Civil engineering structure health intelligent monitoring system based on digital twinning

Abstract

The application relates to the field of civil engineering and discloses a civil engineering structure health intelligent monitoring system based on digital twinning, which comprises the following modules: a multi-field fusion perception and topology coding module, a reversible digital twinning construction and anomaly injection module, an anomaly-driven active perception configuration reconstruction module, a virtual-real collaborative observation backfilling and twinning state correction module, a dynamic evolution boundary and risk trajectory generation module, and a strategy generation, execution and re-verification closed loop module. By introducing digital twinning technology and real-time sensor data fusion, the application can create a virtual model that dynamically and in real time reflects the health state of a civil engineering structure. This enables the monitoring system to capture changes and abnormalities in the structure in real time, improves monitoring accuracy, and avoids the accuracy reduction caused by data lag and discrete sampling in traditional monitoring methods.

Description

Technical Field

[0001] This invention relates to the field of civil engineering technology, specifically to an intelligent monitoring system for the health of civil engineering structures based on digital twins. Background Technology

[0002] Digital twin technology, a rapidly developing emerging technology in recent years, has been widely applied in various fields, such as manufacturing, aerospace, and urban management. By creating virtual models corresponding to physical entities, digital twin technology can reflect the real-time state changes of the physical entities and perform predictions and optimizations. In the field of civil engineering, the introduction of digital twin technology provides a new solution for structural health monitoring. By acquiring sensor data of the physical structure in real time and comparing and extrapolating with the virtual model, comprehensive monitoring and dynamic evaluation of the structure can be achieved.

[0003] Existing digital twin systems rely heavily on traditional static models or preset thresholds for anomaly detection and health assessment, which cannot cope with complex and dynamic structural changes. They also lack effective reverse engineering capabilities to identify potential damage or anomalies in the structure. Furthermore, many existing systems rely primarily on linear physical models or standardized computational methods when optimizing models, which are prone to accuracy deviations when faced with nonlinear changes. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a digital twin-based intelligent health monitoring system for civil engineering structures, which solves the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a digital twin-based intelligent monitoring system for the health of civil engineering structures, comprising: The multi-field fusion sensing and topology coding module is used to collect multiple physical quantities through multiple sensors set on civil engineering structures, generate multi-field spatiotemporal data tensors at discrete spatial locations and discrete times of the structure, and construct a structural topology coding matrix that includes structural node connection relationships and material properties. The reversible digital twin construction and anomaly injection module is used to construct a digital twin model based on multi-field spatiotemporal data tensors and structural topological coding matrices. By constructing reversible evolution operators and generating twin state sequences, it performs state deduction and reverse deduction, and at the same time generates anomaly injection residual fields. An anomaly-driven active sensing configuration reconstruction module is used to analyze the anomaly propagation path, generate the sensing gain field and active excitation scheme, and calculate the observation reachable domain based on the twin state sequence, reversible evolution operator and anomaly injection residual field. The virtual-real collaborative observation backfill and twin state correction module is used to generate a virtual observation field in an unobservable region under the premise of a known observation reachable region, observe and obtain real and virtual observation data, and correct the twin state sequence based on the real and virtual observation data, while generating a confidence field. The dynamic evolution boundary and risk trajectory generation module is used to construct evolution boundary functions based on the corrected twin state sequence, confidence field and anomaly injection residual field, and generate risk trajectories at the component level; The strategy generation, execution, and re-verification closed-loop module is used to generate monitoring strategies, operation strategies, maintenance strategies, and digital twin-side strategies based on the evolutionary boundary function, risk trajectory set, and corrected twin state sequence. It also adjusts the parameters of the preceding modules through a feedback mechanism to form adaptive closed-loop control.

[0006] Preferably, the multi-field fusion sensing and topology coding module includes physical quantities such as vibration, strain, temperature, humidity, and corrosion potential.

[0007] Preferably, the reversible evolution operator is used to describe the evolution relationship of twin states between adjacent time steps.

[0008] Preferably, the abnormality-driven active perception configuration reconstruction module calculates the abnormality influence weight of each structural component based on the propagation analysis of the abnormality injection residual field, and forms an abnormality influence weight matrix. Based on the abnormality influence weight matrix, it generates a sensor-related sensing gain field to adjust the sensor's sampling frequency, range, sensitivity, and filtering bandwidth.

[0009] Preferably, the structural topology encoding matrix includes a connection relationship matrix and an attribute matrix. The connection relationship matrix describes the connection relationships between structural nodes, and the attribute matrix describes the physical attributes of each structural unit.

[0010] Preferably, the anomaly-driven active sensing configuration reconstruction module further analyzes the propagation path of the anomaly in the structure based on the twin state sequence and the anomaly injection residual field, determines the degree of anomaly in key parts, and generates a sensing gain field based on the analysis results to optimize the deployment and configuration of monitoring sensors.

[0011] Preferably, the virtual observation field is extrapolated based on twin state sequences to fill data blind spots.

[0012] Preferably, the feedback mechanism includes receiving execution feedback data of monitoring, operation and maintenance strategies during system execution, and adjusting the parameters of the front-end modules in real time based on the feedback data to optimize system performance and form adaptive closed-loop control.

[0013] Preferably, the monitoring strategy is used to optimize sensor deployment, sampling frequency, and monitoring range to ensure that high-risk areas are effectively monitored; The operational strategy is used to generate control measures such as load limits and speed limits based on the structural health assessment results, so as to ensure the safety of the structure during operation. The maintenance strategy is used for regular inspections, local reinforcement, and re-inspections to ensure the long-term stability of the structure. The digital twin side strategy is used to update and optimize the parameters of the digital twin model, improve the accuracy and adaptability of the model, and ensure a high-precision description of the structural health status during long-term operation.

[0014] A digital twin-based intelligent monitoring method for the health of civil engineering structures includes the following steps: S1: Collect various physical quantity data of civil engineering structures through multi-field fusion sensing and topology coding modules, and generate multi-field spatiotemporal data tensors and structural topology coding matrices; S2: A digital twin state sequence is generated through a reversible digital twin construction and anomaly injection module, and an anomaly injection residual field is generated based on the difference between the observed data and the inference results; S3: Generates the sensing gain field and active excitation scheme through the abnormal-driven active sensing configuration reconstruction module, and adjusts the sensor sampling configuration; S4: By integrating real and virtual observation data through the virtual-real collaborative observation backfill and twin state correction module, the twin state is corrected to generate a credibility field; S5: The dynamic evolution boundary and risk trajectory generation module generates evolution boundary functions and risk trajectories to reflect the evolution trend of structural risks. S6: Generates and executes monitoring, operation, maintenance, and digital twin-side strategies through a closed-loop module of strategy generation, execution, and re-verification. Based on the strategy execution results, it performs parameter feedback and re-verification on the front-end modules to ensure the system's adaptability and optimization.

[0015] This invention provides an intelligent monitoring system for the health of civil engineering structures based on digital twins. It has the following beneficial effects: 1. This invention introduces digital twin technology and real-time sensor data fusion to create a dynamic, real-time virtual model that reflects the health status of civil engineering structures. This enables the monitoring system to capture structural changes and anomalies in real time, improving monitoring accuracy and avoiding the decrease in accuracy caused by data lag and discrete sampling in traditional monitoring methods.

