high-load threshold of the total resource capacity, the third-priority task is downgraded. Downgrade processing includes increasing the simulation time step, coarsening the finite element mesh by one level, or temporarily skipping the simulation batch of low-risk poles, resuming the simulation once resource reserves are restored. The bidirectional feedback verification of the 3D virtual device model parameters and the evaluation model parameters specifically involves: when the feature library undergoes a geometric incremental update that causes changes in the geometric parameters of the 3D virtual device model of the pole, the changed geometric parameters are automatically transmitted. The input preprocessing module of the freeze-thaw damage quantification model updates the feature vector and spatial neighborhood coupling edges. When the risk quantification prediction value output by the freeze-thaw damage quantification model after weight adaptive adjustment shows a systematic deviation in the region, a regression model is constructed by collecting the deviation of all marked poles in the region and their material constitutive parameters in the 3D model. The material constitutive parameter correction amount minimizes the expected value of the overall deviation in the region through Bayesian optimization inverse solution. The correction amount is synchronously applied to the material property fields of the 3D models of all poles in the region and the damage parameters of the forward solver. At the same time, the material constitutive parameters in the multi-factor coupled freeze-thaw damage quantification mathematical model are updated based on the corrected material constitutive parameters. The relevant inputs of the material degradation attenuation term are used to feed the corrected material parameters back into the mathematical model. The three-dimensional evolution effect evaluation includes: simulation-measurement consistency evaluation, which statistically analyzes the consistency rate between the measured damage level and the simulated predicted damage level of the faulty poles within a preset time period; computing power input-output ratio evaluation, which statistically analyzes the ratio of the total computing power consumed within the evolution cycle to the improvement value of the regional risk detection rate before and after the cycle, and compares it with a preset historical average efficiency baseline. The total computing power is statistically analyzed using the number of hours the central processing unit cores run as the quantification unit; and feature library quality evaluation, which checks the update frequency of feature records for each pole and generates supplementary inspection tasks for poles with feature gaps.

[0047] Compared with the prior art, the technical effects of the present invention are as follows:

[0048] 1. By using weighted fusion of multiple reference stations in the spatial neighborhood and time decay interpolation, combined with the extraction and application of micro-topography correction coefficients, the spatial bias and temporal asynchrony of input data caused by the sparseness of monitoring stations, differences in sampling frequency, and microclimate differentiation are effectively reduced. The fusion weights comprehensively consider the spatial distance decay and observation time difference decay characteristics. Missing data segments are reconstructed by the spatiotemporal weighted mean of multiple reference stations. Micro-topography correction uses a high-resolution digital elevation model to extract topographic attributes and outputs local correction coefficients through a pre-trained model to refine the macro-meteorological data, thereby significantly narrowing the statistical bias of freeze-thaw cycle counts.

[0049] 2. A multi-factor coupled freeze-thaw damage quantitative mathematical model was constructed by combining the main damage term with multiple correction factors. The main damage term is composed of the sum of the freeze-thaw energy dissipation accumulation term, the icing-induced additional damage term, and the material degradation attenuation term. It comprehensively expresses the damage contributions from three aspects: temperature cycle driving, icing load superposition, and time-varying attenuation of material resistance. Then, it is multiplied by the crack correction factor, modal stiffness correction factor, and Bayesian posterior probability correction factor in turn. The apparent crack characterization, structural stiffness degradation, and prior information of spatial neighborhood historical failure statistics are incorporated into the assessment in a nonlinear manner. The detection capability of severe damage and failure level is significantly improved compared with traditional methods.

[0050] 3. By organically integrating the mechanism-driven mathematical model with the data-driven four-branch neural network, the cumulative term of freeze-thaw energy dissipation and the time-varying attenuation weighted value of layered moisture content output by the mathematical model are used as the physical prior feature inputs of the freeze-thaw energy product branch and the soil frost heave spatial field branch of the neural network, respectively. This realizes the effective injection of physical knowledge into the data-driven model. With the three-type working condition hierarchical cyclic weight adjustment strategy, hierarchical progressive full parameter update is performed under high sample equilibrium working condition. Under the working condition of local sample sparseness, hierarchical Bayesian strong prior constraints and virtual sample generation attenuation mechanism are used to prevent overfitting. Under the working condition of new landed area without faults in the whole domain, physical guidance self-supervised adaptation is used to achieve zero sample cold start. The three strategies are automatically switched with the change of sample distribution, which ensures the generalization ability and prediction stability of the model under complex scenarios such as long-tailed sample distribution, zero label in new landed area and distribution drift. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0052] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0053] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0054] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0055] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0056] Example 1:

[0057] like Figure 1 As shown, taking a power distribution network in a frigid region as an example, there are 2,153 concrete poles in operation in this area, with an average of about 82 freeze-thaw cycles per year. The extreme low temperature in winter can reach -41°C. Temperature sensors, layered soil moisture probes (at three depths of 30cm, 60cm, and 90cm), frost heave displacement gauges, and triaxial acceleration vibration sensors are deployed on each pole. UAVs are used to collect visible light inspection images every month, and airborne lidar acquires a digital elevation model once a year before winter.

[0058] When acquiring the vibration modal frequency shift, a triaxial accelerometer is installed on the top of the pole, with a sampling frequency of 200Hz. Environmental excitation response signals are collected for 10 minutes each time. After bandpass filtering from 0.5 to 20Hz, the first three natural frequencies are identified using the random subspace method. The initial measurement value after initial installation is used as the reference frequency. Crack feature parameters were extracted using the U-Net semantic segmentation network, which was trained on 10,000 historical pole inspection images labeled with pixel-level crack tags. The loss function was a weighted combination of cross-entropy loss and Dice coefficient, and the segmentation accuracy IoU reached 0.87.

[0059] When step two is executed, the target pole The spatial neighborhood radius is set to 15km, and the spatial attenuation bandwidth is... Take 10km as an example, and the time decay constant. Taking a 24-hour period, for a 48-hour period of sensor loss, the root mean square error of the reconstructed temperature sequence was 1.8℃ using weighted average interpolation from four surrounding reference stations. In the micro-topography correction, the training labels of the lightweight gradient boosting tree model were derived from 30 utility poles in the region where tower-level meteorological sensors have been deployed. The deviation between their measured minimum winter temperature and the minimum temperature of the nearest macro-meteorological station during the same period was used as the regression target. After training, the model was generalized to the entire region. For example, the temperature cycle factor correction coefficient of a utility pole located on a north slope was 1.12, indicating that the actual freeze-thaw temperature difference at that location was 12% higher than the regional average. This correction value was substituted into the subsequent coupled model to compensate for the deviation caused by the slope aspect.

[0060] In step three, the model computation uses a step-by-step estimation strategy to determine the hyperparameters of the coupled model: first, the expectation-maximization algorithm is used to estimate the hyperparameters of the coupled model. to Perform joint maximum likelihood estimation, and then... , and The parameters are independently estimated based on the statistical distribution of crack characteristics and the distribution of modal migration characteristics, respectively. Other hyperparameters are fixed during each solution step, thus decomposing the high-dimensional parameter space into multiple low-dimensional subspaces for sequential solution. The water content exponential decay coefficient is included in this calculation. The value is 1.8, reflecting the nonlinear amplification effect of moisture content on frost heave drive, and the ice thickness index. The value is 1.2, reflecting the sublinear effect of icing thickness. These parameters are the optimal values ​​obtained for this region using the step-by-step estimation strategy described above. Different regions can be recalibrated based on actual data. In a particular assessment, the effective freeze-thaw cycle cooling amplitude... Freeze duration h, the volumetric water content of the three soil layers were 0.22, 0.28, and 0.31, respectively, and the frost heave displacement was... Valid number of loops Calculations show that the time-varying attenuation weighted value of stratified moisture content... Accumulated freeze-thaw energy dissipation Additional damage caused by icing Material degradation and attenuation item The total damage factor is 0.46; the crack correction factor is 1.05 due to the maximum crack width of 0.3mm (critical value 0.5mm); the modal correction factor is 1.16 due to the frequency shift of 4.8%; and the Bayesian correction factor is 0.92 due to the 4 faulty poles in the neighborhood. The assessment result falls into the severe range, which is consistent with the actual inspection that found multiple longitudinal cracks in the pole and excessive tilt rate.

