A method and system for early warning of surface cracks of a UHPC bridge

Abstract

The application provides a UHPC bridge surface crack early warning method and system, the region of the UHPC bridge is classified by grading, according to the region after grading, the multi-source data of each region is collected, the differentiated image and multi-physical field data collection strategy is executed for different classification regions, the spatial-temporal alignment of regional data is completed, the regional UHPC exclusive crack feature library and improved U-Net identification algorithm are constructed, the crack accurate identification and parameter extraction are realized, the regional independent crack expansion-multi-physical field LSTM crack expansion model is established, the quantitative evaluation of the structure health in the region is completed in combination with the finite element inversion, and finally the comprehensive evaluation and damage evolution prediction of the whole bridge are realized through weighted fusion, since the health indexes for evaluating the health states of single region and whole bridge are accurately calculated, the crack early warning can be effectively carried out, and the bridge can be maintained in advance.

Description

Technical Field

[0001] This invention belongs to the field of bridge health early warning technology, and specifically relates to a method and system for early warning of surface cracks in UHPC bridges. Background Technology

[0002] Ultra-High Performance Concrete (UHPC) has been widely used in bridge engineering due to its superior properties such as ultra-high strength, high toughness, and high durability, and has become a core material for the reinforcement and renovation of long-span bridges. Cracks are a key indicator affecting the service safety, durability, and service life of UHPC bridge structures. In particular, the microcracks (width <0.1mm) unique to UHPC materials directly characterize the internal damage evolution of the structure through their initiation and propagation. Accurate crack monitoring, damage assessment, and evolution prediction are core requirements for the operation and maintenance management of UHPC bridges throughout their entire life cycle.

[0003] The existing UHPC bridge crack monitoring technology has the following core shortcomings: 1. Imbalance in the uniform monitoring mode across the entire area: Existing technologies mostly adopt a uniform high-precision and high-frequency monitoring scheme across the entire area. On the one hand, the accuracy of microcrack identification in the core stress area of ​​the bridge cannot be guaranteed in a targeted manner, and it is easy to miss detections. On the other hand, non-critical low-risk areas generate massive amounts of redundant data, resulting in high computing power consumption and poor real-time monitoring, making it impossible to achieve a balance between accuracy and efficiency.

[0004] 2. The division of monitoring areas is highly subjective: the existing division of monitoring areas relies heavily on manual experience and lacks a quantitative and scientific classification method that is adapted to the material properties of UHPC and the stress characteristics of bridges. It cannot adapt to the dynamic changes in the damage state and environmental load during the service of bridges, and it does not consider the connectivity effect of crack propagation in adjacent areas. The area division is isolated and does not conform to the objective law of crack evolution along the stress path.

[0005] 3. Insufficient crack identification accuracy: Existing technologies mostly use a unified crack identification model and general feature library across the entire domain, without considering the essential differences in the initiation mechanism and morphological characteristics of cracks in different areas of UHPC bridges. For microcracks in the high-density, low-porosity background of UHPC, there are serious problems of missed detection and false detection, which cannot meet the accuracy requirements of microcrack monitoring in UHPC bridges.

[0006] 4. Large errors in damage prediction and assessment: Existing technologies mostly construct a unified crack propagation-multiphysics coupling model across the entire domain, without considering the differences in the sensitivity of crack propagation to physical field parameters in different regions. This results in serious "averaging errors" in the models and low accuracy in crack propagation prediction. At the same time, the whole-bridge health assessment mostly adopts a simple averaging method, which does not reflect the decisive impact of core high-risk areas on structural safety. The assessment results cannot be directly connected with the engineering operation and maintenance needs. Summary of the Invention

[0007] Based on this, the present invention provides a method and system for early warning of surface cracks on UHPC bridges, which aims to provide accurate early warning of surface cracks on UHPC bridges.

[0008] A first aspect of this invention provides a method for early warning of surface cracks on a UHPC bridge, the method comprising: S1, the area of ​​the UHPC bridge is divided into hierarchical regions, and multi-source data of each region is collected according to the hierarchical regions. The multi-source data includes image data. S2, based on the image characteristics of different regions, perform differentiated image preprocessing and spatiotemporal alignment processing in sequence to construct hierarchical regional data; S3. Construct a dedicated crack feature sub-library based on the graded regional data, determine the recognition accuracy requirements for different regions, design the U-Net crack recognition algorithm, and use the algorithm to identify cracks. S4. Based on the identification results, cluster the cracks in each region. Based on the crack clustering results, determine the main controlling factors of crack propagation in different regions and construct an independent LSTM crack propagation model for each region. S5 constructs finite element sub-models with differentiated accuracy for different graded regions, combines measured crack parameters with physical field data to complete damage parameter inversion, calculates the structural health index of different regions, and conducts single-region early warning based on the first early warning mechanism; S6. Based on the sub-regional structural health index and regional risk weight, calculate the comprehensive structural health index of the entire bridge through weighted fusion, and conduct a full-bridge early warning according to the second early warning mechanism; S7. Construct a full-bridge damage evolution prediction model based on the LSTM crack propagation model, and output long-term and short-term damage evolution prediction results. The long-term and short-term damage evolution prediction results include regional structural health indices and a comprehensive full-bridge structural health index at future time nodes. The regional structural health index in S5 and the comprehensive full-bridge structural health index in S6 are used to train the full-bridge damage evolution prediction model.

[0009] Furthermore, step S1 includes: A multi-dimensional crack risk assessment index system adapted to the characteristics of UHPC materials was constructed, and the comprehensive weight of the index was determined by subjective and objective coupling weighting. Based on the comprehensive weight of the aforementioned indicators, and combined with the normal cloud model, the basic regional risk is quantified, and the basic regional risk value is calculated. Based on the basic risk value of the region, a spatial lag model is introduced to correct the transmission effect of crack propagation in adjacent regions, and the final dynamic risk value of the region is obtained. The final dynamic risk value of the region is used for adaptive clustering to complete the classification of the monitoring region; Based on the hierarchical division of regions, multi-source data is collected for each region, including image data.

[0010] Furthermore, the multi-dimensional crack risk assessment index system includes 4 criterion layers and 13 quantitative indicators, specifically: C1 is a criterion layer for inherent structural properties and stress characteristics, including component safety level coefficient, UHPC stress level coefficient, construction detail complexity coefficient, and UHPC material degradation coefficient. The intrinsic risk criterion layer for C2 crack initiation and propagation includes fatigue load effect coefficient, historical damage deterioration coefficient, construction quality defect coefficient, and multi-field coupling sensitivity coefficient. C3 service environment and external excitation influence criteria layer, including temperature gradient amplitude coefficient, environmental corrosion level coefficient, and vehicle load impact coefficient; C4 spatial correlation impact characteristic criterion layer, including distance coefficient between adjacent high-risk areas and crack connectivity probability coefficient; All indicators are positive; the higher the indicator value, the higher the risk of cracks.

