An iterative inversion control system for compaction construction quality based on machine-road collaboration

Abstract

The present invention is a compaction construction quality iterative inversion control system based on machine-road collaboration. The system includes a roller, a calculation and processing unit, and a mobile terminal. A displacement and acceleration sensor is installed on the roller. The specific iterative inversion process is as follows: Connect the data of the roller, the calculation and processing unit, and the mobile terminal in real-time to transmit the sensor signals of the roller and the return values of the calculation and processing unit, and provide immediate feedback during the compaction construction process. Establish a composite neural network, which includes two parts, neural network A and neural network B. The two parts are trained separately, and the output of neural network A is connected to the input of neural network B. Establish a numerical simulation model, connect the output of neural network B to the numerical simulation model, and perform iterative inversion correction. The system continuously monitors the amplitude change of the mechanical properties of the roadbed during the compaction process, and obtains the real-time compaction index of the soil through the iterative inversion correction finite element calculation embedded in the composite neural network.

Description

Technical Field

[0001] The present invention relates to the field of intelligent compaction construction control for roadbeds, and in particular to an iterative inversion control system for compaction construction quality based on machine-road collaboration. Background Art

[0002] During highway construction, problems such as over- or under-compaction during roadbed compaction can easily lead to road defects such as cracking and collapse during the post-construction operation phase. Therefore, real-time monitoring of compaction during the roadbed compaction phase, along with control of the subsequent compaction process and improved quality, can directly improve the quality of the resulting road.

[0003] In traditional compaction processes, quality control relies primarily on pre-construction plans and post-construction testing. However, in practice, it's often impossible to follow the pre-construction plan completely. Post-construction testing, performed after compaction is complete, doesn't provide a reliable basis for ongoing process control.

[0004] In intelligent compaction, thanks to the application of sensors, various data from the construction process can be obtained in real time. Many companies and researchers use the acceleration response of the roller-soil system during the rolling process (such as the amplitude of the vibrating wheel before and after rolling) as an indicator of compaction, and attempt to use function fitting to determine the correlation between this and soil properties (such as porosity and elastic modulus). However, soil is a complex mixture, and the particle arrangement, pore shape, and connection characteristics of the soil will change during the rolling process. Simply using a single indicator to determine compaction quality and link soil property parameters has limitations. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the technical problem that the present invention intends to solve is to provide an iterative inversion control system for compaction construction quality based on machine-road collaboration, which is used for real-time monitoring of roadbed compaction quality. The system can continuously monitor the changes in the amplitude of the roller-roadbed first harmonic signal caused by changes in the mechanical properties of the roadbed during the compaction process, and obtain real-time compaction indicators of the soil by combining physical parameters such as gradation and porosity that can be directly obtained from fill-related data, and embedding iterative inversion modified finite element calculations through a composite neural network.

[0006] The technical solution of the present invention to solve the technical problem is:

[0007] An iterative inversion control system for compaction quality based on machine-road collaboration includes a roller, a computing processing unit, and a mobile terminal. The roller is equipped with displacement and acceleration sensors. The specific iterative inversion process is as follows:

[0008] A. Interconnect the roller, the computing and processing unit, and the mobile terminal for data transmission in real time, and transmit the sensor signals of the roller and the return values of the computing and processing unit instantaneously during the compaction construction process for immediate feedback.

[0009] B. Establish a composite neural network, which includes two parts: neural network A and neural network B. The two parts are trained separately. Neural network A takes CMV, soil density, and gradation as input terminals and soil compaction degree as the output terminal.

[0010] Neural network B takes soil compaction degree, soil density, and gradation as input terminals and cohesion, friction angle, elastic modulus, and Poisson's ratio as output terminals.

[0011] The output of neural network A is connected to the input of neural network B.