[0016] 2. This invention achieves forward and backward deduction through reversible evolution operators, which can accurately predict the future state of the structure and reverse deduce its historical state when the structure changes. This method can help the system more accurately identify potential anomalies, damages or risks, improve the sensitivity and accuracy of anomaly detection, and thus provide a more effective early warning mechanism for civil engineering structures.

[0017] 3. This invention, through the adaptive optimization capability of the digital twin model, can automatically adjust model parameters and optimize monitoring strategies based on real-time monitoring data and changes in structural health status. Furthermore, the digital twin-side strategy supports continuous model updates and optimization, enabling full lifecycle management of civil engineering structures. This ensures that structural health monitoring remains efficient and accurate throughout the entire lifecycle, adapting to the needs of structural and environmental changes. Attached Figure Description

[0018] Figure 1 This is a system framework diagram of the present invention. Detailed Implementation

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example: Please see the appendix Figure 1 This invention provides a digital twin-based intelligent monitoring system for the health of civil engineering structures, including a multi-field fusion sensing and topology coding module, a topology coding-based reversible digital twin construction and anomaly injection module, an anomaly-driven active sensing configuration reconstruction module, a virtual-real collaborative observation backfilling and twin state correction module, a dynamic evolution boundary and risk trajectory generation module, and a strategy generation, execution, and re-verification closed-loop module.

[0021] This system is applicable to civil engineering projects such as bridges, tunnels, and building structures. The system forms a complete data processing chain through the sequential connection of its modules and achieves a closed-loop operation through parameter feedback after strategy execution. During operation, the system continuously collects observation data on various physical fields of the structure and constructs a reversible digital twin model under structural topology coding constraints to achieve structural state estimation, anomaly identification, active monitoring configuration, risk boundary determination, and strategy generation.

[0022] In the overall system architecture, the multi-field fusion sensing and topology coding module serves as the data input, acquiring physical quantities of the structure at different spatial locations and times, and generating a structural topology coding matrix. This topology coding matrix describes the connection relationships and material properties between nodes and structural units, providing fundamental constraints for the construction of the reversible digital twin model.

[0023] The multi-field spatiotemporal data tensor output by the multi-field fusion sensing and topology coding module is denoted as: ; in, Indicates position and time stress, Indicates position and time The response, Represents the temperature field. Represents the humidity field. Represents corrosion potential, symbol This represents other collectable structural physical fields. The structural topology encoding matrix is ​​denoted as: This module is used to record the connection relationships between structural nodes and structural units, as well as material properties. The reversible digital twin construction and anomaly injection module based on topological coding takes a multi-field spatiotemporal data tensor and a structural topological coding matrix as input to construct a digital twin state sequence. The twin state sequence is denoted as: ; in, Indicates the first At that moment, Indicates that digital twins are at any moment The structural state, This represents a digital twin model. This module further constructs a reversible evolution operator to enable forward and reverse extrapolation of the twin state over time.

[0024] The reversible evolution operator is denoted as: It satisfies the following forward and reverse operational relationships: ; ; This invention constructs anomaly-injected residual fields by exploiting the differences between actual observation data and reversible deduction results. Recorded as: ; in, Indicates the position of the structure ,time The actual observation state.

[0025] The anomaly-driven active sensing configuration reconstruction module generates a component anomaly impact weight matrix, sensing gain field, active excitation scheme, and observation reachability domain based on the digital twin state sequence, reversible evolution operator, and anomaly injection residual field output by the reversible digital twin construction and anomaly injection module. The observation reachability domain is used to describe the observability of each spatial region of the structure in the next observation period, providing constraints for virtual observation backfilling. The virtual-real collaborative observation backfilling and twin state correction module spatially partitions real observation data and digital twin inference results under the constraint of the observation reachability domain. Real observations are used within the reachability domain, while virtual observation backfilling is performed in inaccessible areas. The two types of observation results are fused based on a spatial mask to form a corrected twin state sequence. This module further generates a confidence field to reflect the reliability of the structural state estimation at different locations.

[0026] The dynamic evolution boundary and risk trajectory generation module constructs an evolution boundary function based on the corrected twin state sequence, confidence field, and anomaly injection residual field. This evolution boundary function defines the state boundaries of the structure in both the time and spatial domains. Based on this boundary function, a set of risk trajectories for the structural components is constructed to characterize the change of risk over time.

[0027] The strategy generation, execution, and re-verification closed-loop module generates monitoring, operation, maintenance, and digital twin-side strategies based on the evolutionary boundary function and risk trajectory set. The observation data obtained after executing these strategies is used to update the parameters of the multi-field fusion sensing and topology coding module, the reversible digital twin construction module, and the active sensing configuration reconstruction module. Through this data feedback, the present invention forms a top-down closed-loop structure, enabling the system to continuously adjust monitoring configurations and optimize digital twin model parameters during long-term operation, achieving an adaptive operation mode for specific structures.

[0028] A method for intelligent monitoring of the health of civil engineering structures based on digital twins includes the following steps: S1. The multi-field fusion sensing and topology coding module synchronously acquires data from vibration sensors, acceleration sensors, strain gauges, temperature sensors, humidity sensors, and corrosion potential sensors deployed on the civil engineering structure. It then performs denoising, time alignment, and coordinate mapping, organizing the observed data into a unified multi-field spatiotemporal data tensor. This module discretizes the structure, generating a set of nodes and structural elements, and constructs a structural topology coding matrix to record node connections and the material properties, cross-sectional dimensions, and construction information of the structural elements. The multi-field spatiotemporal data tensor and the structural topology coding matrix serve as inputs to the reversible digital twin construction and anomaly injection module, providing structural constraints and a physical quantity basis for subsequent twin construction.

[0029] S2. The reversible digital twin construction and anomaly injection module constructs a digital twin model based on multi-field spatiotemporal data tensors and structural topology encoding matrices, generating a twin state sequence containing state variables such as displacement, velocity, internal force, and stress. This module establishes forward and backward evolution relationships between adjacent time steps using reversible evolution operators, ensuring temporal consistency of the twin states. During online system operation, the module compares real-time observation data with prediction results based on reversible evolution operators, constructing an anomaly injection residual field to identify structural response deviations that are difficult to explain by the twin model. The twin state sequence, reversible evolution operators, and anomaly injection residual field serve as inputs to the anomaly-driven active sensing configuration reconstruction module.

[0030] S3. The anomaly-driven active sensing configuration reconstruction module utilizes twin state sequences, reversible evolution operators, anomaly injection residual fields, and structural topology coding matrices to calculate the component anomaly impact weights at the structural component level and generate a cumulative anomaly impact index within a given time window. Based on the component anomaly impact and sensor installation locations, the module generates a sensing gain field and adjusts the sensor sampling frequency, range, sensitivity, and filtering bandwidth. Simultaneously, it generates an active excitation scheme based on structural dynamic response characteristics to enhance the observability of high-anomaly regions. Based on the sensing gain field, the active excitation scheme, and the relationship between structural dynamic response and sensor deployment, the module calculates the observability index for discrete structural locations, forming the observation reachability domain for the next observation cycle. The component anomaly impact weights, sensing gain field, active excitation scheme, and observation reachability domain serve as inputs to the virtual-real collaborative observation backfill and twin state correction module.

