Bridge bearing capacity monitoring method and system based on digital twinning

Abstract

The invention provides a bridge bearing capacity monitoring method and system based on digital twinning, and the method comprises the steps: constructing a digital twinning finite element model, and identifying the parameters of a time-varying material constitutive model; continuously updating the model to obtain a synchronous twin model; calculating a theoretical load effect on a vehicle prediction path, and identifying a temporary load condition data packet; outputting a virtual future load scene by using a multi-level probability model and an extreme value generator; injecting the temporary load condition data packet and the virtual future load scene into a synchronous twin model to obtain structural response statistical distribution; and calculating a safety evaluation index, and performing decision optimization by utilizing reinforcement learning to obtain an optimal management and maintenance intervention scheme. According to the method, the digital model and the physical entity can be synchronously evolved in real time, the probabilistic risk of the structure under the future random load can be predicted and evaluated, intelligent decision support is provided, and dynamic, online and predictive evaluation of the bridge bearing capacity is achieved.

Description

Technical Field

[0001] This invention relates to the field of bridge monitoring technology, and in particular to a bridge load-bearing capacity monitoring method and system based on digital twins. Background Technology

[0002] The load-bearing capacity of a bridge is a core indicator for assessing its structural safety. Traditional bridge load-bearing capacity assessments mainly rely on two methods: static calculations based on design drawings and standard formulas, and on-site verification through load tests. These methods have fundamental limitations: they are usually static and infrequent, and cannot reflect the real-time, dynamic evolution of the structure's load-bearing capacity under actual operating conditions (material aging, damage accumulation, and randomly changing traffic and environmental loads).

[0003] With the development of structural health monitoring technology, it has become possible to acquire structural responses in real time through sensors, providing a data foundation for dynamic assessment of load-bearing capacity. However, existing assessment methods based on monitoring data still face the following shortcomings: 1. The model is disconnected from reality and cannot reflect the actual state: Although finite element (FE) models of structures can be built for analysis, these models are usually based on idealized design parameters. They cannot reflect the dynamic changes in stiffness and mechanical behavior of real structures caused by temperature changes, material property degradation, or minor damage in real time and accurately. This "digital gap" between the model and reality limits the credibility of simulation results based on static models.

[0004] 2. Assessments based on deterministic loads lack the ability to predict future risks: Most existing real-time assessment methods verify the response to known vehicle loads that have already passed. While this ex-post analysis model can determine whether it was safe just now, it cannot make forward-looking risk predictions for upcoming or future more dangerous random traffic loads (such as heavy vehicle convoys or abnormal congestion) or extreme environmental loads (such as strong gusts of wind).

[0005] 3. Limited range of safety evaluation indicators, failing to quantify failure probability: Traditional safety assessments typically calculate a deterministic safety factor or safety margin based on allowable stress or displacement. While intuitive, this method cannot quantify the actual failure risk of a structure under the combined influence of multiple uncertainties (such as the randomness of loads and the variability of material parameters). It can only answer whether a structure is safe, but not the more profound, probability-based reliability question of what the probability of unsafety is. Summary of the Invention

[0006] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a bridge load-bearing capacity monitoring method and system based on digital twins. This method can not only synchronize the evolution of digital models and physical entities in real time, but also predict and assess the probabilistic risks of the structure under future random loads, providing intelligent decision support and realizing dynamic, online and predictive assessment of the bridge's load-bearing capacity.

[0007] To achieve the above objectives, the present invention provides the following solution: a bridge bearing capacity monitoring method and system based on digital twins, comprising: Based on the bridge structure, original monitoring data of multi-physics fields are obtained, a digital twin finite element model is constructed, and then the time-varying material constitutive model parameters in the digital twin model are identified online using a physical information neural network. Using the time-varying material constitutive model parameters, the material property matrix and stiffness matrix of the digital twin finite element model are continuously updated to obtain a synchronous twin model; Based on the current state of the vehicle, the theoretical load effect on the predicted path of the vehicle is calculated, and then the temporary load condition data packet is identified based on the theoretical load effect. Based on historical dynamic weighing data of vehicles, a virtual future load scenario including normal and extreme working conditions is output using a multi-level probability model and an extreme value generator. The temporary load case data package and the virtual future load scenario are injected into the synchronous twin model to perform load simulation and obtain the structural response statistical distribution. Using the corresponding statistical distribution of the structure, safety assessment indicators are calculated, and then reinforcement learning is used to optimize the decision based on the safety assessment indicators to obtain the optimal management and intervention plan.

