A bridge targeted reinforcement scheme generation method and system based on digital twinning

Abstract

The present application relates to the technical field of bridge engineering structure health monitoring, in particular to a bridge targeted reinforcement scheme generation method and system based on digital twinning. The method comprises the following steps: identifying the inherent weak areas of the bridge structure to obtain the structural weak point features; extracting the dynamic features of the structural vibration signal through adaptive time-frequency analysis to obtain dynamic abnormal features; reconstructing the real-time stress field, combining the real-time stress field, the structural weak point features and the dynamic abnormal features to comprehensively evaluate the bridge health state; analyzing the cumulative influence of the environment history on the structure according to the environmental damage influence and the dynamic abnormal features; and generating a bridge targeted reinforcement scheme through multi-objective optimization and reinforcement learning based on the geometric deformation features, the bridge health state and the cumulative influence. The present application generates a targeted reinforcement scheme by fusing multiple source technologies, and is suitable for the whole life cycle health management and precise reinforcement decision of highway, railway and municipal bridges.

Description

Technical Field

[0001] This invention relates to the field of bridge engineering structural health monitoring technology, specifically to a method and system for generating targeted reinforcement schemes for bridges based on digital twins. Background Technology

[0002] As a core component of transportation infrastructure, the structural safety and service performance of bridges directly affect the traffic efficiency and public safety of transportation networks. With the increase of service life, bridges are susceptible to damage such as cracks, deformation, and stress concentration caused by environmental erosion (temperature, humidity, corrosion), load effects (traffic load, wind load) and material deterioration. Therefore, reinforcement and maintenance are necessary to extend the service life of bridges.

[0003] Existing bridge reinforcement scheme generation technologies have the following key shortcomings:

[0004] Low accuracy in identifying structural vulnerabilities: Traditional methods rely on human experience or single detection data (such as crack detection), which cannot integrate multi-source knowledge such as design specifications, historical damage, and similar cases, and are difficult to handle uncertain information such as "crack depth is about 5mm", resulting in large deviations in vulnerability location;

[0005] Poor adaptability to dynamic anomaly detection: Using fixed wavelet basis or single modal parameters (such as natural frequency) to analyze vibration signals cannot adapt to non-stationary signals generated by the superposition of traffic flow and wind loads on bridges, and it is difficult to capture weak dynamic anomalies (such as high-frequency vibration changes caused by bearing aging).

[0006] Distortion in the quantification of environmental damage impact: Pearson correlation analysis links environmental factors with damage, but cannot distinguish between causal relationships and confounding interferences (such as "simultaneous increase of high temperature and traffic load" may easily misjudge high temperature as the main damage factor), resulting in deviations in the calculation of the contribution of environmental impact.

[0007] Insufficient accuracy in geometric deformation quantification: Relying on point measurement methods such as total stations, it is impossible to cover complex bridge surfaces (such as curved beams), and point cloud processing only focuses on three-dimensional reconstruction, lacking multi-dimensional topographic analysis of normal vector-curvature-displacement field, and subtle deformations (such as 10mm sag at mid-span) are easily missed.

[0008] Stress field monitoring has significant limitations: point stress measurement using strain gauges cannot obtain the stress distribution over the entire area, and finite element models are mostly updated offline and statically, making it impossible to dynamically reconstruct the stress field by combining real-time displacement data, and making it difficult to capture stress concentration in key areas such as supports.

[0009] Poor handling of conflicts in comprehensive health assessment: When integrating multi-source monitoring information (vulnerability, dynamic anomaly, stress), the uncertainty and conflict of data (such as "vulnerability points indicate poor health, stress data indicate good health") are not taken into account, which can easily lead to misjudgment of health status;

[0010] The reinforcement schemes lack targeting: they fail to integrate the propagation path of historical environmental impacts with the structural health status, adopt a one-size-fits-all approach to reinforcement (such as full-bridge corrosion protection), and neglect core sensitive components (such as environmentally sensitive supports), resulting in high reinforcement costs and poor effectiveness.

[0011] In summary, existing technologies cannot achieve integrated decision-making that combines accurate fusion of multi-source information, deep perception of structural status, and targeted generation of reinforcement schemes. There is an urgent need for a full-process intelligent method based on digital twins to solve the core problems of inaccurate positioning, imprecise quantification, and suboptimal decision-making in the generation of bridge reinforcement schemes. Summary of the Invention

[0012] To address the shortcomings of existing methods and the needs of practical applications, and in order to solve the aforementioned problems, this invention provides a method for generating targeted reinforcement schemes for bridges based on digital twins, comprising the following steps:

[0013] The process involves identifying inherently vulnerable areas in the bridge structure and obtaining structural vulnerability features; extracting dynamic features of structural vibration signals through adaptive time-frequency analysis to obtain dynamic anomaly features; reconstructing the real-time stress field; and comprehensively assessing the bridge's health status by combining the real-time stress field, the structural vulnerability features, and the dynamic anomaly features; analyzing the cumulative impact of environmental history on the structure based on environmental damage effects and the dynamic anomaly features; and generating targeted reinforcement schemes for the bridge through multi-objective optimization and reinforcement learning based on geometric deformation features, the bridge's health status, and the cumulative impact.

[0014] This invention uses knowledge graphs and fuzzy logic to accurately locate vulnerable points, breaking through the limitations of traditional experience-based judgment and laying the foundation for targeted reinforcement. Adaptive wavelet and modal analysis enable real-time early warning of vibration anomalies, capturing potential damage. Causal inference quantifies environmental damage, avoiding misjudgments based on correlations and supporting environmental fatigue-resistant design. Point cloud technology quantifies geometric deformation with high precision, compensating for blind spots in traditional measurements. DIC and finite element updates enable full-domain stress monitoring, locating areas exceeding limits. Evidence theory integrates multi-source information, avoiding the one-sidedness of single indicators. Graph neural networks aggregate the cumulative effects of historical environmental data, considering component correlations. Finally, multi-objective optimization and reinforcement learning generate operable solutions, forming a complete chain of identification-monitoring-evaluation-solution, solving problems such as low accuracy, information conflict, and incomplete coverage in traditional technologies, significantly improving the targeting and effectiveness of bridge reinforcement, and ensuring structural safety.