[0011] Furthermore, in the step of constructing a multi-dimensional crack risk assessment index system adapted to the characteristics of UHPC materials, and determining the comprehensive weight of the index through subjective-objective coupling weighting, the step of determining the comprehensive weight of the index through subjective-objective coupling weighting includes: The indicators under the same criterion level are sorted by importance, the importance ratio of adjacent indicators is determined, the subjective weight of each indicator is calculated recursively and normalized to obtain the global subjective weight. The indicators are standardized by 0-1, the standard deviation and conflict of the indicators are calculated, the information content of the indicators is obtained, and the objective weight of the indicators is obtained after normalization. The global subjective weight and the objective weight of the indicator are coupled by a multiplicative synthesis method to obtain the comprehensive weight of the indicator.

[0012] Furthermore, the step of quantifying the regional basic risk and calculating the regional basic risk value based on the comprehensive weight of the indicators and the normal cloud model includes: Determine the normal cloud digital characteristics of each indicator, and calculate the cloud membership degree of each indicator in each region using a positive normal cloud generator; Based on the comprehensive weight of the aforementioned indicators and cloud membership, the basic risk value for each region is calculated.

[0013] Furthermore, in the step of introducing a spatial lag model to correct the transmission effect of crack propagation in adjacent areas based on the basic risk value of the region, and obtaining the final dynamic risk value of the region, a spatial weight matrix based on adjacency relationship and distance attenuation is constructed and row standardization is performed to calculate the final dynamic risk value of the region. The calculation formula is: ; Where M is the total number of regions, and ρ is the spatial transmission coefficient. For the elements of the standardized spatial weight matrix, This represents the base risk value for the k-th region. This represents the base risk value for the l-th region.

[0014] Furthermore, in the U-Net crack identification algorithm, for the k-th monitoring region, the total loss function L k The calculation formula is: ; in, The weighting coefficient for the k-th region is related to the final dynamic risk value of that region. Matching, L ce L is the cross-entropy loss function, used for pixel-level classification accuracy control. dice The Dice loss function is used to control the accuracy of crack overall contour recognition.

[0015] Furthermore, in step S4, the step of constructing the LSTM crack propagation model includes: For the k-th region, the correlation between crack propagation rate and each physical field parameter is calculated by grey relational analysis. Physical field parameters with a correlation greater than a preset value are selected as the main controlling factors for crack propagation in this region, and a set of main controlling physical field parameters is constructed. The crack parameters and the master physical field parameters in the corresponding region are separated into trend terms, periodic terms and random terms using the STL decomposition method; Using the trend and periodic terms of the physical field parameters after the corresponding region decomposition as inputs, and the trend and periodic terms of the crack parameters as outputs, a regional LSTM crack propagation model is trained to obtain the basic prediction values. Based on the latest measured data and the basic predicted values ​​for each monitoring period, the parameters of the LSTM crack propagation model are corrected in real time using the Bayes formula. The dynamic optimization of the LSTM crack propagation model for the corresponding region is completed in each monitoring period to complete the construction of the LSTM crack propagation model.

[0016] Furthermore, step S5 includes: For different graded regions, finite element sub-models with different precision are constructed. The crack parameters and measured physical field data of the corresponding regions are substituted into the finite element sub-models. With the goal of minimizing the residual between the simulation value and the measured value, the material damage reduction coefficient of the region is obtained by inversion. Based on the material damage reduction factor and the final dynamic risk value of the region, the structural health index of the corresponding region is calculated.

[0017] A second aspect of this invention provides a UHPC bridge surface crack early warning system for implementing the UHPC bridge surface crack early warning method described in the first aspect, the system comprising: The segmentation module is used to hierarchically segment the area of ​​the UHPC bridge, and collect multi-source data for each area based on the segmented area. The multi-source data includes image data. The first construction module is used to perform differentiated image preprocessing and spatiotemporal alignment processing in sequence according to the image characteristics of different regions in order to construct hierarchical regional data; The identification module is used to construct a dedicated crack feature sub-library based on the graded regional data, determine the identification accuracy requirements of different regions, design the U-Net crack identification algorithm, and identify cracks through the algorithm. The second construction module is used to cluster cracks in each region based on the identification results, determine the main controlling factors of crack propagation in different regions based on the crack clustering results, and construct an independent LSTM crack propagation model for the corresponding region. The first calculation module is used to construct finite element sub-models with different precision for different graded regions, combine measured crack parameters and physical field data to complete damage parameter inversion, and calculate the structural health index of the sub-regions. The second calculation module is used to calculate the overall structural health index of the entire bridge by weighted fusion based on the sub-regional structural health index and regional risk weight, and to complete the comprehensive evaluation of the overall health status of the entire bridge. The prediction module is used to construct a full-bridge damage evolution prediction model based on the LSTM crack propagation model and output long-term and short-term damage evolution prediction results. The long-term and short-term damage evolution prediction results include regional structural health indices and a comprehensive full-bridge structural health index at future time nodes. The regional structural health index in the first calculation module and the comprehensive full-bridge structural health index in the second calculation module are used to train the full-bridge damage evolution prediction model.

[0018] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the UHPC bridge surface crack early warning method provided in the first aspect.

[0019] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the UHPC bridge surface crack early warning method provided in the first aspect.

[0020] This invention provides a method and system for early warning of surface cracks on a UHPC bridge. The method involves hierarchically dividing the UHPC bridge into different regions and collecting multi-source data for each region. Differentiated image and multiphysics data acquisition strategies are implemented for different hierarchical regions to achieve spatiotemporal alignment of regional data. A regional UHPC-specific crack feature library and an improved U-Net recognition algorithm are constructed to achieve accurate crack identification and parameter extraction. A regionally independent crack propagation-multiphysics LSTM crack propagation model is established, and combined with finite element inversion, a quantitative assessment of the structural health of each region is completed. Finally, a weighted fusion method is used to achieve a comprehensive assessment of the entire bridge and predict damage evolution. Because a health index is accurately calculated to assess the health status of a single region and the entire bridge, effective crack early warning can be provided, allowing for timely bridge maintenance. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the implementation of a UHPC bridge surface crack early warning method according to Embodiment 1 of the present invention. Figure 2 This is a structural block diagram of a UHPC bridge surface crack early warning system provided in Embodiment 2 of the present invention; Figure 3 This is a structural block diagram of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0022] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0023] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0025] Example 1 According to an embodiment of the present invention, a method for early warning of surface cracks in UHPC bridges is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0026] This first embodiment provides a method for early warning of surface cracks in UHPC bridges, which can be used in electronic devices, such as computers. Please refer to... Figure 1 , Figure 1 The flowchart of a method for early warning of surface cracks on a UHPC bridge provided in Embodiment 1 of the present invention is shown, specifically including steps S1 to S7.

[0027] S1, the area of ​​the UHPC bridge is divided into hierarchical regions, and multi-source data of each region is collected according to the hierarchical regions. The multi-source data includes image data.

