Method and device for detecting density of asphalt pavement

Abstract

The present application relates to a kind of density detection method and density detection device of asphalt pavement, the method comprises: obtaining the raw material parameter of asphalt pavement;Raw material parameters include asphalt dielectric constant, aggregate dielectric constant, asphalt density, aggregate density, asphalt content and asphalt mixture maximum theoretical density;The road surface dielectric constant of the asphalt pavement is determined by ground penetrating radar;The raw material parameter and the road surface dielectric constant are input into density prediction model, to output the road surface density of the asphalt pavement;Wherein, the density prediction model is obtained based on XGBoost algorithm training.This application improves the intelligent level and detection accuracy of asphalt pavement detection while realizing nondestructive testing, effectively promotes the intelligentization of asphalt pavement construction, reduces the labor cost and test consumption in the detection process.

Description

Technical Field

[0001] This invention belongs to the field of non-destructive testing technology, and specifically relates to a density testing method and device for asphalt pavement. Background Technology

[0002] Compaction degree is a key technical indicator for controlling the construction quality of asphalt pavement. It is calculated by combining the in-situ density of the asphalt pavement and the density of indoor Marshall specimens, and measures the difference between the designed porosity and the actual porosity of the pavement. Therefore, in-situ density testing of asphalt pavement is a crucial technology for controlling pavement construction quality. Traditional methods for testing asphalt pavement density mainly include destructive testing and non-destructive testing. Destructive testing primarily involves drilling pavement core samples for indoor density testing. While this method is considered to have high accuracy, it suffers from drawbacks such as low testing efficiency, discontinuous measuring points, and damage to the pavement structure. Among traditional non-destructive testing methods, nuclear density meters use radioactive materials, posing a risk to the surrounding environment and operators, and are prohibited in my country. Existing nuclear-free density meters are safer than nuclear density meters, but still suffer from discontinuous measuring points and low testing efficiency, providing only limited pavement density data.

[0003] In recent years, density prediction models based on classic electromagnetic hybrid models have been widely used, the most classic of which include the complex refractive index (CRIM) model, the Rayleigh model, and the Bottcher model. More complex density prediction models have also been established and developed in recent years, such as the Al-Qadi, Lahouar, and Leng (ALL) model, the influence and shape optimization (ISO) model, and the density prediction (PM) model. However, due to the complexity of pavement structure types and material compositions, and the significant environmental differences that greatly affect density detection results, theoretical calculation models are difficult to apply to all roads.

[0004] In summary, existing methods for testing the density of asphalt pavement still have the following problems: 1. Traditional destructive testing methods require drilling pavement core samples, which damages the pavement and affects the integrity of the pavement structure; 2. Traditional non-destructive testing methods are single-point tests, with low testing efficiency and cannot achieve continuous and rapid testing; 3. The prediction accuracy of traditional density prediction models cannot meet engineering requirements and cannot be applied to the density testing requirements of newly built asphalt pavements. Summary of the Invention

[0005] The purpose of this invention is to solve the above-mentioned problems in the prior art and provide a more accurate and faster asphalt pavement density detection solution.

[0006] This invention discloses a method for detecting the density of asphalt pavement, comprising the following steps:

[0007] Obtain the raw material parameters of asphalt pavement; the raw material parameters include asphalt dielectric constant, aggregate dielectric constant, asphalt density, aggregate density, asphalt content, and maximum theoretical density of asphalt mixture;

[0008] The dielectric constant of the asphalt pavement was determined by ground-penetrating radar.

[0009] The raw material parameters and the road surface dielectric constant are input into the density prediction model to output the road surface density of the asphalt pavement; wherein the density prediction model is trained based on the XGBoost algorithm.

[0010] Optionally, the step of determining the dielectric constant of the asphalt pavement using ground-penetrating radar includes:

[0011] The ground-penetrating radar acquires the first electromagnetic wave signal reflected by the calibration component.

[0012] The ground-penetrating radar acquires the second electromagnetic wave signal reflected by the asphalt pavement.

[0013] The road surface dielectric constant is calculated based on the first electromagnetic wave signal and the second electromagnetic wave signal.

[0014] Optionally, the calibration component is a calibration copper plate.

[0015] Optionally, the density prediction model is trained through the following steps:

[0016] A training dataset for historical asphalt pavement is obtained, the training dataset including raw material training parameters and pavement training density of the historical asphalt pavement; wherein, the raw material training parameters are used to characterize a variety of known parameter combinations associated with the raw materials of the historical asphalt pavement; and the pavement training density is used to characterize the actual density of the historical asphalt pavement.

