A method, system and device for dynamic monitoring of subgrade modulus of resilience

Abstract

The application relates to the technical field of resilience modulus monitoring, in particular to a dynamic monitoring method, system and equipment for resilience modulus of a roadbed, which comprises the following steps: for each working condition, preparing multiple groups of soil sample, respectively performing multistage loading dynamic triaxial tests on the soil sample, and generating hysteresis curves of each strain level; obtaining discrete coefficients of each soil sample at each strain level, correcting initial dynamic elastic modulus extracted from the hysteresis curves, and establishing a nonlinear constitutive model responding to state changes of the soil; generating a simulation data set containing multiple characteristic vectors and resilience modulus thereof, and using the simulation data set to train an integrated learning model; and at a roadbed construction site, using the integrated learning model to predict the resilience modulus according to the obtained characteristic vectors, and evaluating the compaction quality of the roadbed. The application improves the monitoring accuracy of the resilience modulus of the roadbed in real time.

Description

Technical Field

[0001] This application relates to the field of resilient modulus monitoring technology, specifically to a dynamic monitoring method, system, and equipment for roadbed resilient modulus. Background Technology

[0002] Subgrade compaction is a crucial process in road engineering, and its quality directly affects the long-term performance and service life of the pavement. The core evaluation indicator is the resilient modulus of the subgrade top surface. The resilient modulus characterizes the compressive strength and deformation recovery capacity of the subgrade under load and is a fundamental basis for pavement structure design.

[0003] Traditional methods for monitoring the resilient modulus of roadbeds mainly rely on intelligent compaction technology to monitor the vibration response of the roller and thus assess the compaction state. However, due to the nonlinearity of the roadbed soil, its modulus and damping characteristics change significantly with strain level and stress state, leading to fundamental errors in inferring the soil state from the vibration response. Secondly, the influence of local soil inhomogeneity causes the same compaction process to elicit different vibration responses at different locations. This means that traditional methods fail to fully consider the nonlinearity and local inhomogeneity of the soil, making them insensitive to changes in local conditions and resulting in inaccurate monitoring of the roadbed resilient modulus, thus affecting the accuracy of the assessment of the roadbed compaction quality. Summary of the Invention

[0004] To address the aforementioned technical problems, a method, system, and device for dynamic monitoring of roadbed resilient modulus are provided to resolve the existing issues.

[0005] The solution to the technical problem presented in this application is to provide a method, system, and equipment for dynamic monitoring of the resilient modulus of roadbed, comprising the following steps: In a first aspect, embodiments of this application provide a method for dynamic monitoring of the resilient modulus of a roadbed, the method comprising the following steps: By controlling the physical and mechanical environmental parameters of the soil to form various test conditions, multiple sets of soil samples are prepared for each condition, and multi-level loading dynamic triaxial tests are performed on them to generate hysteresis curves for each strain level. The study analyzes the deviation of the hysteresis curves of soil samples at each strain level from their own fitted curves, as well as their deviation from the fitted curves of other soil samples under the same working conditions. The dispersion coefficients of each soil sample at each strain level are obtained. The initial dynamic elastic modulus extracted from the hysteresis curves is corrected, and the dynamic shear modulus and damping ratio are determined based on the theory of elasticity. The parameters of the nonlinear constitutive model are fitted, and a nonlinear constitutive model responding to changes in soil state is established. In finite element simulation, a nonlinear constitutive model is used as the material property, and the soil physical parameters and compaction process parameters are changed to simulate the vibration compaction process. The vibration response characteristics are analyzed and the compaction index characterizing the compaction state of the subgrade is calculated. Combined with the soil physical parameters and compaction process parameters, feature vectors are constructed. Simulated vehicle moving loads are applied to the subgrade that has completed the compaction simulation, the resilient modulus of the top surface of the subgrade is calculated, and a simulation dataset containing multiple feature vectors and their resilient moduli is generated for training the ensemble learning model. Based on the vibration response of the road roller, compaction process parameters, and soil physical parameters at the construction site, feature vectors are acquired in real time, and the resilient modulus is predicted using an integrated learning model to evaluate the compaction quality of the subgrade.

[0006] Preferably, obtaining the dispersion coefficients of each soil sample at each strain level includes: For each group of soil samples under each test condition, all data points distributed on the hysteresis curves of each strain level are fitted with an ellipse. The fitted ellipse is obtained and the fitting error is calculated as the first characteristic value. The difference between the hysteresis curve of each strain level and the fitted ellipse of the other soil samples at the same strain level is calculated as the relative error. The difference between the relative error and the first characteristic value is used as the second characteristic value. The correlation between the first characteristic value and the second characteristic value of each group of soil samples under each test condition and the other soil samples at all strain levels is calculated as the first correlation and the second correlation, respectively. The mean values ​​of the first correlation and the second correlation are positively mapped as the synergy coefficient. For each group of soil samples under each test condition, the first and second characteristic values ​​of each strain level are weighted and fused to obtain the weighted error. The discrete coefficient is positively correlated with the weighted error, but negatively correlated with the synergistic coefficient.

[0007] Preferably, the initial dynamic elastic modulus extracted from the hysteresis curve is corrected, including: For each group of soil samples under each test condition, the slope of the line connecting the two endpoints of the hysteresis curve at each strain level is extracted and used as the initial dynamic elastic modulus. Then the first The first test condition Group of soil samples in the first The adjusted dynamic elastic modulus corresponding to each strain level The calculation formula is: ,in, The initial dynamic elastic modulus, For the first The first test condition Group of soil samples in the first Normalized discrete coefficients for each strain level This is the preset adjustment range.

[0008] Preferably, the step of determining the dynamic shear modulus and damping ratio, fitting the parameters of the nonlinear constitutive model, and establishing a nonlinear constitutive model responding to changes in soil state includes: For each group of soil samples under each test condition, the dynamic shear modulus ratio and damping ratio are calculated based on the theory of elasticity using the adjusted dynamic elastic modulus at each strain level. The mean of the dynamic shear modulus and the mean of the damping ratio of all soil samples under each test condition at the same strain level are used as the dynamic shear modulus and damping ratio at each strain level under each test condition. The axial dynamic strain corresponding to each strain level is converted into shear strain. The shear strain, dynamic shear modulus, and damping ratio of each strain level are combined into a three-dimensional array. The three-dimensional array of all strain levels under each test condition is substituted into the Davidenkov model for fitting to obtain all model parameters. For the soil physical parameters and mechanical environment parameters under all test conditions, with them as independent variables and each model parameter as dependent variable, a predictive model of each model parameter with respect to the soil physical parameters and mechanical environment parameters is constructed, thereby establishing the Davidenkov model associated with the soil state as a nonlinear constitutive model.