[0028] Specifically, a multi-dimensional crack risk assessment index system adapted to the characteristics of UHPC materials is constructed, and the comprehensive weight of the indexes is determined through subjective and objective coupling weighting. The multi-dimensional crack risk assessment index system includes four criterion layers and 13 quantitative indicators, specifically: C1 is a criterion layer for inherent structural properties and stress characteristics, including component safety level coefficient, UHPC stress level coefficient, construction detail complexity coefficient, and UHPC material degradation coefficient. The intrinsic risk criterion layer for C2 crack initiation and propagation includes fatigue load effect coefficient, historical damage deterioration coefficient, construction quality defect coefficient, and multi-field coupling sensitivity coefficient. C3 service environment and external excitation influence criteria layer, including temperature gradient amplitude coefficient, environmental corrosion level coefficient, and vehicle load impact coefficient; C4 spatial correlation impact characteristic criterion layer, including distance coefficient between adjacent high-risk areas and crack connectivity probability coefficient; All indicators are positive; the higher the indicator value, the higher the risk of cracks. In this embodiment of the invention, the component safety level coefficient is quantified according to the highway bridge and culvert design specifications: 1.0 for key components of extra-large bridges, 0.9 for key components of large bridges, 0.8 for key components of medium and small bridges, 0.3~0.7 for minor components, and 0.1~0.2 for auxiliary components; the UHPC stress level coefficient is σ / σ cr σ is the principal tensile stress under the most unfavorable load combination. cr The measured cracking tensile stress of UHPC (a core characteristic parameter of UHPC); the structural detail complexity coefficient is set to 1.0 for weak structures of UHPC, 0.9 for wet joints / splicing interfaces, 0.8 for prestressed anchorage zones, 0.8 for variable cross-section corners, and 0.3 for ordinary straight sections; the UHPC material degradation coefficient is... f 实测 f represents the measured compressive strength of UHPC / the measured fiber bond strength. 设计 To design the compressive strength / fiber bond strength of UHPC; the fatigue load effect factor is the number of vehicle load cycles already borne divided by the fatigue crack life of UHPC (based on UHPC fatigue test curve fitting); the historical damage degradation factor is... b is the proportionality coefficient, w max The maximum historical crack width is given by , v is the average crack propagation rate, and 0 is used for no historical damage; the construction quality defect coefficient is based on the construction characteristics of UHPC, with 1.0 for unqualified fiber dispersion, 0.9 for insufficient pouring, 0.6 for excessive surface flatness, and 0.2 for no defects; the multi-field coupling sensitivity coefficient is the average correlation between crack propagation and strain / temperature / displacement calculated based on grey relational analysis, with higher correlation values ​​resulting in larger values, ranging from 0 to 1; the temperature gradient amplitude coefficient is the ratio of the region's annual maximum positive and negative temperature gradient amplitude to the design value; the environmental corrosion level coefficient is 1.0 for nearshore chloride environment, 0.8 for industrial corrosion environment, and 0.3 for general atmospheric environment; the vehicle load impact coefficient is 1.0 for driveway wheel tracks, 0.7 for non-wheel track driveways, and 0.7 for sidewalks / The flange plate is set to 0.2; the distance coefficient between adjacent high-risk areas is the reciprocal normalized value of the minimum distance to the identified high-risk areas, the closer the distance, the larger the value, and the value ranges from 0 to 1; the crack connectivity probability coefficient is the crack connectivity expansion probability calculated based on the crack direction and spacing of adjacent areas, and the value ranges from 0 to 1, and is 0 if there are no adjacent cracks. More specifically, the steps for determining the comprehensive weight of indicators through subjective and objective weighting include: Indicators under the same criterion level are ranked by importance, the importance ratio of adjacent indicators is determined, the subjective weight of each indicator is recursively calculated and normalized to obtain the global subjective weight. In this embodiment of the invention, experts rank indicators under the same criterion level by importance: , Indicates "higher importance", where m is the number of indicators included in the current criterion layer, and z... t For the t-th indicator within the criterion layer, the importance ratio of adjacent indicators is expressed as: ; This represents the importance ratio between the t-th and (t+1)-th indicators. The unnormalized subjective weight of the t-th indicator within the criterion layer. The unnormalized subjective weight of the (t+1)th indicator within the criterion layer; Recursively calculate the weight of the final indicator: ; The weights of all indicators within the criterion layer are obtained through reverse recursion: ; Finally, after global normalization, the global subjective weight of the j-th indicator is obtained. : ; Let J be the global subjective weight of the j-th indicator, and J be the total number of indicators. The indicators are standardized using 0-1 methods, and their standard deviation and conflict are calculated to obtain the information content of the indicators. After normalization, the objective weights of the indicators are obtained. Specifically, for the original values ​​of the k-th region and the j-th indicator... Standardize 0-1: ; For standardized values, This represents the minimum original value of the k-th region and the j-th indicator. Given the maximum value of the original value of the j-th indicator in the k-th region, calculate the standardized mean of the j-th indicator: ; M represents the total number of regions. For the standardized mean of the j-th indicator, calculate the standard deviation (variability) of the indicator: ; Let j be the standard deviation of the indicator, and calculate the conflict between indicators: ; Q j Due to the conflict of indicators, Calculate the information content of the indicators based on the Pearson correlation coefficient between indicator j and indicator l: ; Normalization yields the objective weight of the j-th indicator: ; Let be the objective weight of the j-th indicator; The global subjective weight and the objective weight of the indicator are coupled using a multiplicative synthesis method to obtain the comprehensive weight of the indicator, which is expressed as: ; The overall weight of the indicators; Based on the comprehensive weights of the aforementioned indicators, and combined with a normal cloud model, the basic regional risk is quantified, and the basic regional risk value is calculated. Specifically, the normal cloud numerical characteristics of each indicator are determined, and the cloud membership degree of each indicator in each region is calculated using a positive normal cloud generator. It should be noted that the normal cloud model uses three numerical characteristics (E... x E n H e A complete description of the uncertainty of the indicator, expected value E x Entropy E is the most representative quantitative value of the indicator, reflecting the central trend of risk. n The fuzziness threshold of the indicator reflects the acceptable range of values, and the hyperentropy H. e The uncertainty of entropy reflects the degree of random fluctuation of the indicator. For the j-th standardized indicator, the cloud digital feature calculation formula is as follows: ; The hyperentropy coefficient, Let j be the expected value of the j-th indicator. Let the entropy of the j-th index be... Let the hyperentropy of the j-th index be... The maximum value of the standardized value. The minimum value of the standardized values; Furthermore, using a forward normal cloud generator, the cloud membership degree of the j-th indicator in the k-th region is calculated, expressed as: ; ; For For expectations, A normally distributed random number with standard deviation. Let be the cloud membership degree of the j-th indicator in the k-th region. This membership degree incorporates both the fuzzy boundary and random fluctuations of the indicator. The mathematical symbol for normal distribution; Based on the comprehensive weight of the aforementioned indicators and cloud membership, the basic risk value for each region is calculated and expressed as follows: ; Based on the basic risk value of the region, a spatial lag model is introduced to correct the transmission effect of crack propagation in adjacent regions, resulting in the final dynamic risk value of the region. Specifically, a spatial weight matrix based on adjacency and distance attenuation is constructed to quantify the spatial correlation strength between regions. W is an M×M matrix, where M is the total number of regions, and the matrix elements are... The calculation formula is: ; d kl Let d be the centroid distance between region k and region l, and d0 be the distance decay threshold. Taking half the span of a single bridge span, W is standardized to obtain the elements of the standardized spatial weight matrix. The sum of the elements in each row is 1; Calculate the final dynamic risk value of the region The calculation formula is: ; Where M is the total number of regions, and ρ is the spatial transmission coefficient, taking values ​​[0,1]. The coefficient is determined based on the UHPC bridge structure and crack propagation characteristics: 0.3 for simply supported beams and 0.5 for continuous beams (cracks in the negative bending moment zone of continuous beams propagate continuously along the beam, resulting in a stronger spatial transmission effect). When ρ=0, it degenerates into a traditional model without spatial correlation. For the elements of the standardized spatial weight matrix, This represents the base risk value for the k-th region. This represents the baseline risk value for the l-th region. The final dynamic risk value of the region is used to complete the classification of the monitoring area through adaptive clustering. Specifically, the K-means++ algorithm is used to realize the adaptive classification of the region. In this embodiment of the invention, the region is divided into at least three types, and the boundaries and mileage coordinates of each region are clearly defined. For example, the first type of core monitoring area includes the tension zone at mid-span of simply supported beam, the negative bending moment zone of continuous beam, the shear compression zone of support, the wet joint / splitting interface, the prestressed anchorage zone, and the parts with historical damage. The second type of secondary critical monitoring area includes the non-shear compression zone of beam web, the non-wheel track zone of bridge deck pavement, and the non-cantilever end of flange plate. The third type of general monitoring area includes the non-sunlight sensitive area on the top surface of beam, the guardrail base, the inner side of flange plate, and other low stress parts. Based on the tiered regional classification, multi-source data is collected for each region. This multi-source data includes image data. It should be noted that differentiated image acquisition equipment, acquisition frequencies, and physical field sensor deployment schemes are matched for different level regions to achieve optimal configuration of monitoring resources. For example, a core monitoring area of ​​category I uses an industrial camera with a resolution of ≥24 megapixels, acquiring data once per hour, and simultaneously deploying dense strain sensors, displacement gauges, and temperature sensors. The sampling frequency is completely synchronized with image acquisition, achieving sub-millimeter-level image resolution and high-frequency multi-physical field synchronous acquisition. A secondary key monitoring area of ​​category II uses an industrial camera with a resolution of ≥12 megapixels, acquiring data once per day, and correspondingly deploying an appropriate number of strain and temperature sensors. The sampling frequency is synchronized with image acquisition. A general monitoring area of ​​category III uses an industrial camera with a resolution of ≥8 megapixels, acquiring data once every 7 days, and deploying only a small number of environmental temperature and humidity sensors for parameter calibration.