[0017] The density prediction model includes a first type of hyperparameters and a second type of hyperparameters;

[0018] A decision tree model is generated based on the first type of hyperparameters, which are used to determine the structure of the decision tree model; the raw material training parameters and the road surface training dielectric constant are used as input data, and the road surface training density is used as the model output target to iteratively train the node classification rules in the decision tree model;

[0019] The decision tree model is iteratively optimized based on the XGBoost algorithm according to the second type of hyperparameters to determine the target classification rule corresponding to the maximum gain of the objective function; the objective function includes an error function term, a leaf node term, an L1 regularization term, and an L2 regularization term, and the second type of hyperparameters includes the number of iterations, the learning rate, the leaf node coefficients, the L1 regularization term coefficients, and the L2 regularization term coefficients.

[0020] Optionally, the first type of hyperparameters and the second type of hyperparameters are optimized using a Bayesian optimization algorithm.

[0021] Optionally, the steps of optimizing the first type of hyperparameters and the second type of hyperparameters using a Bayesian optimization algorithm include:

[0022] Obtain the value range corresponding to each parameter type in the first type of hyperparameters and the second type of hyperparameters;

[0023] Select the error function and Bayesian iteration settings for the Bayesian optimization algorithm;

[0024] Based on the obtained parameter types, value ranges, and error function types, Bayesian optimization iterations are performed to obtain multiple current optimal parameter value combinations and multiple error function values ​​corresponding to the current optimal parameter value combinations. The parameter value combination corresponding to the minimum error function value during the Bayesian optimization process is selected as the optimal optimization combination of the first type of hyperparameters and the second type of hyperparameters.

[0025] Optionally, the steps for obtaining the current optimal parameter value combination and the corresponding error function value are as follows: select multiple initial observation points, set a maximum number of observation points, use a probabilistic surrogate model to predict the function distribution based on the initial observation points, and obtain the current optimal parameter value combination; replace the values ​​of the first type of hyperparameters and the second type of hyperparameters in the density prediction model with the current optimal parameter value combination, and use the training dataset to calculate the error function value in the density prediction model after replacing the values ​​of the first type of hyperparameters and the second type of hyperparameters, thereby obtaining the error function value corresponding to the current optimal parameter value combination.

[0026] On the other hand, the present invention also discloses a density detection device for asphalt pavement, comprising:

[0027] The material parameter acquisition module is used to acquire the raw material parameters of asphalt pavement.

[0028] A dielectric constant determination module is used to determine the dielectric constant of the asphalt pavement using ground-penetrating radar.

[0029] The density prediction module is used to input the raw material parameters and the road surface dielectric constant into the density prediction model to output the road surface density of the asphalt pavement; wherein, the density prediction model is trained based on the XGBoost algorithm.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. This invention utilizes material parameters from the asphalt pavement design stage and the dielectric constant of the compacted asphalt pavement to detect asphalt pavement density without damaging the pavement. Compared with traditional methods, this invention maintains the structural integrity of the asphalt pavement when detecting the density of newly constructed asphalt pavements.

[0032] 2. This invention can achieve rapid and continuous detection of asphalt pavement density. Compared with traditional methods, it can obtain more density information of newly built asphalt pavement and more effectively ensure the construction quality of asphalt pavement.

[0033] 3. This invention uses the XGBoost algorithm to train the density prediction model of asphalt pavement, which improves the level of intelligence in asphalt pavement inspection, effectively promotes the intelligent construction of asphalt pavement, and reduces labor costs and test consumption in the inspection process.

[0034] 4. This invention combines Bayesian optimization with the XGBoost algorithm to predict asphalt pavement density, ensuring that the hyperparameters in the density prediction model maintain the best combination and matching, further improving the accuracy of asphalt pavement density prediction, and further narrowing the gap between the predicted value and the true value. Attached Figure Description

[0035] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0036] Figure 1 A schematic flowchart of the asphalt pavement density detection method in Embodiment 1 of the present invention is shown;

[0037] Figure 2 A schematic flowchart illustrating the determination of the road surface dielectric constant in Embodiment 1 of the present invention is shown;

[0038] Figure 3 A schematic diagram of the pavement density at different locations of the asphalt pavement obtained in Embodiment 1 of the present invention is shown;

[0039] Figure 4 A schematic flowchart illustrating the use of a training density prediction model in Embodiment 1 of the present invention is shown.