[0031] S4. The virtual-real collaborative observation backfilling and twin state correction module divides the structural space into observable and unobservable regions based on the observable reach domain. In the observable region, it uses real observation data; in the unobservable region, it generates a virtual observation field using the twin state from the previous time step and reversible evolution operators. The real and virtual observations are then fused according to spatial location to construct a virtual-real collaborative observation field. Based on the error between the virtual-real collaborative observation field and the original twin state, the module corrects the twin state using a state update gain matrix, obtaining a corrected twin state sequence. Simultaneously, the module constructs a confidence field based on observation density, anomaly injection residual size, and virtual observation backfilling ratio to describe the reliability of state estimation at different locations in the structural space. The corrected twin state sequence and confidence field serve as inputs to the dynamic evolution boundary and risk trajectory generation module.

[0032] S5. The dynamic evolution boundary and risk trajectory generation module constructs a state margin index based on the corrected twin state sequence, an uncertainty factor based on the credibility field, and an anomaly intensity index based on the anomaly injection residual field. These three indices are then weighted and combined to form an evolution boundary function, which describes the comprehensive risk level at different locations within the structural space at different times. The module calculates component risk boundary indices at the structural component level and forms a risk trajectory set over time, reflecting the evolutionary trend of component risk. The risk boundary indices and risk trajectories can distinguish between short-term impact risks and long-term cumulative risks. The evolution boundary function, component risk boundary indices, and risk trajectory set serve as inputs to the strategy generation, execution, and re-verification closed-loop module.

[0033] S6. The strategy generation, execution, and re-verification closed-loop module constructs a set of strategies, including monitoring strategies, operation strategies, maintenance strategies, and digital twin-side strategies, based on the evolutionary boundary function, component risk boundary indicators, risk trajectory sets, and corrected twin state sequences. This module calculates component strategy strength factors based on component risk boundary indicators and risk increments, mapping these factors to corresponding monitoring, operation, and maintenance strategy configurations. It also generates digital twin-side parameter adjustment schemes based on model deviation metrics and credibility fields, used to update digital twin model parameters and reversible evolution operator parameters. Various strategies are executed through the monitoring system, operation control system, and maintenance management system. After the next data collection cycle, the module calculates parameter feedback adjustments based on the actual effect of strategy execution, feeding this feedback to the multi-field fusion sensing and topology coding module, the reversible digital twin construction and anomaly injection module, and the anomaly-driven active sensing configuration reconstruction module, thus forming a closed-loop adaptive and self-evolving mechanism throughout the entire system process.

[0034] The following section, with reference to the accompanying diagrams, will provide a detailed explanation of the implementation principles of each of the above steps and the logic modules involved: The field fusion sensing and topology coding module includes a sensor array for acquiring observation data of multiple physical fields, a data acquisition unit for performing data acquisition and preprocessing, and a processing unit for generating multi-field spatiotemporal data tensors and structural topology coding matrices. In one specific embodiment, the multi-field fusion sensing and topology coding module collects various physical quantities of the civil engineering structure by deploying vibration sensors, acceleration sensors, strain gauges, temperature sensors, humidity sensors, and corrosion potential sensors on key components. Sensor deployment locations include near supports, mid-span locations, and connection node areas of components such as beams, columns, slabs, piers, and cables. The data acquisition unit synchronizes all types of sensors with a unified clock, acquires multi-channel observation data using a preset sampling frequency, and performs noise reduction, detrending, and outlier removal processing on the acquired data. The multi-field fusion sensing and topology coding module organizes the preprocessed observation data into a multi-field spatiotemporal data tensor. The multi-field spatiotemporal data tensor is denoted as: ; in, Indicates spatial location in terms of structure. Indicates time, Indicates position and time stress, Indicates position and time The response, Indicates position and time temperature, Indicates position and time humidity, Indicates position and time The corrosion potential, denoted by the symbol ..., represents other physical field quantities acquired by the multi-field fusion sensing and topology coding module as needed. The multi-field fusion sensing and topology coding module uses spatial interpolation, coordinate mapping, and temporal alignment to uniformly map data from different sensor channels to discrete locations on the structure. and time This forms a complete multi-field spatiotemporal data tensor; The multi-field fusion sensing and topology coding module discretizes civil engineering structures, dividing them into nodes and structural elements. Nodes represent the geometric positions of the structure, and structural elements represent the component elements connecting two or more nodes. Based on the discretization results, the module constructs a structural topology coding matrix. The structural topology coding matrix is ​​denoted as: ; in, For the connection matrix, Attribute matrix. Connection matrix. The elements are denoted as: ; in, and Number the nodes. Attribute matrix. The elements are denoted as: ; in, Number the structural units. For attribute dimension indexing, Representing structural units The Each attribute value includes cross-sectional dimensions, material strength grade, material elastic modulus, density, prestressing state parameters, and structural type identifier.

[0035] The multi-field fusion sensing and topology coding module maps the physical location of the sensors to discrete structural nodes, constructing a mapping relationship from sensor channels to nodes. The sensor observation vector is denoted as: ; in, Number the sensor. For the first At that moment, Indicates the first Each sensor at time The observed values. The multi-field fusion sensing and topology coding module establishes a mapping matrix based on the correspondence between sensor installation locations and structural node coordinates, and... Mapped to the corresponding position in the multi-field spatiotemporal data tensor ,in Indicates the first The system identifies discrete node locations. For locations where sensors are not directly deployed, the multi-field fusion sensing and topology coding module uses an interpolation method based on the responses of neighboring nodes or structural units to generate complete multi-field data.

[0036] The multi-field fusion sensing and topology coding module, through the aforementioned data organization and topology coding process, provides a unified multi-field observation data representation for civil engineering structures in both temporal and spatial dimensions. Furthermore, it provides connection relationships between nodes and structural units, as well as material and structural attribute information, in the form of a structural topology coding matrix. The multi-field spatiotemporal data tensor and the structural topology coding matrix serve as inputs to the topology-based reversible digital twin construction and anomaly injection module, providing fundamental data support for subsequent digital twin model construction, reversible time extrapolation, and anomaly injection residual field generation.

[0037] In one specific embodiment, The reversible digital twin construction and anomaly injection module includes a twin modeling unit for constructing a digital twin model, a reversible evolution operator generation unit for realizing reversible time evolution, a training unit for performing model training and parameter calibration, and an anomaly residual calculation unit for generating anomaly injection residual fields. The reversible digital twin construction and anomaly injection module takes the multi-field spatiotemporal data tensor and structural topological coding matrix output by the multi-field fusion sensing and topological coding module as input. The multi-field spatiotemporal data tensor is denoted as... The structural topology encoding matrix is ​​denoted as The reversible digital twin construction and anomaly injection module, through twin modeling units, combines multi-field spatiotemporal data with structural topological constraints to construct a digital twin model describing the state of civil engineering structures. The digital twin model at time... Output the twin state sequence, denoted as: ; in, For the first At that moment, For at any time The twin state vector, For digital twin models, This is the parameter set for the digital twin model. Twin state vector. It can include the representation of state variables such as structural node displacement, node velocity, component internal force, component stress level, and crack index in a unified state space. The reversible digital twin construction and anomaly injection module constructs reversible evolution operators in the digital twin state space through a reversible evolution operator generation unit.