[0061] In step four, the freeze-thaw energy product branch receives not only the freeze-thaw cycle temperature difference sequence and the freezing duration sequence, but also the freeze-thaw energy dissipation accumulation term output from step three. As a priori physical feature, the proportion of fault samples in this area is about 7%, with only 11 samples of severe or worse damage. Based on the fact that the proportion of fault samples is lower than the preset threshold and that fault labels exist, the system determines that the sample is locally sparse and starts a virtual sample generation loop. Starting with the parameters of 3 mild and 8 moderate freeze-thaw damaged poles, the loop increases by 5 times with each freeze-thaw cycle, and the frost heave displacement increases synchronously with a linear proportional relationship. The loop is then input into the finite element forward modeling solver to generate 50 synthetic severe and failed samples with modal frequency shifts exceeding 10% or crack widths exceeding the critical value. The initial weight of the virtual samples is set to 0.8. As 3 real severe fault samples are added in winter, the weight of the virtual samples decreases exponentially to 0.5. After the model is updated, the detection rate on the validation set is improved.

[0062] Step 5 involves performing 3D reconstruction and simulation of the pole. In the Mazars damage model, the initial threshold for the tensile damage scalar is taken as the ultimate tensile strain of the concrete. The initial threshold for compressive damage scalar is taken as the ultimate compressive strain of concrete. The softening curve exponent is set to 0.8. The damage scalar used is determined to be the larger value between the tensile damage scalar and the compressive damage scalar. It is predicted that the pole top displacement will increase to 32mm and the modal frequency will decrease by 3.2Hz by the end of the next winter, triggering an incremental update of the damage level. The predicted crack texture is thermally added to the crack feature layer of the feature library. In the next evaluation cycle, the system extracts the updated crack feature parameters of the pole from the feature library as input to the mathematical model and neural network. Subsequent on-site inspections found that the crack propagation location predicted by the simulation was close to the actual location of the newly added crack.

[0063] Step 6: Construct the topology graph and calculate the risk product index in edge weight calculation. Stiffness ratio index Using the severely damaged pole as the source point, a critical weak path was found that passes through 4 moderately damaged poles to reach the healthy pole on the main line.

[0064] In step seven, the bidirectional feedback verification collected the deviation of the marked poles in the region, obtained the material constitutive parameter correction through Bayesian optimization inverse solution, and synchronized it to the material property field of the 3D model and the damage parameters of the forward solver. At the same time, based on the corrected material constitutive parameters, the material degradation attenuation term in the mathematical model of step three was updated. The relevant input parameters are used to feed the corrected material parameters back to the mechanism model, thereby achieving a closed-loop synchronization between the mechanism model and the three-dimensional simulation model.

[0065] Example 2:

[0066] like Figure 1 As shown, taking the area with 2153 utility poles as the object, a control group was set up using the traditional single-factor freeze-thaw assessment scheme. This scheme only calculates the cumulative negative temperature value based on the daily minimum temperature of the weather station. When the cumulative negative temperature value exceeds the preset experience threshold, it is judged as high risk, and is combined with manual inspection and visual confirmation. The experimental group adopted the complete process of this invention, and recorded 47 freeze-thaw damage failures in the past five years.

[0067] Data input completeness dimension: The control group had only 4 effective features, while the experimental group had 37. Risk value calculation accuracy dimension: The control group detected 29 faults, with a detection rate of 61.7% and 83 false alarms, with a false alarm rate of 3.9%; the experimental group detected 42 faults, with a detection rate of 89.4% and a false alarm rate of 1.2%. Small sample / zero sample fit dimension: A sub-region within the area that had only been operating for 2 years and had only 3 minor faults was selected. The control group did not provide any warnings for the subsequent 2 severe tilt faults, while the experimental group provided warnings for the 9 faults before the fault occurred. Within 0 days, the risk quantification value is output, indicating the entry into the severe zone, and work orders are automatically dispatched. In terms of damage prediction lead time: the control group averaged 18 days in advance, while the experimental group averaged 67 days in advance, extending the warning window by approximately 3.7 times. In terms of regional operation and maintenance costs: the control group had an annual inspection manpower input of approximately 420 person-days, resulting in 3 sudden pole collapses due to missed reporting, with an average delay of 4.2 days in handling and a loss of approximately 370,000 yuan; the experimental group reduced manpower to 185 person-days, and no pole collapses occurred through pre-reinforcement, resulting in a reduction of approximately 56% in overall operation and maintenance costs.

