Self-adaptive pre-stressed plate girder mechanical state forecasting method based on digital twinning

Abstract

The invention discloses a self-adaptive pre-stressed plate girder mechanical state forecasting method based on digital twinning, and the method comprises the steps: obtaining state parameters of a current plate girder through a pre-erected multi-source sensor network; noise reduction, abnormal value elimination, data completion and standardization processing are sequentially carried out on the state parameters, the processed state parameters of the current plate girder are sent into a digital twinborn model, and actually measured prestress is obtained through calculation; acquiring a simulated prestress in the digital twin model, and calculating an error between the simulated prestress and the actually measured prestress; updating the digital twin model according to the error; and based on the updated digital twin model, predicting the structural state of the plate girder by using a prediction algorithm. According to the method, the digital twin model of the plate girder is continuously updated according to the actually detected data, the prestress is output in real time through the digital twin model, and the structural state of the plate girder is predicted through a prediction algorithm.

Description

Technical Field

[0001] This invention relates to the field of information technology in civil engineering, and in particular to an adaptive method for predicting the mechanical state of prestressed slab beams based on digital twins. Background Technology

[0002] Prestressed slab beam structures are widely used in engineering fields such as highway bridges, railway bridges, and crane beams in industrial plants. During their long-term service, they must withstand the coupled effects of multiple factors such as material deterioration, load action, environmental erosion, and construction defects, which lead to problems such as concrete shrinkage and creep, corrosion of prestressing tendons, and reduction in structural bearing capacity, seriously affecting service safety and lifespan.

[0003] Existing technologies rely heavily on static models and periodic manual inspections, which have significant drawbacks: static models are difficult to dynamically match the actual state of the structure and are out of touch with the service process; the inspection methods are outdated and cannot capture changes in structural mechanics in real time; performance forecasts lack multi-source data fusion and adaptive adjustment mechanisms, resulting in insufficient accuracy; and maintenance decisions rely on experience-based judgments, which are highly unreliable and make it difficult to achieve precise operation and maintenance throughout the entire life cycle.

[0004] The aforementioned problems lead to untimely safety warnings and inaccurate assessments of remaining life of prestressed slab beam structures, which not only increase safety hazards but also waste operation and maintenance costs. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an adaptive prestressed slab beam mechanical state prediction method based on digital twin, which addresses the shortcomings of the prior art. The method continuously updates the digital twin model of the slab beam based on the actual detected data, outputs the prestress in real time through the digital twin model, and predicts the structural state of the slab beam through a prediction algorithm.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: an adaptive prestressed slab beam mechanical state prediction method based on digital twins, comprising: Step 1: Obtain the current state parameters of the slab beam through a pre-established multi-source sensor network; Step 2: Input the current state parameters of the slab beam into the digital twin model to calculate the measured prestress; Step 3: Obtain the simulated prestress in the digital twin model and calculate the error between the simulated prestress and the measured prestress; Step 4: Update the digital twin model based on the error; Step 5: Based on the updated digital twin model, use a prediction algorithm to predict the structural state of the slab beam.

[0007] Furthermore, between step 1 and step 2, the following steps are performed: The state parameters described in step 1 are sequentially subjected to noise reduction, outlier removal, data completion, and standardization, and the processed state parameters are then fed back into step 1.

[0008] Furthermore, in step 4, updating the digital twin model based on the error includes: The error is expressed as a percentage. Determine whether the error is less than or equal to a set percentage; When the error is less than or equal to the set percentage, maintain the current digital twin model; When the error exceeds a set percentage, update the current digital twin model.

[0009] Furthermore, updating the current digital twin model includes: Start the optimization algorithm; The algorithm iteratively searches within the parameter space of the current digital twin model to adjust the parameter combination of the current digital twin model. Update the parameters of the current digital twin model based on the adjusted parameter combination; Based on the parameters of the current digital twin model, the digital twin model is re-output to obtain the updated digital twin model.

[0010] Furthermore, the optimization algorithm is a particle swarm optimization algorithm.

[0011] Furthermore, the iterative search within the parameter space of the current digital twin model using an optimization algorithm includes: Identify key parameters within the parameter space of the current digital twin model; The algorithm iteratively searches for key parameters within the parameter space.

[0012] Furthermore, the identification of key parameters within the parameter space of the current digital twin model includes: Collect several input parameters that may affect the mechanical state of the structure, and select a set number of output response indicators from the input parameters to determine the reasonable value range of each input parameter; The Morris screening method was used to initially screen out the top 30% of candidate key parameters with the greatest impact. Then, the Sobol method was used to accurately analyze and calculate the first-order sensitivity index and the total sensitivity index of each parameter. Parameters with a first-order sensitivity index ≥ 10% were identified as key parameters.

[0013] Furthermore, the set percentage is 3%.