[0015] Optionally, identifying inherently vulnerable areas of the bridge structure and obtaining structural vulnerability point characteristics includes the following steps:

[0016] This invention integrates knowledge graphs and fuzzy logic. Knowledge graphs extract information such as structural components and attributes through entity recognition and construct a knowledge graph by combining it with relationship mining to build a structured knowledge system, thus solving the problem of fragmented multi-source heterogeneous data. Fuzzy logic quantifies uncertain damage information through membership functions and, combined with expert rule reasoning, addresses the problem of "imprecisely descriptive" data, overcoming the shortcomings of existing technologies that use either approach alone. Effectively avoiding the bias and blindness of identification and ensuring targeted reinforcement is a key prerequisite for precise bridge maintenance.

[0017] Optionally, the step of extracting dynamic features of structural vibration signals through adaptive time-frequency analysis to obtain dynamic anomaly features includes the following steps:

[0018] This invention optimizes the adaptive wavelet basis function; based on the optimized wavelet basis, signal decomposition is performed, and real-time modal parameters are extracted from the decomposed signal. Dynamic anomalies deviating from the normal state are identified based on changes in these modal parameters. Through adaptive wavelet basis optimization, this invention dynamically adapts to non-stationary vibration signals of bridges; combining modal parameters with multi-dimensional features of wavelet entropy, it accurately captures weak dynamic anomalies, avoiding missed detections. This enables real-time monitoring and anomaly warning of the structural dynamic state, promptly detecting vibration characteristic changes caused by potential damage. Simultaneously, the output dynamic anomaly features provide crucial data for subsequent comprehensive health status assessment, and the vibration transmission relationship provides weighting criteria for constructing a structural diagram model aggregating environmental historical influences. This is a vital link connecting real-time monitoring and subsequent analysis, overcoming the limitations of traditional time-frequency analysis and isolated applications of structural dynamics.

[0019] Optionally, the reconstructing of the real-time stress field includes the following steps:

[0020] This invention acquires surface displacement through image analysis and constructs a high-precision surface displacement field. It then optimizes finite element model parameters and reconstructs the stress field by combining the high-precision surface displacement field with the optimized finite element model. This allows for real-time reconstruction of the internal stress field of the structure, effectively overcoming the limitations of traditional point sensors (such as strain gauges) and achieving full-domain stress monitoring. It accurately captures stress concentration or over-limit areas that are difficult to cover using traditional methods. By dynamically coupling real-time data with the model, it eliminates offline static model errors, ensuring the timeliness and accuracy of stress field analysis. This provides real-time, full-domain stress data support for bridge safety assessment, timely warnings of potential structural risks, and precise location data for subsequent targeted reinforcement, solving the industry challenge of non-contact, full-domain real-time stress monitoring for large bridges.

[0021] Optionally, the comprehensive assessment of the bridge's health status by combining the real-time stress field, the structural vulnerability features, and the dynamic anomaly features includes the following steps:

[0022] This invention normalizes structural vulnerability features, dynamic anomaly features, and real-time stress field features to establish a Bridge Health Assessment (BPA) spatial mapping rule. Based on multi-source information fusion and evidence theory, the health status of bridges is comprehensively assessed according to the BPA spatial mapping rule. This invention unifies the scale of three heterogeneous features—structural vulnerability (inherent risk), dynamic anomaly (dynamic hidden danger), and real-time stress field (current stress)—through normalization and BPA spatial mapping, solving the problem of difficulty in integrating multi-dimensional data. Furthermore, evidence theory is used to handle information conflicts (such as the contradiction of high vulnerability but temporarily normal stress), quantifying uncertainty and avoiding the one-sidedness of single-feature assessment. This achieves a leap from "local feature judgment" to "comprehensive assessment across the entire domain," accurately outputting the bridge's health level and the root causes of core problems, clarifying priorities for subsequent targeted reinforcement, avoiding misjudgments or omissions caused by relying on a single indicator, and ensuring the comprehensiveness and reliability of health assessment.

[0023] Optionally, the step of analyzing the cumulative impact of environmental history on the structure based on the environmental damage effects and the dynamic anomaly characteristics includes the following steps:

[0024] A structural graph model is constructed, and its structure is enhanced and parameters are initialized using the dynamic anomaly features. An asymmetric transition matrix is ​​built based on the causal network diagram of environmental damage effects, thereby improving the damage propagation model. This damage propagation model is then used to analyze the cumulative impact of environmental history on the structure. This invention enhances the structural graph model and initializes parameters using dynamic anomaly features, and constructs an asymmetric transition matrix based on the causal relationship of environmental damage to improve the damage propagation model. This allows for accurate capture of the cumulative impact of environmental history on the structure, quantification of the asymmetric risk transmission effect between components, and provides component-level cumulative risk data for subsequent targeted reinforcement, avoiding incomplete reinforcement due to neglecting cumulative effects.

[0025] Optionally, obtaining the environmental damage impact includes the following steps:

[0026] Based on environmental factors and structural damage indicators, a time series is established; a propensity score matching method is introduced to establish a causal relationship network diagram; and based on causal inference and time series analysis, the impact of the environmental damage is obtained. This invention, by distinguishing the causal relationship between environmental factors and structural damage and eliminating confounding variables such as traffic loads, accurately quantifies the contribution of the environment to damage, precisely locates key environmental impacts and their extent, and provides a basis for targeted reinforcement such as resistance to temperature fatigue.

[0027] Optionally, the step of generating a targeted bridge reinforcement scheme based on geometric deformation features, the bridge health status, and the cumulative impact through multi-objective optimization and reinforcement learning includes the following steps:

[0028] This invention constructs optimization objectives and constraints; employs a hierarchical state-space modeling method to build a dynamic reward mechanism; and generates targeted reinforcement schemes for bridges through multi-objective optimization. By dynamically adapting complex constraints (such as the coupling of environmental influences and deformation) through reinforcement learning, and combining multi-objective optimization to balance multi-dimensional needs, this invention generates precise reinforcement schemes that consider effectiveness, cost, and feasibility, achieving targeted rather than comprehensive reinforcement.

[0029] Optionally, extracting the geometric deformation features includes the following steps:

[0030] Enhancement processing is performed on bridge point cloud data scanned by lidar; morphological features are extracted based on the processed point cloud data; and geometric deformation features are extracted based on the morphological features. This invention achieves high-precision three-dimensional quantization of structural geometry through multi-dimensional analysis of normal vector, curvature, and displacement fields, combined with a region fusion algorithm to automatically identify deformation hotspots. It can capture local deformations that are difficult to cover by traditional measurement methods (such as total stations).