[0068] The above quantitative comparison shows that the present invention demonstrates improvements in detection rate, false alarm rate, zero-sample adaptability, early warning lead time, and operation and maintenance economy.

[0069] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired network and / or wireless network. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives (SSDs).

[0070] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0071] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0072] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only one method, and other division methods may exist in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual couplings or direct couplings or communication connections shown or discussed may be through some interfaces; the indirect couplings or communication connections between apparatuses or units may be electrical, mechanical, or other forms.

[0073] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0074] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0075] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-factor coupled method for quantitatively assessing the risk of freeze-thaw damage to utility poles in frigid regions, executed by a computer system, characterized in that... Includes the following steps: The study acquires multi-source heterogeneous data on utility poles in frigid regions, including time-series temperature data, soil frost heave displacement data, layered soil volumetric moisture content data, pole concrete strength and reinforcement parameters, ice thickness and duration data, topographic and micro-geographical attribute data, groundwater level dynamic monitoring data, and historical maintenance fault label data. Additionally, it extracts apparent crack feature parameters from circumferential visible light images of utility poles collected by drones using a crack segmentation model, and extracts vibration mode frequency offsets from vibration response signals collected by accelerometers deployed on the poles using a working modal analysis algorithm. Adaptive spatiotemporal fusion filtering is performed on the multi-source heterogeneous data. Outlier removal, missing value interpolation, and spatiotemporal alignment are performed using spatial neighborhood reference station weighted fusion and time decay window. Micro-topography correction coefficients are extracted to locally correct the temperature cycle factor and soil moisture content factor, and a standardized freeze-thaw meta-dataset is generated. Based on the standardized freeze-thaw metadata, a multi-factor coupled freeze-thaw damage quantification mathematical model is constructed. The main damage term is generated based on freeze-thaw energy dissipation, icing load, and material degradation. The main damage term is then corrected using apparent crack characteristic parameters, structural modal characteristics, and historical fault statistical characteristics to obtain a single-bar freeze-thaw damage risk quantification value. Based on the value range of the risk quantification value, four risk levels are defined. The output of the mathematical model also includes a freeze-thaw energy dissipation accumulation term and a time-varying decay weighted value of stratified moisture content, which serve as physical prior features for subsequent steps. A four-branch multilayer feedforward neural network is constructed, comprising a freeze-thaw energy product branch, a soil frost heave spatial field branch, a crack visual feature branch, and a modal stiffness offset branch. A backpropagation algorithm with Bayesian prior constraints is used to adaptively adjust the neuron connection weights of each branch. The working condition type is determined based on the proportion of fault samples in the region and the presence of fault labels, and the corresponding hierarchical cyclic weight adjustment strategy is invoked according to the working condition type. The input of the freeze-thaw energy product branch includes the accumulated freeze-thaw energy dissipation term, and the input of the soil frost heave spatial field branch includes the time-varying decay weighted value of the layered moisture content. A three-dimensional virtual equipment model of the utility pole is constructed using multi-source inspection data. Multimodal freeze-thaw damage features are mapped to the corresponding spatial locations in the model. A finite element forward modeling solver embedded in the three-dimensional virtual equipment model of the utility pole is invoked to perform virtual forward simulation of freeze-thaw damage with thermo-hydraulic-mechanical three-field coupling. Based on the comparison between the simulation results and the existing feature library, incremental updates of the feature library at the geometric, damage, or modal levels are triggered. The feature library is used to uniformly store and manage the multimodal features of each utility pole, including measured features and predicted features generated by simulation. Before each risk assessment, the latest feature vector of the corresponding utility pole is extracted from the feature library as the input of the mathematical model and neural network. A freeze-thaw damage topology model and adjacency matrix are constructed based on the risk quantification value and material stiffness characteristics of all poles in the region. On the freeze-thaw damage topology model, weak paths with high-risk poles as the source point and risk values ​​of all nodes along the path exceeding a preset threshold are searched to identify key weak paths in the region. The three-priority computing power dynamic scheduling mechanism optimizes resources for the adaptive iteration of the neural network, three-dimensional virtual simulation and deduction, and regional topology risk analysis. It also utilizes bidirectional feedback verification between the parameters of the three-dimensional virtual device model and the evaluation model parameters, as well as multi-dimensional evolution effect evaluation of simulation-measurement consistency, computing power input-output ratio and feature library quality to achieve full-domain closed-loop iteration.

