Pavement structure response monitoring and total distribution calculation method based on digital twinborn technology

Abstract

The invention provides a pavement structure response monitoring and total distribution calculation method based on a digital twinborn technology. The method comprises the following steps: S1, constructing a road structure and a digital twinborn model thereof; s2, an embedded sensor is arranged in the road structure, and real-time structure response data of the road structure at the feature position under the vehicle load effect are obtained; s3, carrying out an on-site loading test, obtaining structure response actual measurement data under the action of a vehicle moving load, and verifying and correcting the digital twinborn model; s4, using finite element calculation software to perform simulation calculation and output structural response data of the sampling points under the action of different load working conditions based on the digital twinborn model; and S5, training a deep learning model by using the simulation response data. According to the method, a digital twinning technology is utilized, a high-fidelity mapping model of a physical entity in a virtual space is constructed, behaviors of the physical entity in a real environment are accurately simulated, and a data basis can be provided for a machine learning response deduction method at low cost.

Description

Technical Field

[0001] The present invention relates to the field of road structure response monitoring, and in particular to a road surface structure response monitoring and fully distributed calculation method based on digital twin technology. Background Art

[0002] Road structure response monitoring based on embedded sensors helps to understand the mechanical state of each structural layer, evaluate the health of the pavement structure, and promptly identify early road structure diseases, thereby formulating a reasonable maintenance plan. However, since existing embedded sensors can only obtain local point response data, a large number of sensors need to be buried to achieve effective structural health status monitoring, which not only increases the monitoring cost, but also damages the bearing capacity of the pavement structure.

[0003] With the development of artificial intelligence technology in recent years, machine learning methods have become an effective tool for solving many prediction problems with their powerful data processing and pattern recognition capabilities. In this regard, relevant studies have proposed that machine learning algorithms can be introduced to achieve an overall assessment of the health status of road structures using response data obtained from a limited number of sensors. However, the current road field lacks actual measured structural response data of real road entities, which cannot support the large amount of training data required to build machine learning models. Summary of the invention

[0004] In order to solve the problems existing in the above-mentioned prior art, the present invention provides a pavement structure response monitoring and full distribution calculation method based on digital twin technology, comprising the following steps: S1. Twin model creation: Construct the road structure and its digital twin model. The digital twin model maps the structural composition, material properties, mechanical behavior and environmental factors of the road structure.

[0005] S2. Real-time structural response monitoring: embedded sensors are placed inside the road structure to obtain real-time structural response data at characteristic positions of the road structure under vehicle loads, and the data is transmitted and stored in the twin database.

[0006] S3. Verify the fidelity of the digital twin model. Conduct on-site loading tests to obtain measured data on the structural response under the vehicle moving load, use the digital twin model to simulate the response data under the same load data, verify and calibrate the digital twin model until the fidelity is no less than the first threshold.

[0007] S4. Digital twin model simulation and data output: Use finite element calculation software based on the digital twin model to simulate and calculate and output the structural response data of the sampling points under different load conditions.

[0008] S5. Construction and verification of the deduction model; using the above digital twin model to simulate the response data and train the deep learning model; by inputting the response data of the sampling points, predict the full distribution of the response on the cross section.

[0009] Furthermore, in S1, a digital twin model of the road structure is established using finite element simulation software to define the geometric dimensions of the road structure, the material composition of each layer, the material constitutive properties, the model boundary conditions and environmental factors, and to construct a digital twin that describes the physical mechanical behavior and characteristics of the road structure.

[0010] Furthermore, in S2, the locations where embedded sensors are deployed include wheel tracks, road edge, cross-section center point, bottom of asphalt surface layer and / or bottom of semi-rigid base layer, and the road structure feature locations where response sensors need to be buried are determined based on the road layer and the type of structural damage that needs to be monitored.

[0011] Furthermore, in S3, when the fidelity of the output result of the digital twin model is lower than the first threshold, the digital twin model is iteratively corrected to adjust its simulation parameters, including material constitutive model parameters, boundary conditions and / or mesh division.

[0012] Furthermore, in S4, several types of vehicle loads are simulated in the virtual model space to realize the calculation and output of the digital twin model's sampling point structural response data under various vehicle load conditions, providing training data for the subsequent deduction model training. Different load conditions should consider the value combination of different vehicle axle numbers, wheelbases, axle weights, tire numbers and sizes, wheelbases, and other factors to restore the various vehicle load conditions in service to the greatest extent possible.