[0040] Figure 5 A schematic flowchart illustrating the training of a density prediction model using a Bayesian optimization algorithm is shown in Embodiment 1 of the present invention.

[0041] Figure 6The diagram illustrates the error of the density prediction model of Embodiment 1 of the present invention compared with other existing models.

[0042] Figure 7 A schematic diagram of the density detection system for asphalt pavement in Embodiment 2 of the present invention is shown.

[0043] Figure label:

[0044] 70. Density detection system for asphalt pavement; 71. Material parameter acquisition module; 72. Dielectric constant determination module; 73. Density prediction module. Detailed Implementation

[0045] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0046] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0047] Example 1

[0048] Please see Figure 1 This invention provides a density detection method for asphalt pavement, which is a density detection method for asphalt pavement based on the limit gradient lifting algorithm, including the following steps:

[0049] S100: Obtain the raw material parameters for asphalt pavement.

[0050] In one example, the raw material parameters include the dielectric constant of asphalt, the dielectric constant of aggregates, the density of asphalt, the density of aggregates, the asphalt content, and the maximum theoretical density of the asphalt mixture, corresponding to the asphalt pavement. Using these raw material parameters allows for more accurate density testing values ​​for asphalt pavements at a lower cost.

[0051] S200. Determine the dielectric constant of the asphalt pavement using ground-penetrating radar.

[0052] Ground penetrating radar (GPR) is a non-destructive geophysical exploration method used for shallow exploration, offering advantages such as speed, non-destructive operation, and continuous detection. In this embodiment, the pavement dielectric constant can be calculated using radar electromagnetic wave signals reflected from calibrated slabs and asphalt pavements, respectively. Specifically, such as... Figure 2 As shown, the dielectric constant of asphalt pavement can be determined through the following steps:

[0053] S210. Obtain the first electromagnetic wave signal reflected by the calibration plate through ground penetrating radar.

[0054] In this embodiment, the ground-penetrating radar (GPR) can be configured to emit radar electromagnetic wave signals approximately vertically downwards. For example, the GPR antenna can be fixed on a bracket so that a portion of the radar electromagnetic wave signal emitted by the GPR can be emitted approximately vertically towards the ground. By placing a calibration plate at a position corresponding to the GPR, it can be ensured that the radar electromagnetic wave signal emitted towards the ground is reflected back to the GPR receiver after passing through the calibration plate. In this embodiment, the electromagnetic wave signal reflected by the calibration plate is referred to as the first electromagnetic wave signal. In one example, the calibration plate can be a copper plate with a thickness of 1-3 mm. This ensures that the distance between the GPR antenna and the surface of the calibration copper plate is not too large while maintaining the uniformity of the copper plate material, thus reducing errors. Preferably, the thickness of the copper plate is set to 1.5 mm.

[0055] S220. The ground-penetrating radar acquires the second electromagnetic wave signal reflected by the asphalt pavement.

[0056] In this step, the calibration plate is moved away from under the ground penetrating radar, so that a portion of the electromagnetic waves emitted by the ground penetrating radar can reach the asphalt road surface and be reflected back to the receiving end of the ground penetrating radar. In this embodiment, the electromagnetic wave signal reflected by the asphalt road surface is called the second electromagnetic wave signal.

[0057] S230. Calculate the road surface dielectric constant based on the first electromagnetic wave signal and the second electromagnetic wave signal.

[0058] The formula for calculating the dielectric constant of road surface is as follows:

[0059]

[0060] Where, ε r A represents the dielectric constant of asphalt pavement. p A0 represents the amplitude of the first electromagnetic wave signal, and A0 represents the amplitude of the second electromagnetic wave signal.

[0061] Through the above steps, this embodiment can obtain the dielectric constant of the asphalt pavement. It should be noted that the first electromagnetic wave signal obtained in this embodiment can include multiple signals. For example, a ground-penetrating radar in a stationary state continuously emits radar electromagnetic waves towards a calibration copper plate within a preset time period (e.g., 1 minute), and receives multiple first electromagnetic wave signals accordingly. Based on this, averaging the multiple first electromagnetic wave signals yields a more accurate first electromagnetic wave signal. The second electromagnetic wave signal obtained in this embodiment can also include multiple signals. Unlike the calibration copper plate, since the asphalt pavement section is relatively long, and considering the potential for unevenness during pavement construction, to ensure that the obtained second electromagnetic wave signal reflects the actual pavement condition as comprehensively as possible, the ground-penetrating radar can be positioned in a straight line along the asphalt pavement to continuously obtain second electromagnetic wave signals at different locations on the asphalt pavement. Substituting the multiple second electromagnetic wave signals into the above formula for calculating the pavement dielectric constant yields multiple pavement dielectric constants.