[0029] S2, based on the image characteristics of different regions, performs differentiated image preprocessing and spatiotemporal alignment processing in sequence to construct hierarchical regional data.

[0030] In this embodiment of the invention, the first type of core monitoring area (Category I area) adopts adaptive histogram equalization + non-local mean denoising + sub-pixel level edge enhancement to retain the UHPC microcrack (width <0.1mm) features; the second type of secondary key monitoring area (Category II area) adopts Gaussian filtering + histogram equalization to balance denoising and feature preservation; the third type of general monitoring area (Category III area) adopts median filtering + contrast stretching to quickly remove environmental noise and improve screening efficiency.

[0031] Spatiotemporal alignment processing includes spatial alignment and temporal alignment. In spatial alignment, the spatial position of images in each region is matched by the BeiDou positioning information acquired from the images and the mileage coordinates of the bridge structure, establishing a one-to-one mapping relationship between bridge structure mileage, region number, and image pixel coordinates. In temporal alignment, the physical field data of the same region and the same acquisition period are synchronized by linear interpolation using the image acquisition time t0 as the time reference, thus completing the temporal dimension alignment.

[0032] Finally, hierarchical regional data is constructed. Understandably, each region corresponds to an independent image-physical field matching dataset. It should be noted that the physical field refers to the collection of various physical quantities that dynamically change with spatial location and time during the service of the UHPC bridge structure, are quantifiable, and directly affect the initiation and propagation of cracks. For example, the categories of physical fields include mechanical physical fields, temperature physical fields, and environmental physical fields. Mechanical physical fields directly reflect the stress / deformation state of the structure, and mechanical physical field parameters can include tensile strain in the tension zone, mid-span deflection, etc. Temperature physical fields generate temperature stress, which induces / exacerbates crack propagation, and temperature physical field parameters can include temperature gradient between the top and bottom plates of the box girder, temperature change rate, etc. Environmental physical fields affect the durability propagation of cracks, and environmental physical field parameters can include environmental humidity, environmental pH value, etc.

[0033] S3. Construct a dedicated crack feature sub-library based on the graded regional data, determine the recognition accuracy requirements for different regions, design the U-Net crack recognition algorithm, and use the algorithm to identify cracks.

[0034] In this embodiment of the invention, based on the material properties of UHPC (Ultra-High-Toughness Polymer), such as high toughness, low porosity, and finer, denser cracks, and combined with the crack initiation mechanism in different regions, three independent crack feature sub-libraries are constructed. Simultaneously, a negative sample library for the corresponding region is matched. Specifically, the first-class crack feature sub-library covers all dimensions of flexural transverse cracks, shear-compression diagonal cracks, anchorage zone splitting cracks, and joint interface cracks, with a focus on annotating the characteristics of UHPC micro-cracks ranging from 0.02mm to 0.1mm, including crack width, length, orientation, branching degree, edge gradient, and texture features. The second-class crack feature sub-library covers the core features of web shrinkage cracks and pavement temperature cracks, taking into account both micro-cracks and macro-cracks (0.05mm to 1mm), with a focus on retaining crack width, length, orientation, and other characteristics. The core features include three types of crack feature sub-libraries: covering surface shrinkage cracks and non-stressed cracks, with a focus on macroscopic cracks (width > 0.1 mm) to reduce the false detection rate of non-stressed micro-cracks. In addition, a regional negative sample library is constructed simultaneously to match the environmental interference features of the corresponding regions: negative samples in the bridge deck area are marked with tire marks and oil stains, negative samples in the beam web area are marked with template joint marks and drips, and negative samples in the beam top surface area are marked with water marks and dust, thereby reducing the false detection rate from the root.