[0038] Starting from the known current state (such as initial stress, strain, etc.) and external load conditions (such as temperature changes, wind loads, etc.), the state of the structure at a certain future moment is deduced step by step using mechanical equations and numerical solutions. The calculation results of each time step (such as stress, displacement, etc.) constitute the twin state sequence of the structure. These state sequences describe the healthy evolution of the structure throughout its entire life cycle. After knowing the future state of the structure, the same mechanical equations and numerical methods are used to perform reverse deduction to calculate the past state of the structure. Combining the results of forward deduction and reverse deduction, the reversible evolution operator is finally obtained.

[0039] The reversible evolution operator is denoted as , Specifically, firstly, based on multi-field spatiotemporal data tensors and structural topological coding matrices, the displacement, stress, strain, crack indices of key component nodes in civil engineering structures, as well as some environmental state quantities, are selected and arranged to form discrete-time twin state vectors. These twin state vectors constitute the single-time state of the twin state sequence and are reversible evolution operators. The effective space is determined by the structural topology encoding matrix. Secondly, constrained by the node connections, element types, and material properties contained in the matrix, and combined with the structural dynamics equations, the structural response under multiple loads is discretized in time using numerical integration. This constructs a fundamental evolutionary mapping within the twin state space, providing a predictive relationship from the twin state at the previous time step to the twin state at the next time step. Based on this, an invertible parameter mapping structure with explicit inverse mapping is introduced after the fundamental evolutionary mapping to transform the fundamental evolution results. This invertible parameter mapping structure ensures a one-to-one correspondence of the overall mapping within the twin state space through block affine transformations or coupled mappings, thus defining the invertible evolution operator. This allows, given a twin state at a certain moment, both through To obtain the twin state in the next moment, and to be able to... The inverse mapping recovers the twin state from the twin state at the next time step to the twin state at the previous time step. Finally, using the twin state sequence within the historical time period and the corresponding real observation data, the parameters in the invertible parameter mapping structure are calibrated, enabling the invertible evolution operator to... While satisfying structural dynamics and topological constraints, it maintains a high degree of consistency with the time evolution behavior of real structures. Used to describe the evolution relationship of twin states between adjacent time steps. The reversible evolution operator satisfies the following forward evolution relationship: ; And the following reverse evolutionary relationships: ; in, In time step to External excitations or operating parameter vectors acting on the structure, For the parameter set of the reversible evolution operator, This is the inverse mapping of the reversible evolution operator. The reversible evolution operator is implemented using a parameterized structure that satisfies the reversibility property. By constraining the network structure or state-space model structure, the forward and inverse mappings maintain consistency in numerical computation. The training unit in the reversible digital twin construction and anomaly injection module is based on historical observation data and structural analysis data, and it optimizes the parameter set of the digital twin model. and the parameter set of the reversible evolution operator Calibration is performed. The training unit extracts continuous time windows from the multi-field spatiotemporal data tensor and generates state sequence samples using the connection relationships and attribute information contained in the structural topological encoding matrix.

[0040] The training unit updates the parameter set by minimizing the twin state prediction error and the time-reversible reconstruction error. and This enables the digital twin model to accurately describe the structural dynamic response under structural topological constraints while maintaining time reversibility. During the online operation phase of the system, the reversible digital twin construction and anomaly injection module receives real-time observation data from the multi-field fusion sensing and topology coding module, and uses the trained digital twin model and reversible evolution operators to predict the twin state. The actual observed state is denoted as , indicating the position and time The actual structural state. The reversible digital twin construction and anomaly injection module compares the actual observed state with the twin state derived using the reversible evolution operator through the anomaly residual calculation unit to construct the anomaly injection residual field.

[0041] In one embodiment, the reversible digital twin construction and anomaly injection module constructs the anomaly injection residual field in the following manner. First, at time t... Obtaining twin state using digital twin model The predicted state is obtained through forward derivation using a reversible evolution operator. Its expression is: ; Then, predict the state Substituting the inverse mapping of the invertible evolution operator, we obtain the reconstructed state: ; The reversible digital twin construction and anomaly injection module constructs an anomaly injection residual field based on the difference between the actual observed state and the reversible deduction results. The anomaly injection residual field is denoted as: ; in, Indicates position and time The abnormal residual vector. For situations where it is necessary to highlight the impact of time reversibility, the abnormal injection residual field can also be combined with the reconstructed state. The biases are jointly measured to assess the deviation of the digital twin model from the real structural behavior under reversible evolution conditions. The reversible digital twin construction and anomaly injection module organizes the anomaly injection residual field into a spatiotemporal field form through anomaly residual calculation units and associates it with the structural topology encoding matrix, enabling the anomaly injection residual field to be located at specific nodes and structural units. Anomaly Injection Residual Field and reversible evolution operators As input to the anomaly-driven active sensing configuration reconstruction module, subsequent modules perform anomaly impact propagation analysis, sensor gain adjustment, and observation reachability domain construction based on the aforementioned quantities. The reversible digital twin construction and anomaly injection module utilizes multi-field spatiotemporal data tensors and structural topology coding matrices to construct a structural digital twin state sequence. Simultaneously, it constructs a reversible evolution operator with forward and reverse mapping capabilities and generates anomaly injection residual fields based on these operators. This provides the necessary state evolution information and anomaly measurement information for the subsequent active sensing configuration reconstruction and virtual-real collaborative observation backfilling modules.

[0042] The anomaly-driven active sensing configuration reconstruction module provided by the present invention includes an anomaly influence analysis unit for calculating the component anomaly influence weight matrix based on the anomaly injection residual field, a sensing configuration adjustment unit for generating a sensing gain field, an excitation configuration unit for generating an active excitation scheme, and an observability analysis unit for constructing the observation reachable domain. In one embodiment, the anomaly-driven active perception configuration reconfiguration module takes as input the twin state sequence, reversible evolution operator, and anomaly injection residual field output by the reversible digital twin construction and anomaly injection module. The twin state sequence is denoted as... The reversible evolution operator is denoted as The abnormal injection residual field is denoted as .in, For the first At that moment, The structure is spatially discrete. Under the constraint of the structural topology encoding matrix, the anomaly-driven active sensing configuration reconstruction module analyzes the propagation of the anomaly injection residual field in the time dimension and structural topology, forming a measure of the anomaly's impact on each structural component. The anomaly-driven active sensing configuration reconstruction module first discretizes the structure into several sets of structural components. ,in For the first Each structural component This refers to the number of structural components. Each structural component... Corresponding to a set of discrete spatial locations .