[0013] Furthermore, in S5, the network structure of the deep learning model is a Transformer deep learning model, the input of the model is the sampling point response data, and the output is the full distribution data of the response; the data set used for deep learning model training is the sampling point response data output by the digital twin model simulation in S4, and the deep learning model is verified using the sampling point response simulation data under the action of external loads of the training data set. When the accuracy of the predicted response data output by the model reaches the second threshold, the deep learning model is considered to have been successfully trained.

[0014] Furthermore, the training steps of the deep learning model include: Dataset division: The response data output by the digital twin model simulation is divided into training set, test set and validation set in a preset ratio.

[0015] Data input processing: The training data is divided into two tensors of the same dimension. Tensor 1 is used as the input tensor to store the response data of known sampling points of specified number and distribution, representing the point response data collected by the embedded sensor; Tensor 2 is used as the target tensor to store the response data of the simulated output, representing the target value of the full distribution of the structural response.

[0016] Model training: Adjust model training parameters, including learning rate, batch size, and / or number of rounds, and train the deep learning model using the training set in several rounds until the model converges.

[0017] Model verification: Use response simulation data under load conditions other than the training set to input into the converged deep learning model, and compare the difference between the output response results and the sampling values ​​under the actual load. When the accuracy of the predicted response data output by the model reaches the second threshold, the deep learning model training is considered successful; otherwise, adjust the parameters of the deep learning model and retrain the model until the accuracy of the predicted response data output by the model reaches the second threshold.

[0018] Furthermore, the ratio of training set, test set and validation set in the data set partition is set to A:B:C; A is set to 0.6-0.98, B is set to 0.01-0.2, and C is set to 0.01-0.2.

[0019] Furthermore, in S3, the collection of measured data on the structural response under the action of vehicle moving load includes: loading a known vehicle moving load on the road section to be monitored where the embedded sensor is buried, conducting several on-site loading tests under the same load, and collecting measured data on the structural response of the vehicle throughout the entire driving process, until the measured response data error is less than a third threshold, and further comparing it with the twin simulation data to verify the accuracy of the model simulation results.

[0020] Furthermore, in S3, the verification method of the simulation degree of the digital twin model is as follows: using finite element simulation software, writing a load definition subroutine, setting the vehicle wheelbase, axle weight, number and size of tires, wheelbase, and driving speed to be consistent with the vehicle in the on-site loading test, loading the simulated pavement structure model, and outputting the response data of the entire loading process. Loading the simulated pavement structure model, outputting the stress and strain data of the entire loading process, comparing the simulated data with the measured data, when the error is within an acceptable range, it can be considered that the twin simulation model has a high fidelity.

[0021] Based on digital twin technology, the present invention constructs a high-fidelity twin model of the road physical structure in virtual space, realizes the simulation of the mechanical behavior of the road structure under different vehicle loads, and provides a data basis for the training of deep learning deduction models. At the same time, this model better solves the problem that machine learning methods are difficult to apply in the field of road engineering due to insufficient monitoring data, and promotes the application of artificial intelligence technology in the field of road engineering.

[0022] By burying embedded sensors in the road structure, the present invention realizes real-time monitoring of structural response data, which helps to better understand the mechanical state inside the structural layer and specify more reasonable maintenance plans; based on the twin model simulation data, a deep learning model for structural response deduction is constructed, which realizes the deduction of the global response distribution of the structure and the assessment of the health status using a limited number of sensor monitoring data, greatly reducing the cost of structural response monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 It is a flow chart of the method of the present invention.

[0025] Figure 2 Schematic diagram of geometric parameters of the road structure model in the embodiment of the present invention.

[0026] Figure 3 Schematic diagram of the axle geometry and axle weight of the vehicle load used in the embodiment of the present invention.

[0027] Figure 4 It is a comparison chart of the digital twin model simulation and measured strain data in an embodiment of the present invention.

[0028] Figure 5 It is a structural diagram of the response deduction deep model of the present invention.