[0062] S300. Input the raw material parameters and the road surface dielectric constant into the density prediction model to output the road surface density of the asphalt pavement; wherein, the density prediction model is trained based on the XGBoost algorithm.

[0063] This invention establishes the relationship between the dielectric properties of composite materials and their internal properties. By back-calculating electromagnetic mixing theory and the volume parameters of asphalt mixtures, it achieves accurate prediction of the density of asphalt pavement composed of composite materials. This embodiment uses a decision tree model as its foundation. By continuously optimizing the structure and classification rules of the current decision tree model, the structure and classification rules are fixed when the iteration stopping condition is met, resulting in the density prediction model of this invention. This density prediction model can directly output the predicted pavement density based on the input raw material parameters and the pavement dielectric constant. When there are multiple pavement dielectric constants, multiple pavement densities can be output respectively. Figure 3 This diagram illustrates the pavement density at different locations on the asphalt pavement obtained in Embodiment 1 of the present invention. From... Figure 3 It can be seen that the pavement density of asphalt pavement varies slightly at different distances, i.e., different locations, but generally remains at 2.5 g / cm³. 3 Fluctuations in the vicinity.

[0064] Figure 4 A schematic flowchart illustrating the training of the density prediction model in Embodiment 1 of the present invention is shown. Figure 4 As shown, the density prediction model in this embodiment is trained through the following steps:

[0065] S410. Obtain a training dataset of historical asphalt pavement, wherein the training dataset includes training parameters of raw materials, training dielectric constant of pavement, and training density of pavement; wherein the training parameters of raw materials include asphalt dielectric constant, aggregate dielectric constant, asphalt density, aggregate density, asphalt content, and maximum theoretical density of asphalt mixture; the training dielectric constant of pavement refers to the surface dielectric constant of historical asphalt pavement; the training density of pavement refers to the density of historical asphalt pavement, that is, the actual density of asphalt pavement.

[0066] The training dataset can be derived from field inspection data of asphalt pavement and indoor density test data. Specifically, the dielectric constant of aggregates can be determined using an LCR network analyzer, the dielectric constant of asphalt is set to 3, the density of asphalt and aggregates is measured using the specific gravity bottle method and the basket method, respectively, the asphalt content is obtained from the Marshall mix design, the maximum theoretical density of asphalt mixture is determined using the indoor vacuum method, the training dielectric constant of asphalt pavement is obtained by 2.2 GHz ground penetrating radar detection, and the density of asphalt pavement is the density of pavement core samples measured by the surface drying method.

[0067] S420, the density prediction model includes first-type hyperparameters and second-type hyperparameters;

[0068] A decision tree model is generated based on a first type of hyperparameters, which are used to determine the structure of the decision tree model. The raw material training parameters and the road surface training dielectric constant are used as input data, and the road surface training density is used as the output target of the density prediction model. The node classification rules in the decision tree model are iteratively trained. In this embodiment, the first type of hyperparameters may include the depth of the decision tree model. Those skilled in the art understand that when the depth is determined, the structure of a decision tree is also determined. For example, a decision tree with a depth of 2 is equivalent to splitting down from the root node once to obtain two leaf nodes; a decision tree with a depth of 3 is equivalent to splitting down from the root node once to obtain a first left node and a first right node, the first left node then splitting down once to obtain a first leaf node and a second leaf node, and the first right node then splitting down once to obtain a third leaf node and a fourth leaf node. It is understood that each node, except for the bottom-level leaf nodes, has its own corresponding classification rules. These classification rules ultimately divide multiple samples in the input data into a specific leaf node at the bottom level of the decision tree model. Furthermore, based on the different input data samples contained in each leaf node, the value of each leaf node can be calculated, thereby calculating the predicted density accordingly. The classification rules corresponding to the above nodes can be determined based on the greedy algorithm in the prior art, which will not be elaborated on in this embodiment.