[0012] C. Establish a numerical simulation model. Connect the output of neural network B to the numerical simulation model. After inputting the compaction degree and soil density after compaction given by the numerical simulation model into neural network B to obtain the new parameters of the old soil layer, use the new parameters of the old soil layer as the initial conditions for the next iterative calculation. Compare the measured compaction degree of the current iteration with the compaction degree calculated by the numerical simulation model. If the difference exceeds the difference threshold, inversely infer and correct the thickness of the model soil layer. Recalculate the compaction degree after compaction with the corrected thickness and then input it into neural network B for iterative calculation to predict the cohesion, friction angle, elastic modulus, and Poisson's ratio required for the next numerical simulation model input. Repeat this process to complete the iterative inversion and correction.

[0013] The training process of the composite neural network is as follows:

[0014] Before the system is used in construction, it is necessary to collect the vibration characteristics of the roller, and conduct relevant tests on the fill soil used in this batch of construction through the laboratory to obtain the inherent parameters of the soil and the parameters required for finite element modeling, and then train the composite neural network.

[0015] CMV is the continuous compaction index, which is calculated by the following formula from the amplitude A1 of the first harmonic signal collected by the roller and the amplitude A0 of the vibration fundamental frequency of the roller:

[0016]

[0017] where c is a constant coefficient.

[0018] Conduct soil mechanics tests on the fill soil samples in the laboratory to obtain a series of data of cohesion, friction angle, elastic modulus, soil density, gradation, and porosity corresponding to the corresponding compaction degrees of the subgrade fill soil used in the project.

[0019] Among them, the soil compaction degree is the ratio of the dry density of the soil to the standard maximum dry density; the initial compaction degree K0 of the unrolled initial fill soil is the ratio of the dry density of the sampled soil to the standard maximum dry density.

[0020] Taking CMV, soil density, gradation, and the corresponding soil compaction degree as a set of data, multiple sets of data are obtained from different batches of fill soil, constituting the dataset for training neural network A;

[0021] Taking soil compaction degree, soil density, gradation, and the corresponding cohesion, friction angle, elastic modulus, and Poisson's ratio as a set of data, multiple sets of data are obtained from different batches of fill soil, constituting the dataset for training neural network B;

[0022] Using their respective datasets to train their respective neural networks to obtain the trained composite neural network.

[0023] The specific process of the iterative inversion correction is as follows:

[0024] The first step is to establish a numerical simulation model

[0025] In the numerical simulation model, the roller model is simulated by a cylinder. The input includes the roller parameters such as the material, size, and vibration characteristics of the roller wheels. According to the requirements of the actual construction plan, the fill thickness and subgrade size are obtained. The boundary of the numerical simulation model is fixed by the subgrade size in the project and ignoring the lateral deformation generated during compaction. The calculation area is divided into blocks at intervals of 50m to characterize the lateral soil differences in the calculation. In the model, the same fill is divided into the same layer of soil units, and the initial thickness of each layer of soil units is the thickness of each fill. Different fills are modeled separately to characterize the changes in soil properties of each layer of fill during multiple fill compactions;

[0026] When setting the initial parameters of the soil material in the numerical simulation model, the unrolled initial fill compaction degree K0 is used as the input condition, and the parameters required for soil modeling within each block are represented by the parameters obtained after passing through neural network B from the unrolled initial fill compaction degree K0;

[0027] Connect the established numerical simulation model and the composite neural network using a modeling script program; then, automatically perform numerical simulation calculations;

[0028] The second step is the machine-road collaborative iterative inversion

[0029] After the composite neural network is trained, during the construction process, the gradation and soil density are input through a mobile device. At the same time, the amplitude of the first harmonic signal of the vibration monitoring of the roller-subgrade system during the nth compaction is collected using the displacement and acceleration sensors on the roller. After data processing, the continuous compaction index CMV is obtained. The CMV, gradation, and soil density are input into neural network A in the composite neural network to obtain the compaction degree K n ′ after this compaction. Take K n ′ as the measured compaction degree after this compaction in actual construction;