[0043] The anomaly-driven active perception configuration reconfiguration module defines the component anomaly impact weights based on the distribution of the anomaly-injected residual field within the structural component. The component anomaly impact weight matrix is ​​denoted as: ; in, Indicates at time Structural components Abnormal influence weights Indicates the location and time The norm of the anomaly injection residual vector. The anomaly-driven active perception configuration reconstruction module performs time accumulation or sliding window statistics on the component anomaly impact weights based on the anomaly injection residual field of multiple consecutive time steps, in order to identify structural components that have a large anomaly impact that persists in the time dimension; In one embodiment, to demonstrate the role of the reversible evolution operator in the anomaly propagation path, the anomaly-driven active perception and reconstruction module utilizes the reversible evolution operator to propagate the twin state forward and backward within a given time window, expanding the anomaly injection residual field at a single time step into a cumulative effect over the time series. For a length of... Within the time window, the anomaly-driven proactive perception and reconstruction module can construct time-extended component anomaly impact weights: ; in, In order to Structural components within the end time window The cumulative anomaly affects the weight. This refers to the length of the time window. The anomaly-driven active perception and reconstruction module sorts and classifies structural components based on the component anomaly impact weight matrix and the cumulative anomaly impact weight, identifying component areas with higher anomaly impact levels. The anomaly-driven active perception and reconstruction module also uses a sensor set... The sensing gain field is generated based on this. The sensing gain field is denoted as: ; in, For the first One sensor, For the number of sensors, Indicates the first Each sensor at time The sampling frequency adjustment coefficient, This represents the range adjustment factor. This represents the sensitivity adjustment coefficient. This represents the filter bandwidth adjustment coefficient. The anomaly-driven active perception and reconstruction module maps the weight of the component's anomaly impact to adjacent sensors based on the installation correspondence between the sensors and structural components. Through a preset allocation function, the anomaly impact weight is converted into the aforementioned gain adjustment coefficients, enabling sensors closer to components with high anomaly impact to obtain higher sampling frequencies and higher sensitivity configurations. The anomaly-driven proactive perception and reconstruction module generates a proactive incentive scheme based on the component anomaly impact weight matrix and twin state sequence through the incentive configuration unit. The proactive incentive scheme is denoted as: ; in, Indicates at time The active excitation vector applied to the structure, Indicates the first Each incentive channel at time The excitation amplitude or excitation parameters, This represents the number of excitation channels. The anomaly-driven active sensing and reconstruction module prioritizes excitation locations near components with significant anomaly impact, enhancing the observability of structural responses related to the anomaly by adjusting excitation amplitude and frequency. The anomaly-driven active sensing and reconstruction module constructs the observation reachability domain through an observability analysis unit. The observation reachability domain is denoted as: ; in, Indicates at time Structurally observable at least as low as the threshold The set of spatial locations For position At any moment Observability indicators A preset observability threshold is set. The observability index can be calculated using the transfer relationship between sensor distribution, sensor gain, and structural dynamic response. In one embodiment, the observability index can be expressed as a sensitivity measure of the sensor response to the structural state. Let... For position With sensors At any moment If the response transfer function value is given, then the observability index can be expressed as: ; in, In order to be with the first Weighting coefficients related to each sensor, For the aforementioned sensing gain vector, the symbol "·" indicates element-wise multiplication or weighting. Norm operations are represented. The anomaly-driven active perception configuration reconstruction module calculates the observability index of each discrete position of the structure according to the above formula, and compares it with a preset threshold to determine the observable reachable domain.

[0044] The anomaly-driven active sensing configuration reconstruction module uses the component anomaly impact weight matrix, sensor gain field, active excitation scheme, and observation reachability domain as inputs to the virtual-real collaborative observation backfill and twin state correction module. The observation reachability domain provides spatial constraints for subsequent virtual observation backfill, while the sensor gain field and active excitation scheme provide parameter basis for the acquisition configuration of the new cycle of observation data. Through the above implementation method, the anomaly-driven active sensing configuration reconstruction module establishes a functional relationship between the anomaly injection residual field and the configuration process of sensor and excitation resources. This enables the civil engineering structural health monitoring system to perform targeted reconstruction of the observation layout and observation capabilities under anomaly-driven conditions, providing basic data and configuration conditions for subsequent virtual-real collaborative correction and risk assessment.

[0045] The virtual-real collaborative observation backfill and twin state correction module provided by the present invention includes a real observation processing unit for processing real observation data of a new period, a virtual observation generation unit for generating a virtual observation field in an observation-inaccessible area, a twin state correction unit for fusing and correcting the digital twin state, and a credibility assessment unit for constructing a credibility field. In one embodiment, the virtual-real collaborative observation backfilling and twin state correction module takes the observation reachability region, sensor gain field, and active excitation scheme output by the anomaly-driven active sensing configuration reconstruction module as input, and simultaneously receives the original twin state sequence output by the reversible digital twin construction and anomaly injection module, as well as the new cycle of real observation data acquired under the sensor gain field and active excitation scheme configuration. The observation reachability region is denoted as... The sensing gain field is denoted as The proactive incentive plan is denoted as The original twin state sequence is denoted as The representation of the actual observation data in the new period at the structural location is denoted as... ,in, Let x be the discrete spatial location of the structure at the k-th time. This is the number of the nth sensor.

[0046] The virtual-real collaborative observation backfill and twin state correction module first divides the structural space based on the observation reachability domain. The observation mask function is denoted as: ; in, Indicates at time The x-axis indicates whether the location is within the reachable region of observation; a value of 1 indicates the location is within the reachable region, and a value of 0 indicates the location is outside the reachable region. The real observation processing unit, based on the sensor gain field and the active excitation scheme, performs denoising, alignment, and coordinate mapping on the raw sensor data acquired in the new cycle, converting the sensor channel data into a real observation field at discrete structural locations. and only Assign values ​​within the specified area.

[0047] The virtual observation generation unit generates a virtual observation field in areas inaccessible by traditional observation. The virtual observation field is denoted as... In one embodiment, for satisfying The virtual observation generation unit performs state deduction based on the twin state sequence of the previous time step or multiple historical time steps and a reversible evolution operator. Let the reversible evolution operator be... The external operating condition vector is The parameter set of the reversible evolution operator is Then the virtual observation generation unit is in the location Virtual observations can be represented as: ; in, Let be the mapping function that maps from the twin state to the observation space. For structural topology encoding matrix, For the parameter set of the mapping function, This is the predicted state obtained by forward extrapolating the twin state from the previous time step using a reversible evolution operator. For conditions satisfying... In this location, the virtual observation generation unit does not overwrite the real observation, and the virtual observation field does not participate in subsequent fusion within this region. Virtual-real collaborative observation backfilling and twin state correction construct a virtual-real collaborative observation field. The virtual-real collaborative observation field is denoted as... , can be represented as: ; Among them, the symbol " " indicates point-by-point multiplication. Indicates at time An observation field is obtained by fusing real and virtual observations using a spatial mask. This expression uses real observations within the observable region and virtual observations in the unobservable region, thus forming a complete observation field across the entire structural space. A twin state correction unit corrects the digital twin state based on the virtual-real co-observation field. Let the observation operator be... Used to map twin states to the observation space, then at time... The observation error can be expressed as: ; in, For position and time The observation error vector, This is the parameter set of the observation operator. The twin state correction unit constructs a state increment term based on the observation error, corrects the original twin state sequence, and obtains the corrected twin state sequence. The corrected twin state sequence is denoted as... Its expression is: ; in, For position and time The state update gain matrix is ​​used to control the magnitude of the effect of observation error on twin state correction.