[0069] The following example, using a single iteration, illustrates how to classify input data using a decision tree model. Assume the i-th training data sample includes {asphalt dielectric constant i, aggregate dielectric constant i, asphalt density i, aggregate density i, asphalt content i, maximum theoretical density of asphalt mixture i, pavement training dielectric constant i, asphalt pavement density i}. This {asphalt dielectric constant i, aggregate dielectric constant i, asphalt density i, aggregate density i, asphalt content i, maximum theoretical density of asphalt mixture i, pavement training dielectric constant i} can be used as input data and fed into the decision tree model i, whose first hyperparameter, Max deep, is determined, for classification. Finally, each parameter in the input data is mapped to a different leaf node of the decision tree model i. By calculating the input data contained in each leaf node, the corresponding predicted density y can be obtained. i The asphalt pavement density i in the above training sample data is compared with the predicted density y. i By comparing the results, we can obtain the error function value of the current decision tree model i.

[0070] S430: The decision tree model is iteratively optimized based on the XGBoost algorithm according to the second type of hyperparameters to determine the target classification rule corresponding to the maximum gain of the objective function; the objective function includes an error function term, a leaf node term, an L1 regularization term, and an L2 regularization term, and the second hyperparameters include the number of iterations, the learning rate, the leaf node coefficients, the L1 regularization term coefficients, and the L2 regularization term coefficients.

[0071] The following formula represents the objective function O based on the XGBoost algorithm in this embodiment of the invention. k The expression:

[0072]

[0073] In the above formula, k represents the current iteration number; γ represents the leaf node coefficient; α represents the L1 regularization coefficient; λ represents the L2 regularization coefficient; n is the number of samples in the training data; T represents the total number of leaf nodes corresponding to the decision tree model in the current iteration; w j y represents the leaf weight of the j-th leaf node in the current decision tree model; i This represents the actual asphalt pavement density corresponding to the i-th training data sample; denoted by , where represents the prediction density output by the current decision tree model; i represents the i-th training data sample; j represents the j-th leaf node in the current decision tree model; and l is the root mean square error function. The error function representing the difference between the predicted and actual values:

[0074] Based on the principle of the XGBoost algorithm, by introducing a second-order Taylor expansion, the above objective function Ok It can ultimately be approximated as:

[0075]

[0076] In the above formula, I j Represents the set of leaf nodes. This represents the first-order partial derivative of the error function. It represents the second-order partial derivative of the error function.

[0077] Based on this, we can determine O k * Determine the information gain of the current decision tree model, and select the node classification rule corresponding to the node with the maximum information gain as the target classification rule of the current decision tree.

[0078] Through the above steps, this embodiment obtains a density prediction model based on the XGBoost algorithm. The structure of this density prediction model is determined by the first type of hyperparameters, and the node classification rule of this density prediction model is determined iteratively by the XGBoost algorithm based on the second type of hyperparameters. For any input set of data containing raw material parameters and the dielectric constant of asphalt pavement, the predicted asphalt pavement density can be output. Compared with existing theoretical calculation models, the density prediction model provided in this embodiment is more applicable to asphalt pavement under different conditions and has higher prediction accuracy.

[0079] In this embodiment of the invention, the first type of hyperparameters and the second type of hyperparameters are determined by a Bayesian optimization algorithm. Figure 5 A schematic flowchart illustrating the training of a density prediction model using a Bayesian optimization algorithm in Embodiment 1 of the present invention is shown. Figure 5 As shown, the steps for training the first and second type of hyperparameters using the Bayesian optimization algorithm include:

[0080] S510: Obtain the parameter types corresponding to the first type of hyperparameters and the second type of hyperparameters, as well as the value range corresponding to each parameter type.

[0081] Specifically, the hyperparameter types optimized using the Bayesian optimization algorithm in this embodiment include: Num boostround: the number of regression trees in the XGBoost model, with a lower bound of 80 and an upper bound of 125; Max depth: the depth of the decision tree, with a lower bound of 5 and an upper bound of 25; Learning rate: the learning rate, with a lower bound of 0.01 and an upper bound of 1; Gamma: the parameter of the leaf nodes, with a lower bound of e. -9 The upper bound is 1000; Reg alpha: L1 regularization term coefficient, with a lower bound of e. -9 The upper bound is 1000; Reg lambda: L2 regularization term coefficient, with a lower bound of e.-9 The upper bound is 1000. The upper and lower bounds of the above value range can be set based on empirical values.

[0082] S520: Select the root mean square error function as the error function and Bayesian iteration setting value for the Bayesian optimization algorithm.