[0030] During the numerical simulation process, the grid deformation, displacements, and stresses of each layer after current compaction can be obtained. The thickness of each layer after grid deformation is recorded as the thickness of the soil layer after compaction. Then, based on the thickness, the compaction degree and soil density are converted to determine the compaction degree and soil density after compaction. At the same time, the thickness of each layer of soil after compaction calculated hereby is used as the initial thickness for the next numerical simulation model calculation. If new soil is filled in the next step, new soil filling units are added to the numerical simulation model and a frictional contact is established between two soil units to form a new numerical simulation model, which is used for the next iterative calculation. During the numerical simulation, the compaction degree K obtained by neural network A in the composite neural network from the CMV data collected by the roller during the nth compaction construction process is n compared with the compaction degree K c obtained from the numerical simulation model. c represents the number of compaction numerical simulation iterations, and c = n. If K c - K n '≤N, the current soil layer thickness, soil density after compaction, and compaction degree after compaction are output. If the difference between the two is not less than N, the soil layer thickness in the numerical simulation model needs to be corrected, and the corrected soil layer thickness is used as the output and the corresponding compaction degree after compaction is converted based on this in sequence. N is the difference threshold.

[0031] The specific process of converting the compaction degree and soil density based on the thickness is as follows:

[0032] The compaction degree K after numerical simulation c is calculated through the compaction degree K before compaction c-1 , the thickness t of the soil layer before compaction c-1 and the thickness t of the soil layer after compaction c as follows:

[0033]

[0034] The thickness t of the soil layer before compaction c-1 is the thickness of the soil layer in the previous iteration, and the thickness t of the soil layer after compaction c is the thickness of each layer of soil after compaction calculated through the numerical simulation model. When c = 1 for the first compaction, the thickness t of the soil layer before compaction c-1 is the thickness t0 of the newly filled soil in the roadbed at the construction site, and the compaction degree K0 before compaction is the compaction degree of the unrolled initial filled soil.

[0035] The soil density ρ of the soil after compaction c is calculated through the soil density ρ before compaction c-1 , the thickness t of the soil layer before compaction c-1 and the thickness t of the soil layer after compaction c as follows: After inputting the compactness and soil density after compaction given by the numerical simulation model into Neural Network B, the new parameters of the old soil layer are obtained, and the new parameters of the old soil layer are used as the initial conditions for the next iterative calculation.

[0036] The process of soil layer thickness correction is as follows: the difference is distributed to each layer of subgrade soil in the numerical simulation model according to the soil layer thickness of the numerical simulation model, and the thickness of each soil layer in the numerical simulation model is corrected to meet the compactness.

[0037] The difference threshold N is 5% of K c of.

[0038] Transmit the calculation results of the numerical simulation model to the roller display or to the display device for display, display the results on the mobile terminal and confirm whether to end the compaction construction. If continuing the construction, perform the numerical calculation of the machine-road collaborative iterative inversion according to the new parameters of the old soil layer after conversion and the newly filled soil layer.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0040] 1. The present invention combines the numerical simulation model with the composite neural network. The cohesion, friction angle, elastic modulus, and Poisson's ratio in the composite neural network are automatically transmitted to the numerical simulation model for subgrade compaction simulation. The output result of the numerical simulation model can also act on the composite neural network in reverse. The data in the process can be displayed in real time, and the compaction parameters and high-precision compaction quality indicators can be obtained in real time from the vibration response of the roller-subgrade system where the soil properties are constantly changing during the compaction process, and it can guide the compaction project in real time during the construction process.

[0041] 2. The correlation between the soil parameters such as CMV, fill soil density, gradation, cohesion, friction angle, elastic modulus, and Poisson's ratio obtained from the amplitudes of the first harmonic signal of the vibration of the roller-subgrade system and the amplitude of the fundamental frequency of the roller vibration in the composite neural network of the present invention is high, and it is continuously iteratively inverted and corrected during construction and calculation, with high calculation accuracy.