[0029] Figure 6 It is a thermal diagram of strain distribution predicted by the deduction model and the actual distribution in the embodiment of the present invention. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] See also Figure 1 The pavement structure response monitoring and full distribution calculation method based on digital twin technology of this embodiment includes the following steps: S1. Twin model creation: Construct a road structure and its digital twin model, and the digital twin model maps the structural composition, material properties, mechanical behavior and environmental factors of the road structure. Preferably, a digital twin model of the road structure is established using finite element simulation software, the geometric dimensions of the road structure, the material composition of each layer, the material constitutive properties, the model boundary conditions and environmental factors are defined, and a digital twin describing the mechanical behavior and characteristics of the road structure entity is constructed.

[0032] In this embodiment, a typical flexible base pavement structure model is established in the Abaqus finite element simulation software. The structure consists of a 36 cm thick asphalt layer (4 cm SMA-13 ​​+ 6 cm AC-20 + 10 cm AC-25 + 16 cm ATB), a 32 cm thick graded gravel layer, and a soil base with a depth of 2 m. The geometric dimensions are 2.25 m*2.68 m*10.35 m. Figure 2 The boundary conditions of the model are: the left side is XSYMM constraint (U1=UR2=UR3=0), the right side is a free plane, and the bottom is a hinge constraint (U1=U2=U3=0). The material parameters of each layer of the model are shown in the following table: Horizon Thickness / cm Young's modulus / MPa Poisson's ratio Asphalt layer 36 10000 (instantaneous) 0.35 Graded crushed stone layer 32 350 0.35 Soil foundation >100 60 0.4 The viscoelastic constitutive relation of the asphalt layer is described by the generalized Maxwell model, and the initial temperature is 20℃.

[0033] S2. Real-time structural response monitoring: embedded sensors are placed inside the road structure to obtain real-time structural response data at characteristic positions of the road structure under vehicle loads, and the data is transmitted and stored in the twin database.

[0034] In this embodiment, the fiber grating sensor is buried at the bottom of the asphalt layer below the wheel track in the driving direction and perpendicular to the driving direction to monitor the tensile strain at the bottom of the asphalt layer in real time, and the response data is transmitted to the twin database. In other embodiments, the road structure feature position where the response sensor needs to be buried is determined according to the road layer and the type of structural damage to be monitored. The positions where the embedded sensors are laid include the wheel track, the edge of the road surface, the center point of the cross section, the bottom of the asphalt surface layer and / or the bottom of the semi-rigid base layer.

[0035] S3. Verify the fidelity of the digital twin model. Conduct on-site loading tests to obtain measured data on the structural response under the vehicle moving load, use the digital twin model to simulate the response data under the same load data, verify and calibrate the digital twin model until the fidelity is no less than the first threshold. The first threshold is selected based on actual needs.

[0036] The simulation degree of the digital twin model is verified by using finite element simulation software, writing a load definition subroutine, setting the vehicle wheelbase, axle weight, number and size of tires, wheelbase, and driving speed to be consistent with the vehicle in the on-site loading test, loading the simulated pavement structure model, and outputting the response data of the entire loading process. Loading the simulated pavement structure model, outputting the stress and strain data of the entire loading process, comparing the simulated data with the measured data, and when the error is within an acceptable range, the twin simulation model can be considered to have a high fidelity.

[0037] In this embodiment, a four-axle truck is used to load the road where the fiber grating sensor is buried at a speed of 60 km / h. The axle weight and axle geometry are as follows: Figure 3 As shown. The strain data of the bottom of the asphalt layer perpendicular to the driving direction when the vehicle load passes through the road section is collected and transmitted to the twin database. The measurement is repeated multiple times until the response error measured multiple times is less than the third threshold value. The third threshold value is selected according to actual needs. In the twin model, the user-defined load subroutine DLOAD in the Abaqus finite element simulation software is used to add a moving load consistent with the measured truck load, and the simulation calculation results of the strain of the bottom of the asphalt layer perpendicular to the driving direction are output and exported to the twin database. Plot the strain change curve of the measured data and the simulation data over time, as shown in Figure 4 As shown in the figure, the error between the twin model simulation data and the measured data meets the accuracy requirements, indicating that the twin model has achieved a high fidelity.

[0038] In other embodiments, when the fidelity of the digital twin model output result is lower than a first threshold, the digital twin model is iteratively corrected and its simulation parameters, including material constitutive model parameters, boundary conditions and / or mesh division, are adjusted until the accuracy of the model output response data meets the expected requirements. In this embodiment, no adjustment is required to meet the requirements.

[0039] S4. Digital twin model simulation and data output: Use finite element calculation software based on the digital twin model to simulate and calculate and output the structural response data of the sampling points under different load conditions.