[0014] Compared with the prior art, the present invention has the following advantages: This invention provides an adaptive prestressed slab beam mechanical state prediction method based on digital twins, effectively overcoming a series of defects in the prior art caused by reliance on static models and periodic manual inspections, such as dynamic mismatch, monitoring lag, insufficient prediction accuracy, and blind maintenance decisions. Its core effect lies in constructing an intelligent operation and maintenance system driven by real-time data, dynamically evolving the model, and self-learning and optimizing the prediction through the closed-loop linkage of the multi-source sensor network and the digital twin model. Specifically, this method can accurately calculate and output the "measured prestress" inversely within the digital twin model based on real-time collected and preprocessed multi-source state parameters, achieving non-destructive and continuous perception of the key mechanical states inside the slab beam, completely changing the intermittent and delayed mode of traditional manual inspections. Furthermore, by continuously comparing the errors between measured prestress and model-simulated prestress, and introducing a threshold-based trigger-based model update mechanism, the digital twin model can adaptively adjust its physical parameters and boundary conditions to follow time-varying processes such as material performance degradation and prestress loss. This ensures high-fidelity synchronous mapping between the model and the actual structure throughout the entire service life, solving the fundamental problem of the disconnect between the static model and reality. Based on this dynamic, high-fidelity digital twin model, combined with prediction algorithms (whose model parameters can be efficiently optimized for key sensitive parameters using optimization algorithms such as particle swarm optimization as described in claims 4-7), it is possible to provide highly reliable forward-looking predictions of the future stress evolution, damage development, and bearing capacity decay of the slab beam. This achieves a shift from "post-event processing" to "pre-event warning," providing a scientific basis for state-based precision maintenance (such as determining the optimal tensioning timing and assessing remaining life), thereby significantly improving the long-term service safety of the structure and achieving overall optimization of the entire life-cycle operation and maintenance costs.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall structure of an adaptive prestressed slab beam mechanical state prediction method based on digital twins provided by the present invention.

[0017] Figure 2 This is a flowchart illustrating the implementation of the present invention. Detailed Implementation

[0018] See Figure 1-2 This invention provides an adaptive method for predicting the mechanical state of prestressed slab beams based on digital twins, such as... Figure 1 As shown, it includes: Step 1: Obtain the current state parameters of the slab beam through a pre-established multi-source sensor network; Step 2: Input the current state parameters of the slab beam into the digital twin model to calculate the measured prestress; Step 3: Obtain the simulated prestress in the digital twin model and calculate the error between the simulated prestress and the measured prestress; Step 4: Update the digital twin model based on the error; Step 5: Based on the updated digital twin model, use a prediction algorithm to predict the structural state of the slab beam.

[0019] The method of this invention aims to construct a closed-loop intelligent system of "perception-mapping-comparison-update-prediction" to overcome the defects of static model disconnection, monitoring lag, and inaccurate prediction in the background technology.

[0020] In step 1, the current state parameters of the slab beam are obtained through a pre-established multi-source sensor network.

[0021] This step forms the foundation for establishing a real-time data link between the digital twin and the real-world entity. The multi-source sensor network is optimized based on the mechanical properties and potential damage modes of the prestressed slab beam. Typically, this network may include, but is not limited to: fiber optic grating sensors or resistance strain gauges for monitoring concrete strain; accelerometers for monitoring vibration characteristics; temperature and humidity sensors for monitoring ambient temperature and humidity; crack gauges for monitoring crack development; and inclinometers or displacement sensors for monitoring macroscopic deformation. These sensors are networked wired or wirelessly, acquiring data synchronously or asynchronously at a set sampling frequency (e.g., once per minute or triggered by load change events). The acquired "state parameters" are a multi-dimensional data vector encompassing direct and indirect indicators reflecting the real-time mechanical behavior of the slab beam, such as strain distribution at mid-span, acceleration time histories at key locations, and ambient temperature values. This step upgrades traditional discrete, periodic manual inspection to continuous, automated, all-time monitoring, providing a data source for subsequent real-time analysis and model updates.

[0022] In step 2, the current state parameters of the slab beam are fed into the digital twin model to calculate the measured prestress.

[0023] The "digital twin model" is a high-fidelity dynamic mapping of the real slab beam in virtual space. Its initial model is constructed based on design drawings, material constitutive relations, boundary conditions, etc., and is usually established using the finite element method. The core innovation of this step lies in "calculating the measured prestress". Traditional methods are difficult to directly and non-destructively measure the effective prestress of the prestressing tendons. This invention uses the digital twin model as a "reverse solver": taking pre-processed multi-source state parameters (such as strain and deflection) reflecting the overall structural response as input, and based on the physical relations of the model (such as mechanical equilibrium equations and constitutive relations), the prestress value that best matches the model response with the actual monitoring data is solved through inversion algorithms (such as model correction methods and neural network inversion). This value is defined as the "measured prestress". This achieves non-destructive, continuous, and real-time perception of the internal key mechanical state (prestress), solving the problem of "inability to capture structural mechanical changes in real time" in the background technology.

[0024] Between step 1 and step 2, the state parameters in step 1 are sequentially subjected to noise reduction, outlier removal, data completion, and standardization, and the processed state parameters are then sent back to step 1.

[0025] The raw state parameters obtained in step 1, when directly used for calculation, may introduce significant errors due to noise and outliers. Therefore, in a preferred embodiment, a data preprocessing step is added between steps 1 and 2: the state parameters obtained in step 1 are sequentially subjected to noise reduction (e.g., using wavelet transform or Kalman filtering), outlier removal (e.g., based on the 3σ criterion), data completion (e.g., using time series interpolation), and standardization. This series of preprocessing operations can significantly improve the quality and consistency of the raw data, laying the foundation for subsequent high-precision calculations.

[0026] In step 3, the simulated prestress in the digital twin model is obtained, and the error between the simulated prestress and the measured prestress is calculated.