[0048] The state update gain matrix is ​​related to the observation reachability region, the sensor gain field, and historical residual statistics. In regions with dense observation data and small outlier residuals, the state update gain matrix can take a larger value; conversely, in regions with sparse observations or large outlier residuals, it can take a smaller value, thus achieving spatial adjustment of the twin state correction magnitude. The confidence assessment unit constructs a confidence field based on the corrected twin state. The confidence field is denoted as... Used to characterize position At any moment The reliability of the state estimation is assessed. The reliability evaluation unit comprehensively considers factors such as observation density, the magnitude of anomaly injection residuals, and the virtual observation backfill ratio to construct a reliability function. In one embodiment, the reliability field can be expressed as: ; in, For the credibility calculation function, For the set of parameters of the credibility function, For position At any moment The observation density index The anomaly injection residual field output by the anomaly injection module for reversible digital twin construction. For position and time The virtual observation backfill ratio index. The observation density index is used to reflect the location within a certain time window. The effective number of observations or observation coverage, and the virtual observation backfill ratio index are used to reflect the degree to which virtual observations are used in the fusion at this location.

[0049] In one embodiment, the credibility assessment unit can construct the credibility function using a linear weighted form. Let the weight coefficients be... The sensitivity parameter is The credibility field can then be specifically written as: ; in, It is an exponential function. Inject the norm of the residual vector into the anomaly. The larger the value, the more comprehensive the observation at that location. The smaller the value, the smaller the abnormal deviation. A smaller value indicates a lower virtual backfill ratio at that location, thus resulting in higher reliability. Weighting coefficients and sensitivity parameters are determined using historical data or engineering experience. Experience has confirmed this. The virtual-real collaborative observation backfilling and twin state correction module will correct the twin state sequence. and credibility field The output is sent to the dynamic evolution boundary and risk trajectory generation module. Simultaneously, a virtual-real collaborative observation field is generated. With observation mask function It can be used for subsequent auxiliary analysis of the evolutionary boundary function construction process and to observe the mask function. This is an indicator function for the observation reachability domain. Through the above implementation method, the virtual-real collaborative observation backfilling and twin state correction module, under the constraint of the observation reachability domain, spatially fuses real observations with virtual observations based on digital twins, selectively corrects the digital twin state, and constructs a confidence field with spatial and temporal resolution, providing basic data and confidence descriptions for subsequent dynamic evolution boundary construction and risk trajectory analysis.

[0050] The dynamic evolution boundary and risk trajectory generation module provided by the present invention may include: a boundary construction unit for constructing an evolution boundary function, a risk trajectory generation unit for generating risk trajectories for key components, and a result output unit for outputting the boundary function and risk trajectory to the subsequent strategy generation, execution and re-verification closed-loop module; The dynamic evolution boundary and risk trajectory generation module takes the corrected twin state sequence and confidence field output by the virtual-real collaborative observation backfilling and twin state correction module as input, and the anomaly injection residual field output by the reversible digital twin construction and anomaly injection module as input. The corrected twin state sequence is denoted as... The credibility field is denoted as The abnormal injection residual field is denoted as ,in, For discrete spatial locations in terms of structure For the first At any given moment, the dynamic evolution boundary and risk trajectory generation module, based on the above inputs, constructs an evolution boundary function describing the structural state safety margin and uncertainty, and forms a risk trajectory at the component level.

[0051] To construct the evolutionary boundary function, the dynamic evolutionary boundary and risk trajectory generation module first calculates the state margin index based on the corrected twin state sequence. The state margin index is defined as follows: Used to reflect location At any moment The degree to which the structural response approaches the design limit or working limit. The state margin index can be calculated based on the ratio or difference between the stress, displacement, crack width, etc., in the corrected twin state and their corresponding allowable values.

[0052] For multiple state components, the state margin index can be represented by a normalized composite index. The dynamic evolution boundary and risk trajectory generation module constructs an uncertainty suppression factor based on the credibility field. Credibility field The value of can be limited to [0,1], with a value closer to 1 indicating higher reliability of the state estimate. To reflect the influence of reliability in the evolutionary boundary function, the dynamic evolutionary boundary and risk trajectory generation module constructs a reliability factor. Its expression can be written as: ; in, Indicates position and time The higher the value, the greater the uncertainty in the state estimate at that location. The dynamic evolution boundary and risk trajectory generation module constructs an anomaly intensity index based on the anomaly injection residual field. The anomaly intensity index is labeled as... Its expression can be written as: ; in, For position and time The norm of the abnormal injection residual vector is used to quantitatively describe the degree of abnormality at that location at the current time.

[0053] The dynamic evolution boundary and risk trajectory generation module combines state margin indices, uncertainty factors, and anomaly intensity indices to construct an evolution boundary function. The evolution boundary function is denoted as... , can be represented as: ; in, These are weighting coefficients, used to control the relative influence of state margin, uncertainty factor, and anomaly intensity on the evolutionary boundary function, respectively. The numerical value is used to characterize the position. At any moment The overall degree of proximity to the boundary.

[0054] When the state margin index, uncertainty factor, and anomaly intensity are large, the value of the evolution boundary function increases, indicating that the location is near the structural safety boundary or in a high-risk state. The dynamic evolution boundary and risk trajectory generation module defines risk metrics at the structural component level. Let the set of structural components be... ,in For the first Each structural component For the number of components, components The corresponding set of spatial locations is denoted as At that moment member Risk boundary is marked as It can be represented as: ; in, Reflecting components At any moment The maximum value of the evolutionary boundary function within the internal spatial location is used to characterize the point with the highest risk level in the component. For cases where the average risk level of the component needs to be considered, a method can also be used within the component's spatial range... Component risk indicators are defined using integration or averaging. The dynamic evolution boundary and risk trajectory generation module serializes the component risk boundary indicators over time, forming a risk trajectory. The risk trajectory is recorded as follows: ; in, For the selected time steps, Representing components A sequence of risk boundary indicators at multiple discrete moments. The dynamic evolution boundary and risk trajectory generation module characterizes the risk evolution process of components by calculating the changing trend of the risk trajectory; To further differentiate between short-term shock risks and long-term gradual risks, the dynamic evolution boundary and risk trajectory generation module can perform derivative or incremental analysis on the risk trajectory. (Component) The risk increment at adjacent time steps is denoted as: ; in, Representing components At any moment Relative to time The risk boundary indicator changes. Based on the risk increment and the absolute value of the risk boundary indicator, the dynamic evolution boundary and risk trajectory generation module can identify components whose risk boundary indicators rise sharply in a short period of time and components whose risk boundary indicators accumulate slowly over a longer period of time.