[0008] Optionally, based on the bridge structure, original multiphysics monitoring data is acquired, a digital twin finite element model is constructed, and then a physical information neural network is used to identify the time-varying constitutive parameters in the digital twin model online, including: A sensor network for structural health monitoring is pre-deployed on the bridge structure, and the raw multi-physics monitoring data of the sensor network is continuously received through the sensor network interface; the sensor network includes strain sensors for measuring mechanical fields, displacement gauges, and temperature sensors for measuring temperature fields. Based on the bridge's design drawings and material properties, a finite element model including the geometric and material parameters of the bridge components is constructed. The deployment locations of the sensor network are mapped to the finite element model, and the correspondence between physical measuring points and virtual measuring points is established to obtain a digital twin finite element model. The original monitoring data of the multiphysics field and the hybrid loss function are used to train the network to obtain the physical information neural network. Then, the time-varying material constitutive model parameters of the digital twin finite element model are inferred using the physical information neural network.

[0009] Optionally, the calculation expression for the hybrid loss function is: ; in, To minimize the data fitting term that matches the difference between the network output and the sensor's measured strain, These are the physical constraint terms embedded in the governing equations of structural mechanics. Weighting coefficients to balance data and physical constraints.

[0010] Optionally, based on the acquired current vehicle state, the theoretical load effect on the predicted vehicle path is calculated, and then, based on the theoretical load effect, temporary load condition data packets are identified, including: At the bridgehead, a dynamic weighing system is used to obtain the real-time vehicle weight, speed, and wheelbase to determine the vehicle's current status. Based on the current state of the vehicle and the geometric linearity of the bridge, the dynamic prediction model is used to predict the path coordinate sequence that the vehicle will pass through in the next tens of seconds step by step, thus obtaining the predicted path. For each position of the vehicle on the predicted path, calculate the theoretical load effect of the predicted vehicle position on the critical cross section of the structure; Based on the theoretical load effect, the corresponding load magnitude, spatial location, and duration are extracted at all vehicle prediction times and locations to identify the spatiotemporal points that cause the most unfavorable effects on the structure. The identified spatiotemporal points are then packaged into temporary load condition data packages.

[0011] Optionally, based on historical dynamic weighing data of vehicles, a multi-level probabilistic model and extreme value generator are used to output virtual future load scenarios including normal and extreme working conditions, including: Based on historical dynamic weighing data of vehicles, a multi-level probability model is constructed. Using Monte Carlo sampling, virtual vehicles and virtual vehicle attributes are generated on the multi-level probability model to obtain a virtual regular traffic flow scenario that conforms to statistical laws. An extreme value generator based on extreme value theory is introduced into the multi-level probability model. The extreme value generator is used to sample and generate extremely overweight vehicles from the historical dynamic weighing data of the vehicles. The extremely overweight vehicles are then injected into the virtual normal traffic flow scenario, and the virtual future moving load time series including normal and extreme conditions is output to obtain the virtual future load scenario.

[0012] Optionally, the multi-level probability model includes a Markov chain model layer, a discrete probability distribution layer, and a conditional probability distribution layer. The Markov model layer is used to calculate the vehicle arrival time interval, the discrete probability distribution layer is used to calculate the frequency of occurrence of vehicles with different axle types, and the conditional probability distribution layer is used to calculate the vehicle weight and speed of each axle type.

[0013] Optionally, the temporary load case data package and the virtual future load scenario are injected into the synchronous twin model to perform load simulation and obtain the structural response statistical distribution, including: The temporary load case data package is injected into the synchronous twin model to calculate the maximum stress value of the key section or key part and output the maximum structural response of a single critical case. The virtual future moving load time series is injected one by one into the synchronous twin model, and for each virtual traffic scenario, the maximum stress and displacement at key locations are calculated, and the maximum response statistical distribution under various random future traffic scenarios is output. By combining the maximum structural response under a single critical condition and the statistical distribution of the maximum response, the statistical distribution of the structural response is obtained.