[0031] Secondly, to efficiently execute the bridge targeted reinforcement scheme generation method based on digital twins provided by this invention, this invention also provides a bridge targeted reinforcement scheme generation system based on digital twins, including a processor, an input device, an output device, and a memory. The processor, input device, output device, and memory are interconnected. The memory stores a computer program containing program instructions. The processor is configured to call the program instructions to execute the bridge targeted reinforcement scheme generation method based on digital twins as described in the first aspect of this invention. The bridge targeted reinforcement scheme generation system based on digital twins of this invention has a compact structure and stable performance, and can stably execute the bridge targeted reinforcement scheme generation method based on digital twins provided by this invention, further improving the overall applicability and practical application capability of this invention. Attached Figure Description

[0032] Figure 1 A flowchart of a bridge targeted reinforcement scheme generation method based on digital twin provided in an embodiment of the present invention;

[0033] Figure 2 This is a framework diagram of a bridge targeted reinforcement scheme generation system based on digital twins, provided for an embodiment of the present invention. Detailed Implementation

[0034] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0035] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0036] Please see Figure 1 To address the aforementioned problems, this invention provides a method for generating targeted reinforcement schemes for bridges based on digital twins, such as... Figure 1 As shown, in one embodiment, the method includes the following steps:

[0037] S1. Identify inherently vulnerable areas in the bridge structure and obtain structural vulnerability features.

[0038] In this embodiment, identifying inherently vulnerable areas of the bridge structure and obtaining structural vulnerability features includes the following steps:

[0039] S11. Perform entity recognition on multi-source heterogeneous data and construct a knowledge graph.

[0040] Using the "component-attribute-damage-specification" quadruple as the core knowledge unit, this method employs a BERT pre-trained model combined with a Conditional Random Field (CRF) algorithm to perform entity recognition on multi-source heterogeneous data such as bridge inspection reports and design drawings. This accurately extracts structural components such as "beams," "supports," and "piers," as well as attribute indicators like "concrete strength" and "steel corrosion rate." In this embodiment, a Graph Neural Network (GNN) model is used to mine relationships between entities during the relation extraction stage, constructing a knowledge graph containing over 100,000 semantic triples. This graph is then stored in a high-performance graph database, and efficient knowledge retrieval and reasoning support are achieved through the Cypher query language.

[0041] S12. Design fuzzy logic rules, identify inherently vulnerable areas of the bridge structure based on fuzzy inference algorithms, and obtain structural vulnerability features.

[0042] To address the uncertainties in bridge structural safety assessments, core assessment indicators such as "damage frequency," "material degradation degree," and "number of load exceedances" are fuzzified. A fuzzy set system is constructed using triangular membership functions to classify each assessment indicator into three semantic levels: "low," "medium," and "high." Taking the damage frequency indicator as an example, 0.1 and 0.3 are used as grading and quantification thresholds to achieve precise semantic transformation of the assessment indicators.

[0043] Furthermore, based on the structural mechanics relationships and historical accident case data contained in the knowledge graph, a fuzzy reasoning system containing multiple expert experience rules is constructed. The rules in this system deeply mine the causal relationships between structural components, such as "if the aging level of the support is judged to be high and the damage frequency level of the beam is medium, then the vulnerability level of the beam-support connection node is rated as high". Through such logical rules, a systematic transformation from qualitative analysis to quantitative assessment is achieved.

[0044] Furthermore, an evaluation engine was built based on the Mamdani fuzzy inference algorithm to obtain structural vulnerability features. First, through a fuzzification step, the physical parameters of bridge structural components, such as measured data of stress, strain, and displacement, were transformed into fuzzy linguistic variables such as "low," "medium," and "high" according to a pre-defined membership function, completing the mapping from precise values ​​to fuzzy sets. For example, for strain data of the bridge beam, when the strain value falls within a certain range, it is assigned the fuzzy attribute of "high strain," and the corresponding membership value is determined.

[0045] The next step is rule matching. Based on the experience and historical data of bridge engineering experts, a comprehensive fuzzy rule base was constructed. These rules are presented in the form of "if...then...", such as "if the beam strain is high and the vibration frequency is abnormal, then the structure is highly vulnerable." By matching the fuzzified input variables with the conditions in the rule base, the corresponding rules are activated.

[0046] Next, a synthesis calculation is performed using the max-min synthesis method to synthesize the conclusions of the activation rules, resulting in a fuzzy output set. This set reflects the probability distribution of different levels of vulnerability.

[0047] Finally, through defuzzification, the fuzzy output set is transformed into precise numerical values, thus quantitatively assessing the vulnerability of each structural component of the bridge and generating a risk score from 0 to 100. A higher score indicates a greater vulnerability of the component and a higher probability of failure.

[0048] Furthermore, the weak points of the bridge structure were precisely located, and a Top-N ranking algorithm was used to select the top 20% of critical components as structural vulnerabilities. Specifically, all components were first sorted in descending order of risk scores, and then the top 20% were selected. For these critical components, not only were their spatial locations detailed, such as the precise geometric location information of "mid-span section of beam 3" and "support node of pier 4," but their structural types were also clearly marked, including beam structures, arch structures, and rigid frame structures. Their core vulnerability characteristics were also analyzed in depth. Specific failure modes included "insufficient fatigue strength," which may be due to the bridge being subjected to alternating loads for a long time, leading to the formation and gradual propagation of microcracks within the material; "attenuation of connection stiffness," commonly seen in the connection parts between components due to long-term wear and corrosion, resulting in decreased connection performance; and "decrease in material elastic modulus," mostly caused by material aging and environmental erosion, leading to a reduction in the mechanical properties of the material. This comprehensive and accurate information provides a solid and reliable basis for subsequent maintenance and reinforcement decisions, helping to formulate targeted, efficient, and reasonable maintenance plans to ensure the safety and stability of the bridge structure.

[0049] S2. Dynamic features of structural vibration signals are extracted through adaptive time-frequency analysis to obtain dynamic anomaly features.

[0050] In this embodiment, the step of extracting dynamic features of structural vibration signals through adaptive time-frequency analysis to obtain dynamic anomaly features includes the following steps:

[0051] S21. Optimize the adaptive wavelet basis function.