[0042] 3. The numerical simulation model is established parametrically. The numerical simulation method can accurately simulate the soil state, obtain the stress and strain at each part of the soil, and the numerical simulation results are output as the final results. The repetitive operations are automatically carried out through software script modeling calculations, ensuring accuracy while improving efficiency, shortening the time used to adapt to the construction speed. And this numerical simulation model can reflect the uneven characteristics of the soil distribution along the transverse and longitudinal directions at the construction site, and can reflect the influence of compaction on the microscopic structure of the soil and the resulting changes in soil properties, achieving the purpose of guiding the compaction construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a schematic diagram of the overall structure of the control system of the present invention;

[0044] Figure 2 This is the flowchart of the composite neural network of the present invention;

[0045] Figure 3 This is the flowchart of the iterative inversion correction numerical simulation of the present invention.

[0046] In the figure, 1 is a mobile terminal, 2 is a computing and processing unit, 3 is a roadbed, and 4 is a road roller. Specific implementation manners

[0047] The following gives specific embodiments of the present invention. The specific embodiments are only used to further illustrate the present invention in detail and do not limit the protection scope of this application.

[0048] The iterative inversion control system for compaction construction quality based on machine-road collaboration of the present invention can provide a reference index for compaction quality during the roadbed compaction process to guide the compaction construction.

[0049] The first part is the overall system layout and function overview.

[0050] The iterative inversion control system for compaction construction quality based on machine-road collaboration is mainly divided into three parts: the road roller 4, the computing and processing unit 2, and the mobile terminal 1. The three parts are connected through a mobile network to transmit data, and parameters can be input through the mobile terminal to monitor and control the calculation process in real time.

[0051] The compaction degree of the roadbed soil can be determined by the amplitude generated during the contact vibration compaction process between the vibrating wheel of the road roller and the soil of the roadbed 3 and the amplitude of the fundamental frequency of the road roller base.

[0052] The road roller mainly includes a signal acquisition module and a result feedback module. The signal acquisition module is mainly displacement and acceleration sensors. The result feedback module is used to transmit data and is connected to the display on the road roller. The transmitted data can be displayed on the display on the road roller or connected to other relevant systems. The displacement and acceleration sensors are a displacement sensor and an acceleration sensor, which are installed at adjacent positions on the road roller.

[0053] The computing and processing unit is mainly responsible for data processing, neural network training and recognition, and finite element modeling calculation, and can be deployed on a workstation or a supercomputing platform. A composite neural network is loaded in the computing and processing unit, and the composite neural network includes neural network A and neural network B.

[0054] The mobile terminal can input necessary parameters through an APP or a web page, view the vibration data uploaded by the road roller in real time, monitor the calculation status and calculation results. The mobile terminal can be a mobile device such as a smart phone, an APP, etc.

[0055] The second part is the training process of the composite neural network.

[0056] Before the system is used during construction, it is necessary to collect the vibration characteristics of the roller, and conduct relevant tests on the fill soil used in this batch of construction through the laboratory to obtain the inherent parameters of the soil body and the parameters required for finite element modeling, and train the composite neural network.

[0057] The vibration data collected based on the vibration characteristics of the roller is processed to obtain the continuous compaction index CMV. The data adopted in this embodiment is CMV, that is, the continuous compaction index (compactness value), which is defined by the amplitude of the first harmonic signal (A1) collected by the roller and the amplitude of the fundamental vibration frequency of the roller (A0):

[0058]

[0059] where c is a constant coefficient.

[0060] In the laboratory, geotechnical tests can be conducted on the fill soil samples to obtain a series of data such as cohesion, friction angle, elastic modulus, soil density, gradation, porosity, etc. corresponding to the corresponding compaction degrees of the subgrade fill soil used in the project.