[0009] Preferably, the process of constructing the feature vector is as follows: Each time the vibration compaction process is simulated, the vertical acceleration of the center of the vibration wheel at each moment is extracted to calculate the compaction index, which includes the compaction value and the compaction modulus. Among them, frequency domain analysis is performed on the vertical acceleration at all times to obtain the spectrum; the ratio of the energy corresponding to the second harmonic to the energy corresponding to the fundamental frequency in the spectrum is calculated, and the product of this ratio and the preset enhancement constant is used as the compaction value; Calculate the dynamic contact force between the vibratory wheel and the subgrade soil. The specific formula is as follows: ,in, For the mass of the vibrating wheel, Vertical acceleration, Indicates the amplitude of the excitation force. Represents the angular frequency of the excitation force. Let t be a cosine function, and let t represent the time variable. The compaction modulus is then solved iteratively using Lundberg contact theory. The soil physical parameters, compaction process parameters, and compaction indices from each vibration compaction simulation are combined to form a feature vector.

[0010] Preferably, the calculation of the resilient modulus of the top surface of the roadbed and the generation of a simulation dataset containing multiple feature vectors and their resilient moduli include: For the subgrade model after each vibration compaction simulation, the maximum dynamic deflection value of the top surface of the subgrade is extracted by simulating the moving load of vehicles. The resilient modulus of the top surface of the subgrade is calculated according to the equivalent deflection method. The feature vectors obtained from multiple simulations and their corresponding resilient moduli are combined to form a simulation dataset.

[0011] Preferably, the training process of the ensemble learning model is as follows: For the simulation dataset, the feature vectors of all samples and their elastic modulus are used as inputs to the Relief-F algorithm to calculate the weight of each feature component on the elastic modulus. Calculate the correlation between any two feature components in the feature vectors of all samples in the simulation dataset; if the correlation is greater than a preset judgment threshold, remove the feature component with the smallest importance weight from the two feature components; otherwise, retain the two feature components and define the filtered feature components as key components; use the key components in the feature vectors of all samples in the simulation dataset and the spring modulus as inputs to train the ensemble learning model.

[0012] Preferably, the evaluation of the compaction quality of the subgrade includes: during the compaction operation, based on the key components in the feature vectors acquired in real time, using the trained ensemble learning model to predict the resilient modulus, generating a resilient modulus distribution cloud map covering the entire subgrade; for the resilient modulus at each location in the cloud map, if it is greater than or equal to a preset threshold, then the compaction at this location is qualified; otherwise, there is under-compaction at this location.

[0013] Secondly, embodiments of this application also provide a dynamic monitoring system for the resilient modulus of a roadbed. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the dynamic monitoring method for the resilient modulus of a roadbed as described above.

[0014] Thirdly, this application also provides a dynamic monitoring device for the resilient modulus of a roadbed, wherein the device stores a computer program, and when the computer program is executed by a processor, it implements the steps of the dynamic monitoring method for the resilient modulus of a roadbed as described above.

[0015] This application has at least the following beneficial effects: This application generates various test conditions by controlling physical parameters such as soil moisture content and compaction degree, as well as mechanical environmental parameters such as confining pressure and loading frequency. This comprehensively covers various situations in actual engineering, providing rich data support for subsequent model building and parameter fitting. It generates stress-strain hysteresis curves at different strain levels, providing detailed data for analyzing the nonlinear behavior of soil. Furthermore, it analyzes the variation characteristics of hysteresis curves of multiple soil samples under the same test condition, calculating the dispersion coefficients of each soil sample at each strain level. The beneficial effect is that it considers the deviation of stress response data corresponding to different soil samples under the same test condition, thereby quantifying the solidification of soil materials. Some non-uniformity and experimental uncertainty exist; the initial dynamic elastic modulus extracted from the hysteresis curve is corrected. This has the advantage of considering the influence of local defects, minor non-uniformity, or test disturbances on the initial dynamic elastic modulus. The initial dynamic elastic modulus is corrected using the coefficient of variation, overcoming parameter extraction errors caused by local material non-uniformity and experimental dispersion, thus more accurately reflecting the actual mechanical behavior of the soil sample. Based on elasticity theory, the dynamic shear modulus and damping ratio are determined, and the parameters of the nonlinear constitutive model are fitted to establish a nonlinear constitutive model responding to changes in soil state. This has the advantage of highlighting the soil stress response in terms of strain amplitude, compaction state, and... The nonlinear characteristics under the coupled effects of multiple factors such as water content allow the nonlinear constitutive model to effectively overcome data discrepancies and distortions caused by local soil heterogeneity, sample preparation variability, and test noise by learning characteristic patterns under different disturbance modes. This enables stable and accurate identification of subgrade dynamic parameters. In finite element simulation, the nonlinear constitutive model is used as a material property, and soil physical parameters and compaction process parameters are changed to simulate the vibration compaction process. Compaction indices characterizing the subgrade compaction state are calculated and eigenvectors are constructed. Simulated vehicle movement loads are applied to the subgrade after compaction simulation, the resilient modulus of the subgrade top surface is calculated, and multiple eigenvectors are generated. The simulation dataset of the resilient modulus of the roadbed has the following advantages: First, it establishes a cross-strain level analytical mechanism between the large strain of vibration compaction and the small strain of vehicle load using a nonlinear constitutive model as a link. This makes it possible to directly and quantitatively obtain the resilient modulus of the roadbed using the vibration signal of the road roller during the construction stage. Second, by virtually reproducing the complete vibration compaction and vehicle load process in finite element simulation and systematically changing the physical parameters of the soil and the compaction process parameters, it simulates a variety of working conditions to generate a massive simulation dataset that covers multiple working conditions and has clear physical and mechanical mechanism support. This solves the problem of weak model generalization ability caused by the small sample size and incomplete working condition coverage of traditional field tests.Based on the vibration response of the road roller, compaction process parameters, and soil physical parameters at the construction site, feature vectors are acquired in real time. An ensemble learning model is then used to predict the resilient modulus and assess the compaction quality of the subgrade. The beneficial effect of this method is that real-time acquisition of process parameters and vibration signals from the construction site is input into a trained ensemble learning model, enabling real-time and continuous prediction and output of the subgrade resilient modulus at different compaction locations. This seamlessly integrates the offline-trained high-precision model into the construction process, achieving real-time, non-destructive, full-section intelligent perception and visualization assessment of the subgrade resilient modulus, thus improving the accuracy of real-time monitoring of the subgrade resilient modulus. Attached Figure Description

[0016] The following section provides a more detailed description of a dynamic monitoring method for the resilient modulus of a roadbed, in conjunction with the accompanying drawings.

[0017] Figure 1 A flowchart illustrating the steps of a method for dynamically monitoring the resilient modulus of a roadbed, as provided in this application embodiment; Figure 2 A flowchart illustrating the steps of the method for obtaining discrete coefficients provided in this application embodiment. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description, in conjunction with the accompanying drawings and implementation examples, provides a method, system, and equipment for dynamic monitoring of roadbed resilient modulus proposed in this application. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0020] Please see Figure 1 The diagram illustrates a flowchart of a dynamic monitoring method for the resilient modulus of a roadbed according to an embodiment of this application. The method includes the following steps: Step 1: Control the physical and mechanical environmental parameters of the soil to form various test conditions. For each test condition, prepare multiple sets of soil samples and conduct multi-level loading dynamic triaxial tests on them to generate hysteresis curves for each strain level.