[0052] In complex industrial environments, vibration signals are often affected by noise interference. To accurately extract effective information, maximizing the signal-to-noise ratio (SNR) of the vibration signal is the core optimization objective. A particle swarm optimization (PSO) algorithm is employed, using cooperation and competition among particles to find the optimal solution. During the optimization process, the initial range for the scale parameter of the wavelet basis is set to 1~10, and the initial range for the displacement parameter is set to 0~1. After multiple rounds of iterative adjustments, an adaptive wavelet basis that fits the signal characteristics is generated. For example, for low-frequency vibration signals, the db4 wavelet has good time-frequency localization characteristics and can effectively capture signal details; while for high-frequency signals, the sym8 wavelet, with its high vanishing moment, can accurately analyze high-frequency components.

[0053] S22. Perform signal decomposition based on the optimized wavelet basis, extract real-time modal parameters from the decomposed signal, and identify dynamic anomalies that deviate from the normal state based on the changes in the modal parameters.

[0054] The vibration signal is decomposed into five levels using an optimized wavelet basis. This process subdivides the signal progressively according to frequency levels, thereby obtaining the detail coefficients and approximation coefficients corresponding to each frequency band. By calculating the energy proportion of each frequency band, the energy distribution of the signal in different frequency bands can be intuitively reflected. For example, when the energy proportion of the 1~5Hz frequency band is ≥60%, it can be preliminarily determined that the equipment is operating normally. The wavelet entropy concept is introduced to quantify the uncertainty of the signal. Under normal operating conditions, the wavelet entropy value is in the range of 0.3~0.5. Exceeding this range may indicate that the equipment is malfunctioning.

[0055] Furthermore, based on the random subspace method (SSI), real-time modal parameters, including natural frequencies f and mode shape vectors φ, are efficiently extracted from the decomposed signal. These real-time parameters are then... With equipment design value By making a comparison, the deviation rate is calculated. When the deviation rate is ≥5%, the equipment is considered to be abnormal. Combining wavelet features for comprehensive judgment, if the wavelet entropy is >0.6, it indicates that the complexity and uncertainty of the signal have increased significantly, and thus outputs a detailed description of the dynamic anomaly features, such as "the natural frequency of Pier No. 2 has decreased by 8%, accompanied by a sudden increase in high-frequency vibration energy", providing an accurate basis for equipment fault diagnosis.

[0056] S3. Reconstruct the real-time stress field, and comprehensively assess the health status of the bridge by combining the real-time stress field, the structural vulnerability characteristics, and the dynamic anomaly characteristics.

[0057] The reconstructed real-time stress field includes the following steps:

[0058] S311. Obtain surface displacement through image analysis and construct a high-precision surface displacement field.

[0059] The Lucas-Kanade algorithm is improved by using a multi-scale optical flow estimation strategy and an adaptive window mechanism. The multi-scale optical flow estimation optimizes the displacement vector layer by layer from coarse to fine by constructing an image pyramid to ensure accurate tracking of feature points in large displacement scenes. The adaptive window mechanism dynamically adjusts the size of the calculation window according to the local texture complexity of the image. Small windows are used in texture-rich areas to improve calculation accuracy, while windows are enlarged in texture-scarce areas to enhance the robustness of the algorithm.

[0060] Based on the optimized algorithm, a high-precision surface displacement field is obtained. and , respectively characterize the displacement changes of the object's surface in the x and y directions with time t.

[0061] Furthermore, based on geometric relationships and fundamental principles of kinematics, the strain field on the object's surface is quantitatively calculated using strain field calculation formulas, satisfying the following:

[0062]

[0063] The deformation rate is represented by this calculation result, which not only visually presents the deformation distribution on the object's surface, but also provides reliable data support for subsequent fatigue life prediction, stress concentration analysis, etc., significantly improving the accuracy of experimental analysis and its engineering application value.

[0064] S312. Optimize the parameters of the finite element model, and reconstruct the stress field by combining the high-precision surface displacement field and the optimized finite element model.

[0065] With the optimization objective of minimizing the error between the DIC measured strain field and the strain field calculated by the finite element model, the response surface methodology is used to iteratively update the parameters of the finite element model. The optimization focuses on key parameters such as the material's elastic modulus E and Poisson's ratio μ, employing the following iterative update formula:

[0066]

[0067] in, This represents the parameters of the finite element model after iteration. This represents the parameters of the finite element model before iteration. This indicates the measured strain in DIC. This indicates that the finite element model is used to calculate strain. The iteration coefficient is initially set to 0.1. Through an adaptive adjustment mechanism, its value is dynamically adjusted according to the error change rate of each iteration to ensure that the model parameter optimization process can converge quickly and avoid oscillations caused by excessive step size, thereby achieving accurate simulation of actual physical behavior by the finite element model.

[0068] Furthermore, based on the optimized high-precision finite element model, tetrahedral elements were used for structural discretization, with element sizes controlled at a fine scale of 0.5-2 mm to ensure the model effectively captures the details of complex structures. According to the fourth strength theory, nonlinear static analysis was conducted using the Mises stress calculation formula in conjunction with the ANSYS Mechanical APDL platform to calculate the Mises stress field distribution within the structure.

[0069] Furthermore, a digital image correlation (DIC) system was introduced for data verification. The displacement data of the structural surface was collected in real time at a sampling frequency of 1000fps. The stress field calculated by the finite element method was calibrated in real time through a dynamic correction model based on the particle swarm optimization (PSO) algorithm. During the calibration process, an adaptive weight allocation strategy was adopted to effectively eliminate cumulative errors caused by error sources such as boundary condition deviations, material parameter uncertainties, and data acquisition noise due to model simplification.

[0070] After multiple rounds of iterative correction, a high-fidelity stress field cloud map with a resolution of 0.1mm × 0.1mm is obtained, and a feature information report is generated simultaneously. For example, in the analysis of a bridge bearing area, stress concentration was identified, with the current peak stress reaching 180MPa, exceeding the design value by 20%, and the stress gradient changing rate exceeding 50MPa / mm within a 10mm range. In another embodiment, this result not only provides a quantitative basis for the safety assessment of engineering structures, but also, through time-series comparative analysis with historical data, constructs a fault early warning model based on an LSTM neural network, enabling early prediction of structural failure.

[0071] In this embodiment, the comprehensive assessment of the bridge's health status by combining the real-time stress field, the structural vulnerability features, and the dynamic anomaly features includes the following steps:

[0072] S321. Normalize the structural weak point characteristics, dynamic anomaly characteristics, and real-time stress field characteristics, and establish BPA space mapping rules.

[0073] For heterogeneous data such as vulnerability score (original value range 0~100), dynamic anomaly deviation rate (0~20%), and stress over-limit rate (0~50%), the min-max normalization method is used to map them to the [0,1] interval to ensure data scale consistency.