[0061] Among them, the soil compaction degree is the ratio of the dry density of the soil body to the standard maximum dry density; the initial compaction degree K0 of the unrolled fill soil is the ratio of the dry density of the sampled soil body to the standard maximum dry density.

[0062] The composite neural network takes the compaction degree as the main intermediate parameter and can be divided into two parts: neural network A and neural network B. The two parts are trained separately. Taking CMV, soil density, gradation and the corresponding soil compaction degree as a set of data, multiple sets of data are obtained from different batches of fill soil to form the data set for training neural network A;

[0063] Taking the soil compaction degree, soil density, gradation and the corresponding cohesion, friction angle, elastic modulus, Poisson's ratio as a set of data, multiple sets of data are obtained from different batches of fill soil to form the data set for training neural network B.

[0064] Such as Figure 2 In, the first part neural network A is used to determine the correlation between the continuous compaction index CMV collected based on the vibration characteristics in the roller - subgrade system and the soil compaction degree. In this embodiment, the neural network A takes CMV, soil density, and gradation as the input end and the soil compaction degree as the output end. The CMV here is calculated from the amplitude of the first harmonic signal monitored by the vibration of the roller - subgrade system and the amplitude A0 of the fundamental vibration frequency of the roller.

[0065] The second part neural network B is used to determine the correlation between the soil compaction degree, gradation, soil density and cohesion, friction angle, elastic modulus, Poisson's ratio. It takes the soil compaction degree, soil density, and gradation as the input end and cohesion, friction angle, elastic modulus, Poisson's ratio as the output end.

[0066] The neural networks A and B can be network forms such as BP network, LSTM, MLP, transformer, etc.

[0067] The third part is the iterative inversion finite element calculation process combined with the composite neural network.

[0068] The first step is to establish a numerical simulation model

[0069] In the numerical simulation model, the roller model is simulated by a cylinder, and the parameters of the roller input by the mobile device are adopted, including the material, size, and vibration characteristics (amplitude of the fundamental vibration frequency of the roller) of the roller wheels. According to the requirements of the actual construction plan, the filling thickness and subgrade size are obtained. The boundary of the numerical simulation model is fixed by the subgrade size in the project and ignoring the lateral deformation generated during compaction. The calculation area is divided into blocks at intervals of 50 m to characterize the lateral soil differences in the calculation. In the model, the same filling is divided into the same layer of soil units, and the initial thickness of each layer of soil units is the thickness of each filling. Different fillings are modeled separately to characterize the changes in the soil properties of each layer of filling during multiple fill compactions.

[0070] When setting the initial parameters of the soil material in the numerical simulation model, the unrolled initial filling compaction degree K0 is used as the input condition, and the parameters required for soil modeling within each block are represented by the cohesion, friction angle, elastic modulus, and Poisson's ratio obtained by the neural network B from the unrolled initial filling compaction degree K0.

[0071] The established numerical simulation model and the composite neural network are connected by a modeling script program, so that the output of the composite neural network can be automatically given to the numerical simulation model, and the numerical simulation model automatically performs numerical simulation calculations.

[0072] The specific modeling script program is related to the finite element software used. Its main function is to perform automatic parametric modeling according to the given parameters, and corresponding script modeling can be written using most commercial software such as ANSYS and ABAQUS.

[0073] The second step is the machine-road collaborative iterative inversion

[0074] After the composite neural network is trained, during the construction process, the gradation and soil density are input through the mobile device. At the same time, the displacement and acceleration sensors on the roller are used to collect the amplitude of the first harmonic signal of the vibration monitoring of the roller-subgrade system during each compaction. After data processing, the continuous compaction index CMV is obtained. The CMV, gradation, and soil density are input into the neural network A in the composite neural network to obtain the compaction degree K n ′ after this compaction, which is used as the measured compaction degree after this compaction in actual construction.