[0021] The roadbed is the supporting structure of a road. To ensure the road's performance, the design and construction of the roadbed must not only meet the overall stability requirements but also strictly control its deformation. The magnitude of the roadbed's elastic deformation is mainly determined by its stiffness. The index characterizing the roadbed's dynamic stiffness is the resilient modulus. Therefore, the resilient modulus is an important parameter for evaluating the subgrade's bearing capacity and deformation capacity.

[0022] Because subgrade soil exhibits typical nonlinearity and strain dependence, its dynamic characteristics change significantly with the amplitude of dynamic strain. Therefore, its resilient modulus varies significantly under different strain levels: under large strain conditions such as vibration compaction, it exhibits low stiffness and a low soil modulus, displaying a easily deformable softened state; while under small strain conditions such as vehicle loads, it exhibits high stiffness, and the soil modulus recovers to a higher level, demonstrating good elastic bearing capacity. Furthermore, this modulus is significantly affected by physical states such as moisture content and compaction degree. To accurately describe this complex characteristic, the Davidenkov model is selected as the skeleton curve to quantify the stiffness changes of subgrade soil under different strains and states, thereby describing the nonlinear dynamic characteristics of the subgrade soil.

[0023] The specific process of the Davidenkov model is as follows: At the target roadbed survey site, standard compaction tests were conducted on the roadbed fill material to determine its maximum dry density and optimum moisture content; It should be noted that the compaction test is a method of understanding the compaction characteristics of soil by using a hammer to compact soil samples. By using different compaction efforts (hammer weight × drop distance × number of hammer blows) to hammer soil samples with different moisture contents and measuring the corresponding dry density, the maximum dry density and optimum moisture content can be obtained. The compaction test is a well-known technique and will not be described in detail here.

[0024] In this embodiment, the hammer weight and drop distance need to be adjusted according to the volume of the compaction barrel. The hammer weight can be selected as 2.5kg or 4.5kg, the drop distance can be selected as 305mm or 457mm, and the number of hammer blows is set to 56. As other implementation methods, the implementer can set them according to the actual situation.

[0025] To study the effects of four factors—moisture content, compaction degree, confining pressure, and loading frequency—on the dynamic properties of soil, a multi-factor, multi-level experimental design was adopted. Based on the maximum dry density and optimum moisture content, various test conditions were formed by controlling the physical and mechanical environmental parameters of the soil and corresponding soil samples were prepared. Multiple groups of soil samples were prepared in parallel under the same test condition. It should be noted that the physical parameters of the soil include moisture content and compaction degree, while the mechanical environmental parameters include confining pressure and loading frequency.

[0026] In this embodiment, the soil sample is a cylinder with a diameter of 10cm and a height of 15cm. As another implementation method, the implementer can set the dimensions according to the actual situation. Two sets of soil samples are prepared under the same test condition, i.e., two soil samples are prepared under the same test condition. As another implementation method, the implementer can set the dimensions according to the actual situation. Secondly, a multi-factor orthogonal test scheme is designed: Multiple influencing factors were selected as experimental variables, including moisture content, compaction degree, confining pressure, and loading frequency; three research levels were set for each influencing factor. Among them, the optimal moisture content (OMC) is used as the benchmark, and three levels of moisture content are set as OMC-2%, OMC, and OMC+2% to cover the typical fluctuation range of moisture content during construction. Based on the maximum dry density, three compaction levels were set: 90%, 93%, and 96%. The compaction degree is the ratio of the actual dry density to the maximum dry density, representing the typical state from under-compaction to full compaction. Three confining pressure levels were set at 30 kPa, 60 kPa, and 90 kPa to simulate different lateral constraint stress states experienced by the shallow, middle, and deep soil layers of the roadbed. The three loading frequency levels are 5Hz, 15Hz, and 30Hz, respectively, to simulate different dynamic loading conditions such as heavy vehicle load, medium frequency vibration, and high frequency compaction by vibratory rollers. Therefore, for the four-factor, three-level experimental system, in order to obtain comprehensive information while minimizing the number of experiments, a multi-factor orthogonal experiment is used, which requires only 9 experiments, including 9 experimental conditions. It should be noted that multi-factor orthogonal experiments are a well-known technique and will not be elaborated here; secondly, since the same test condition includes 2 soil samples, a total of 18 soil samples are included.

[0027] Soil samples were tested using a strain-controlled triaxial apparatus. A graded loading method under undrained conditions was adopted. According to the test conditions, a specified confining pressure was applied to the soil samples for isobaric consolidation to simulate the in-situ stress state in the subgrade. After consolidation, loads of different strain levels were applied sequentially. Cyclic loads were applied to the soil sample at each strain level. During the last application of cyclic load, the dynamic stress and dynamic strain of the soil sample were recorded in real time to generate hysteresis curves for each strain level. In this embodiment, the axial dynamic strain amplitude varies across the entire strain range, wherein the entire strain range is set as follows: Therefore, the axial dynamic strain amplitudes corresponding to each strain level are respectively: , , 0.5× , At each strain level, the soil sample is subjected to 10 cycles of loading. The first 9 cycles are to allow the soil response to reach a stable repeating state. In the 10th cycle, the dynamic stress and dynamic strain of the soil sample are recorded in real time. The acquisition frequency of dynamic stress and dynamic strain is 1000Hz. As an alternative implementation method, the implementer can set the frequency according to the actual situation.

[0028] It should be noted that when applying 10 cycles of load to the soil sample, the strain level of each cycle is constant; secondly, the process of plotting the hysteresis curve is as follows: for a soil sample, the dynamic stress and dynamic strain of the soil sample are recorded in real time based on each strain level, and a hysteresis curve is plotted with dynamic strain as the abscissa and dynamic stress as the ordinate. Therefore, the hysteresis curves of each group of soil samples at each strain level under each test condition can be obtained.

[0029] Step 2: Analyze the deviation of the hysteresis curves of the soil samples at each strain level from their own fitted curves, as well as from the fitted curves of other soil samples under the same working conditions. Obtain the dispersion coefficients of each soil sample at each strain level, correct the initial dynamic elastic modulus extracted from the hysteresis curves, and determine the dynamic shear modulus and damping ratio based on the theory of elasticity. Fit the parameters of the nonlinear constitutive model to establish a nonlinear constitutive model that responds to changes in soil state.

[0030] Secondly, due to the nonlinearity of soil, a nonlinear relationship also exists between each strain level and the stress in the soil response when plotting hysteresis curves. This nonlinear relationship may lead to different stress responses obtained from two groups of soil samples under the same strain level and test conditions. If only the experimental results of a single group of soil samples are used, the obtained curve may introduce errors due to the differences in nonlinear responses between soil samples.