[0027] In a digital twin model, based on its current parameter settings (such as material elastic modulus, initial prestress value, boundary constraint stiffness, etc.), forward mechanical calculations can yield the theoretical prestress distribution under the current load and environmental conditions; this is the "simulated prestress." The difference between the simulated prestress and the "measured prestress" obtained from step 2 at the same location and time is calculated and quantified as error. This error essentially reflects the accuracy of the current digital twin model in representing the true mechanical state of the physical structure. The existence of this error indicates that, due to time-varying effects such as material performance degradation, prestress loss, and damage accumulation, the initial or previous version of the model has gradually become "out of sync" with the physical structure.

[0028] In step 4, the digital twin model is updated based on the error.

[0029] This step is the key mechanism for achieving the "adaptive" and "dynamic evolution" functions of this method, ensuring the synchronization of the digital twin model with the physical structure throughout its entire lifecycle. This is achieved through the following methods: The error is expressed as a percentage. Determine whether the error is less than or equal to a set percentage; When the error is less than or equal to the set percentage, maintain the current digital twin model; When the error exceeds a set percentage, update the current digital twin model.

[0030] First, the error calculated in step 3 is expressed as a percentage, for example, (|simulated value - measured value| / measured value) × 100%. Second, it is determined whether this percentage is less than or equal to a set threshold (3%). This threshold is a parameter that has undergone engineering trade-offs; too small a threshold leads to frequent and unnecessary model updates, resulting in a high computational load; too large a threshold causes model updates to lag and lose fidelity. A 3% threshold strikes a good balance between forecast accuracy and computational efficiency. Finally, the judgment logic is executed: if the error is ≤3%, the current model is considered to still have sufficient fidelity and is maintained, awaiting the next round of data input; if the error is >3%, the model update process is triggered.

[0031] The updating of the current digital twin model includes: Start the optimization algorithm; The algorithm iteratively searches within the parameter space of the current digital twin model to adjust the parameter combination of the current digital twin model. Update the parameters of the current digital twin model based on the adjusted parameter combination; Based on the parameters of the current digital twin model, the digital twin model is re-output to obtain the updated digital twin model.

[0032] Specifically, 4a: Activate the optimization algorithm. This algorithm automatically finds the optimal parameter combination that makes the model output (simulated prestress) more closely approximate the measured data. 4b: Iteratively search within the parameter space of the current digital twin model using the optimization algorithm, adjusting the model parameter combination. The parameter space contains all model input parameters that may affect the calculation results, such as the concrete elastic modulus, creep coefficient, effective area of ​​prestressed tendons, and anchorage loss coefficient. 4c: Update the corresponding parameter values ​​in the digital twin model based on the optimal parameter combination found by the optimization algorithm. 4d: Rerun or compile the digital twin model using the updated parameters to obtain a new model version with a higher degree of consistency with the current measured data, i.e., the "updated digital twin model".

[0033] The optimization algorithm is the Particle Swarm Optimization (PSO) algorithm. The PSO algorithm simulates the foraging behavior of bird flocks, guiding the search through historical optimal information of individuals and the group. It features strong global search capabilities, fast convergence speed, and suitability for multi-parameter optimization problems, and can efficiently complete parameter correction for complex digital twin models.

[0034] Furthermore, the iterative search within the parameter space of the current digital twin model using an optimization algorithm includes: Identify key parameters within the parameter space of the current digital twin model; The algorithm iteratively searches for key parameters within the parameter space.

[0035] To improve optimization efficiency and accuracy, and avoid wasting computational resources on irrelevant or minor parameters, the search process can be further optimized (corresponding to claim 6): First, identify the key parameters in the parameter space of the current digital twin model, i.e., those parameters that have the most significant impact on the output response (such as prestress and deflection). Then, iteratively search within this reduced key parameter space using an optimization algorithm (such as particle swarm optimization). This significantly reduces the dimensionality of the optimization problem and accelerates the convergence speed.

[0036] In addition, the identification of key parameters within the parameter space of the current digital twin model includes: Collect several input parameters that may affect the mechanical state of the structure, and select a set number of output response indicators from the input parameters to determine the reasonable value range of each input parameter; The Morris screening method was used to initially screen out the top 30% of candidate key parameters with the greatest impact. Then, the Sobol method was used to accurately analyze and calculate the first-order sensitivity index and the total sensitivity index of each parameter. Parameters with a first-order sensitivity index ≥ 10% were identified as key parameters.

[0037] One method for identifying key parameters is a two-stage global sensitivity analysis, specifically: Phase 1 (Coarse Screening): Collect all input parameters that may affect the structural mechanical state (e.g., 10-20) and determine their reasonable value ranges. Select core output response indicators (e.g., mid-span prestress, maximum deflection). Using the Morris screening method, through a limited number of random trajectory calculations, quickly evaluate the average influence of each parameter on the output, and screen out the parameters with the highest influence to form a "candidate key parameter set".

[0038] The second stage (refined analysis) involves using the Sobol method to perform precise variance decomposition on the candidate key parameter set. Through extensive sampling (such as Monte Carlo sampling), the first-order sensitivity index (measuring its individual effect) and the overall sensitivity index (measuring its individual effect and interaction with other parameters) are calculated for each parameter. Parameters with a first-order sensitivity index ≥ 10% are ultimately identified as "key parameters." This combined approach ensures both comprehensive key parameter identification and computational efficiency.

[0039] In this step, based on the updated digital twin model, a prediction algorithm is used to predict the structural state of the slab beam.