[0074] Furthermore, a strict BPA space mapping rule is established. Taking the vulnerability index as an example, when “vulnerability ≥ 80”, the basic probability allocation (BPA) for “poor health level” is assigned to 0.8, and the BPA for “moderate” status is 0.2. If “60 ≤ vulnerability < 80”, then the BPA for “moderate” is set to 0.7, and for “good” it is 0.3. The data is accurately converted into the evidence theory space through quantitative rules.

[0075] S322. Based on multi-source information fusion and evidence theory, comprehensively evaluate the health status of the bridge according to the BPA spatial mapping rule.

[0076] The Yager combination rule is introduced as the core processing strategy. When the conflict coefficient between multiple sources of evidence is greater than 0.5, the conflict handling mechanism is triggered, and then the conflict is redistributed.

[0077] Taking bridge monitoring as an example, assuming the "Poor vulnerability indication" BPA is 0.8 and the "Poor stress over-limit indication" BPA is 0.7, and the conflict coefficient K>0.5, according to the Yager rule, the conflicting part (i.e. the difference between 1-0.8-0.7+0.8×0.7) is assigned to the "unknown" state. After fusion, the BPA of the "poor" state is updated to 0.9, effectively avoiding erroneous decisions caused by high conflicting evidence.

[0078] Furthermore, a five-level health status evaluation system (Excellent / Good / Average / Poor / Dangerous) is constructed, and detailed BPA threshold mapping standards are established. For example, BPA ≥ 0.8 corresponds to "Dangerous," and 0.6~0.8 corresponds to "Poor," ensuring that the evaluation results are intuitive and interpretable. In the embodiment, the bridge health status assessment results not only indicate the health level (e.g., "Poor health level of beam No. 3"), but also deeply relate to core issues (e.g., "excessive stress at weak points + abnormal dynamic characteristics"), while providing comparisons of abnormal values ​​for key indicators (e.g., stress exceedance rate reaches 35%, exceeding the safety threshold by 15%), providing a comprehensive basis for maintenance decisions.

[0079] S4. Based on the environmental damage impact and the aforementioned dynamic anomaly characteristics, analyze the cumulative impact of environmental history on the structure.

[0080] In one optional embodiment, obtaining the environmental damage impact includes the following steps:

[0081] S411. Establish a time series based on environmental factors and structural damage indicators.

[0082] For environmental factors such as hourly temperature series T(t) and structural damage indicators such as concrete beam crack width series W(t), a wavelet denoising algorithm is first used to remove outliers and suppress high-frequency noise in the original monitoring data, and then missing values ​​are filled in using cubic spline interpolation. Based on this, a vector autoregressive (VAR) model incorporating multiple variables such as temperature, humidity, and traffic load is constructed, with a maximum lag order of 7.

[0083] Furthermore, Granger causality tests are used to verify the lagged correlations between variables. For example, by calculating the Pearson correlation coefficient between T(t-7) and W(t), the influence of historical temperature changes on the current crack development is quantified, and a lagged correlation diagram is drawn to visually demonstrate the dynamic correlation characteristics.

[0084] S412. Introduce the propensity score matching method to establish a causal relationship network diagram.

[0085] The propensity score matching (PSM) method was introduced, using traffic load intensity and maintenance measures as covariates to eliminate confounding factors and accurately analyze the causal impact of environmental factors on structural damage. A virtual intervention experiment was designed using the Do-Calculus causal inference framework: assuming an artificial increase of 5°C in the daily temperature range, the average treatment effect (ATE) on crack width was calculated through counterfactual analysis. The difference-in-differences (DID) method was used to compare the data of the intervention and control groups, ultimately identifying key causal relationships. For example, when the daily temperature range consistently exceeded 15°C, the probability of the weekly crack width growth rate exceeding 0.1 mm in concrete structures significantly increased by 23%. A causal network diagram was established to visualize the action path.

[0086] S413. Based on causal inference and time series analysis, the environmental damage impact is obtained.

[0087] Using the Causal Forest nonparametric regression algorithm, a systematic impact assessment of environmental factors such as temperature, humidity, and wind speed was conducted. A decision tree ensemble model was constructed through 500 bootstrap sampling iterations to quantify the average treatment effect (ATE) of each factor on structural damage. For example, the model output shows that for every 1°C increase in diurnal temperature range, the weekly crack width growth rate increases by an average of 0.02 mm; and for every 10% decrease in humidity, the crack propagation rate increases by 0.008 mm / week. The final environmental damage contribution radar chart clearly shows that temperature cycles contribute 60% to crack propagation, humidity changes contribute 30%, and other factors account for the remaining 10%, providing a quantitative basis for structural health monitoring decisions.

[0088] This invention introduces causal inference, and through intervention analysis to remove confounding variables, it achieves causal quantification of environmental impact, clarifies the specific degree of impact of environmental factors on damage, and provides a basis for targeted reinforcement (such as temperature fatigue resistance design).

[0089] In this embodiment, the step of analyzing the cumulative impact of environmental history on the structure based on the environmental damage effects and the dynamic anomaly characteristics includes the following steps:

[0090] S421. Construct a structural graph model, and enhance the graph structure and initialize parameters through the dynamic anomaly features.

[0091] Using large structural components such as bridges as nodes, such as "beam 1" and "support A," the connections between components are abstracted as edges. The weights of the edges are set as vibration transmission coefficients, which are derived through in-depth analysis and calculation of the structural dynamic anomaly characteristics. Based on this, a complete graph structure G=(V,E,W) is constructed. Here, the node set V represents all structural components, the edge set E represents the connections between components, and the weight set W quantifies the intensity of vibration transmission, providing a basic framework for subsequent analysis.

[0092] It is understandable that in complex bridge structures, the interactions between components are dynamic and multifaceted. Therefore, edge weights are optimized based on the vibration transmission coefficients of dynamic anomaly characteristics. For example, based on the constructed graph structure G=(V,E,W), the vibration transmission coefficient of "pier 2-beam 3" (e.g., 0.78) is multiplied by the environmental damage transmission probability of this connection (e.g., temperature influence transmission rate 0.65) to obtain the comprehensive weight of the edge, thereby quantifying the transmission efficiency of the influence between components. By integrating the vibration response in the bridge structure with the influence of environmental factors on the connection, the edge weights can more accurately reflect the actual correlation strength between components.