[0031] Furthermore, the hysteresis curve is the relationship between dynamic stress and dynamic strain, reflecting the nonlinear behavior of soil under cyclic loading. Ideally, this curve is a symmetrical ellipse, indicating that the relationship between stress and strain is stable and predictable. However, in actual experiments, due to various factors such as local heterogeneity of the soil, minor variations in sample preparation, and testing noise, the hysteresis curve may deviate from the ideal state, mainly manifested in the following three deviations: In a single test, data points are not closely distributed on the ideal elliptical curve, but rather exhibit a certain degree of dispersion; the elliptical shape of the hysteresis curve may shift between different groups of soil samples at the same strain level; and the elliptical shape of the hysteresis curve may change between different strain levels.

[0032] The flowchart of the method for obtaining the discrete coefficients provided in this application embodiment is as follows: Figure 2 As shown.

[0033] Based on the above analysis, the fitting differences of the hysteresis curves of different groups of soil samples at the same strain level under each test condition are analyzed, specifically as follows: For each group of soil samples under each test condition, all data points distributed on the hysteresis curves of each strain level are fitted with an ellipse, the fitted ellipse is obtained and the fitting error is calculated, which is used as the first feature value. In this embodiment, the least squares method is used for ellipse fitting. The least squares method is a well-known technique and will not be described in detail here. Secondly, the fitting error is measured by the mean absolute error. The calculation of the mean absolute error is a well-known technique and will not be described in detail here.

[0034] For each group of soil samples under each test condition, the difference between the hysteresis curve of each strain level and the fitted ellipse of the other soil samples under the same strain level is calculated as the relative error, and the difference between the relative error and the first characteristic value is used as the second characteristic value. In this embodiment, since each test condition includes two sets of soil samples, denoted as sample A and sample B respectively, for the hysteresis curve of sample A at each strain level, the fitted ellipse of sample B at that strain level is obtained. For each strain level, the residual between the dynamic stress at each dynamic strain on the hysteresis curve of sample A and the corresponding predicted dynamic stress at the same dynamic strain on the fitted ellipse of sample B is calculated. The average absolute error is calculated using all residuals as the relative error. Then, the absolute value of the difference between the first characteristic value and the relative error is used as the second characteristic value.

[0035] It should be noted that the larger the first eigenvalue, the greater the difference between the fitted ellipse and the actual hysteresis curve, indicating a poor fitting effect and reflecting that the mechanical behavior of the soil at that strain level is more complex or has greater heterogeneity. Since the two groups of soil samples at the same strain level under each test condition may be affected by the heterogeneity of the soil samples, the center position or shape of the fitted ellipse may deviate. The larger this deviation is, that is, the larger the relative error is, it indicates that the mechanical behavior of this group of soil samples is inconsistent with that of other samples. In other words, the larger the obtained second eigenvalue is, it indicates that there may be greater heterogeneity or sample preparation differences between the two groups of soil samples under this test condition, which will affect the accuracy of the calculation of the resilient modulus.

[0036] Furthermore, the differences in the variation of the first and second characteristic values ​​of different soil samples under each test condition at all strain levels were analyzed, and the coefficient of variation was calculated, specifically: For each test condition, the correlation between the first characteristic value of each group of soil samples and the other soil samples under all strain levels is calculated as the first correlation. The correlation between the second characteristic value of each group of soil samples and the other soil samples under all strain levels is calculated as the second correlation. In this embodiment, the degree of correlation is measured by calculating the Pearson correlation coefficient. The calculation of the Pearson correlation coefficient is a well-known technique and will not be described in detail here. As other implementation methods, implementers may use other methods of the prior art, such as Euclidean distance, etc. This embodiment does not impose any special restrictions on this.

[0037] A positive mapping is performed between the mean values ​​of the first and second relevance, which is used as the synergy coefficient. In this embodiment, the specific process of positive mapping is as follows: the sum of the mean and the value 1 is used as the result of positive mapping, and through the process of positive mapping, the result of positive mapping is made to be a non-negative number.

[0038] It should be noted that the smaller the first correlation, the lower the similarity of the first characteristic value of each group of soil samples to other samples. The smaller the second correlation, the lower the similarity of the second characteristic value of each group of soil samples to other samples. This reflects that there are significant differences in the deviation of the mechanical behavior of different soil samples under the same test conditions at different strain levels. The smaller the obtained synergy coefficient, the greater the difference in the mechanical behavior of different soil samples under the same test conditions at different strain levels.

[0039] For each group of soil samples under each test condition, the first and second characteristic values ​​at each strain level are weighted and fused to obtain the weighted error. Under each test condition, the coefficient of variation of each group of soil samples at each strain level was positively correlated with the weighted error, but negatively correlated with the coefficient of synergy. It should be noted that a positive correlation means that the dependent variable increases as the independent variable increases and decreases as the independent variable decreases; a negative correlation means that the dependent variable decreases as the independent variable increases and increases as the independent variable decreases.

[0040] In this embodiment, the specific formula for calculating the discrete coefficients is as follows: in, For the first The first test condition Group of soil samples in the first The coefficient of variation for each strain level For the first The first test condition Group of soil samples in the first The first characteristic value of each strain level For the first The first test condition Group of soil samples in the first The second characteristic value of each strain level, For the first The first test condition Synergy coefficient of the soil samples. For the first Preset weights for each strain level, To avoid a denominator of 0, the parameter tuning factor is set to 0.1 in this embodiment. In other implementations, the implementer can set the value according to the actual situation. Indicates the weighted error. This represents the normalization function; in this embodiment, the maximum-minimum normalization method is used to normalize the coefficients of variation of all soil samples under all test conditions, wherein the range of the normalized coefficients of variation is... Among them, the maximum-minimum normalization method is a well-known technique and will not be elaborated here.

[0041] The weights of the five stress levels are 0.1, 0.15, 0.2, 0.25, and 0.4, respectively. As other implementation methods, implementers can set them according to the actual situation.

[0042] It should be noted that the larger the synergy coefficient, the more similar the behavior patterns of the two groups of soil samples are throughout the strain history under the same test conditions, and the better the synergy. The larger the weighted error, the greater the combined influence of the first and second eigenvalues ​​at each strain level, indicating that the mechanical behavior of the soil sample is more complex, reflecting that the stress data quality of the soil sample at that strain level is worse and the inherent unreliability is higher. The higher the dispersion coefficient, the worse the quality of the dynamic stress and dynamic strain data recorded at that strain level, and the higher its own dispersion, higher essential differences between samples, and isolated behavior patterns, reflecting that the mechanical behavior of the soil sample has greater dispersion and heterogeneity at different strain levels.

[0043] Therefore, considering the local heterogeneity of the soil and the random dispersion of the experimental data, the initial dynamic elastic modulus estimated under a single test is adjusted using the coefficient of variation, specifically as follows: For each group of soil samples under each test condition, the slope of the line connecting the two endpoints of the hysteresis curve at each strain level is extracted and used as the initial dynamic elastic modulus. It should be noted that the endpoints are the two data points on the hysteresis curve where the horizontal coordinates reach their maximum and minimum values, respectively, and the slope between the two points is taken as the initial dynamic elastic modulus.