[0040] Once an updated digital twin model highly synchronized with the current physical structure is obtained, it can be used for forward-looking predictions. The "prediction algorithm" can be based on extrapolation from the physical model or combined with data-driven models (such as ARIMA time series analysis or LSTM long short-term memory networks). Specifically, the following steps are taken: inputting possible future environmental load sequences (such as predicted temperature changes and traffic load spectra) or preset extreme conditions (such as design live loads) into the updated digital twin model, running forward mechanical analysis or combining it with the prediction algorithm for extrapolation, thereby obtaining a prediction of the structural state of the slab beam over a future period. These states can include: the stress time history evolution of prestressed tendons, stress distribution at key concrete sections, deflection development, fatigue damage accumulation, and even the decay curve of remaining bearing capacity. This fundamentally changes the passive situation in the background technology where "performance prediction lacks multi-source data fusion and adaptive adjustment mechanisms, resulting in insufficient accuracy" and "maintenance decisions rely on experience-based judgment," realizing a shift from "post-event processing" to "pre-event warning," providing accurate and scientific decision-making basis for determining the optimal maintenance timing and assessing remaining lifespan.

[0041] In specific implementation, such as Figure 2 As shown, it includes the following four levels.

[0042] Employing a four-layer closed-loop architecture, each layer works collaboratively to achieve intelligent prediction throughout the entire process of "perception-modeling-update-forecasting": (1) Physical entity layer: The core is the physical structure of prestressed slab beams and the multi-source sensor network deployed on them.

[0043] (2) Data interaction layer: responsible for the acquisition, transmission and preprocessing of sensor data, and realizes real-time data upload through 5G wireless communication technology.

[0044] (3) Digital twin model layer: This is the core layer of the system, including the parameterized initial model, key parameter identification module and adaptive correction engine.

[0045] (4) Application service layer: Provides a visual management interface and multi-functional application services, including health status classification early warning, short-term and long-term mechanical status forecast.

[0046] The following sections will introduce these four aspects in detail: Physical entity layer: The core is the physical structure of the prestressed slab beam and the multi-source sensor network deployed on it.

[0047] The physical entity layer consists of the prestressed slab beam physical structure and a multi-source sensor network. Its core function is to achieve comprehensive, real-time, and accurate perception of the structural mechanical state and environmental conditions.

[0048] (1) Physical structure of prestressed slab beam This invention is applicable to various prestressed slab beam structures, including prestressed concrete T-beams, I-beams, box girders, slab beams, composite beams, and steel-concrete composite prestressed slab beams, with spans ranging from 5 to 50 meters. It covers scenarios such as highway bridges, railway bridges, urban rail transit bridges, prestressed crane beams for industrial plants, prestressed floor slabs for large-span public buildings, and prestressed concrete pipe piles.

[0049] (2) Multi-source sensor network design Multi-source sensor networks develop customized sensor deployment schemes based on factors such as structural type, span size, stress characteristics, and service environment.

[0050] ① Sensor type selection and technical parameters Fiber Bragg grating strain gauge: It adopts distributed fiber Bragg grating sensing technology, with a measurement range of -1500-3000με, resolution ≤1με, accuracy ≤±1με, operating temperature range of -40℃-85℃, protection level IP67, and is suitable for long-term monitoring of strain distribution in key parts of structures. Piezoelectric stress sensor: Based on the piezoelectric effect principle, the measurement range is 0-50MPa, the resolution is ≤0.01MPa, the accuracy is ≤±0.5MPa, and the response frequency is ≥1kHz. It is suitable for monitoring dynamic stress changes in structures. Laser displacement gauge: It adopts the principle of laser triangulation, with a measurement range of 0-500mm, resolution ≤0.001mm, accuracy ≤±0.01mm, and sampling frequency ≥10Hz. It is suitable for monitoring beam deflection, settlement and horizontal displacement. High-precision temperature and humidity sensor: Temperature measurement range -40℃-85℃, accuracy ≤±0.1℃; Humidity measurement range 0-100%RH, accuracy ≤±2%RH; Sampling frequency ≥1Hz; used to monitor changes in ambient temperature and humidity. Chloride ion concentration sensor: Measurement range 0-10000ppm, resolution ≤1ppm, accuracy ≤±5%FS, suitable for monitoring chloride ion concentration in coastal areas or chloride-polluted environments, and predicting the risk of concrete carbonation and steel corrosion; Dynamic weighing equipment: It adopts piezoelectric or bent plate weighing sensors, with a weighing range of 0-100t, accuracy ≤±2%, response time ≤10ms, and can monitor the weight and axle load distribution of passing vehicles in real time. Beidou positioning module: positioning accuracy ≤1cm (static), ≤5cm (dynamic), data update frequency 1-5Hz, used to monitor the overall displacement and deformation of the beam.

[0051] ② Collaboration of different types of sensors: The distributed monitoring of fiber optic strain gauges is combined with the dynamic response monitoring of piezoelectric stress sensors, and the local deflection measurement of laser displacement gauges is mutually verified with the overall displacement monitoring of Beidou positioning modules.