[0040] In the virtual model space, several types of vehicle loads are simulated to realize the calculation and output of the digital twin model's sampling point structural response data under various vehicle load conditions, providing training data for subsequent deduction model training. Different load conditions should consider the value combination of different vehicle axles, wheelbases, axle weights, tire numbers and sizes, wheelbases and other factors to restore the various vehicle load conditions in service to the greatest extent possible.

[0041] In this embodiment, by investigating and consulting relevant specifications, considering the single-axle two-wheel load with a wheelbase range of 1.30~1.90m and an axle weight range of 7000~24000kg, the wheelbase and axle weight are combined to obtain a total of 595 vehicle load conditions. By writing a Python script program, the established road structure twin model is used in the Abaqus finite element simulation software to calculate and output the strain distribution data in the 1.75*4.35m area at the bottom of the asphalt layer under the action of the above 595 vehicle loads at a speed of 50km / h, and store it in the twin database.

[0042] S5. Construction and verification of the deduction model; using the above digital twin model to simulate the response data and train the deep learning model; by inputting the response data of the sampling points, predict the full distribution of the response on the cross section.

[0043] In this embodiment, the network structure of the deep learning model is a Transformer deep learning model. The input of the model is the sampling point response data, and the output is the full distribution data of the response. The data set used for deep learning model training is the sampling point response data output by the digital twin model simulation in S4. The deep learning model is verified using the sampling point response simulation data under the action of external loads of the training data set. When the accuracy of the predicted response data output by the model reaches the second threshold, the deep learning model is considered to be successfully trained. The selection of the second threshold is determined according to actual needs.

[0044] In this embodiment, a Transformer deep learning model is constructed, and the network structure is as follows Figure 5 The input of the model is four columns of 40 known sampling point response data, simulating the response data that can be collected by four fiber Bragg grating sensors. Through the response full distribution deduction model, the response deduction results of 100 points evenly distributed in the study area are output.

[0045] Write a Python program to complete the model training. The steps are as follows: Data set division: After normalizing the strain data output by the twin model simulation, in this embodiment, it is divided into a training set, a test set, and a validation set at a ratio of 0.6:0.2:0.2. In other embodiments, the ratio of the training set, the test set, and the validation set can also be set to 0.6-0.98:0.01-0.2:0.01-0.2. When the amount of original data is large, the proportion of the training set can be increased to obtain a more accurate model.

[0046] Data input processing: The normalized training data is divided into two tensors of the same dimension. Tensor 1 is used as the input tensor to store the response data of known sampling points of specified number and distribution, representing the point response data collected by a limited number of embedded sensors; Tensor 2 is used as the target tensor to store all the strain data of the simulation output, representing the target value of the full distribution of structural strain.

[0047] Model training: Set the model training parameters to a learning rate of 0.005, a batch size of 32, and a number of rounds of 50. Use the training data set to train the deep learning model until the model loss value tends to a stable small value, indicating that the model has converged and the training is completed.

[0048] Model verification: Use the response simulation data of a vehicle with a wheelbase of 2.0m and an axle weight of 20,000kg outside the training data set to input into the converged model to obtain the strain distribution data predicted by the deduction model. The target strain distribution data and the deduction prediction data are plotted into a thermal diagram, such as Figure 6 shown.

[0049] Depend on Figure 6 It can be seen that the strain distribution obtained by the model is highly consistent with the true distribution. The regression square sum R of the model prediction value is further calculated. 2 It is 0.9381, indicating that the trained deduction model has achieved a high prediction accuracy. The pavement structure response monitoring and fully distributed deduction method based on digital twin technology proposed in the present invention can use a small number of sensors to achieve high-precision deduction of the global response and thus evaluate the overall health status of the road structure.

[0050] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A pavement structure response monitoring and fully distributed calculation method based on digital twin technology, characterized in that: The following steps are involved: S1. Digital twin model creation: Construct the road structure and its digital twin model, which maps the structural composition, material properties, mechanical behavior and environmental factors of the road structure; S2. Real-time structural response monitoring: embedded sensors are placed inside the road structure to obtain real-time structural response data at characteristic positions of the road structure under vehicle loads, and the data is transmitted and stored in the twin database; S3. Verify the fidelity of the digital twin model. Conduct on-site loading tests to obtain measured data on the structural response under vehicle moving loads, use the digital twin model to simulate the response data under the same load data, verify and calibrate the digital twin model until the fidelity is no less than the first threshold. S4. Digital twin model simulation and data output: Use finite element calculation software based on the digital twin model to simulate and calculate and output the structural response data of the sampling points under different load conditions; S5. Construction and verification of the deduction model; using the above digital twin model to simulate the response data and train the deep learning model; by inputting the response data of the sampling points, predict the full distribution of the response on the cross section.