[0044] The formula for calculating the adjusted dynamic elastic modulus is: in, For the first The first test condition Group of soil samples in the first The adjusted dynamic elastic modulus corresponding to each strain level The initial dynamic elastic modulus, For the first The first test condition Group of soil samples in the first Normalized discrete coefficients for each strain level To preset the adjustment range, avoid Too large and dominant Therefore, the preset adjustment range is set to... In this embodiment, the preset adjustment range is 0.15. In other implementation methods, the implementer can set it according to the actual situation.

[0045] It should be noted that since the initial dynamic elastic modulus is calculated based on the slope of the line connecting the tips of the hysteresis curve, when the soil structure is heterogeneous (i.e., the larger the coefficient of variation), relying solely on the slope of the tips may overestimate the actual bearing capacity of the soil. Therefore, when the data quality is poor and the soil exhibits high dispersion, the initial dynamic elastic modulus becomes less reliable and more likely to be artificially inflated. The reduction force is measured to more accurately reflect the actual mechanical behavior of the soil sample.

[0046] For each group of soil samples under each test condition, the dynamic shear modulus ratio and damping ratio are calculated using the adjusted dynamic elastic modulus at each strain level. In this embodiment, based on the theory of elasticity, the mapping relationship between dynamic shear modulus, dynamic elastic modulus, and Poisson's ratio is established. Therefore, the dynamic shear modulus... The calculation formula is: in, The adjusted dynamic elastic modulus. The Poisson's ratio is the soil mass. The Poisson's ratio can be determined empirically based on the soil type. For example, the Poisson's ratio of saturated clay is 0.5, that of sand is 0.3 to 0.35, and that of loess is 0.44. As other implementation methods, the implementer can set the value according to the actual situation.

[0047] Damping ratio The hysteresis curve is usually measured by its area, and the specific calculation formula is as follows: in, The area enclosed by the hysteresis curve. It is the elastic strain energy.

[0048] It should be noted that the calculation of dynamic shear modulus and damping ratio are well-known techniques and will not be elaborated here.

[0049] The mean dynamic shear modulus and the mean damping ratio of all soil samples under the same strain level were calculated for each test condition, and were used as the dynamic shear modulus and damping ratio for each strain level under each test condition. Furthermore, the Davidenkov model is used to describe the nonlinear relationship between the dynamic shear modulus ratio and the damping ratio as a function of shear strain, and its expressions are as follows: in, For the maximum dynamic shear modulus, For nominal shear strain, B and B are the fitting coefficients related to soil properties; The axial dynamic strain corresponding to each strain level is converted into shear strain, and the shear strain, dynamic shear modulus and damping ratio of each strain level under each test condition are combined into a three-dimensional array. For each test condition, the three-dimensional arrays of all strain levels were substituted into the Davidenkov model for fitting, and all model parameters were obtained. These model parameters include: maximum dynamic shear modulus, nominal shear strain, and fitting coefficients. and B; For all test conditions, soil physical parameters and mechanical environment parameters are used as independent variables and model parameters as dependent variables. Predictive models of each model parameter with respect to soil physical parameters and mechanical environment parameters are constructed, thereby establishing nonlinear constitutive models related to test variables. In this embodiment, the independent variables are moisture content, compaction degree, confining pressure, and loading frequency. Each model parameter is used as the dependent variable. Multiple linear stepwise regression is performed on the independent and dependent variables of all test conditions to obtain the multiple regression equation as the prediction model. Multiple linear stepwise regression is a well-known technique and will not be described in detail here.

[0050] Therefore, based on the soil state in real time, i.e., the soil physical parameters and mechanical environment parameters, these can be substituted into the corresponding prediction models to obtain all the model parameters of the Davidenkov model. Then, by substituting the model parameters back into the Davidenkov formula, the nonlinear constitutive model of the subgrade soil state can be obtained.

[0051] It should be noted that the Davidenkov model is a well-known technique and will not be elaborated upon here.

[0052] Thus, the nonlinear constitutive model of the subgrade soil state is obtained.

[0053] Step 3: In the finite element simulation, the nonlinear constitutive model is called as the material property, and the soil physical parameters and compaction process parameters are changed to simulate the vibration compaction process. The vibration response characteristics are analyzed and the compaction index characterizing the compaction state of the subgrade is calculated. Combined with the soil physical parameters and compaction process parameters, feature vectors are constructed. The simulated vehicle moving load is applied to the subgrade that has completed the compaction simulation, the resilient modulus of the top surface of the subgrade is calculated, and a simulation dataset containing multiple feature vectors and their resilient moduli is generated for training the ensemble learning model.

[0054] To accurately reflect the nonlinear dynamic characteristics of subgrade soil as strain level changes, finite element simulation was performed on the subgrade, and the VUMAT subroutine was developed based on the Fortran language. The calculation logic of the VUMAT subroutine is as follows: Elastic parameter update: ABAQUS software was selected as the simulation platform, and the above nonlinear constitutive model was embedded. In each increment step, the main program input the real-time strain increment, and the subroutine calculated the shear strain according to the strain state of the previous increment step. According to the embedded nonlinear constitutive model, the tangent shear modulus and damping ratio were updated in real time. The updated tangent shear modulus and damping ratio were used as inputs, and the tangent bulk modulus was set. The elastic test stress was calculated using the generalized Hooke's law. The tangential bulk modulus is typically 2 to 5 times the tangential shear modulus. In this embodiment, the tangential bulk modulus is 2 times the tangential shear modulus.

[0055] Plastic Correction in Principal Stress Space: To characterize the plastic deformation of the subgrade under large excitation force, the Mohr-Coulomb yield criterion is introduced. This criterion defines the yield surface by soil cohesion and internal friction angle. To solve the problem of singularity points in the yield surface in general stress space, this embodiment adopts the principal stress space reflection algorithm to convert the elastic trial stress into principal stress vector. According to the Mohr-Coulomb yield criterion, it is determined whether the principal stress state exceeds the yield surface. If it does, plastic yielding occurs. According to the flow law and setting the shear-diffusivity angle of the soil, the stress that exceeds the boundary is pulled back to the yield surface. The specific pull-back position is determined by calculating the discriminant factor to determine the specific pull-back area. The updated principal stress is iteratively solved. It should be noted that the specific stress pull-back location is divided into four regions: the reference yield surface, the intersection of the two yield surfaces, and the vertex of the yield surface. In this embodiment, the range of the internal friction angle of the soil is: In this embodiment, the internal friction angle of the soil is set to 30 degrees, and the value range of the soil cohesion is [value missing]. In this embodiment, the soil cohesion is taken as 30 kPa, and the soil shear-swell angle ranges from [value missing]. In this embodiment, the shear-expansion angle of the soil is 10 degrees. In other implementation methods, the implementer can set the value according to the actual situation.