[0052] ③ Sensor deployment plan (focusing on covering stress concentration areas, deformation-sensitive areas, and areas prone to damage): Mid-span region: Deploy 2-4 fiber optic strain gauges (uniformly distributed along the cross-sectional height), 1 piezoelectric stress sensor, and 1 laser displacement gauge to monitor mid-span strain, stress, and deflection; In the 1 / 4 span and 3 / 4 span areas: two fiber optic strain gauges and one laser displacement gauge are installed in each area to monitor the strain and deflection at the quarter points of the beam. Support area: Each area is equipped with 2 fiber optic strain gauges, 1 pressure sensor, and 1 laser displacement gauge to monitor support reaction force, strain, and settlement; Prestressed anchorage ends: Each anchorage end is equipped with 3-4 fiber optic strain gauges, 1 temperature and humidity sensor, and 1 chloride ion concentration sensor to monitor stress distribution, environmental parameters, and corrosion risk at the anchorage end. For critical joint areas (such as the connection points of multi-span beams): install 2 fiber optic strain gauges and 1 laser displacement gauge to monitor the strain and relative displacement at the joints; Sensor deployment principles: For slab beams with a span of ≤10m: no less than 10 sensors shall be installed in each span, with key locations such as mid-span, supports, and anchorage ends accounting for ≥60%; For slab beams with spans of 10-20m: the number of sensors installed in each span shall not be less than 15, with key parts accounting for ≥70%; For slab beams with spans of 20-50m: each span should have no fewer than 20 sensors, with key components accounting for ≥80%; In special environments (coastal corrosion, high-altitude freeze-thaw cycles) or heavy-load traffic scenarios: increase the sensor deployment density by 20% on the basis of the above.

[0053] ④ Sensor installation and protection installation process The strain sensors are adhesive-mounted. Before installation, the beam surface is ground, cleaned, and dried to ensure a surface flatness error of ≤0.1mm. Special structural adhesive is used for bonding, with a uniform thickness (0.2-0.5mm). After bonding, the surface is allowed to cure for ≥24 hours at a controlled temperature of 5-35℃. The displacement sensors are bracket-mounted, with expansion bolts or chemical anchors used for fixation. Vibration damping pads are added to the mounting brackets to reduce the impact of vehicle vibration on measurement accuracy, ensuring a secure installation. The sensor measurement direction is aligned with the displacement direction. The environmental sensors are externally mounted, equipped with sunshades and rainproof covers to avoid direct exposure to sunlight or rain, raising the protection level to IP68. Power supply: A "solar + lithium battery" backup power system is used. Wired transmission sensors are equipped with a stable mains power supply, while wireless sensors are equipped with low-power lithium batteries, ensuring continuous operation of the sensor network for at least 72 hours in extreme weather or power outage conditions to avoid data interruption.

[0054] ⑤ Sensor calibration and debugging Static calibration: The strain sensor is calibrated using a standard strain gauge and a tensile testing machine to ensure that the measurement accuracy meets the requirements; The stress sensor is calibrated using a standard pressure source; The displacement sensor is calibrated using a standard displacement stage; Dynamic calibration: A vibration table is used to calibrate the dynamic response of the dynamic sensor to ensure the measurement accuracy of the sensor under dynamic loads; System debugging: Start the data acquisition system and run it continuously for 72 hours to check the stability, integrity and accuracy of the sensor data acquisition.

[0055] Data interaction layer: responsible for the acquisition, transmission and preprocessing of sensor data, and realizes real-time data upload through 5G wireless communication technology.

[0056] The data interaction layer is a key link connecting the physical entity layer, the digital twin model layer, and the application service layer. It is responsible for data collection, preprocessing, transmission, storage, and distribution, ensuring the real-time performance, reliability, and availability of the data.

[0057] (1) Data acquisition module The data acquisition module adopts a distributed acquisition architecture, consisting of a data acquisition terminal, an acquisition controller, and a synchronization clock.

[0058] Data acquisition terminal: Each data acquisition terminal can connect to 8-16 sensors, supporting multiple signal types such as analog signals, digital signals, and fiber optic signals. The sampling frequency can be dynamically adjusted according to the type of monitoring parameter. The sampling frequency for strain and stress parameters is 100-500Hz, the sampling frequency for deflection, temperature, and environmental parameters is 1-10Hz, and the positioning data update frequency is 1-5Hz. Data Acquisition Controller: It adopts an industrial-grade microcontroller as the core controller, which has data caching, logic control and communication functions. It can realize centralized control of multiple data acquisition terminals and supports remote configuration and adjustment of acquisition parameters. Synchronization clock: The system adopts a combination of BeiDou time synchronization and local clock synchronization to ensure that the time synchronization accuracy of all data acquisition terminals is ≤1ms, thus guaranteeing the time consistency of data from different sensors. (2) Data preprocessing module The processing workflow is "noise reduction - outlier removal - data completion - standardization". Noise reduction processing: For random noise in sensor-acquired data, the Kalman filter algorithm is suitable for time-varying data such as strain, stress, and vibration acceleration under dynamic loads, while the wavelet threshold noise reduction algorithm is suitable for static or slowly changing data such as temperature, humidity, and chloride ion concentration, thus improving the targeting of noise reduction. Outlier removal: The 3σ criterion is used to identify outlier data. When a data value exceeds the mean by ± 3 times the standard deviation, it is judged as an outlier and removed. At the same time, a trend consistency check is added. For data that does not exceed the 3σ range but deviates significantly from the trend of adjacent data (the rate of change exceeds twice the normal range), it is marked as suspicious data and retained. After verification with multiple sets of subsequent data, a decision is made on whether to remove it to avoid accidentally deleting valid data. For continuous abnormal data caused by sensor failure, mark it as fault data and trigger an alarm; Data completion: For a small amount of missing data (missing rate ≤5%) caused by communication interruption or momentary sensor failure, use linear interpolation to complete it; For data with a missing rate between 5% and 20% and consecutive missing data length ≤ 10 sampling points, an LSTM neural network model is used for data completion, with a data completion error ≤ 5%. For data with a missing rate exceeding 20% ​​or with consecutive missing data lengths > 10 sampling points, no completion will be performed; the data will be marked as missing and an alarm will be triggered. Standardization processing: Monitoring data of different types and dimensions are converted into dimensionless data of a unified standard using the z-score standardization formula. (in The original data, The mean of the dataset. The standardization process (which is the standard deviation of the dataset) is performed after data preprocessing and before model input, to facilitate subsequent data fusion and model computation.