[0014] Optionally, using the corresponding statistical distribution of the structure, a safety assessment index is calculated, and then, based on the safety assessment index, reinforcement learning is used for decision optimization to obtain the optimal management and intervention plan, including: Based on the maximum structural response under a single critical condition, a dynamic safety margin is calculated. It is then determined whether this safety margin is lower than a preset warning threshold. If so, a high-level warning message is issued via the output interface. The expression for calculating the dynamic safety margin is: ; in, To determine the preset allowable stress of the material for a single, real load event, For real-time stress; Based on the statistical distribution of the maximum response, the number of times the allowable stress is exceeded is counted to estimate the failure probability. Then, the time-varying reliability index is calculated using the failure probability to obtain a quantitative index of the overall safety level under the current structural state and for future uncertain load scenarios. The calculation expression for the time-varying reliability index is as follows: ; in, It is the inverse cumulative distribution function of the standard normal distribution. The failure probability of the structure; The goal is to minimize the total expected cost over the entire lifecycle. Based on the goal and the time-varying reliability index, a reinforcement learning agent is used to output the optimal maintenance intervention plan. The optimal maintenance intervention plan includes real-time early warning information, tiered early warning results, and maintenance intervention suggestions.

[0015] This invention also provides a bridge load-bearing capacity monitoring system based on digital twins, comprising: The model parameter identification module is used to acquire original monitoring data of multi-physics fields based on the bridge structure, construct a digital twin finite element model, and then use a physical information neural network to identify the time-varying material constitutive model parameters in the digital twin model online. The digital twin modeling module is used to continuously update the material property matrix and stiffness matrix of the digital twin finite element model using the parameters of the time-varying material constitutive model, so as to obtain a synchronous twin model. The temporary load condition module is used to calculate the theoretical load effect on the predicted path of the vehicle based on the current state of the vehicle, and then identify the temporary load condition data packet based on the theoretical load effect. The Future Load Scenario Module is used to output virtual future load scenarios, including normal and extreme conditions, based on historical dynamic weighing data of vehicles and using multi-level probability models and extreme value generators. The load simulation module is used to inject the temporary load condition data package and the virtual future load scenario into the synchronous twin model to perform load simulation and obtain the structural response statistical distribution. The intervention plan module is used to calculate the safety assessment index using the corresponding statistical distribution of the structure, and then use reinforcement learning to optimize the decision based on the safety assessment index to obtain the optimal management intervention plan. The model parameter identification module, the digital twin modeling module, the temporary load condition module, the future load scenario module, the load simulation module, and the intervention scheme module are interconnected.

[0016] This invention discloses the following technical effects by providing a bridge bearing capacity monitoring method and system based on digital twins: 1. Achieving a "vivid" digital twin, solving the problem of model-reality disconnect: This invention utilizes online identification technologies such as physical information neural networks to enable the digital twin model to learn and reflect real-time, dynamic changes in the mechanical behavior of physical structures caused by factors such as temperature and damage. This completely breaks through the limitations of the static and idealized nature of traditional finite element models, achieving deep synchronous evolution between the model and reality, and greatly improving the credibility of simulation analysis results.

[0017] 2. A leap from "post-event verification" to "proactive risk prediction": This invention, through the application of random load generation and extreme value theory, moves the system beyond passive analysis of past loads. It can proactively and probabilistically simulate potentially more dangerous extreme load scenarios in the future and evaluate the structure's response under these scenarios. This transforms load-bearing capacity monitoring from post-event diagnosis to pre-event early warning, enabling preventative maintenance.

[0018] 3. This invention elevates the assessment from "deterministic safety factor" to "probabilistic time-varying reliability": By introducing probabilistic simulation and time-varying reliability analysis, it upgrades safety assessment from a simple "yes / no" judgment to a scientific system capable of quantifying uncertainty and assessing real failure risks. The output reliability index... β It can more profoundly and comprehensively reflect the true safety level of a structure in a complex operating environment than the traditional "safety margin", and can further drive optimal decision-making based on the whole life cycle cost.

[0019] 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

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the method flow provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the data flow provided in an embodiment of the present invention; Figure 3 A schematic diagram of the online model identification process provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the load processing and scenario generation process provided in an embodiment of the present invention; Figure 5 A schematic diagram illustrating the structural response analysis and load-bearing capacity assessment process provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the load-bearing capacity assessment and decision-making process provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of sensor deployment provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the system architecture provided for an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of 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.