[0035] To address the varying recognition accuracy requirements across different regions, a multi-scale feature fusion module is incorporated into the encoder of the U-Net network to capture subtle crack features in UHPCs. The core of this module is a region-specific adaptive weighted loss function. In this embodiment, the U-Net network employs a symmetrical encoder-decoder architecture. The encoder consists of sequentially connected convolutional layers, batch normalization layers, activation function layers, and max-pooling downsampling layers. A multi-scale feature fusion module is introduced at the encoder's end to enhance the extraction of fine-grained features from UHPC micro-cracks. The decoder gradually upsamples through transposed convolutional layers to restore the feature map size and then uses skip connections to fused the high-resolution shallow features of the encoder's corresponding layers with the deep semantic features of the decoder. A 1×1 convolutional layer at the network's end maps and outputs a binary classification feature map of the cracks. Furthermore, an adaptive weighted loss function associated with regional risk values ​​is introduced during network training to achieve differentiated control of crack recognition accuracy across different monitoring regions. For the k-th monitoring area, the total loss function L k The calculation formula is: ; in, The weighting coefficient for the k-th region is related to the final dynamic risk value of that region. Matching, L ce L is the cross-entropy loss function, used for pixel-level classification accuracy control. dice The Dice loss function is used to control the accuracy of overall crack contour recognition. The expression for the cross-entropy loss function is: ; N is the total number of pixels in the image, y h For the true label of pixel h (crack is 1, background is 0), p h To predict the probability that pixel h is a crack; The expression for the Dice loss function is: .

[0036] S4. Based on the identification results, the cracks in each region are clustered. Based on the crack clustering results, the main controlling factors of crack propagation in different regions are determined, and an independent LSTM crack propagation model for the corresponding region is constructed.

[0037] Specifically, based on the identification results, a differentiated parameter extraction strategy is adopted to complete the regional crack quantification. For example, for Class I areas, the maximum width, average width, length, direction, area, and number of branches of the crack are extracted; for Class II areas, the core parameters of the maximum width, length, and direction of the crack are extracted; for Class III areas, only the core parameters of macroscopic cracks with a width > 0.1 mm are extracted. Then, crack clustering is performed on the identification results of each area, and cracks that are spatially adjacent and have the same direction in the same area are clustered into the same crack group to obtain cracks of the same type.

[0038] Furthermore, to construct an LSTM crack propagation model, specifically for the k-th region, the correlation between crack propagation rate and various physical field parameters is calculated using grey relational analysis. Physical field parameters with a correlation greater than a preset value are selected as the main controlling factors for crack propagation in that region, and a set of main controlling physical field parameters is constructed. The formula for calculating the correlation is: ; For correlation, T is the number of monitoring data samples, and v u Let be the crack propagation rate in the u-th monitoring cycle. For the u-th monitoring period The normalized values ​​of the physical field parameters, where c is the resolution coefficient. The minimum global difference Finding the global maximum difference is understandable, and filtering is necessary. Physical field parameters with a value ≥0.6 are used as the main controlling factors for crack propagation in this region, and a set of main controlling physical field parameters for the corresponding region is constructed. The crack parameters and the master physical field parameters in the corresponding region are separated into trend terms, periodic terms and random terms using the STL decomposition method; Using the trend and periodic terms of the physical field parameters after decomposition of the corresponding region as input, and the trend and periodic terms of the crack parameters as output, a regional LSTM crack propagation model is trained to obtain basic prediction values. In this embodiment of the invention, the LSTM crack propagation model uses the time series of the trend and periodic terms of the regional master physical field parameters after STL decomposition as the model input features. First, the input time series data is scaled through a normalization layer, and then passed into a multi-layer stacked LSTM hidden layer. Each LSTM unit captures the nonlinear time series dependency between physical field changes and crack propagation through the collaborative operation of forget gate, input gate, cell state update and output gate. After the output of the hidden layer, the Dropout layer is connected to suppress model overfitting. Then, the high-dimensional time series features are reduced and fitted through a fully connected mapping layer. Finally, the basic prediction values ​​of the trend and periodic terms of the expansion parameters such as crack width and length of the corresponding region are output through a linear output layer, forming a dedicated time series prediction model adapted to the damage evolution law of a single monitoring area. Based on the latest measured data and the basic predicted values ​​for each monitoring period, the parameters of the LSTM crack propagation model are corrected in real time using the Bayes formula. Dynamic optimization of the LSTM crack propagation model for the corresponding region is completed in each monitoring period to finalize the construction of the LSTM crack propagation model. The Bayes formula is: ; θ represents the parameters to be corrected (crack propagation rate, physical field sensitivity coefficient), and D represents the latest monitoring data for this region. For the prior distribution of parameters, It is the likelihood function, constructed based on the error between the measured and predicted values.

[0039] S5 constructs finite element sub-models with differentiated accuracy for different graded regions, combines measured crack parameters with physical field data to complete damage parameter inversion, calculates the structural health index of different regions, and conducts single-region early warning based on the first early warning mechanism.

[0040] Specifically, finite element sub-models with varying precision are constructed for different graded regions. For example, region I uses refined solid element modeling with a mesh size ≤ 5 mm to accurately simulate crack geometry; region II uses shell element modeling with a mesh size ≤ 20 mm; and region III uses simplified beam element modeling with a mesh size ≤ 50 mm. Crack parameters and measured physical field data for the corresponding regions are substituted into the finite element sub-models. With the goal of minimizing the residual between simulated and measured values, the material damage reduction coefficient D for that region is obtained through inversion. k The objective function is expressed as: ; These are the physical field parameters obtained from finite element simulation. These are actual measured values ​​on-site. Based on the material damage reduction factor and the final dynamic risk value of the region, the structural health index of the corresponding region is calculated using the following formula: ; SHI k Let k be the structural health index of the k-th region. The closer the value is to 1, the better the health status.

[0041] In the embodiments of the present invention, a type of region SHI k <0.75, Class II area SHI k <0.6, Class III area SHI k When the value is less than 0.5, a tiered early warning is triggered for the corresponding area.

[0042] S6. Based on the sub-regional structural health index and regional risk weight, calculate the comprehensive structural health index of the entire bridge through weighted fusion, and issue a full-bridge early warning according to the second early warning mechanism.

[0043] The formula for calculating the overall structural health index of the bridge is as follows: ; ; The overall structural health index of the bridge is represented by M, where M is the total number of regions and S is the total number of regions. k The percentage of the area of ​​the k-th region to the total monitored area of ​​the entire bridge. This represents the overall weight of the k-th region.

[0044] In the embodiments of the present invention, SHI total When the value is less than 0.8, a comprehensive bridge-wide early warning is triggered, prompting a full-scale inspection, repair, and reinforcement.