[0059] (3) Communication transmission module The communication transmission module adopts a "5G+LoRa" hybrid communication architecture. 5G communication is used to transmit high-frequency, high-capacity dynamic monitoring data (such as strain, stress, and vibration data), with a transmission delay of ≤100ms; LoRa communication is used to transmit low-frequency, small-capacity environmental data (such as temperature, humidity, and chloride ion concentration), reducing power consumption and communication latency ≤1s; When the 5G signal is interrupted, it automatically switches to LoRa communication to transmit critical data, ensuring communication continuity; (4) Data storage module The data storage module adopts a hybrid storage architecture of time-series database and relational database to meet the storage needs of different types of data.

[0060] Database selection and division of labor: The time series database uses InfluxDB to store real-time monitoring data and historical data of model evolution, with a data retention period of no less than 50 years; the relational database uses MySQL to store structured data such as basic information, sensor parameters, maintenance records, and health assessment reports. Data backup: A dual backup mechanism of "local backup + cloud backup" is adopted. Local backup is performed daily and cloud backup is performed weekly. The backup data retention period is no less than 10 years.

[0061] Digital twin model layer: This is the core layer of the system, including a parameterized initial model, a key parameter identification module, and an adaptive correction engine.

[0062] (1) The composition of digital twin models The digital twin model of this invention is a multi-dimensional, multi-scale, dynamically evolving virtual model with the following characteristics: Multi-dimensional mapping: Achieving comprehensive and accurate mapping of physical entities in terms of geometric, material, mechanical, load, and environmental dimensions; Multi-scale modeling: A multi-scale modeling method combining macro- and micro-scale models is adopted. The macro-scale model is constructed using rod and shell elements to calculate the overall stress, deflection, and vibration characteristics of the structure. The micro-model targets key areas such as mid-span, anchorage ends, and joints, using solid elements to simulate the micro-distribution of concrete aggregates, steel bars, and prestressed tendons. It accurately captures local stress concentration, crack initiation and propagation processes, balancing model accuracy and computational efficiency. Dynamic consistency: Through real-time data interaction and adaptive correction, the dynamic consistency error between the model and the physical entity is kept within 3%. Self-learning evolution: Based on monitoring data and historical data, continuously optimize model parameters and algorithms to improve model accuracy.

[0063] (2) Construction of the parameterized initial model ① Geometric parameter modeling: Establish a geometric model based on BIM design drawings, construction drawings, etc., with geometric dimension error ≤0.1mm and modeling accuracy of key parts ≤0.05mm, to ensure geometric consistency between the model and the physical entity; ② Material parameter modeling: This includes concrete materials, prestressed tendon materials, ordinary steel reinforcement materials, and other materials. Material parameters should prioritize test data from the construction phase. If test data is unavailable, recommended values ​​from specifications should be used. Parameter correction interfaces should be reserved to facilitate subsequent adjustments based on monitoring data. ③ Prestress parameter modeling: Prestress parameter modeling adopts the initial strain method or equivalent load method to simulate the prestress application process, ensuring the accuracy of prestress distribution; ④ Boundary constraint parameter modeling: including fixed hinge support, sliding hinge support, elastic support, and other constraints; ⑤ Load parameter modeling: The initial model considers conventional loads and environmental loads, and a load library is preset (including conventional vehicle loads, crowd loads, wind loads, temperature loads, etc.). Subsequently, traffic and meteorological dynamic data are obtained in real time through API interface to realize dynamic updates of load parameters, laying the foundation for subsequent dynamic load updates.

[0064] (3) Key parameter identification module Specifically, a combination of the Morris screening method and the Sobol method, which balances analytical efficiency and accuracy, is adopted: First, collect a set of 20-50 input parameters that may affect the mechanical state of the structure, such as geometry, materials, and prestressing. Then, select 3-5 core output response indicators, such as mid-span deflection and maximum principal stress, and determine the reasonable range of ±30% of the design value of each input parameter. The Morris screening method was used to initially screen out the candidate key parameters with the largest impact. Then, the Sobol method was used to accurately analyze and calculate the first-order sensitivity index and the total sensitivity index of each parameter. Parameters with a first-order sensitivity index ≥ 10% were identified as key parameters and a database was established. Finally, the reliability of the analysis is verified by selecting 1-2 key parameters and adjusting them multiple times within their range, and comparing the finite element analysis results with the experimental results. If the error exceeds 5%, the range of parameters or response index of the sensitivity analysis is adjusted, and the Morris screening method and Sobol method are executed again until the results are reliable. Key parameter dynamic update mechanism: The global sensitivity analysis is re-executed every 6 months, and the list of key parameters is adjusted in combination with the structural performance degradation (such as increased steel corrosion and deeper concrete carbonation) to ensure the targeted nature of model correction.