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] like Figure 1 , Figure 2 As shown, this invention provides a bridge load-bearing capacity monitoring method and system based on digital twins, including: Step 1, as follows Figure 3 As shown, based on the bridge structure, original monitoring data of multiphysics fields are obtained, a digital twin finite element model is constructed, and then the time-varying material constitutive model parameters in the digital twin model are identified online using a physical information neural network. Step 1 includes: A sensor network for structural health monitoring is pre-deployed on the bridge structure, and the raw multi-physics monitoring data of the sensor network is continuously received through the sensor network interface; the sensor network includes strain sensors for measuring mechanical fields, displacement gauges, and temperature sensors for measuring temperature fields.

[0025] Based on the bridge's design drawings and material properties, a finite element model including the geometric and material parameters of the bridge components is constructed. The deployment locations of the sensor network are mapped into the finite element model, and the correspondence between physical measuring points and virtual measuring points is established to obtain a digital twin finite element model.

[0026] The original monitoring data of the multiphysics field and the hybrid loss function are used to train the network to obtain the physical information neural network. Then, the time-varying material constitutive model parameters of the digital twin finite element model are inferred using the physical information neural network.

[0027] The calculation expression for the hybrid loss function is as follows: ; in, To minimize the data fitting term that matches the difference between the network output and the sensor's measured strain, These are the physical constraint terms embedded in the governing equations of structural mechanics. Weighting coefficients to balance data and physical constraints.

[0028] By optimizing this hybrid loss function, the physical information neural network can infer the parameters of the time-varying material constitutive model that best explains the current observation data, such as the stiffness reduction factor reflecting damage accumulation or the thermal expansion coefficient reflecting temperature effects.

[0029] Step 2: Using the time-varying material constitutive model parameters, continuously update the material property matrix and stiffness matrix of the digital twin finite element model to obtain a synchronous twin model.

[0030] Specifically, these parameters identified by the physical information neural network are immediately used to update the core parameter matrix of the digital twin finite element model, thereby enabling the digital twin model to reflect the real performance state of the physical structure in real time.

[0031] Step 3, as follows Figure 4 , Figure 5 As shown, based on the current state of the vehicle, the theoretical load effect on the predicted path of the vehicle is calculated, and then the temporary load condition data packet is identified based on the theoretical load effect.

[0032] Step 3 includes: At the bridgehead, a dynamic weighing system is used to obtain the real-time vehicle weight, speed, and wheelbase to determine the vehicle's current state. Based on the vehicle's current state and the bridge's geometric linearity (slope, curvature, etc.), a dynamic prediction model is used to predict the path coordinate sequence that the vehicle will traverse in the next tens of seconds step by step, thus obtaining the predicted path.

[0033] For each position of the vehicle on the predicted path, calculate the theoretical load effect (such as bending moment) on key structural sections (such as mid-span) caused by the predicted vehicle position.

[0034] Based on the theoretical load effect, the corresponding load magnitude, spatial location, and duration are extracted at all vehicle prediction times and locations to identify the spatiotemporal points that cause the most unfavorable effects on the structure. The identified spatiotemporal points are then packaged into temporary load condition data packages.

[0035] Step 4, as follows Figure 4 , Figure 5 As shown, based on the vehicle's historical dynamic weighing data, a multi-level probability model and an extreme value generator are used to output virtual future load scenarios including normal and extreme working conditions. Step 4 includes: Based on historical dynamic weighing data of vehicles, a multi-level probability model is constructed. Using Monte Carlo sampling, virtual vehicles and virtual vehicle attributes are generated on the multi-level probability model to obtain a virtual regular traffic flow scenario that conforms to statistical laws. The multi-level probability model includes a Markov chain model layer, a discrete probability distribution layer, and a conditional probability distribution layer. The Markov model layer is used to calculate the vehicle arrival time interval, the discrete probability distribution layer is used to calculate the occurrence frequency of vehicles with different axle types, and the conditional probability distribution layer is used to calculate the vehicle weight and speed of each axle type.

[0036] An extreme value generator based on extreme value theory is introduced into the multi-level probability model. The extreme value generator is used to sample and generate extremely overweight vehicles from the historical dynamic weighing data of the vehicles. The extremely overweight vehicles are then injected into the virtual normal traffic flow scenario, and the virtual future moving load time series including normal and extreme conditions is output to obtain the virtual future load scenario.