[0056] The elastic test stress, soil internal friction angle, soil cohesion, and soil shear-diffusivity angle are used as inputs. The yield surface parameters and the number of iterations are set, and the Newton-Raphson iterative algorithm is used to calculate the discriminant factor. In this embodiment, the range of values ​​for the yield surface parameter is: In this embodiment, the yield surface parameter is set to The number of iterations is set to 30. As another implementation method, the implementer can set it according to the actual situation. Secondly, the Mohr-Coulomb yield criterion, principal stress space mapping algorithm and Newton-Raphson iterative algorithm are all well-known technologies, and will not be described in detail here.

[0057] Damping force introduction and stress update: Taking the dynamic damping ratio and loading frequency as inputs, based on Rayleigh linear damping theory, the dynamic damping ratio calculated by the Davidenkov constitutive model is converted into the stiffness proportional damping coefficient; taking the stiffness proportional damping coefficient, tangent shear modulus and tangent bulk modulus as inputs, the actual stress tensor of the current analysis step is calculated using the existing incremental algorithm and fed back to the main program.

[0058] It should be noted that Rayleigh's linear damping theory is a well-known technique and will not be elaborated upon here.

[0059] Therefore, a finite element model of the roadbed was established. In order to accurately simulate the propagation of vibration waves and eliminate the interference of boundary reflected waves, the core loading region of the model adopted reduced integral solid elements (C3D8R), and the surrounding boundary was defined as infinite elements (CIN3D8). By embedding a user-defined material subroutine into the finite element solver, the subroutine ensures that the stiffness and damping energy of the soil are updated in real time and nonlinearly with strain level and physical state during simulation. The vibration compaction process of the roadbed is simulated by applying a periodic excitation force to the vibratory wheel of the road roller, and the vertical acceleration of the center of the vibratory wheel at each moment is extracted. In this embodiment, the periodic excitation force is: ,in, Indicates the amplitude of the excitation force. Represents angular frequency. The function is a sine function, and t represents the time variable. Secondly, the simulation duration is 7s, of which the first 2s is the self-weight settlement equilibrium and the last 5s is the stable vibration. Therefore, the vertical acceleration at each moment in the stable vibration stage is extracted.

[0060] The compaction index is calculated using vertical acceleration, and the compaction index includes compaction value and compaction modulus; The calculation process for the compaction value is as follows: Frequency domain analysis was performed on the vertical acceleration at all times to obtain the spectrum. In this embodiment, the Fast Fourier Transform (FFT) is used to obtain the spectrum. The FFT is a well-known technique and will not be described in detail here.

[0061] Calculate the ratio of the energy corresponding to the second harmonic to the energy corresponding to the fundamental frequency in the spectrum diagram, and multiply it by the preset enhancement constant as the compaction value; In this embodiment, the second harmonic refers to twice the fundamental frequency. Furthermore, the enhancement constant is set to 300. As for other implementation methods, the implementer can set it according to the actual situation.

[0062] Secondly, the calculation process for the compaction modulus is as follows: First, calculate the dynamic contact force between the vibratory wheel and the subgrade soil. The specific formula is as follows: in, For dynamic contact force, For the mass of the vibrating wheel, Vertical acceleration, Indicates the amplitude of the excitation force. The frequency of the excitation force is represented by the angular frequency. Let be a cosine function, and t represent the time variable; where vertically upward is defined as the positive direction. Since the compaction modulus of the soil is unknown, an initial compaction modulus is set. In this embodiment, the initial compaction modulus is set to 50 MPa. ; The contact width between the vibratory wheel and the soil is solved iteratively using the Lundberg formula. Specifically: in, For the first Contact width in the next iteration The radius of the vibrating wheel, Poisson's ratio of the soil For the first Compaction modulus of soil in the next iteration The axial width of the vibrating wheel; Contact width Substituting into the Lundberg settlement formula for back calculation: in, For the first Compaction modulus of soil in the next iteration This represents the amount of soil deformation. It is a logarithmic function with the natural constant as the base; Among them, the soil deformation is the settlement of the center of the vibrating wheel during the stable vibration stage output by the simulation platform during the simulation process, which is the instantaneous vertical displacement of the soil.

[0063] The iteration is stopped when the relative error between the compaction modulus calculated in two adjacent iterations is less than the preset tolerance. Specifically: relative error The calculation formula is: ; like If the iteration converges, then let ; like Then let Continue iterating through the above steps; among them, Preset tolerance; In this embodiment, a preset tolerance is used. The value is 0.01. As for other implementation methods, the implementer can set it according to the actual situation.

[0064] It should be noted that Lundberg's theory is a well-known technique and will not be elaborated upon here.

[0065] In the simulation platform, the vibratory wheel of the road roller is removed and a standard vehicle axle load is applied. The moving load is simulated using the VDLOAD subroutine. The maximum dynamic deflection value of the top surface of the roadbed is extracted. Combined with the equivalent deflection method, the resilient modulus of the top surface of the roadbed is calculated in reverse. In this embodiment, the standard vehicle axle load is a single axle dual wheel set with a ground pressure of 0.7MPa. The VDLOAD subroutine is used to control the wheel load to slide across the road surface at 20m / s. As for other implementation methods, the implementer can set it according to the actual situation. Secondly, the vertical displacement time history curve of the roadbed surface directly below the load center point is monitored in real time through simulation. The maximum downward displacement peak value in the curve represents the maximum dynamic deflection value.

[0066] Secondly, based on Boussinesq's theory, the resilient modulus is calculated using the equivalent deflection method, specifically as follows: in, The resilient modulus of the top surface of the roadbed. This refers to the tire's ground pressure. The equivalent circle radius of the tire. Poisson's ratio of the soil This represents the maximum dynamic deflection value.

[0067] It should be noted that the Boussinesq theory is a well-known technique and will not be elaborated upon here.

[0068] The vibration compaction process of the roadbed is simulated by changing the input soil physical parameters and compaction process parameters to calculate the compaction index. The soil physical parameters, compaction process parameters and compaction index input each time are combined into a feature vector, and the resilient modulus under the working condition is calculated. By repeating the above simulation process, multiple feature vectors and their corresponding resilient moduli are generated to form a simulation dataset. Each feature vector and its corresponding resilient modulus in the simulation dataset is used as a sample. It should be noted that the physical parameters of the soil include moisture content and degree of compaction, while the compaction process parameters are the amplitude and frequency of the excitation force; among them, the excitation frequency is... ,in, ω is the angular frequency of the excitation force; the simulation dataset contains a large number of operating conditions.