[0065] (4) Adaptive correction Correction objective: To minimize the normalized root mean square error (NRMSE), ensuring that NRMSE ≤ 3% and the change over three consecutive iterations ≤ 0.1%, while also guaranteeing that the peak stress error ≤ 5% and the extreme deflection error ≤ 3%; Optimization algorithm: The core algorithm is particle swarm optimization. The population size and number of iterations are determined based on the parameter dimension (when the parameter dimension is ≤6, the population size is 50 and the number of iterations is 200; when the parameter dimension is >6, the population size is 80-100 and the number of iterations is 300). If the target is not met, the genetic algorithm is switched. If the target is still not met, Bayesian update is started. Triggering mechanism: The normal correction cycle is 1-7 days (shortened to 1 day when the data variation coefficient is >15%), and immediate correction is triggered by extreme loads, maintenance construction and other events; A new feature, "Model Prediction Error Exceeds Standard Trigger Correction," has been added. When the extreme value error of stress and deflection in the short-term mechanical state prediction exceeds 5%, the correction process will be automatically initiated.

[0066] Calibration process: Input real-time monitoring data and current key parameters, iteratively optimize until the error requirements are met, and update the model parameters; Verification of calibration results: After calibration, the model accuracy is verified using independent monitoring data (last 24 hours) that were not involved in the calibration, ensuring that NRMSE ≤ 3% and single index error ≤ 5%; otherwise, calibration is restarted.

[0067] (5) Model Evolution Module Geometric morphology evolution: Based on laser displacement gauge and Beidou positioning data, the model node coordinates are adjusted every 7 days to simulate the cumulative deformation of the structure; when cracks are detected, crack elements are added to the model, and the relationship between crack propagation rate and stress field is set to simulate crack development. Material property evolution: By combining time-varying models and monitoring data, parameters such as concrete strength and effective stress of prestressed tendons are updated to reflect material degradation. An exponential decay model is used to measure concrete strength degradation. (in Let be the concrete strength at time t. (where k is the initial strength and k is the attenuation coefficient, dynamically adjusted based on monitoring data of temperature, humidity, and chloride ion concentration); the effective stress loss of the prestressed tendon is calculated using both relaxation and corrosion models to ensure that the evolution of material properties is consistent with reality. Evolution of mechanical properties: The model's stiffness, bearing capacity, dynamic characteristics, etc. are updated synchronously to ensure consistency with the mechanical state of the physical entity.

[0068] Application Service Layer: Provides a visual management interface and multi-functional application services, including health status classification and early warning, short-term and long-term mechanical status forecasts.

[0069] (1) Mechanical state prediction module Short-term extreme load response forecast (1-72 hours) Input data: Future weather data, traffic load data and temporary load information are obtained through API interface, combined with real-time correction parameters of digital twin model; Load transformation: meteorological data is transformed into temperature load, wind load, and additional bridge deck load; traffic load is transformed into moving load series or equivalent uniformly distributed load form, and temporary load is transformed into time-varying boundary conditions that can be identified by the model according to actual working conditions. Prediction algorithm: Based on the corrected digital twin model, an explicit integral algorithm is used for transient dynamic analysis, with the time step set to 1e-4-1e-3s to ensure the capture of the structure's dynamic response; Output results: Output stress cloud map, deflection curve, vibration acceleration time history and extreme value data for the next hour, with time accuracy ≤1min and stress and deflection prediction error ≤5%, providing data support for real-time safety early warning; Forecast result credibility assessment: Based on the model calibration accuracy and the completeness of the input data, the credibility level of the forecast result is output (high, medium, low). When the credibility is below 70%, the user is prompted to supplement data or strengthen monitoring.

[0070] Long-term performance degradation prediction (5-30 years) Input data: Integrate the calibration parameters of the digital twin model, current mechanical state and material property data, combined with long-term environmental forecast data and long-term traffic load trends, and supplement "historical maintenance records".

[0071] Core model: Coupled material time-varying model, including concrete shrinkage and creep, prestressed tendon corrosion evolution model, concrete carbonation model (CO2 diffusion equation) and material strength decay formula (considering the effects of temperature, humidity and chemical corrosion). Prediction process: The quasi-static nonlinear time history analysis method is adopted, with an annual time step, to iteratively update material properties and load conditions, and calculate the changes in structural mechanical parameters; Results output: Generates load-bearing capacity degradation curves, prestress loss curves, and deflection growth trends; clarifies the remaining structural life (the time point when the load-bearing capacity drops to 80% of the design value) and damage evolution trends; load-bearing capacity prediction error ≤10%; remaining life prediction error ≤15%. Critical warning value for performance degradation (maintenance reminder triggered when load-bearing capacity drops to 90% of design value); prediction result update frequency: long-term prediction results are updated quarterly based on the latest monitoring data.

[0072] (2) Health status assessment module Based on short-term forecast extreme values ​​and long-term prediction results, a multi-dimensional assessment system is established in conjunction with normative standards to achieve structural health status classification and accurate early warning.

[0073] The evaluation index system includes the following indicators: mechanical performance indicators (stress, deflection, prestress loss, etc.), damage state indicators (crack size, carbonization depth, steel corrosion area, etc.), durability indicators (chloride ion concentration, environmental corrosion level), and safety reserve indicators (bearing capacity reserve coefficient). Indicator weight determination: The weights of the four categories of indicators are determined by the Analytic Hierarchy Process (AHP), with mechanical performance indicators accounting for 40%, damage status indicators accounting for 30%, durability indicators accounting for 20%, and safety reserve indicators accounting for 10%. The weights can be adjusted according to the structural type and service environment.