2. The method for quantitatively assessing the risk of freeze-thaw damage to utility poles in frigid regions based on multi-factor coupling as described in claim 1, characterized in that, The adaptive spatiotemporal fusion filtering uses the spatial location of the target pole as an index to select multiple reference stations within a preset spatial radius as spatial neighbor reference stations. A fusion weight is assigned to the observations of each reference station. This fusion weight is proportional to the product of the spatial distance attenuation factor and the time attenuation factor, and is normalized using the sum of the corresponding products of all reference stations as the denominator. The spatial distance attenuation factor decays exponentially with the square of the spatial distance, and the attenuation rate is controlled by the spatial attenuation bandwidth. The time attenuation factor decays exponentially with the absolute value of the observation time difference, and the attenuation rate is controlled by the time attenuation constant. For missing... The data segment is filled by interpolation using the spatiotemporal weighted average of multiple reference stations; the micro-topography correction coefficient is extracted by using a high-resolution digital elevation model obtained by airborne lidar to extract the altitude, slope, aspect, sky visibility factor, and topographic humidity index at each pole. The above topographic attributes are input into a pre-trained lightweight gradient boosting tree model, which outputs the local correction coefficients of the temperature cycle factor and the soil moisture content factor. The training labels of the lightweight gradient boosting tree model are derived from the deviation between the measured local temperature of the poles equipped with tower-level meteorological sensors and the temperature of the macro-meteorological station in the region.

3. The method for quantitatively assessing the risk of freeze-thaw damage to utility poles in frigid regions based on multi-factor coupling as described in claim 1, characterized in that, The specific structure of the multi-factor coupled freeze-thaw damage quantitative mathematical model is as follows: freeze-thaw energy dissipation accumulation term. The calculation formula is: In the formula, , As a validity discriminant factor, when and If the value is 1, then the value is 0; otherwise, the value is 0. The total number of effective freeze-thaw cycles, For the first The temperature drop in the next cycle, For the duration of the freeze, For the first Soil volumetric moisture content, For the first Soil layer thickness The total soil integral depth, This is the displacement due to frost heave. and These are the minimum temperature difference threshold for effective freeze-thaw cycles and the minimum freezing duration threshold, respectively. to For hyperparameters, , This is a normalized reference value; Additional damage from icing ,in To assess the maximum icing thickness during the time period, Reference value for icing thickness For hyperparameters, material degradation attenuation term in , These are the measured reinforcement ratio and the measured compressive strength, respectively. , For the corresponding nominal value, to For hyperparameters, crack correction factor ,in For the corresponding critical value, For weighting coefficients and power exponents, the modal stiffness correction factor is... ,in This is the measured natural frequency offset. The initial reference natural frequency, Here, represents the gain coefficient, and represents the Bayesian posterior probability correction factor. ,in , This is a priori pseudo-counting. This represents the number of faulty pole samples within the neighborhood. The final risk quantification value is the total number of pole samples within the neighborhood. The hyperparameters of the above-mentioned coupled models are obtained from the maximum likelihood estimation of historical failure samples using the expectation-maximization algorithm; the value range of the four risk levels is divided as follows: a risk quantification value less than 0.25 is mild freeze-thaw damage, greater than or equal to 0.25 and less than 0.50 is moderate freeze-thaw damage, greater than or equal to 0.50 and less than 0.75 is severe freeze-thaw damage, and greater than or equal to 0.75 is failure.