[0045] In other embodiments of the present invention, based on the independent LSTM crack propagation model for each region, a cross-regional linkage module is added to quantify the transmission effect of crack propagation in adjacent regions, thereby achieving accurate prediction of crack cross-regional spread risk. Specifically, this includes: Based on the elements of the standardized spatial weight matrix And correlation, calculate the transmission coefficient of crack propagation in adjacent areas, the calculation formula is: ; The conduction coefficient is the crack transmission intensity from region k to region l. , These represent the maximum dominant physical field correlation degree (characterizing the regional crack propagation activity) for cracks in regions k and l, respectively. Based on the output results and transmission coefficients of the independent LSTM crack propagation models for each region, a cross-regional coordinated evolution model is constructed to correct the crack propagation prediction results for each region. The cross-regional coordinated evolution model is expressed as follows: ; To correct the crack propagation rate of region k at a future time Δt, , The original predicted crack propagation rates of the LSTM crack propagation model for regions k and l are respectively, where M is the total number of regions. It can be understood that the crack propagation rate can be the propagation rate of crack length or width. A cross-regional diffusion risk threshold is set, and the transmission coefficient is compared with the cross-regional diffusion risk threshold. Simultaneously, the single-region early warning situation is correlated to determine the diffusion risk. For example, when... Furthermore, crack k in region is at least moderately damaged (SHI). k When the value is less than 0.75, it is judged as "high diffusion risk", triggering a cross-regional joint early warning to prevent the crack from spreading from the core area to the secondary critical area and the general area.

[0046] S7. Construct a full-bridge damage evolution prediction model based on the LSTM crack propagation model, and output long-term and short-term damage evolution prediction results. The long-term and short-term damage evolution prediction results include regional structural health indices and a comprehensive full-bridge structural health index at future time nodes. The regional structural health index in S5 and the comprehensive full-bridge structural health index in S6 are used to train the full-bridge damage evolution prediction model.

[0047] In this embodiment of the invention, based on the LSTM crack propagation model of each region of the full bridge and combined with a multi-task deep learning framework, a full bridge damage evolution prediction model is constructed. The model takes into account the predicted crack parameters and master physical field parameters of each region, and outputs the regional SHI (Screen Hierarchy Indicators) for the next 1 month, 3 months, 6 months, and 1 year. k With the whole bridge SHI total This enables short- and long-term predictions of damage evolution.

[0048] It should be noted that the full-bridge damage evolution prediction model adopts a multi-task deep learning architecture. It takes the physical field temporal features of each monitoring area decomposed by STL, crack propagation temporal data, regional final dynamic risk value and spatial correlation features as the overall input. First, it performs unified feature extraction and dimension normalization on the heterogeneous data of the multi-region of the full bridge through a shared feature encoding layer. Then, it connects to the independent LSTM crack propagation model branches of each region to learn the independent temporal laws of damage evolution in different regions simultaneously. Subsequently, it performs adaptive fusion of the output features of each LSTM crack propagation model branch based on regional risk weight and spatial transmission coefficient through a spatial weighted fusion layer to explore the damage transmission correlation between regions. Finally, it outputs the crack parameters, structural health index and the overall structural health index of the full bridge in parallel at different time nodes in the future through a multi-layer fully connected fitting layer and a regression output layer.

[0049] In summary, the UHPC bridge surface crack early warning method in the above embodiments of the present invention divides the UHPC bridge into hierarchical regions and collects multi-source data for each region. Differentiated image and multiphysics data acquisition strategies are implemented for different hierarchical regions to achieve spatiotemporal alignment of regional data. A regional UHPC-specific crack feature library and an improved U-Net recognition algorithm are constructed to achieve accurate crack identification and parameter extraction. A regionally independent crack propagation-multiphysics LSTM crack propagation model is established, and combined with finite element inversion, a quantitative assessment of the structural health of each region is completed. Finally, a weighted fusion is used to achieve a comprehensive assessment of the entire bridge and damage evolution prediction. Because the health index used to assess the health status of a single region and the entire bridge is accurately calculated, crack early warning can be effectively performed, allowing for early maintenance of the bridge.

[0050] Example 2 Please see Figure 2 , Figure 2 This is a structural block diagram of a UHPC bridge surface crack early warning system 200 provided in Embodiment 2 of the present invention. This UHPC bridge surface crack early warning system 200 is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0051] Specifically, the UHPC bridge surface crack early warning system 200 includes: a segmentation module 21, a first construction module 22, an identification module 23, a second construction module 24, a first calculation module 25, a second calculation module 26, and a prediction module 27, wherein: The segmentation module 21 is used to hierarchically segment the area of ​​the UHPC bridge and collect multi-source data for each area based on the segmented area. The multi-source data includes image data. The first construction module 22 is used to perform differentiated image preprocessing and spatiotemporal alignment processing in sequence according to the image characteristics of different regions in order to construct hierarchical regional data; The identification module 23 is used to construct a dedicated crack feature sub-library based on the hierarchical region data, determine the identification accuracy requirements for different regions, design the U-Net crack identification algorithm, and identify cracks using this algorithm. In the U-Net crack identification algorithm, for the k-th monitoring region, the total loss function L... k The calculation formula is: ; in, The weighting coefficient for the k-th region is related to the final dynamic risk value of that region. Matching, L ce L is the cross-entropy loss function, used for pixel-level classification accuracy control. dice The Dice loss function is used to control the accuracy of overall crack contour recognition. The second construction module 24 is used to cluster cracks in each region based on the identification results, determine the main controlling factors of crack propagation in different regions based on the crack clustering results, and construct an independent LSTM crack propagation model for the corresponding region. The first calculation module 25 is used to construct finite element sub-models with differentiated accuracy for different graded regions, combine measured crack parameters and physical field data to complete damage parameter inversion, calculate the structural health index of different regions, and conduct single-region early warning according to the first early warning mechanism. The second calculation module 26 is used to calculate the comprehensive health index of the entire bridge structure by weighted fusion based on the sub-regional structural health index and regional risk weight, and to conduct a full-bridge early warning according to the second early warning mechanism; The prediction module 27 is used to construct a full-bridge damage evolution prediction model based on the LSTM crack propagation model and output long-term and short-term damage evolution prediction results. The long-term and short-term damage evolution prediction results include regional structural health indices and a comprehensive full-bridge structural health index at future time nodes. The regional structural health index in the first calculation module and the comprehensive full-bridge structural health index in the second calculation module are used to train the full-bridge damage evolution prediction model.