[0069] Furthermore, using the simulation dataset, the feature components in the feature vector are filtered, specifically as follows: For the simulation dataset, the feature vectors of all samples and their elastic modulus are used as inputs to the Relief-F algorithm to calculate the weight of each feature component on the elastic modulus. In this embodiment, the Relief-F algorithm is a well-known technology and will not be described in detail here. Secondly, the Relief-F algorithm sets the number of nearest neighbor samples to 8 and the number of iterations to an integer multiple of the number of samples in the simulation dataset. In this embodiment, the number of samples in the simulation dataset is 100, so the number of iterations is 200. As other implementation methods, implementers can set it according to the actual situation. It should be noted that the feature components are the components in the feature vector.

[0070] Calculate the correlation between any two feature components in the feature vectors of all samples in the simulation dataset; In this embodiment, the degree of correlation is measured by calculating the Pearson correlation coefficient, which is a well-known technique and will not be described in detail here.

[0071] If the relevance is greater than the preset judgment threshold, the feature component with the smallest importance weight is removed from any two feature components; otherwise, the two feature components are retained, and the filtered feature components are defined as key components. In this embodiment, the preset judgment threshold is 0.85. As for other implementation methods, the implementer can set it according to the actual situation.

[0072] Furthermore, to overcome the limitations of single machine learning algorithms in terms of generalization ability and stability, in this embodiment, the ensemble learning model is a constructed two-layer stacked ensemble regression model, which is trained using key components in the simulation dataset. Specifically, the two-layer stacked ensemble regression model is as follows: The first layer of base learners selects three algorithms with strong differences. In this embodiment, ANN, LightGBM, and RF are selected as the three base learners to capture the feature patterns of different dimensions in the data. For LightGBM: Set the maximum depth of the tree and the number of leaves. In this embodiment, the maximum depth of the tree ranges from [value missing]. Therefore, it is set to 7, and the range of the number of leaves is [value missing]. Therefore, it is set to 63. As another implementation method, the implementer can set it according to the actual situation. The key components in the feature vector are used as input, and LightGBM is used to obtain the prediction result. For ANN: Set the number of hidden layer neurons. In this embodiment, the number of hidden layer neurons is 1 to 2 times the total number of components in the feature vector, so it is set to 12. Set the optimizer to Adam and the loss function to cross-entropy loss. As for other implementation methods, the implementer can set them according to the actual situation. Take the key components in the feature vector as input and use ANN to obtain the prediction result. For RF, the number of decision trees, the maximum depth of the trees, and the minimum number of samples required for node splits are set; in this embodiment, the range of the number of decision trees is: Therefore, it is set to 100, and the maximum depth of the tree can be within the range of values. Therefore, it is set to 20, and the range of the minimum number of samples for node splitting is... Therefore, it is set to 3. As for other implementation methods, the implementer can set it according to the actual situation. The key components in the feature vector are used as input, and the prediction result is obtained by using RF. The second-layer meta-learner uses ridge regression as the meta-learner, taking the prediction results of the three base learners as secondary input features, setting the regularization strength, and performing weighted fusion through regularized linear regression to finally output the predicted resilience modulus corresponding to the roadbed; wherein, in this embodiment, the value range of the regularization strength is [value missing]. Therefore, it is set as As another implementation method, the implementer can set it according to the actual situation.

[0073] The key components of the feature vectors of all samples in the simulation dataset and the spring modulus are used as inputs to train the two-layer stacked ensemble regression model, which is then used as the trained ensemble learning model.

[0074] It should be noted that ANN, LightGBM, and RF models are all well-known technologies and will not be elaborated upon here.

[0075] Thus, the trained ensemble learning model is obtained.

[0076] Step 4: Based on the vibration response of the road roller, compaction process parameters, and soil physical parameters at the construction site, acquire feature vectors in real time, use an integrated learning model to predict the resilient modulus, and evaluate the compaction quality of the subgrade.

[0077] For the roadbed at the construction site, by installing a high-precision RTK-GPS positioning system and vibration acceleration sensors on the vibratory roller, the following data can be collected in real time during the compaction operation of the roadbed, including: The location coordinates and travel speed of the road roller are collected in real time to accurately locate the current compaction position and record the movement status of the road roller; The vertical acceleration of the vibratory roller is collected; and the compaction process parameters of the roller, including the excitation frequency and excitation force amplitude, are read in real time from the roller control bus. Using the vertical acceleration of the vibratory wheel at each moment and within the preset time period before it, the compaction index is calculated according to the above process; In this embodiment, the preset time period is 5 seconds. As for other implementation methods, the implementer can set it according to the actual situation.

[0078] The physical parameters of the subgrade soil are obtained by TDR method or oven drying method; It should be noted that the TDR method or drying method is a well-known technology and will not be elaborated upon here.

[0079] The compaction process parameters, compaction indexes, and soil physical parameters acquired in real time are used to form a feature vector. Its key components are used as the input of the trained double-layer stacked integrated regression model to predict the resilient modulus in real time. During compaction operations, the continuously predicted resilient modulus is mapped to the compaction location of the subgrade, and a color coding technique is used to generate a resilient modulus surface distribution cloud map covering the entire subgrade. In this embodiment, the color coding technology is a well-known technology and will not be described in detail here.

[0080] The resilient modulus at various locations in the resilient modulus distribution cloud map is determined to assess the compaction quality of the roadbed. Specifically: For the resilient modulus distribution cloud map, if the resilient modulus at each compaction location is greater than or equal to a preset threshold, then the compaction at this location is qualified; otherwise, there is undercompaction at this location. In this embodiment, the process of obtaining the preset threshold is as follows: it is set according to the roadbed design specifications or construction requirements. Therefore, the preset threshold is 40 MPa. As for other implementation methods, the implementer can set it according to the actual situation.

[0081] Construction workers supplement the compaction of under-compacted areas, thereby achieving real-time, efficient, and precise control over the quality of roadbed construction.

[0082] This application also provides a dynamic monitoring system for the resilient modulus of a roadbed, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described methods for dynamic monitoring of the resilient modulus of a roadbed.

[0083] Based on the same inventive concept as the above method, this application embodiment also provides a dynamic monitoring device for the resilient modulus of a roadbed. The device stores a computer program, and when the computer program is executed by a processor, it implements the steps of any one of the above-described methods for dynamic monitoring of the resilient modulus of a roadbed.

[0084] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0085] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0086] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of this application, without departing from the content of the technical solution of this application, shall fall within the protection scope of the technical solution of this application.