[0055] The dynamic evolution boundary and risk trajectory generation module summarizes the evolution boundary functions at the overall structural level to form the overall structural risk distribution. Let the overall structural risk function be denoted as... It can be represented as a set of component risk boundary indicators: ; in, Used to describe at time The risk status of each component provides component-level and structure-level risk inputs for the subsequent strategy generation module.

[0056] The dynamic evolution boundary and risk trajectory generation module will generate the evolution boundary function. Component risk boundary indicators and risk trajectory set The output is sent to the strategy generation, execution, and re-verification closed-loop module. Based on the above output, the strategy generation, execution, and re-verification closed-loop module generates monitoring strategies, operation strategies, maintenance strategies, and digital twin-side strategies, and updates the parameters of the preceding modules through strategy execution feedback.

[0057] Through the above implementation methods, the dynamic evolution boundary and risk trajectory generation module constructs an evolution boundary function based on the corrected twin state, confidence field, and anomaly injection residual field, and generates a risk trajectory with time dimension at the component level. This enables dynamic boundary characterization and risk evolution representation of the health status of civil engineering structures, providing quantitative risk basis for subsequent strategy generation and closed-loop system operation.

[0058] The closed-loop module for strategy generation, execution, and re-verification provided by this invention may include: a strategy generation unit for generating multi-level structural health management strategies, a strategy execution unit for executing strategies in monitoring systems and civil engineering structure operation systems, and a parameter feedback and re-verification unit for re-verifying and adjusting parameters of the preceding modules based on the execution results. In one embodiment, the policy generation, execution, and re-verification closed-loop module takes the evolutionary boundary function and risk trajectory set output by the dynamic evolutionary boundary and risk trajectory generation module as its main input, and the corrected twin state sequence output by the virtual-real collaborative observation backfill and twin state correction module as its auxiliary input. The evolutionary boundary function is denoted as... The risk trajectory set is denoted as The corrected twin state sequence is denoted as ,in, For discrete spatial locations in terms of structure For the first At that moment, For the first in the set of structural components Each component.

[0059] The strategy generation, execution, and re-verification closed-loop module, at the component level, refers to component risk boundary indicators and risk trajectory sequences to classify each component into strategy levels. The component risk boundary indicator is marked as... This is derived from the calculation results of the dynamic evolution boundary and risk trajectory generation module. The strategy generation, execution, and re-verification closed-loop module is based on... The numerical value and the trend of risk trajectory over time are used to divide the components into different risk level intervals, and to assign strategy configuration parameters of different intensities to each risk level interval.

[0060] In one embodiment, the policy generation, execution, and re-verification closed-loop module constructs a policy set. The policy set is denoted as: ; in, To monitor a subset of strategies, For a subset of operational strategies, To maintain the policy subset, This refers to a subset of strategies for the digital twin side. (Monitoring strategy subset) Used to describe decisions such as sensor addition, sensor shutdown, sensor placement adjustment, and sampling scheme adjustment; a subset of operational strategies. Used to describe operational control commands such as load limit control, speed limit control, and time-based traffic control; maintains a subset of strategies. Used to describe local re-inspection, supplementary inspection, and structural reinforcement plans; a subset of digital twin-side strategies. Used to describe the triggering conditions and magnitudes for adjusting parameters of reversible evolution operators, retraining or recalibrating digital twin models.

[0061] The strategy generation, execution, and re-verification closed-loop module generates strategy strength factors at the component level. Let the component strategy strength factor be denoted as... Its expression can be written as: ; in, and These are the weighting coefficients. For components At any moment Risk boundary indicators For the risk increment of adjacent time steps, Component strategy strength factor Used to comprehensively reflect the current risk level of components and its rate of change; The closed-loop module for strategy generation, execution, and re-verification maps strategy strength factors to configuration parameters for monitoring, operational, and maintenance strategies. Regarding monitoring strategies, components... The corresponding monitoring strategy configuration can be represented as: ; in, Representing components At any moment The corresponding monitoring strategy configuration vector, To monitor the policy mapping function, The monitoring strategy configuration vector, which serves as a set of control parameters, may include sensor density adjustment parameters, sampling frequency adjustment parameters, and filter setting parameters. Regarding operational strategies, the strategy generation, execution, and re-verification closed-loop module provides operational strategy parameters to the operation control system based on the overall structural risk distribution and risk changes in key components. The operational strategy configuration can be represented as: ; in, For a moment Operational strategy configuration vector, For a moment The overall risk function set of the structure The operational strategy control parameter set includes parameters such as load limit coefficient, speed limit coefficient, and traffic control time window. Regarding maintenance strategies, the strategy generation, execution, and re-verification closed-loop module generates maintenance priorities and recommended maintenance measures based on component risk trajectories and strategy strength factors. The maintenance strategy configuration can be represented as: ; in, For components Maintenance strategy configuration vector, To maintain the policy mapping function, To maintain the set of strategy control parameters, the maintenance strategy configuration vector may include information such as whether to conduct local re-inspection, whether to organize supplementary inspection, and whether to formulate a reinforcement plan; Digital twin side strategy subset This is used to adjust the parameters of the digital twin model and the parameters of the reversible evolution operator. Let the set of parameters of the digital twin model and the set of parameters of the reversible evolution operator be collectively denoted as the system parameter vector. The parameter adjustment increment is denoted as Rules for adjusting parameters in the strategy generation, execution, and re-verification closed-loop module: ; in, Adjust the mapping function for the parameters. Adjust the set of control parameters for the parameters. For at any time A model bias metric is calculated based on the corrected twin state and real observations. The parameter adjustment mapping function controls the parameter adjustment magnitude according to the model bias and the confidence field, ensuring the digital twin model maintains consistency with the real structural response during long-term operation. The policy execution unit distributes the above policy set to the relevant execution systems.

[0062] The monitoring strategy, through the monitoring system configuration interface, operates on the multi-field fusion sensing and topology coding module and the anomaly-driven active sensing configuration reconstruction module to update the sensor deployment scheme and sampling configuration. The operation strategy, through the structural operation control system interface, operates on the traffic scheduling or load control system. The maintenance strategy, through the operation and maintenance management system interface, is transformed into detection and reinforcement plans. The digital twin-side strategy, through the model management interface, operates on the reversible digital twin construction and anomaly injection module to retrain or calibrate model parameters and reversible evolution operator parameters. The parameter feedback and re-verification unit calculates the strategy execution effect evaluation index based on the new cycle of observation data after strategy execution. Let the evolution boundary change evaluation index be denoted as... It can be represented as: ; in, For the evaluation function of evolutionary boundary changes, and These are the evolution boundary functions for adjacent time steps. The parameter backflow and re-verification unit evaluates the effectiveness of the strategy in reducing risk boundary indicators and slowing down the rate of risk growth based on the changes in the evolution boundary functions before and after strategy execution. The parameter feedback and re-verification unit maps the evaluation results into corrections to the parameters of the preceding modules. For the multi-field fusion sensing and topology coding module, the parameter feedback and re-verification unit adjusts the sensor deployment strategy and sampling scheme configuration parameters based on the monitoring strategy execution effect and changes in observation coverage. For the reversible digital twin construction and anomaly injection module, the parameter feedback and re-verification unit updates the digital twin model parameters and reversible evolution operator parameters based on changes in model bias and confidence field. For the anomaly-driven active sensing configuration reconstruction module, the parameter feedback and re-verification unit corrects the component anomaly impact weight update rules and sensor gain field generation rules based on anomaly propagation paths and changes in the observation reachability domain.