4. The method for quantitatively assessing the risk of freeze-thaw damage to utility poles in frigid regions based on multi-factor coupling as described in claim 1, characterized in that, In the four-branch multilayer feedforward neural network: the freeze-thaw energy product branch receives an input feature subset consisting of the freeze-thaw cycle temperature difference sequence, the freezing duration sequence, and the freeze-thaw energy dissipation accumulation term; the soil frost heave spatial field branch receives an input feature subset consisting of the time-varying sequence of layered soil moisture content, the frost heave displacement sequence, and the time-varying attenuation weighted value of the layered moisture content; the crack visual feature branch receives an input feature subset consisting of crack feature parameters; and the modal stiffness offset branch receives an input feature subset consisting of the natural frequency offset and the damping ratio change. After each branch independently completes the nonlinear mapping from the input feature subset to the high-dimensional latent space, the outputs of the four branches are concatenated by the fully connected fusion layer and input into the shared latent layer. The gating fusion coefficients between the four branches are trainable scalar parameters that are multiplied by the outputs of each branch before they are input into the shared latent layer, which are used to adjust the contribution ratio of the outputs of each branch during fusion. The backpropagation algorithm with Bayesian prior constraints introduces a penalty term in the loss function to shrink the weights towards the prior mean. The prior distribution adopts a Gaussian distribution, and its hyperparameters are estimated by the empirical Bayesian method through joint training of historical data from multiple regions.

5. The method for quantitatively assessing the risk of freeze-thaw damage to utility poles in frigid regions based on multi-factor coupling as described in claim 1, characterized in that, The crack feature parameters are extracted as follows: the visible light image is segmented at the pixel level using a deep convolutional neural network based on the U-Net architecture, and a crack region mask is output. The maximum crack width, crack distribution density, and average crack orientation angle are extracted from the mask as crack feature parameters. The training data of the deep convolutional neural network are historical pole inspection images labeled with pixel-level crack labels, and the loss function is a weighted combination of cross-entropy loss and Dice coefficient.

6. The method for quantitatively assessing the risk of freeze-thaw damage to utility poles in frigid regions based on multi-factor coupling as described in claim 4, characterized in that, The hierarchical cyclic weight adjustment strategy includes: under high sample equilibrium conditions, freezing the parameters of the bottom convolutional and attention layers of the crack visual feature branch and the modal stiffness offset branch, updating the weights and biases of the high-level fully connected layers of the two branches, performing adaptive learning rate updates based on the weight Fisher information matrix for all hidden layer weights of the freeze-thaw energy product branch and the soil frost heave spatial field branch, and applying L1 sparsity constraints to the gated fusion coefficients to achieve automatic branch pruning and reactivation; under locally sparse sample conditions, using a hierarchical Bayesian method to bind the weight parameters of each branch to a shared super-prior distribution, implementing elastic weight consolidation for the crack visual feature branch and the modal stiffness offset branch, and initiating virtual sample generation. The system generates synthetic severe samples and synthetic failure samples by extrapolating the concrete freeze-thaw damage constitutive model along the dimensions of freeze-thaw cycle number and frost heave displacement amplitude. The weight of virtual samples in the loss function gradually decreases exponentially with the accumulation of real samples. Under the condition of a fault-free new construction area, a meta-model jointly trained by multiple regions is loaded. A finite element forward model solver is used as the teacher network and the four-branch multilayer feedforward neural network is used as the student network. Physical guidance and self-supervised adaptation are performed using local measured temperature sequences and soil parameters to generate a batch of virtual sample sets covering four levels of damage as initial training data. After the first real fault label is generated, the system gradually transitions from the cold start stage to the semi-supervised incremental update stage.

7. The method for quantitatively assessing the risk of freeze-thaw damage to utility poles in frigid regions based on multi-factor coupling as described in claim 6, characterized in that, The specific method for extrapolating and generating synthetic severe samples and synthetic failure samples in the virtual sample generation loop is as follows: taking the environmental parameters and material parameters of poles with mild and moderate freeze-thaw damage in the area as the starting point, the parameters are increased by a preset increment step along the dimension of the number of freeze-thaw cycles, and by a linear proportional relationship between the freeze-thaw displacement amplitude and the number of freeze-thaw cycles along the dimension of the freeze-thaw displacement amplitude. Each set of incremented parameters is input into the finite element forward modeling solver, and the corresponding theoretical modal frequency offset value and crack width expansion are calculated. When the modal frequency offset value obtained by forward modeling exceeds 10% or the crack width exceeds the critical crack width, the feature vector corresponding to the set of parameters is marked as a synthetic severe sample or a synthetic failure sample.