2. The pavement structure response monitoring and fully distributed calculation method based on digital twin technology according to claim 1 is characterized in that: In S1, the digital twin model of the road structure is established using finite element simulation software to define the geometric dimensions of the road structure, the material composition of each layer, the material constitutive properties, the model boundary conditions and environmental factors.

3. The pavement structure response monitoring and full distribution calculation method based on digital twin technology according to claim 1 is characterized in that: In S2, the locations where the embedded sensors are arranged include the wheel track, the edge of the road surface, the center point of the cross section, the bottom of the asphalt surface layer and / or the bottom of the semi-rigid base layer.

4. The pavement structure response monitoring and fully distributed calculation method based on digital twin technology according to claim 1 is characterized in that: In S3, when the fidelity of the digital twin model output result is lower than a first threshold, the digital twin model is iteratively corrected to adjust its simulation parameters, including material constitutive model parameters, boundary conditions and / or mesh division.

5. The pavement structure response monitoring and fully distributed calculation method based on digital twin technology according to claim 1 is characterized in that: In S4, several types of vehicle loads are simulated to realize the calculation and output of the structural response data of the sampling points of the digital twin model under various vehicle load conditions.

6. The pavement structure response monitoring and fully distributed calculation method based on digital twin technology according to claim 1 is characterized in that: In S5, the network structure of the deep learning model is the Transformer deep learning model. The input of the model is the sampling point response data, and the output is the full distribution data of the response. The data set used for deep learning model training is the sampling point response data output by the digital twin model simulation in S4. The deep learning model is verified using the sampling point response simulation data under the action of external loads in the training data set. When the accuracy of the predicted response data output by the model reaches the second threshold, the deep learning model is considered to have been successfully trained.

7. The pavement structure response monitoring and full distribution calculation method based on digital twin technology according to claim 6 is characterized in that: The training steps of a deep learning model include: Dataset division: The response data output by the digital twin model simulation is divided into training set, test set and validation set according to the preset ratio; Data input processing: The training data is divided into two tensors of the same dimension. Tensor 1 is used as the input tensor to store the response data of the known sampling points of the specified number and distribution, representing the point response data collected by the embedded sensor; Tensor 2 is used as the target tensor to store the response data of the simulated output, representing the target value of the full distribution of the structural response; Model training: Adjust model training parameters, including learning rate, batch size, and / or number of rounds, and train the deep learning model using the training set for several rounds until the model converges; Model verification: Use response simulation data under load conditions other than the training set to input into the converged deep learning model, and compare the difference between the output response results and the sampling values ​​under the actual load. When the accuracy of the predicted response data output by the model reaches the second threshold, the deep learning model training is considered successful; otherwise, adjust the parameters of the deep learning model and retrain the model until the accuracy of the predicted response data output by the model reaches the second threshold.

8. The pavement structure response monitoring and full distribution calculation method based on digital twin technology according to claim 7 is characterized in that: The ratio of training set, test set and validation set in the data set partition is set to A:B:C; A is set to 0.6-0.98, B is set to 0.01-0.2, and C is set to 0.01-0.

2.

9. The pavement structure response monitoring and fully distributed calculation method based on digital twin technology according to claim 1 is characterized in that: In S3, the collection of measured data of structural response under the action of vehicle moving load includes: conducting several on-site loading tests under the same load on the road section to be monitored where the embedded sensors are buried, and collecting measured data of structural response of the vehicle during the entire driving process until the error of the measured response data is less than a third threshold.

10. The pavement structure response monitoring and full distribution calculation method based on digital twin technology according to claim 1 is characterized in that: In S3, the verification method of the simulation degree of the digital twin model is: use finite element simulation software to write a load definition subroutine, set the vehicle wheelbase, axle weight, number and size of tires, wheelbase, and driving speed to be consistent with the vehicle in the on-site loading test, load the simulated pavement structure model, and output the response data of the entire loading process.

Images