[0063] Through the above implementation methods, the strategy generation, execution, and re-verification closed-loop module generates multi-level strategies for monitoring, operation, maintenance, and digital twin based on the evolutionary boundary function and risk trajectory set. The strategy execution results are then applied to the multi-field fusion sensing and topology coding module, the reversible digital twin construction and anomaly injection module, and the anomaly-driven active sensing configuration reconstruction module in the form of parameter feedback. This forms a closed-loop adaptive adjustment mechanism that runs through the entire structural health monitoring process, enabling the system to have self-correction and self-evolution capabilities based on operational feedback.

[0064] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital twin-based intelligent monitoring system for the health of civil engineering structures, characterized in that: include: The multi-field fusion sensing and topology coding module is used to collect multiple physical quantities through multiple sensors set on civil engineering structures, generate multi-field spatiotemporal data tensors at discrete spatial locations and times of the structure, and construct a structural topology coding matrix containing connection relationship matrices and attribute matrices. The reversible digital twin construction and anomaly injection module is used to construct a digital twin model based on the multi-field spatiotemporal data tensor and the structural topology encoding matrix, and to construct a reversible evolution operator to generate a twin state sequence, perform state deduction and reverse deduction, and generate anomaly injection residual field. An anomaly-driven active sensing configuration reconstruction module is used to analyze the anomaly propagation path, generate a sensing gain field and an active excitation scheme, and calculate the observation reachable domain based on the twin state sequence, the reversible evolution operator, and the anomaly injection residual field. The virtual-real collaborative observation backfill and twin state correction module is used to obtain the observation-unreachable area and generate a virtual observation field under the premise of known observation reachable domain, observe and obtain real observation and virtual observation data, and correct the twin state sequence based on real observation and virtual observation data, while generating a confidence field. The dynamic evolution boundary and risk trajectory generation module is used to construct evolution boundary functions based on the corrected twin state sequence, confidence field and anomaly injection residual field, and generate risk trajectories at the component level; The strategy generation, execution, and re-verification closed-loop module is used to generate monitoring strategies, operation strategies, maintenance strategies, and digital twin-side strategies based on the evolutionary boundary function, risk trajectory, and the corrected twin state sequence, and adjust the parameters of the preceding modules through a feedback mechanism.

2. The intelligent monitoring system for the health of civil engineering structures based on digital twins according to claim 1, characterized in that: The multi-field fusion sensing and topology coding module includes various physical quantities such as vibration, strain, temperature, humidity, and corrosion potential.

3. The intelligent monitoring system for the health of civil engineering structures based on digital twins according to claim 1, characterized in that: The reversible digital twin construction and anomaly injection module constructs reversible evolution operators through a reversible evolution operator generation unit, specifically including: Based on the multi-field spatiotemporal data tensor and the structural topology coding matrix, the nodal displacements, stresses, strains, crack indices, and environmental state quantities of civil engineering structures are selected and arranged to form single-time twin states in the twin state sequence, which are used as the input and output spaces of the reversible evolution operator. Based on the node connection relationships, element types, and material properties represented by the structural topology coding matrix, and combined with the structural dynamics equations, a basic evolutionary mapping from the twin state at the previous time step to the twin state at the next time step is constructed using the time discretization method. After the basic evolutionary mapping, an invertible parameter mapping structure with explicit inverse mapping is set to transform the output of the basic evolutionary mapping so that the entire time evolution process maintains a one-to-one correspondence in the twin state space, thereby forming an invertible evolutionary operator.

4. The intelligent monitoring system for the health of civil engineering structures based on digital twins according to claim 1, characterized in that: The abnormality-driven active perception configuration reconstruction module calculates the abnormality impact weight of each structural component based on the propagation analysis of the abnormality injection residual field, and forms an abnormality impact weight matrix. Based on the abnormality impact weight matrix, it generates a sensor-related sensing gain field to adjust the sensor's sampling frequency, range, sensitivity, and filtering bandwidth.

5. The intelligent monitoring system for the health of civil engineering structures based on digital twins according to claim 1, characterized in that: The connection matrix is ​​used to describe the connection relationships between structural nodes, and the attribute matrix is ​​used to describe the physical attributes of each structural unit.

6. The intelligent monitoring system for the health of civil engineering structures based on digital twins according to claim 1, characterized in that: The abnormality-driven active sensing configuration reconstruction module further analyzes the propagation path of the abnormal influence in the structure based on the twin state sequence and the abnormal injection residual field, determines the degree of abnormality in key parts, and generates a sensing gain field based on the analysis results to optimize the deployment and configuration of monitoring sensors.

7. The intelligent monitoring system for the health of civil engineering structures based on digital twins according to claim 1, characterized in that: The virtual observation field is derived based on twin state sequences.

8. The intelligent monitoring system for the health of civil engineering structures based on digital twins according to claim 1, characterized in that: The feedback mechanism includes receiving execution feedback data of monitoring, operation and maintenance strategies during system execution, and adjusting the parameters of the front-end modules in real time based on the feedback data.

9. The intelligent monitoring system for the health of civil engineering structures based on digital twins according to claim 1, characterized in that: The monitoring strategy is used to optimize sensor deployment, sampling frequency, and monitoring range; The operational strategy is used to generate load and speed limit control measures based on the structural health assessment results. The maintenance strategy is used for regular inspections, local reinforcement, and re-inspections; The digital twin side strategy is used to update and optimize the parameters of the digital twin model, ensuring the accuracy of the digital twin model parameters in describing the structural health status during long-term operation.

10. A method for intelligent monitoring of the health of civil engineering structures based on digital twins, implemented based on the intelligent monitoring system for the health of civil engineering structures based on digital twins as described in any one of claims 1-9, characterized in that: Includes the following steps: S1: Collect various physical quantity data of civil engineering structures through multi-field fusion sensing and topology coding modules, and generate multi-field spatiotemporal data tensors and structural topology coding matrices; S2: A digital twin state sequence is generated through a reversible digital twin construction and anomaly injection module, and an anomaly injection residual field is generated based on the difference between the observed data and the inference results; S3: Generates the sensing gain field and active excitation scheme through the abnormal-driven active sensing configuration reconstruction module, and adjusts the sensor sampling configuration; S4: By integrating real and virtual observation data through the virtual-real collaborative observation backfill and twin state correction module, the twin state is corrected to generate a credibility field; S5: The dynamic evolution boundary and risk trajectory generation module generates evolution boundary functions and risk trajectories to reflect the evolution trend of structural risks. S6: Generates and executes monitoring strategies, operation strategies, maintenance strategies, and digital twin-side strategies through a closed-loop module of strategy generation, execution, and re-verification, and performs parameter feedback and re-verification on the front-end modules based on the strategy execution results.

Images