8. The method for quantitatively assessing the risk of freeze-thaw damage to utility poles in frigid regions based on multi-factor coupling as described in claim 1, characterized in that, The construction process of the 3D virtual equipment model of the utility pole is as follows: Visible light inspection images from UAVs, point clouds from airborne LiDAR, and metadata from handheld terminal records are aggregated. After spatiotemporal consistency verification, statistical outlier removal filtering and moving least squares upsampling are performed on the LiDAR point cloud. A region growing segmentation algorithm based on the geometric prior of the utility pole is used to extract the main point clusters of the utility pole. The central axis of the utility pole is fitted, and orthogonal cross-sections are cut along the central axis. Each cross-section is fitted with a closed B-spline curve contour, and then skinned and lofted to generate a parametric 3D surface model of the main body of the utility pole. Visible light images are used to generate auxiliary component models through motion-recovery structure multi-view reconstruction and Poisson surface reconstruction. After iterative nearest-point precise registration and fusion, a complete 3D virtual equipment model of the utility pole is formed. The thermal-hydraulic-mechanical three-field coupling has… In the finite element forward modeling solver, the temperature field solves the transient heat conduction equation containing the latent heat of phase change, the moisture field solves the unsaturated Richards equation, and the mechanical field solves the linear elastic stress equilibrium equation considering the strain of the frost-swelled body. The freeze-thaw damage evolution of concrete materials is described by the Mazars damage model. Within each freeze-thaw cycle step, the three fields are solved sequentially and coupled to update the mesh node displacement, stress, and isotropic damage scalar, and the elastic modulus of the damage element is reduced proportionally according to the damage scalar. The Mazars damage model drives the evolution of damage variables with equivalent strain, and characterizes the material stiffness degradation through two independent damage scalars: tensile damage scalar and compressive damage scalar. The larger value of the tensile damage scalar and the compressive damage scalar is taken as the damage scalar used for subsequent judgment.

9. The method for quantitatively assessing the risk of freeze-thaw damage to utility poles in frigid regions based on multi-factor coupling as described in claim 8, characterized in that, The determination logic for the incremental update of the three-level feature library is as follows: the first level is geometric change detection, which compares the horizontal displacement of the top of the pole obtained by simulation with the displacement recorded in the feature library. When the displacement increment exceeds the preset deformation increment threshold, geometric incremental update is triggered, the full model version is upgraded and the corresponding node coordinate transformation matrix and skin surface shape are updated. The second level of damage propagation detection extracts the set of units in the simulation cloud map whose damage scalar exceeds the preset damage propagation threshold. This set is then compared with the previously recorded set of damaged units in the feature library using a union-intersection operation. When the ratio of the area of ​​the newly added damaged unit to the total surface area of ​​the pole exceeds the preset area ratio threshold, a damage-level incremental update is triggered. The predicted crack texture is then hot-applied to the crack feature layer, and the source of its generation is recorded as the simulation deduction. The third level of modal migration detection compares the natural frequency change rate obtained from the simulation deduction with the most recently measured modal migration in the feature library. When the difference exceeds the preset modal difference tolerance, a modal-level incremental update is triggered. This is written to the modal attribute field in real time, along with the update time and simulation confidence level marker.

10. The method for quantitatively assessing the risk of freeze-thaw damage to utility poles in frigid regions using a multi-factor coupled approach as described in claim 1, characterized in that... The vibration mode frequency offset is obtained as follows: a triaxial acceleration sensor is installed at the top or upper middle part of the pole to collect the vibration response signal of the pole under environmental excitation. After the vibration response signal is preprocessed by detrending and bandpass filtering, the first three natural frequencies and damping ratio of the pole are extracted by the random subspace mode recognition algorithm. The difference between the current measured natural frequency and the initial reference natural frequency of the pole is taken as the vibration mode frequency offset.

Images