[0052] Furthermore, in some optional embodiments of the present invention, the dividing module 21 includes: The construction unit is used to build a multi-dimensional crack risk assessment index system adapted to the characteristics of UHPC materials, and to determine the comprehensive weight of the indexes through subjective and objective coupling weighting. The multi-dimensional crack risk assessment index system includes 4 criterion layers and 13 quantitative indicators, specifically: C1 is a criterion layer for inherent structural properties and stress characteristics, including component safety level coefficient, UHPC stress level coefficient, construction detail complexity coefficient, and UHPC material degradation coefficient. The intrinsic risk criterion layer for C2 crack initiation and propagation includes fatigue load effect coefficient, historical damage deterioration coefficient, construction quality defect coefficient, and multi-field coupling sensitivity coefficient. C3 service environment and external excitation influence criteria layer, including temperature gradient amplitude coefficient, environmental corrosion level coefficient, and vehicle load impact coefficient; C4 spatial correlation impact characteristic criterion layer, including distance coefficient between adjacent high-risk areas and crack connectivity probability coefficient; All indicators are positive; the higher the indicator value, the higher the risk of cracks. The first calculation unit is used to quantify the regional basic risk based on the comprehensive weight of the indicators and in combination with the normal cloud model, and calculate the regional basic risk value. The correction unit is used to correct the transmission effect of crack propagation in adjacent areas based on the basic risk value of the region by introducing a spatial lag model, thereby obtaining the final dynamic risk value of the region. It also constructs a spatial weight matrix based on adjacency and distance attenuation, performs row standardization, and calculates the final dynamic risk value of the region. The calculation formula is: ; Where M is the total number of regions, and ρ is the spatial transmission coefficient. For the elements of the standardized spatial weight matrix, This represents the base risk value for the k-th region. This represents the baseline risk value for the l-th region. Clustering units are used to classify the monitoring area by adaptively clustering the final dynamic risk value of the area. The acquisition unit is used to acquire multi-source data for each region according to the hierarchical division of the region, and the multi-source data includes image data.

[0053] Furthermore, in some optional embodiments of the present invention, the building unit includes: The first calculation subunit is used to sort the indicators under the same criterion layer by importance, determine the importance ratio of adjacent indicators, recursively calculate and normalize the subjective weight of each indicator, and obtain the global subjective weight. The second calculation subunit is used to perform 0-1 standardization on the indicators, calculate the standard deviation and conflict of the indicators, obtain the information content of the indicators, and obtain the objective weight of the indicators after normalization. The third calculation subunit is used to couple the global subjective weight and the objective weight of the indicator using a multiplicative synthesis method to obtain the comprehensive weight of the indicator.

[0054] Furthermore, in some optional embodiments of the present invention, the first computing unit includes: The fourth calculation subunit is used to determine the normal cloud digital characteristics of each indicator and calculate the cloud membership degree of each indicator in each region through a positive normal cloud generator. The fifth calculation subunit is used to calculate the basic risk value of each region based on the comprehensive weight of the indicators and the cloud membership degree.

[0055] Furthermore, in some optional embodiments of the present invention, the second building module 24 includes: The filtering unit is used to calculate the correlation between crack propagation rate and each physical field parameter for the k-th region through grey relational analysis, and to filter physical field parameters with a correlation greater than a preset value as the main controlling factors of crack propagation in that region, thus constructing a set of main controlling physical field parameters. The decomposition unit is used to separate the trend term, periodic term, and random term from the crack parameters and the master physical field parameters of the corresponding region using the STL decomposition method. The training unit is used to train the regional LSTM crack propagation model with the trend and periodic terms of the physical field parameters after the corresponding region decomposition as input and the trend and periodic terms of the crack parameters as output, so as to obtain the basic prediction value. The calibration unit is used to perform real-time calibration of the LSTM crack propagation model parameters using the Bayes formula based on the latest measured data and the basic predicted values ​​for each monitoring cycle. The dynamic optimization of the LSTM crack propagation model for the corresponding region is completed in each monitoring cycle to complete the construction of the LSTM crack propagation model.

[0056] Furthermore, in some optional embodiments of the present invention, the first computing module 25 includes: The inversion unit is used to construct finite element sub-models with different accuracies for different graded regions. The crack parameters and measured physical field data of the corresponding region are substituted into the finite element sub-model. With the goal of minimizing the residual between the simulation value and the measured value, the material damage reduction coefficient of the region is obtained by inversion. The second calculation unit is used to calculate the structural health index of the corresponding region based on the material damage reduction coefficient and the final dynamic risk value of the region.

[0057] Example 3 In another aspect, the present invention also proposes an electronic device, please refer to [link to relevant documentation]. Figure 3 The electronic device shown is an embodiment of the present invention, including a memory 20, a processor 10, and a computer program 30 stored in the memory and executable on the processor. When the processor 10 executes the computer program 30, it implements the UHPC bridge surface crack early warning method as described above.

[0058] In some embodiments, the processor 10 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to run program code stored in memory 20 or process data, such as executing access restriction programs.

[0059] The memory 20 includes at least one type of readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 20 can be an internal storage unit of an electronic device, such as the hard disk of the electronic device. In other embodiments, the memory 20 can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. Furthermore, the memory 20 can include both internal and external storage units of the electronic device. The memory 20 can be used not only to store application software and various types of data of the electronic device, but also to temporarily store data that has been output or will be output.

[0060] It should be pointed out that, Figure 3 The structure shown does not constitute a limitation on the electronic device. In other embodiments, the electronic device may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0061] This invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the UHPC bridge surface crack early warning method described above.

[0062] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0063] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0064] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0065] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0066] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for early warning of surface cracks in UHPC bridges, characterized in that, The method includes the following steps: S1, the area of ​​the UHPC bridge is divided into hierarchical regions, and multi-source data of each region is collected according to the hierarchical regions. The multi-source data includes image data. S2, based on the image characteristics of different regions, perform differentiated image preprocessing and spatiotemporal alignment processing in sequence to construct hierarchical regional data; S3. Construct a dedicated crack feature sub-library based on the graded regional data, determine the recognition accuracy requirements for different regions, design the U-Net crack recognition algorithm, and use the algorithm to identify cracks. S4. Based on the identification results, cluster the cracks in each region. Based on the crack clustering results, determine the main controlling factors of crack propagation in different regions and construct an independent LSTM crack propagation model for each region. S5 constructs finite element sub-models with differentiated accuracy for different graded regions, combines measured crack parameters with physical field data to complete damage parameter inversion, calculates the structural health index of different regions, and conducts single-region early warning based on the first early warning mechanism; S6. Based on the sub-regional structural health index and regional risk weight, calculate the comprehensive structural health index of the entire bridge through weighted fusion, and conduct a full-bridge early warning according to the second early warning mechanism; S7. Construct a full-bridge damage evolution prediction model based on the LSTM crack propagation model, and output long-term and short-term damage evolution prediction results. The long-term and short-term damage evolution prediction results include regional structural health indices and a comprehensive full-bridge structural health index at future time nodes. The regional structural health index in S5 and the comprehensive full-bridge structural health index in S6 are used to train the full-bridge damage evolution prediction model.