[0075] During the numerical simulation process, the grid deformation, displacements and stresses at various locations after current compaction can be obtained. The thickness of each layer after grid deformation is recorded as the thickness of the soil layer after compaction. Then, based on the thickness, the compaction degree and soil density are converted to determine the compaction degree and soil density after compaction. At the same time, the thickness of each layer of soil after compaction calculated hereby is used as the initial thickness for the next numerical simulation model calculation. If new fill is to be carried out next time, new fill units are added to the numerical simulation model and a frictional contact is established between two soil units to form a new numerical simulation model, and the new numerical simulation model is used for the next iterative calculation. New fill units are established based on the thickness of the new fill of the subgrade at the construction site input by the mobile terminal (the control of the fill thickness is in the construction plan and is not the content of this invention).

[0076] The specific conversion process is as follows:

[0077] Since the lateral boundaries are fixed, the thickness change rate in the calculation is the soil volume change rate. The compaction degree (K c ) after compaction in the numerical simulation can be obtained by converting the compaction degree (K c-1 ) before compaction, the thickness (t c-1 ) of the soil layer before compaction and the thickness (t c ) of the soil layer after compaction. c represents the number of iterations of the compaction numerical simulation:

[0078]

[0079] The thickness t c-1 of the soil layer before compaction is the thickness of the soil layer in the previous iteration, and the thickness t c of the soil layer after compaction is the thickness of each layer of soil after compaction calculated by the numerical simulation model. When it is the first compaction (c = 1), the thickness t0 of the soil layer before compaction is the thickness of the new fill of the subgrade at the construction site, and the compaction degree K0 before compaction is the compaction degree of the unrolled initial fill.

[0080] Similarly, the soil densities (ρ c-1 , ρ c ) before and after soil compaction can also be obtained by conversion according to the thickness:

[0081]

[0082] After inputting the compaction degree and soil density after compaction given by the numerical simulation model into the neural network B, the new parameters of the old soil layer are obtained, and the new parameters of the old soil layer are used as the initial conditions for the next iterative calculation. The soil layer thickness and other result data such as stresses and strains at various locations are available for viewing on the mobile terminal.

[0083] During the numerical simulation, the compaction degree K n ′ obtained from the CMV data collected by the roller during the corresponding compaction construction process (c = n) through the neural network A in the composite neural network is compared with the compaction degree K obtained from the numerical simulation modelc Compare (n represents the actual number of rolling passes); if the numerical difference between the two is less than the difference threshold N (5% of the calculated value K c )(K c -K n ′ ≤ N), then output the current soil layer thickness, the density of the compacted soil body, and the degree of compaction after compaction; if the difference is not less than N, it is necessary to correct the soil layer thickness in the numerical simulation model, use the corrected soil layer thickness as the output, and calculate the corresponding degree of compaction after compaction according to the above conversion process. The degree of compaction in the numerical simulation is inversed with the degree of compaction output by Neural Network A in actual construction, and then input into Neural Network B for iterative calculation to predict the cohesion, friction angle, elastic modulus, and Poisson's ratio required for the next input of the numerical simulation model. Repeat this process to complete the iterative inversion correction.

[0084] The process of correcting the soil layer thickness is as follows: Distribute the difference to each layer of the subgrade soil body in the numerical simulation model according to the soil layer thickness of the numerical simulation model (that is, distribute the difference according to the number of soil layers and the thickness of each soil layer in the existing numerical simulation model) to correct the thickness of each soil layer in the numerical simulation model to meet the degree of compaction. The difference can be positive or negative, and the difference (K c -K′ n ) is weighted and distributed to each soil layer thickness.

[0085] In the third step, transmit the calculation result to the roller display or to other systems, display the result on the mobile device and confirm whether to end the compaction construction. If continuing the construction, perform numerical calculations as in the second step with the new parameters of the old soil layer (cohesion, friction angle, elastic modulus, Poisson's ratio, soil layer thickness, density) after conversion and the newly filled soil layer.