[0037] The extreme value generator's role is to generate extreme events: during Monte Carlo simulations, it pauses sampling from the conventional probability model with a very small probability, instead sampling from a generalized Pareto distribution model to generate one or more "once-in-a-century" level extremely overweight vehicles, and injects them into the virtual traffic flow to generate more challenging extreme load scenarios. The probability distribution of the generalized Pareto distribution model is as follows: ; in, For position parameters, For scale parameters, For shape parameters.

[0038] Step 5, as follows Figure 4 , Figure 5 As shown, the temporary load case data package and the virtual future load scenario are injected into the synchronous twin model to perform load simulation and obtain the structural response statistical distribution.

[0039] Step 5 includes: The temporary load case data package is injected into the synchronous twin model to calculate the maximum stress value of the key section or key part and output the maximum structural response of a single critical case.

[0040] The virtual future moving load time series is injected one by one into the synchronous twin model, and for each virtual traffic scenario, the maximum stress and displacement at key locations are calculated, and the maximum response statistical distribution under various random future traffic scenarios is output.

[0041] By combining the maximum structural response under a single critical condition and the statistical distribution of the maximum response, the statistical distribution of the structural response is obtained.

[0042] Step 6, as follows Figure 6 , Figure 7 As shown, the safety assessment index is calculated using the corresponding statistical distribution of the structure, and then the optimal management and intervention plan is obtained by using reinforcement learning based on the safety assessment index.

[0043] Step 6 includes: Based on the maximum structural response under a single critical condition, a dynamic safety margin is calculated. It is then determined whether this safety margin is lower than a preset warning threshold. If so, a high-level warning message is issued via the output interface. The expression for calculating the dynamic safety margin is: ; in, To determine the preset allowable stress of the material for a single, real load event, This represents real-time stress.

[0044] Based on the statistical distribution of the maximum response, the number of times the allowable stress is exceeded is counted to estimate the failure probability. Then, the time-varying reliability index is calculated using the failure probability to obtain a quantitative index of the overall safety level under the current structural state and for future uncertain load scenarios. The calculation expression for the time-varying reliability index is as follows: ; in, It is the inverse cumulative distribution function of the standard normal distribution. This represents the failure probability of the structure.

[0045] The goal is to minimize the total expected cost over the entire lifecycle. Based on the goal and the time-varying reliability index, a reinforcement learning agent is used to output the optimal maintenance intervention plan. The optimal maintenance intervention plan includes real-time early warning information, tiered early warning results, and maintenance intervention suggestions.

[0046] The working principle of reinforcement learning agents is as follows: It will receive reliability metrics Considering the current state of the environment, based on this state, it selects an optimal action from an action space containing multiple preset maintenance measures (such as "no operation," "speed limit," and "maintenance"). The selection is based on a life-cycle cost model library, solved to minimize the total expected cost. The goal is to optimize the system and output optimal management intervention recommendations. ; in, For the cost of intervention measures, To be consistent with the aforementioned reliability index Related risk costs.

[0047] Ultimately, the agent outputs an optimal management and maintenance intervention recommendation that balances current security risks with future maintenance costs. This system can be deployed on one or more servers, high-performance workstations, or local processing centers.

[0048] like Figure 8 As shown, the present invention also provides a bridge load-bearing capacity monitoring system based on digital twins, comprising: The model parameter identification module is used to acquire original monitoring data of multi-physics fields based on the bridge structure, construct a digital twin finite element model, and then use a physical information neural network to identify the time-varying material constitutive model parameters in the digital twin model online. The digital twin modeling module is used to continuously update the material property matrix and stiffness matrix of the digital twin finite element model using the parameters of the time-varying material constitutive model, so as to obtain a synchronous twin model. The temporary load condition module is used to calculate the theoretical load effect on the predicted path of the vehicle based on the current state of the vehicle, and then identify the temporary load condition data packet based on the theoretical load effect. The Future Load Scenario Module is used to output virtual future load scenarios, including normal and extreme conditions, based on historical dynamic weighing data of vehicles and using multi-level probability models and extreme value generators. The load simulation module is used to inject the temporary load condition data package and the virtual future load scenario into the synchronous twin model to perform load simulation and obtain the structural response statistical distribution. The intervention plan module is used to calculate the safety assessment index using the corresponding statistical distribution of the structure, and then use reinforcement learning to optimize the decision based on the safety assessment index to obtain the optimal management intervention plan. The model parameter identification module, the digital twin modeling module, the temporary load condition module, the future load scenario module, the load simulation module, and the intervention scheme module are interconnected.