[0093] Furthermore, the initial weights of the nodes are initialized by directly using the environmental contribution of each component (e.g., "temperature contribution of support No. 4 is 62%) as the initial value of the nodes in the PageRank algorithm, ensuring that the initial state closely matches the actual damage to the components. Different components are affected by environmental factors to varying degrees. In this way, the algorithm can more realistically simulate the damage state of various bridge components under environmental influences from the initial stage.

[0094] S422. Construct an asymmetric transition matrix based on the causal network diagram of the environmental damage impact, and then improve the damage propagation model. Analyze the cumulative impact of environmental history on the structure through the damage propagation model.

[0095] Based on the directionality of structural influence, an asymmetric transition matrix is ​​constructed according to the causal relationship network diagram of the environmental damage influence to improve the symmetric transition matrix in the PageRank algorithm.

[0096] Specifically, based on the causal relationship network diagram (such as "temperature damage → support aging → beam deformation"), an asymmetric transition matrix is ​​constructed. ,in This represents the probability that damage is transmitted from component i to component j. This represents the probability from j to i, and the two probabilities may not be equal (e.g., , This allows for a more accurate description of the unidirectional or asymmetric propagation characteristics of damage between different components in a bridge structure.

[0097] Furthermore, environmental contribution is used as a node feature, and the powerful feature learning capability of Graph Convolutional Networks (GCNs) is leveraged to deeply mine the influence weights between nodes. For example, through training, a quantitative result such as "the influence weight of environmental damage at support A on beam 1 is 0.6" can be obtained. During training, the error between the predicted influence and the measured dynamic anomaly is used as the loss function. By continuously adjusting the model parameters, the GNN model is optimized to more accurately capture the mutual influence relationships between structural components, providing a reliable basis for risk assessment.

[0098] Furthermore, considering that the impact of the environment on bridge structures is not constant but rather accumulates over time and decays, a time decay factor is incorporated. (Values ​​range from 0.8 to 0.95, with smaller values ​​for longer damage histories) Corrected transfer probability ,satisfy: ,in Indicates the time elapsed since the damage occurred.

[0099] The relationships between components in a bridge structure are complex, involving not only direct connections but also numerous implicit relationships. A solution is to integrate implicit relationship learning using a graph neural network (GNN): inputting the graph structure into a GNN model (such as a GCN), training it based on dynamic anomaly propagation data, and outputting the implicit influence weights between components (e.g., "the weight of beam 1 indirectly influencing pier 5 is 0.21"), which are then added to the transition matrix. This paper addresses the issue of indirect influences not explicitly represented in topological graphs. By leveraging the powerful learning capabilities of GNNs, it uncovers potential connections between components that are difficult to represent in traditional topological graphs, thus improving the damage propagation model.

[0100] Furthermore, the iterative formula of the PageRank algorithm satisfies:

[0101]

[0102] in, This represents the value of node i in the (n+1)th iteration. This represents the damping coefficient (set to 0.85). This represents the initial value of node i. This represents all nodes that point to node i. This represents all the nodes that node j points to.

[0103] After each iteration, invalid propagation paths (edges with weights < 0.05) are filtered out using a dynamic anomaly threshold (e.g., a vibration deviation rate of 5%) to reduce computational redundancy. During the iteration process, by continuously updating node weights and filtering paths based on the dynamic anomaly threshold, key damage propagation paths can be effectively focused on, improving computational efficiency.

[0104] Furthermore, based on the cumulative environmental impact values ​​of each component, a ranking of cumulative impact values ​​is generated: components are sorted in descending order, and the top 30% are selected as "environmentally sensitive core components," such as "Support No. 4 (cumulative impact value 0.89) and Mid-span of Beam No. 3 (0.76)." These core components are key parts of the bridge structure that are more susceptible to environmental impact and damage, providing important reference for subsequent targeted reinforcement.

[0105] The damage propagation model outputs propagation path characteristics, and by combining the weight changes during the iteration process, key propagation chains are extracted, such as "temperature cycling → aging of support No. 4 → stress concentration in beam No. 3," and the cumulative impact percentage of each path is marked (e.g., this path accounts for 42% of the total impact). This clearly presents the specific propagation paths of damage caused by environmental factors in bridge structures and the degree of influence of each path, which helps to deepen the understanding of bridge damage mechanisms.

[0106] Correlation analysis of dynamic anomaly characteristics: By matching the propagation path with the location of dynamic anomalies (such as "high-frequency vibration of beam No. 3"), the "environment-dynamic coupling influence characteristics" are output, such as "cumulative temperature influence leads to aggravated vibration anomaly of beam No. 3, contributing 73%". This correlation analysis can reveal the coupling relationship between environmental factors and bridge dynamic response, providing a strong basis for comprehensively assessing the health status of bridge structures.

[0107] This invention replaces uniform initial weights with environmental contribution, aligning with actual structural damage. It constructs an asymmetric transition matrix and a time decay factor to adapt to the directionality and timeliness of damage propagation. Furthermore, it integrates GNN to uncover implicit influences, overcoming the explicit correlation limitations of topological graphs. This achieves a leap from "single-point assessment" to "global propagation aggregation" of historical environmental impacts, accurately locating environmentally sensitive components and impact paths, and providing targeted support for reinforcement schemes to "resist environmental damage."

[0108] S5. Based on geometric deformation characteristics, the bridge health status, and the cumulative impact, a targeted reinforcement scheme for the bridge is generated through multi-objective optimization and reinforcement learning.

[0109] Extracting the geometric deformation features includes the following steps:

[0110] S511. Enhance the point cloud data of the bridge scanned by lidar.

[0111] First, a statistical filtering algorithm is used to calculate the distance distribution from each point in the point cloud data to its neighboring points. Points with a mean distance exceeding three times the standard deviation are identified as noise points and deleted, effectively improving the purity of the point cloud data.

[0112] Then, a region growing algorithm is used to accurately segment key bridge components such as beams and piers based on the geometric features and spatial distribution of the point cloud, providing a foundation for subsequent analysis.

[0113] Finally, the preprocessed point cloud data was registered with the bridge design model using the ICP (Iterative Closest Point) algorithm. The registration error was strictly controlled within ≤5mm to ensure a high degree of consistency between the point cloud data and the design model.

[0114] S512. Extract morphological features based on the processed point cloud data.

[0115] In this embodiment, the surface normal vector and curvature of the processed point cloud data are calculated, and the Gaussian curvature is set to be >0.01mm⁻ 1 The region is identified as a convex deformation region, capturing the morphological features of the point cloud at the microscopic level.