[0086] The parts not described in this invention are applicable to the prior art.

Claims

1. An iterative inversion control system for compaction quality based on machine-road collaboration, characterized in that: The system includes a roller, a computing processing unit, and a mobile terminal. The roller is equipped with displacement and acceleration sensors. The specific iterative inversion process is: A. Connect the roller, computing unit, and mobile terminal data to transmit the roller's sensor signals and the computing unit's return values in real time, providing instant feedback during the compaction process. B. Establishing a composite neural network, the composite neural network includes two parts, neural network A and neural network B, which are trained separately. Neural network A uses CMV, soil density, and gradation as input and soil compaction as output; Neural network B takes soil compaction, soil density, and gradation as input, and cohesion, friction angle, elastic modulus, and Poisson's ratio as output; The output of neural network A is connected to the input of neural network B; C. Establish a numerical simulation model and connect the output of neural network B to the numerical simulation model. Input the compaction degree and soil density after compaction given by the numerical simulation model into neural network B. The new parameters of the old soil layer are obtained and used as the initial conditions for the next iterative calculation. Compare the current measured compaction degree with the compaction degree calculated by the numerical simulation model. If the difference exceeds the difference threshold, inversely correct the model soil layer thickness. Recalculate the compaction degree after compaction with the corrected thickness and input it into neural network B for iterative calculation. Predict the cohesion, friction angle, elastic modulus, and Poisson's ratio required for the next numerical simulation model input. Repeat this process to complete the iterative inversion correction. The training process of the composite neural network is: Before the system is used in construction, the vibration characteristics of the roller must be collected, and relevant laboratory tests must be conducted on the fill used in this batch of construction to obtain the inherent parameters of the soil and the parameters required for finite element modeling and train the composite neural network; CMV is the continuous compaction index, which is calculated by the amplitude A1 of the first harmonic signal collected by the roller and the amplitude A0 of the roller vibration fundamental frequency using the following formula: Where c is a constant coefficient; Conduct soil mechanics tests on fill samples in the laboratory to obtain a series of data on cohesion, friction angle, elastic modulus, soil density, gradation, and porosity corresponding to the compaction degree of the roadbed fill used in the project; Among them, soil compaction is the ratio of soil dry density to standard maximum dry density; the uncompacted initial fill compaction K0 is the ratio of sampled soil dry density to standard maximum dry density; Taking CMV, soil density, gradation and corresponding soil compaction as a set of data, multiple sets of data are obtained from different batches of fill to form the data set for training neural network A; The soil compaction degree, soil density, gradation and corresponding cohesion, friction angle, elastic modulus and Poisson's ratio are used as a set of data. Multiple sets of data are obtained from different batches of fill soil to form the data set for training neural network B. The respective neural networks are trained using their respective data sets to obtain a trained composite neural network.