[0074] The four-level health grading standard is as follows: Normal level: Monitoring / forecast data ≤ 80% of the standard threshold, good structural performance, no obvious damage; Attention level: Data is between 80% and 95% of the standard threshold, performance basically meets the standard; Warning level: Data is between 95% and 105% of the standard threshold, close to the limit; Alarm level: Data > 105% of the standard threshold, serious damage exists.

[0075] Assessment and early warning response process: Collect monitoring and forecast data → Standardize processing → Single indicator rating → Determine weights using analytic hierarchy process (AHP) → Determine comprehensive health level → Generate assessment report; Cross-validate manually detected data with model forecast data. If the assessment result error exceeds 10%, readjust the assessment indicators or weights. Response mechanism: Normal level: A health report is generated every six months, and routine maintenance is performed according to the plan; Attention level: Generate a health report once a month, increase sensor sampling frequency by 50%, and strengthen monitoring of key areas; Warning level: A health report will be generated weekly, a special test will be completed within 3 months, a maintenance plan will be developed and implemented as appropriate; Alarm levels: Real-time push of early warning information, emergency detection initiated within 24 hours, access restricted or use suspended within 72 hours, emergency maintenance initiated; automatic linkage with the emergency command system.

[0076] (3) Visual Management Module Core visualization function: 3D model visualization Data visualization: Displays real-time monitoring data and forecast curves in the form of line charts, bar charts, dashboards, etc., and provides data statistical analysis and report export functions; Early warning management: Centrally display early warning information, support filtering and tracking of processing progress, and push notifications through multiple channels such as SMS, APP, and email; Mobile adaptation: Supports access from mobile devices such as mobile apps and tablets, enabling functions such as real-time push notifications of warning information, convenient query of 3D models, and maintenance progress tracking; Historical status backtracking: Supports querying the structural mechanical state, health level, and model parameters at any point in time (such as 1 year ago or 3 years ago), which is convenient for analyzing the trend of structural performance degradation. Data export and sharing: Supports exporting monitoring data, forecast results, and health reports to Excel and PDF formats, and sharing data with bridge maintenance platforms, traffic management systems, etc. through interfaces to achieve collaborative management.

[0077] The operation and management functions support multi-user permission allocation (administrators, monitoring personnel, maintenance personnel, etc.), can remotely configure system parameters, and have data query, backup, deletion and log management functions, which facilitates system maintenance and troubleshooting.

[0078] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any simple modifications, alterations, or equivalent structural changes made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for predicting the mechanical state of a prestressed slab beam based on digital twins, characterized in that, include: Step 1: Obtain the current state parameters of the slab beam through a pre-established multi-source sensor network; Step 2: Input the current state parameters of the slab beam into the digital twin model to calculate the measured prestress; Step 3: Obtain the simulated prestress in the digital twin model and calculate the error between the simulated prestress and the measured prestress; Step 4: Update the digital twin model based on the error; Step 5: Based on the updated digital twin model, use a prediction algorithm to predict the structural state of the slab beam.

2. The adaptive prestressed slab beam mechanical state prediction method based on digital twin as described in claim 1, characterized in that, Between step 1 and step 2, the following steps are performed: The state parameters described in step 1 are sequentially subjected to noise reduction, outlier removal, data completion, and standardization, and the processed state parameters are then fed back into step 1.

3. The adaptive prestressed slab beam mechanical state prediction method based on digital twin as described in claim 1, characterized in that, In step 4, updating the digital twin model based on the error includes: The error is expressed as a percentage. Determine whether the error is less than or equal to a set percentage; When the error is less than or equal to the set percentage, maintain the current digital twin model; When the error exceeds a set percentage, update the current digital twin model.

4. The adaptive prestressed slab beam mechanical state prediction method based on digital twin as described in claim 3, characterized in that, The update of the current digital twin model includes: Start the optimization algorithm; The algorithm iteratively searches within the parameter space of the current digital twin model to adjust the parameter combination of the current digital twin model. Update the parameters of the current digital twin model based on the adjusted parameter combination; Based on the parameters of the current digital twin model, the digital twin model is re-output to obtain the updated digital twin model.

5. The adaptive prestressed slab beam mechanical state prediction method based on digital twin as described in claim 4, characterized in that, The optimization algorithm is the particle swarm optimization algorithm.

6. The adaptive prestressed slab beam mechanical state prediction method based on digital twin as described in claim 4, characterized in that, The iterative search within the parameter space of the current digital twin model using an optimization algorithm includes: Identify key parameters within the parameter space of the current digital twin model; The algorithm iteratively searches for key parameters within the parameter space.

7. The adaptive prestressed slab beam mechanical state prediction method based on digital twin according to claim 3, characterized in that, The identification of key parameters within the parameter space of the current digital twin model includes: Collect several input parameters that may affect the mechanical state of the structure, and select a set number of output response indicators from the input parameters to determine the reasonable value range of each input parameter; The Morris screening method was used to initially screen out the top 30% of candidate key parameters with the greatest impact. Then, the Sobol method was used to accurately analyze and calculate the first-order sensitivity index and the total sensitivity index of each parameter. Parameters with a first-order sensitivity index ≥ 10% were identified as key parameters.

8. The adaptive prestressed slab beam mechanical state prediction method based on digital twin according to claim 3, characterized in that, The set percentage is 3%.

Images