[0116] And displacement field analysis, by calculating the distance difference between each point and the corresponding position in the design model, constructs a displacement field, which intuitively presents the deviation of each part of the bridge structure from the design state.

[0117] There is also deformation zone extraction. Using the region fusion method, continuous areas with displacement > 10mm are defined as deformation zones, which can accurately locate areas where bridge structures may have problems.

[0118] S513. Extract the geometric deformation features based on the morphological features.

[0119] Specifically, for the extracted deformation zone, its maximum displacement value is accurately calculated, such as "beam No. 3 sags 15mm at mid-span", to describe the degree of deformation with intuitive data.

[0120] Deformation rate assessment involves comparing and analyzing current deformation data with historical monitoring data to calculate the deformation rate, such as "an average annual increase of 3 mm," to understand the development trend of deformation.

[0121] The feature report generation integrates information such as the location, range, magnitude, and growth trend of the deformation zone, and outputs a complete geometric deformation feature report, providing detailed data support for bridge maintenance and safety assessment.

[0122] This invention solves the problem of efficiently quantifying minute deformations on complex surfaces of large bridges by using multi-dimensional analysis of normal vector, curvature, and displacement fields, combined with a region fusion algorithm to automatically identify deformation hotspots and achieve millimeter-level registration with the design model.

[0123] In this embodiment, the process of generating a targeted bridge reinforcement scheme based on geometric deformation features, the bridge's health status, and the cumulative impact through multi-objective optimization and reinforcement learning includes the following steps:

[0124] S521. Construct optimization objectives and constraints.

[0125] A multi-dimensional objective function system based on engineering requirements is constructed, specifically including: ① Improved health level: The health level of the structure after reinforcement must be improved by at least two levels under existing evaluation standards (e.g., from "hazardous" to "qualified"), and quantified through comprehensive indicators such as structural bearing capacity testing and durability assessment; ② Enhanced environmental adaptability: The response of the structure to environmental effects such as wind loads and temperature and humidity changes is used as the evaluation basis, requiring that the resistance of key performance indicators (such as structural displacement and stress fluctuation) be improved by no less than 40%; ③ Precise deformation control: The deformation of key parts of the structure (such as the mid-span of beams and the top of columns) is strictly controlled within 5mm, and monitored in real time by high-precision displacement sensors; ④ Cost constraints: Ensure that the total reinforcement cost does not exceed 80% of the project budget, covering the entire lifecycle costs of materials, labor, and equipment. Simultaneously, process feasibility constraints are set, such as that during carbon fiber bonding construction, the surface flatness of the reinforced component must be controlled within 2mm, and the concrete base strength grade must be no less than C15, providing technical boundaries for subsequent scheme design.

[0126] S522. A dynamic reward mechanism is constructed using a hierarchical state-space modeling method.

[0127] A hierarchical state-space modeling method is adopted, using the triplet "structural component-strengthening measure-working condition" as the state representation unit (e.g., "beam 1-carbon fiber bonding-wind load condition"). A dynamic reward mechanism is constructed by combining structural health monitoring data and simulation results. Positive rewards are given when the strengthening measures improve structural performance indicators beyond a threshold; otherwise, penalties are imposed. A deep neural network is trained based on the Proximal Policy Optimization (PPO) algorithm. Through more than 5000 iterations, the policy network converges to a stable state, ultimately outputting initial strengthening measure recommendations for each component, including material type and construction process parameters (e.g., "For beam 3, it is recommended to bond 3 layers of carbon fiber, each layer 0.11mm thick, with a bonding spacing of 10cm").

[0128] S523. Generate targeted reinforcement schemes for bridges through multi-objective optimization.

[0129] A non-dominated sorting genetic algorithm with an elitist strategy (NSGA-Ⅲ) is introduced. The initial scheme generated by reinforcement learning is used as the seed population, and parameter boundaries are set according to engineering design specifications (e.g., the number of carbon fiber layers ranges from 1 to 5, and the cross-sectional increase does not exceed 30% of the original size). Through more than 200 generations of evolutionary operations, trade-offs are made across six dimensions, including cost, reinforcement effect, and construction difficulty, ultimately generating a Pareto front solution set containing 10-15 schemes. Each scheme is accompanied by a performance comparison radar chart, visually demonstrating the performance of different schemes on various objectives (e.g., Scheme A: cost 500,000 yuan, health level improved by 2 levels, construction period 12 days; Scheme B: cost 700,000 yuan, health level improved by 3 levels, construction period 18 days).

[0130] A multi-attribute decision-making model was established, comprehensively considering engineering experience indicators such as the urgency of the construction period, the difficulty coefficient of construction, and the maintenance cost, and a secondary screening of the Pareto solution set was conducted. The entropy weight method was used to determine the weight of each indicator, and the analytic hierarchy process (AHP) was combined to compare and select the scheme, and finally the implementation scheme with the highest priority was determined. The output includes a 3D BIM construction diagram, process flow chart and detailed technical parameters: for example, "Three layers of 50cm wide carbon fiber cloth (model CF-300) are pasted at the mid-span of beam No. 3, and the No. 2 pot bearing (bearing capacity 5000kN) is replaced simultaneously. The construction is carried out in three stages, with a total construction period of 15 days. It is expected that the structural health level will be improved from 'poor' to 'good'. An intelligent monitoring system is set up to realize real-time tracking of the reinforcement effect."

[0131] Please see Figure 2In an embodiment, to efficiently execute the bridge targeted reinforcement scheme generation method based on digital twins provided by the present invention, the present invention also provides a bridge targeted reinforcement scheme generation system based on digital twins, comprising: an input device, an output device, a processor, and a memory, wherein the input device, output device, processor, and memory are interconnected, and the memory contains program instructions for the steps of the bridge targeted reinforcement scheme generation method based on digital twins. The bridge targeted reinforcement scheme generation system based on digital twins of the present invention has a compact structure and stable performance, and can stably execute the bridge targeted reinforcement scheme generation method based on digital twins of the present invention, further improving the overall applicability and practical application capability of the present invention.

[0132] In this embodiment, the processor may be a central processing unit, but it can also be other general-purpose processors, digital signal processors, application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (OPGs), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. Input devices can be used to acquire data. Output devices can be used to output the results obtained by storing program instructions contained in a computer program in the memory provided by this invention. The memory may include read-only memory and random access memory (RAM), and provides instructions and data to the processor. A portion of the memory may also include non-volatile random access memory (RAM).