2. The iterative inversion control system for compaction quality based on machine-road collaboration according to claim 1 is characterized in that: The specific process of the iterative inversion correction is: The first step is to establish a numerical simulation model The roller model in the numerical simulation model uses a cylinder simulation, and the roller parameters including the roller wheel material, size, and vibration characteristics are input. The fill thickness and roadbed size are obtained according to the actual construction plan requirements. The boundary of the numerical simulation model is fixed by the roadbed size in the project and the lateral deformation generated by compaction is ignored. The calculation area is divided into blocks at intervals of 50m to represent the lateral soil differences in the calculation. In the model, the same fill is divided into the same layer of soil units. The initial thickness of each layer of soil units is the thickness of each fill. Different fills are modeled separately to represent the changes in soil properties of each layer during multiple fill compactions. When setting the initial parameters of the soil material in the numerical simulation model, the uncompacted initial fill compaction K0 is used as the input condition. The parameters required for soil modeling in each block are represented by the parameters obtained by the uncompacted initial fill compaction K0 after passing through the neural network B. The established numerical simulation model and the composite neural network are connected by using a modeling script program; after that, the numerical simulation calculation is automatically performed; The second step is machine-road collaborative iterative inversion After the training of the composite neural network is completed, the gradation and soil density are input through the mobile device during the construction process. At the same time, the displacement and acceleration sensors on the roller are used to collect the amplitude of the first harmonic signal of the roller-roadbed system vibration monitoring in the nth compaction. After data processing, the continuous compaction index CMV is obtained. The CMV, gradation and soil density are input into the composite neural network A to obtain the compaction degree K after this rolling. n ′, K n ' is the actual compaction degree measured after rolling in actual construction; During the numerical simulation process, the deformation of each layer of mesh after current compaction, the displacement and stress of each location can be obtained, and the thickness of each layer of mesh after deformation is recorded as the thickness of the soil layer after compaction. Then, the compaction degree and soil density are converted based on the thickness to determine the compaction degree and soil density after compaction. At the same time, the thickness of each layer of soil after compaction calculated in this way is used as the initial thickness for the next numerical simulation model calculation. If backfilling is performed next time, new backfill units will be added to the numerical simulation model and friction contact will be established between the two layers of soil units to form a new numerical simulation model. The new numerical simulation model will be used for the next iterative calculation. During the numerical simulation, the CMV data collected by the roller during the nth compaction construction process is used to obtain the compaction degree K after this rolling through the neural network A in the composite neural network. n ′ and the compaction degree K obtained by the numerical simulation model c For comparison, c represents the number of iterations of compaction numerical simulation, c = n; if K c -K n ′≤N, the current soil layer thickness, soil density after compaction and compaction degree after compaction are output; if the difference between the two is not less than N, the soil layer thickness in the numerical simulation model needs to be corrected, and the corrected soil layer thickness is used as the output and the corresponding compaction degree after compaction is converted in turn; N is the difference threshold.

3. The iterative inversion control system for compaction quality based on machine-road collaboration according to claim 2 is characterized in that: The specific process of converting the degree of compaction and soil density based on thickness is: Compaction degree K after numerical simulation c By compaction degree K before compaction c-1 , soil layer thickness before compaction t c-1 Thickness of soil layer after compaction t c The conversion is: Thickness of soil layer before compaction t c-1 is the soil layer thickness of the last iteration, and the thickness of the soil layer after compaction is t c is the thickness of each layer of soil after compaction calculated by the numerical simulation model; when it is the first compaction c = 1, the thickness of the soil layer before compaction t c-1 is the thickness of new fill soil on the roadbed at the construction site t0, and the compaction degree before compaction K0 is the compaction degree of the initial fill soil before rolling; Soil density after compaction ρ c The density of soil before compaction ρ c-1 , soil layer thickness before compaction t c-1 Thickness of soil layer after compaction t c The conversion is: After the compaction degree and soil density after compaction given by the numerical simulation model are input into the neural network B, the new parameters of the old soil layer are obtained and used as the initial conditions for the next iterative calculation.

4. The iterative inversion control system for compaction quality based on machine-road collaboration according to claim 2 is characterized in that: The soil layer thickness correction process is: distribute the difference to each layer of roadbed soil in the numerical simulation model according to the soil layer thickness of the numerical simulation model, and correct the thickness of each soil layer in the numerical simulation model to meet the compaction degree.

5. The iterative inversion control system for compaction quality based on machine-road collaboration according to claim 2 is characterized in that: The difference threshold N is K c 5% of.

6. The iterative inversion control system for compaction quality based on machine-road collaboration according to claim 2 is characterized in that: The calculation results of the numerical simulation model are transmitted to the roller display or to the display device for display. The results are displayed on the mobile terminal and it is confirmed whether to end the compaction construction. If the construction is to continue, the numerical calculation of the machine-road collaborative iterative inversion is continued according to the converted new parameters of the old soil layer and the newly added fill layer.

Images