2. The method for early warning of surface cracks in UHPC bridges according to claim 1, characterized in that, Step S1 includes: A multi-dimensional crack risk assessment index system adapted to the characteristics of UHPC materials was constructed, and the comprehensive weight of the index was determined by subjective and objective coupling weighting. Based on the comprehensive weight of the aforementioned indicators, and combined with the normal cloud model, the basic regional risk is quantified, and the basic regional risk value is calculated. Based on the basic risk value of the region, a spatial lag model is introduced to correct the transmission effect of crack propagation in adjacent regions, and the final dynamic risk value of the region is obtained. The final dynamic risk value of the region is used for adaptive clustering to complete the classification of the monitoring region; Based on the hierarchical division of regions, multi-source data is collected for each region, including image data.

3. The method for early warning of surface cracks in UHPC bridges according to claim 2, characterized in that, The multi-dimensional crack risk assessment index system includes 4 criterion layers and 13 quantitative indicators, specifically: C1 is a criterion layer for inherent structural properties and stress characteristics, including component safety level coefficient, UHPC stress level coefficient, construction detail complexity coefficient, and UHPC material degradation coefficient. The intrinsic risk criterion layer for C2 crack initiation and propagation includes fatigue load effect coefficient, historical damage deterioration coefficient, construction quality defect coefficient, and multi-field coupling sensitivity coefficient. C3 service environment and external excitation influence criteria layer, including temperature gradient amplitude coefficient, environmental corrosion level coefficient, and vehicle load impact coefficient; C4 spatial correlation impact characteristic criterion layer, including distance coefficient between adjacent high-risk areas and crack connectivity probability coefficient; All indicators are positive; the higher the indicator value, the higher the risk of cracks.

4. The method for early warning of surface cracks in UHPC bridges according to claim 3, characterized in that, The step of constructing a multi-dimensional crack risk assessment index system adapted to the properties of UHPC materials and determining the comprehensive weight of the indexes through subjective and objective coupling weighting includes the following steps: The indicators under the same criterion level are sorted by importance, the importance ratio of adjacent indicators is determined, the subjective weight of each indicator is calculated recursively and normalized to obtain the global subjective weight. The indicators are standardized by 0-1, the standard deviation and conflict of the indicators are calculated, the information content of the indicators is obtained, and the objective weight of the indicators is obtained after normalization. The global subjective weight and the objective weight of the indicator are coupled by a multiplicative synthesis method to obtain the comprehensive weight of the indicator.

5. The method for early warning of surface cracks on UHPC bridges according to claim 4, characterized in that, The step of quantifying the regional basic risk based on the comprehensive weight of the indicators and combining it with the normal cloud model to calculate the regional basic risk value includes: Determine the normal cloud digital characteristics of each indicator, and calculate the cloud membership degree of each indicator in each region using a positive normal cloud generator; Based on the comprehensive weight of the aforementioned indicators and cloud membership, the basic risk value for each region is calculated.

6. The method for early warning of surface cracks on UHPC bridges according to claim 5, characterized in that, In the step of obtaining the final dynamic risk value of the region by introducing a spatial lag model to correct the transmission effect of crack propagation in adjacent regions based on the basic risk value of the region, a spatial weight matrix based on adjacency relationship and distance attenuation is constructed and row standardization is performed to calculate the final dynamic risk value of the region. The calculation formula is: ； Where M is the total number of regions, and ρ is the spatial transmission coefficient. For the elements of the standardized spatial weight matrix, This represents the base risk value for the k-th region. This represents the base risk value for the l-th region.

7. The method for early warning of surface cracks on UHPC bridges according to claim 6, characterized in that, In the U-Net crack detection algorithm, for the k-th monitoring area, the total loss function L k The calculation formula is: ； in, The weighting coefficient for the k-th region is related to the final dynamic risk value of that region. Matching, L ce L is the cross-entropy loss function, used for pixel-level classification accuracy control. dice The Dice loss function is used to control the accuracy of crack overall contour recognition.

8. The method for early warning of surface cracks on UHPC bridges according to claim 7, characterized in that, In step S4, the steps for constructing the LSTM crack propagation model include: For the k-th region, the correlation between crack propagation rate and each physical field parameter is calculated by grey relational analysis. Physical field parameters with a correlation greater than a preset value are selected as the main controlling factors for crack propagation in this region, and a set of main controlling physical field parameters is constructed. The crack parameters and the master physical field parameters in the corresponding region are separated into trend terms, periodic terms and random terms using the STL decomposition method; Using the trend and periodic terms of the physical field parameters after the corresponding region decomposition as inputs, and the trend and periodic terms of the crack parameters as outputs, a regional LSTM crack propagation model is trained to obtain the basic prediction values. Based on the latest measured data and the basic predicted values ​​for each monitoring period, the parameters of the LSTM crack propagation model are corrected in real time using the Bayes formula. The dynamic optimization of the LSTM crack propagation model for the corresponding region is completed in each monitoring period to complete the construction of the LSTM crack propagation model.

9. The method for early warning of surface cracks on UHPC bridges according to claim 8, characterized in that, Step S5 includes: For different graded regions, finite element sub-models with different precision are constructed. The crack parameters and measured physical field data of the corresponding regions are substituted into the finite element sub-models. With the goal of minimizing the residual between the simulation value and the measured value, the material damage reduction coefficient of the region is obtained by inversion. Based on the material damage reduction factor and the final dynamic risk value of the region, the structural health index of the corresponding region is calculated.

10. A UHPC bridge surface crack early warning system, characterized in that, For implementing the UHPC bridge surface crack early warning method as described in any one of claims 1-9, the system comprises: The segmentation module is used to hierarchically segment the area of ​​the UHPC bridge, and collect multi-source data for each area based on the segmented area. The multi-source data includes image data. The first construction module is used to perform differentiated image preprocessing and spatiotemporal alignment processing in sequence according to the image characteristics of different regions in order to construct hierarchical regional data; The identification module is used to construct a dedicated crack feature sub-library based on the graded regional data, determine the identification accuracy requirements of different regions, design the U-Net crack identification algorithm, and identify cracks through the algorithm. The second construction module is used to cluster cracks in each region based on the identification results, determine the main controlling factors of crack propagation in different regions based on the crack clustering results, and construct an independent LSTM crack propagation model for the corresponding region. The first calculation module is used to construct finite element sub-models with different precision for different graded regions, combine measured crack parameters and physical field data to complete damage parameter inversion, calculate the structural health index of different regions, and conduct single-region early warning according to the first early warning mechanism. The second calculation module is used to calculate the comprehensive structural health index of the entire bridge through weighted fusion based on the sub-regional structural health index and regional risk weight, and to conduct a full-bridge early warning according to the second early warning mechanism. The prediction module is used to construct a full-bridge damage evolution prediction model based on the LSTM crack propagation model and output long-term and short-term damage evolution prediction results. The long-term and short-term damage evolution prediction results include regional structural health indices and a comprehensive full-bridge structural health index at future time nodes. The regional structural health index in the first calculation module and the comprehensive full-bridge structural health index in the second calculation module are used to train the full-bridge damage evolution prediction model.

Images