[0133] In one possible implementation, the memory may include a stored program area and a stored data area. The stored program area may store the operating system and applications required for at least one function; the stored data area may store data created during use. Furthermore, the memory may include read-only memory and random access memory, and provides instructions and data to the processor. The memory stores the operating system and operating instructions, executable modules, or data structures, or subsets thereof, or extended sets thereof. The operating instructions may include various operation instructions for implementing various operations. The operating system may include various system programs for implementing various basic tasks and handling hardware-based tasks.

[0134] The embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for generating a bridge targeted reinforcement scheme based on digital twins.

[0135] The storage medium can include various media that can store program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0136] In summary, this invention uses knowledge graphs and fuzzy logic to accurately locate vulnerable points, overcoming the limitations of traditional experience-based judgment and laying the foundation for targeted reinforcement. Adaptive wavelet and modal analysis enable real-time early warning of vibration anomalies, capturing potential damage. Causal inference quantifies environmental damage, avoiding misjudgments and supporting environmental fatigue-resistant design. Point cloud technology quantifies geometric deformation with high precision, compensating for blind spots in traditional measurements. DIC and finite element updates enable full-domain stress monitoring, locating areas exceeding limits. Evidence theory integrates multi-source information, avoiding the one-sidedness of single indicators. Graph neural networks aggregate the cumulative effects of historical environmental data, considering component correlations. Finally, multi-objective optimization and reinforcement learning generate operable solutions, forming a complete chain of identification-monitoring-evaluation-solution, solving problems such as low accuracy, information conflict, and incomplete coverage in traditional technologies. This significantly improves the targeting and effectiveness of bridge reinforcement, ensuring structural safety.

[0137] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the present invention.

Claims

1. A method for generating targeted reinforcement schemes for bridges based on digital twins, characterized in that, Includes the following steps: Identify inherently vulnerable areas in bridge structures and obtain characteristics of structural vulnerability points; Dynamic features of structural vibration signals are extracted through adaptive time-frequency analysis to obtain dynamic anomaly characteristics; The real-time stress field is reconstructed, and the bridge's health status is comprehensively assessed by combining the real-time stress field, the structural vulnerability characteristics, and the dynamic anomaly characteristics. Based on the impact of environmental damage and the aforementioned dynamic anomaly characteristics, the cumulative impact of environmental history on the structure is analyzed; Based on geometric deformation characteristics, the bridge's health status, and the cumulative impact, a targeted reinforcement scheme for the bridge is generated through multi-objective optimization and reinforcement learning. The reconstructed real-time stress field includes the following steps: Surface displacement is obtained through image analysis, and a high-precision surface displacement field is constructed. The parameters of the finite element model are optimized, and the stress field is reconstructed by combining the high-precision surface displacement field and the optimized finite element model. The analysis of the cumulative impact of environmental history on the structure based on the environmental damage and the dynamic anomaly characteristics includes the following steps: A structural graph model is constructed, and the graph structure is enhanced and parameters are initialized using the aforementioned dynamic anomaly features; Based on the causal network diagram of the environmental damage impact, an asymmetric transition matrix is ​​constructed to improve the damage propagation model. This damage propagation model is then used to analyze the cumulative impact of environmental history on the structure, satisfying the following conditions: in, This represents the value of node i in the (n+1)th iteration. Indicates the damping coefficient. This represents the initial value of node i. This represents all nodes that point to node i. This represents all the nodes that node j points to. This represents the probability that damage is transmitted from component j to component i. This represents the corrected transition probability. Indicates the time decay factor. Indicates the time elapsed since the damage occurred.

2. The method for generating targeted bridge reinforcement schemes based on digital twins according to claim 1, characterized in that, The process of identifying inherently vulnerable areas in a bridge structure and obtaining structural vulnerability features includes the following steps: Entity recognition is performed on multi-source heterogeneous data to construct a knowledge graph; Design fuzzy logic rules and use fuzzy inference algorithms to identify inherently vulnerable areas of the bridge structure and obtain structural vulnerability features.

3. The method for generating targeted bridge reinforcement schemes based on digital twins according to claim 1, characterized in that, The method of extracting dynamic features of structural vibration signals through adaptive time-frequency analysis to obtain dynamic anomaly features includes the following steps: Optimize adaptive wavelet basis functions; Signal decomposition is performed based on the optimized wavelet basis, real-time modal parameters are extracted from the decomposed signal, and dynamic anomalies deviating from the normal state are identified based on the changes in the modal parameters.

4. The method for generating targeted bridge reinforcement schemes based on digital twins according to claim 1, characterized in that, The comprehensive assessment of bridge health status, combining the real-time stress field, structural vulnerability features, and dynamic anomaly features, includes the following steps: Normalize the structural weak point characteristics, dynamic anomaly characteristics, and real-time stress field characteristics, and establish BPA space mapping rules; Based on multi-source information fusion and evidence theory, the health status of bridges is comprehensively evaluated according to the BPA spatial mapping rule.

5. The method for generating targeted bridge reinforcement schemes based on digital twins according to claim 1, characterized in that, Obtaining the environmental damage impact includes the following steps: Establish time series based on environmental factors and structural damage indicators; A propensity score matching method is introduced to establish a causal relationship network diagram; The environmental damage impact was obtained based on causal inference and time series analysis.

6. The method for generating targeted bridge reinforcement schemes based on digital twins according to claim 1, characterized in that, The method for generating targeted bridge reinforcement schemes based on geometric deformation features, bridge health status, and cumulative effects through multi-objective optimization and reinforcement learning includes the following steps: Construct optimization objectives and constraints; A dynamic reward mechanism is constructed using a hierarchical state-space modeling method; A targeted reinforcement scheme for the bridge is generated by solving a multi-objective optimization problem.

7. The method for generating targeted bridge reinforcement schemes based on digital twins according to claim 6, characterized in that, Extracting the geometric deformation features includes the following steps: Enhancement processing is performed on the bridge point cloud data scanned by lidar; Based on the processed point cloud data, shape features are extracted; Based on the morphological features, the geometric deformation features are extracted.

8. A bridge targeted reinforcement scheme generation system based on digital twins, characterized in that, The bridge targeted reinforcement scheme generation system based on digital twins includes: an input device, an output device, a processor, and a memory. The input device, output device, processor, and memory are interconnected. The memory includes program instructions, which are used to execute the bridge targeted reinforcement scheme generation method based on digital twins according to any one of claims 1-7.

Images