[0049] Therefore, this invention provides a bridge load-bearing capacity monitoring method and system based on digital twins, which can not only synchronize the evolution of digital models and physical entities in real time, but also predict and assess the probabilistic risks of the structure under future random loads, provide intelligent decision support, and realize dynamic, online and predictive assessment of bridge load-bearing capacity.

[0050] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0051] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for monitoring the bearing capacity of bridges based on digital twins, characterized in that, include: Based on the bridge structure, original monitoring data of multi-physics fields are obtained, a digital twin finite element model is constructed, and then the time-varying material constitutive model parameters in the digital twin model are identified online using a physical information neural network. Using the time-varying material constitutive model parameters, the material property matrix and stiffness matrix of the digital twin finite element model are continuously updated to obtain a synchronous twin model; Based on the current state of the vehicle, the theoretical load effect on the predicted path of the vehicle is calculated, and then the temporary load condition data packet is identified based on the theoretical load effect. Based on historical dynamic weighing data of vehicles, a virtual future load scenario including normal and extreme working conditions is output using a multi-level probability model and an extreme value generator. The temporary load case data package and the virtual future load scenario are injected into the synchronous twin model to perform load simulation and obtain the structural response statistical distribution. Using the corresponding statistical distribution of the structure, safety assessment indicators are calculated, and then reinforcement learning is used to optimize the decision based on the safety assessment indicators to obtain the optimal management and intervention plan.

2. The bridge bearing capacity monitoring method based on digital twin according to claim 1, characterized in that, Based on the bridge structure, original multiphysics field monitoring data are acquired, a digital twin finite element model is constructed, and then a physical information neural network is used to identify the time-varying constitutive parameters in the digital twin model online, including: A sensor network for structural health monitoring is pre-deployed on the bridge structure, and the raw multi-physics monitoring data of the sensor network is continuously received through the sensor network interface; the sensor network includes strain sensors for measuring mechanical fields, displacement gauges, and temperature sensors for measuring temperature fields. Based on the bridge's design drawings and material properties, a finite element model including the geometric and material parameters of the bridge components is constructed. The deployment locations of the sensor network are mapped to the finite element model, and the correspondence between physical measuring points and virtual measuring points is established to obtain a digital twin finite element model. The original monitoring data of the multiphysics field and the hybrid loss function are used to train the network to obtain the physical information neural network. Then, the time-varying material constitutive model parameters of the digital twin finite element model are inferred using the physical information neural network.

3. The bridge bearing capacity monitoring method based on digital twin according to claim 2, characterized in that, The calculation expression for the hybrid loss function is as follows: ； in, To minimize the data fitting term that matches the difference between the network output and the sensor's measured strain, These are the physical constraint terms embedded in the governing equations of structural mechanics. Weighting coefficients to balance data and physical constraints.

4. The bridge bearing capacity monitoring method based on digital twin according to claim 3, characterized in that, Based on the current vehicle state, the theoretical load effect on the predicted vehicle path is calculated. Then, based on the theoretical load effect, temporary load condition data packets are identified, including: At the bridgehead, a dynamic weighing system is used to obtain the real-time vehicle weight, speed, and wheelbase to determine the vehicle's current status. Based on the current state of the vehicle and the geometric linearity of the bridge, the dynamic prediction model is used to predict the path coordinate sequence that the vehicle will pass through in the next tens of seconds step by step, thus obtaining the predicted path. For each position of the vehicle on the predicted path, calculate the theoretical load effect of the predicted vehicle position on the critical cross section of the structure; Based on the theoretical load effect, the corresponding load magnitude, spatial location, and duration are extracted at all vehicle prediction times and locations to identify the spatiotemporal points that cause the most unfavorable effects on the structure. The identified spatiotemporal points are then packaged into temporary load condition data packages.