Claims

1. A method for dynamic monitoring of the resilient modulus of roadbed, characterized in that, The method includes the following steps: Multiple test conditions were constructed by controlling the physical and mechanical environmental parameters of the soil. For each condition, multiple sets of soil samples were prepared and subjected to multi-level loading dynamic triaxial tests to generate hysteresis curves for each strain level. The study analyzes the deviation of the hysteresis curves of soil samples at each strain level from their own fitted curves, as well as their deviation from the fitted curves of other soil samples under the same working conditions. The dispersion coefficients of each soil sample at each strain level are obtained. The initial dynamic elastic modulus extracted from the hysteresis curves is corrected, and the dynamic shear modulus and damping ratio are determined based on the theory of elasticity. The parameters of the nonlinear constitutive model are fitted, and a nonlinear constitutive model responding to changes in soil state is established. In finite element simulation, a nonlinear constitutive model is used as the material property, and the soil physical parameters and compaction process parameters are changed to simulate the vibration compaction process. The vibration response characteristics are analyzed and the compaction index characterizing the compaction state of the subgrade is calculated. Combined with the soil physical parameters and compaction process parameters, feature vectors are constructed. Simulated vehicle moving loads are applied to the subgrade that has completed the compaction simulation, the resilient modulus of the top surface of the subgrade is calculated, and a simulation dataset containing multiple feature vectors and their resilient moduli is generated for training the ensemble learning model. Based on the vibration response of the road roller, compaction process parameters, and soil physical parameters at the construction site, feature vectors are acquired in real time, and the resilient modulus is predicted using an integrated learning model to evaluate the compaction quality of the subgrade.

2. The method for dynamic monitoring of roadbed resilient modulus as described in claim 1, characterized in that, The obtained dispersion coefficients of each soil sample at each strain level include: For each group of soil samples under each test condition, all data points distributed on the hysteresis curves of each strain level are fitted with an ellipse. The fitted ellipse is obtained and the fitting error is calculated as the first characteristic value. The difference between the hysteresis curve of each strain level and the fitted ellipse of the other soil samples at the same strain level is calculated as the relative error. The difference between the relative error and the first characteristic value is used as the second characteristic value. The correlation between the first characteristic value and the second characteristic value of each group of soil samples under each test condition and the other soil samples at all strain levels is calculated as the first correlation and the second correlation, respectively. The mean values ​​of the first correlation and the second correlation are positively mapped as the synergy coefficient. For each group of soil samples under each test condition, the first and second characteristic values ​​of each strain level are weighted and fused to obtain the weighted error. The discrete coefficient is positively correlated with the weighted error, but negatively correlated with the synergistic coefficient.

3. The method for dynamic monitoring of roadbed resilient modulus as described in claim 1, characterized in that, The initial dynamic elastic modulus extracted from the hysteresis curve is corrected, including: For each group of soil samples under each test condition, the slope of the line connecting the two endpoints of the hysteresis curve at each strain level is extracted and used as the initial dynamic elastic modulus. Then the first The first test condition Group of soil samples in the first The adjusted dynamic elastic modulus corresponding to each strain level The calculation formula is: ,in, The initial dynamic elastic modulus, For the first The first test condition Group of soil samples in the first Normalized discrete coefficients for each strain level This is the preset adjustment range.

4. The method for dynamic monitoring of roadbed resilient modulus as described in claim 3, characterized in that, The process of determining the dynamic shear modulus and damping ratio, fitting the parameters of the nonlinear constitutive model, and establishing a nonlinear constitutive model that responds to changes in soil state includes: For each group of soil samples under each test condition, the dynamic shear modulus ratio and damping ratio are calculated based on the theory of elasticity using the adjusted dynamic elastic modulus at each strain level. The mean of the dynamic shear modulus and the mean of the damping ratio of all soil samples under each test condition at the same strain level are used as the dynamic shear modulus and damping ratio at each strain level under each test condition. The axial dynamic strain corresponding to each strain level is converted into shear strain. The shear strain, dynamic shear modulus, and damping ratio of each strain level are combined into a three-dimensional array. The three-dimensional array of all strain levels under each test condition is substituted into the Davidenkov model for fitting to obtain all model parameters. For the soil physical parameters and mechanical environment parameters under all test conditions, they are used as independent variables and the model parameters are used as dependent variables to construct a predictive model of each model parameter with respect to the soil physical parameters and mechanical environment parameters. Thus, the Davidenkov model associated with the soil state is established as a nonlinear constitutive model.

5. The method for dynamic monitoring of roadbed resilient modulus as described in claim 1, characterized in that, The process of constructing the feature vector is as follows: Each time the vibration compaction process is simulated, the vertical acceleration of the center of the vibration wheel at each moment is extracted to calculate the compaction index, which includes the compaction value and the compaction modulus. Among them, frequency domain analysis is performed on the vertical acceleration at all times to obtain the spectrum; the ratio of the energy corresponding to the second harmonic to the energy corresponding to the fundamental frequency in the spectrum is calculated, and the product of this ratio and the preset enhancement constant is used as the compaction value; Calculate the dynamic contact force between the vibratory wheel and the subgrade soil. The specific formula is as follows: ,in, For the mass of the vibrating wheel, Vertical acceleration, Indicates the amplitude of the excitation force. The frequency of the excitation force is represented by the angular frequency. Let t be a cosine function, and let t represent the time variable. The compaction modulus is then solved iteratively using Lundberg contact theory. The soil physical parameters, compaction process parameters, and compaction indices from each vibration compaction simulation are combined to form a feature vector.

6. The method for dynamic monitoring of roadbed resilient modulus as described in claim 5, characterized in that, The calculation of the resilient modulus of the top surface of the roadbed and the generation of a simulation dataset containing multiple feature vectors and their resilient moduli include: For the subgrade model after each vibration compaction simulation, the maximum dynamic deflection value of the top surface of the subgrade is extracted by simulating the moving load of vehicles. The resilient modulus of the top surface of the subgrade is calculated according to the equivalent deflection method. The feature vectors obtained from multiple simulations and their corresponding resilient moduli are combined to form a simulation dataset.

7. The method for dynamic monitoring of roadbed resilient modulus as described in claim 1, characterized in that, The training process of the ensemble learning model is as follows: For the simulation dataset, the feature vectors of all samples and their elastic modulus are used as inputs to the Relief-F algorithm to calculate the weight of each feature component on the elastic modulus. Calculate the correlation between any two feature components in the feature vectors of all samples in the simulation dataset; if the correlation is greater than a preset judgment threshold, remove the feature component with the smallest importance weight from the two feature components; otherwise, retain the two feature components and define the filtered feature components as key components; use the key components in the feature vectors of all samples in the simulation dataset and the spring modulus as inputs to train the ensemble learning model.

8. The method for dynamic monitoring of roadbed resilient modulus as described in claim 7, characterized in that, The evaluation of the compaction quality of the roadbed includes: during the compaction operation, based on the key components in the feature vectors acquired in real time, using the trained ensemble learning model to predict the resilient modulus, generating a resilient modulus distribution cloud map covering the entire roadbed; for the resilient modulus at each location in the cloud map, if it is greater than or equal to a preset threshold, then the compaction at this location is qualified; otherwise, there is under-compaction at this location.

9. A dynamic monitoring system for the resilient modulus of roadbed, characterized in that, The system includes a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that the processor executes the computer program to implement the steps of the dynamic monitoring method for the resilient modulus of a roadbed as described in any one of claims 1-8.

10. A dynamic monitoring device for the resilient modulus of a roadbed, wherein the device stores a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the dynamic monitoring method for the resilient modulus of roadbed as described in any one of claims 1-8.

Images