5. The bridge bearing capacity monitoring method based on digital twin according to claim 4, characterized in that, Based on historical dynamic weighing data of vehicles, a multi-level probabilistic model and extreme value generator are used to output virtual future load scenarios including normal and extreme working conditions, including: Based on historical dynamic weighing data of vehicles, a multi-level probability model is constructed. Using Monte Carlo sampling, virtual vehicles and virtual vehicle attributes are generated on the multi-level probability model to obtain a virtual regular traffic flow scenario that conforms to statistical laws. An extreme value generator based on extreme value theory is introduced into the multi-level probability model. The extreme value generator is used to sample and generate extremely overweight vehicles from the historical dynamic weighing data of the vehicles. The extremely overweight vehicles are then injected into the virtual normal traffic flow scenario, and the virtual future moving load time series including normal and extreme conditions is output to obtain the virtual future load scenario.

6. The bridge bearing capacity monitoring method based on digital twin according to claim 5, characterized in that, The multi-level probability model includes a Markov chain model layer, a discrete probability distribution layer, and a conditional probability distribution layer. The Markov model layer is used to calculate the vehicle arrival time interval, the discrete probability distribution layer is used to calculate the frequency of occurrence of vehicles with different axle types, and the conditional probability distribution layer is used to calculate the vehicle weight and speed of each axle type.

7. A bridge bearing capacity monitoring method based on digital twins according to claim 6, characterized in that, The temporary load case data package and the virtual future load scenario are injected into the synchronous twin model to perform load simulation and obtain the structural response statistical distribution, including: The temporary load case data package is injected into the synchronous twin model to calculate the maximum stress value of the key section or key part and output the maximum structural response of a single critical case. The virtual future moving load time series is injected one by one into the synchronous twin model, and for each virtual traffic scenario, the maximum stress and displacement at key locations are calculated, and the maximum response statistical distribution under various random future traffic scenarios is output. By combining the maximum structural response under a single critical condition and the statistical distribution of the maximum response, the statistical distribution of the structural response is obtained.

8. The bridge bearing capacity monitoring method based on digital twin according to claim 7, characterized in that, Using the corresponding statistical distribution of the structure, safety assessment indicators are calculated. Then, based on these indicators, reinforcement learning is used for decision optimization to obtain the optimal management and intervention plan, including: Based on the maximum structural response under a single critical condition, a dynamic safety margin is calculated. It is then determined whether this safety margin is lower than a preset warning threshold. If so, a high-level warning message is issued via the output interface. The expression for calculating the dynamic safety margin is: ； in, To determine the preset allowable stress of the material for a single, real load event, For real-time stress; Based on the statistical distribution of the maximum response, the number of times the allowable stress is exceeded is counted to estimate the failure probability. Then, the time-varying reliability index is calculated using the failure probability to obtain a quantitative index of the overall safety level under the current structural state and for future uncertain load scenarios. The calculation expression for the time-varying reliability index is as follows: ； in, It is the inverse cumulative distribution function of the standard normal distribution. The failure probability of the structure; The goal is to minimize the total expected cost over the entire lifecycle. Based on the goal and the time-varying reliability index, a reinforcement learning agent is used to output the optimal maintenance intervention plan. The optimal maintenance intervention plan includes real-time early warning information, tiered early warning results, and maintenance intervention suggestions.

9. A bridge load-bearing capacity monitoring system based on digital twins, characterized in that, include: The model parameter identification module is used to acquire original monitoring data of multi-physics fields based on the bridge structure, construct a digital twin finite element model, and then use a physical information neural network to identify the time-varying material constitutive model parameters in the digital twin model online. The digital twin modeling module is used to continuously update the material property matrix and stiffness matrix of the digital twin finite element model using the parameters of the time-varying material constitutive model, so as to obtain a synchronous twin model. The temporary load condition module is used to calculate the theoretical load effect on the predicted path of the vehicle based on the current state of the vehicle, and then identify the temporary load condition data packet based on the theoretical load effect. The Future Load Scenario Module is used to output virtual future load scenarios, including normal and extreme conditions, based on historical dynamic weighing data of vehicles and using multi-level probability models and extreme value generators. The load simulation module is used to inject the temporary load condition data package and the virtual future load scenario into the synchronous twin model to perform load simulation and obtain the structural response statistical distribution. The intervention plan module is used to calculate the safety assessment index using the corresponding statistical distribution of the structure, and then use reinforcement learning to optimize the decision based on the safety assessment index to obtain the optimal management intervention plan. The model parameter identification module, the digital twin modeling module, the temporary load condition module, the future load scenario module, the load simulation module, and the intervention scheme module are interconnected.

Images