[0083] S530: Based on the obtained parameter types, value ranges, and error function types, perform Bayesian optimization. When the Bayesian optimization reaches the iterative set value, select the parameter type and value combination corresponding to the minimum error function value during the Bayesian optimization process as the optimal optimization combination of the first type of hyperparameters and the second type of hyperparameters.

[0084] Specifically, the Bayesian optimization process is as follows: Multiple initial observation points (e.g., 20) are selected, and a maximum number of observation points is set (e.g., 320). Based on the initial observation points, a probabilistic surrogate model is used to predict the distribution of the function, obtaining the current optimal combination of parameter values. The current optimal combination of parameter values ​​is used to replace the values ​​of the first and second types of hyperparameters in the current density prediction model (i.e., the XGBoost model). Based on the training dataset, the XGBoost model is used to obtain the predicted density value of the asphalt pavement, and the error function value is calculated with the actual density value to obtain the error function value corresponding to the current optimal combination of parameter values. The next observation is performed based on the observation points selected by the acquisition function. When the number of observation points reaches the maximum number, the Bayesian iteration ends, obtaining multiple error function values ​​and the current parameter value combination corresponding to each error function value. At this point, the parameter combination with the smallest error function is selected as the optimal combination of hyperparameter values. This optimal combination is used as the fixed values ​​of the first and second types of hyperparameters in the density prediction model.

[0085] By following the steps above, we can ensure that the combination of hyperparameters in the density prediction model is more reasonable, thereby further improving the accuracy of the probability prediction model. Figure 6 The diagram illustrates the error of the density prediction model in Embodiment 1 compared to other existing models. Figure 6The horizontal axis represents the six existing theoretical calculation models and the two machine learning models provided in this embodiment of the invention. The six existing theoretical calculation models include the ALL model, PM model, ISO model, CRIM model, Rayleigh model, and Bottcher model. The two models provided in Embodiment 1 of this invention are a density prediction model trained solely on the XGBoost algorithm and a density prediction model XGBoost-B trained simultaneously on the XGBoost and Bayesian optimization algorithms. It can be seen that the density prediction model trained solely on the XGBoost algorithm has an error percentage of 0.8% compared to the other six existing models, which is a significant decrease. Furthermore, the density prediction model XGBoost-B, trained simultaneously on the XGBoost and Bayesian optimization algorithms, has an error of only 0.2%, a substantial decrease compared to the density prediction model trained solely on the XGBoost algorithm.

[0086] Example 2

[0087] This embodiment provides a density detection device for asphalt pavement. This density detection device is an asphalt pavement density detection system 70 based on the limit gradient lifting algorithm, using the aforementioned asphalt pavement density detection method, such as... Figure 7 As shown, it includes a material parameter acquisition module 71, a dielectric constant determination module 72, and a density prediction module 73. Wherein:

[0088] The material parameter acquisition module 71 is suitable for acquiring raw material parameters of asphalt pavement, wherein the raw material parameters are used to characterize a combination of known parameters associated with the raw materials of the asphalt pavement.

[0089] The dielectric constant determination module 72 is suitable for determining the dielectric constant of the asphalt pavement by ground penetrating radar;

[0090] The density prediction module 73 is suitable for inputting the raw material parameters and the road surface dielectric constant into the density prediction model to output the road surface density of the asphalt road surface; wherein the density prediction model is trained based on the XGBoost algorithm.

[0091] This invention utilizes material parameters from the asphalt pavement design stage and the dielectric constant of the compacted asphalt pavement to train an asphalt pavement density prediction model using the XGBoost algorithm, thereby achieving asphalt pavement density detection. This invention improves the intelligence and accuracy of asphalt pavement inspection while achieving non-destructive testing, effectively promoting intelligent asphalt pavement construction and reducing labor costs and testing expenses during the inspection process.

[0092] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for detecting the density of asphalt pavement, characterized in that, Includes the following steps: Obtain the raw material parameters of asphalt pavement; the raw material parameters include asphalt dielectric constant, aggregate dielectric constant, asphalt density, aggregate density, asphalt content, and maximum theoretical density of asphalt mixture; The dielectric constant of the asphalt pavement was determined by ground-penetrating radar. The raw material parameters and the road surface dielectric constant are input into the density prediction model to output the road surface density of the asphalt pavement; wherein, the density prediction model is trained based on the XGBoost algorithm; The density prediction model is trained through the following steps: A training dataset for historical asphalt pavement is obtained, the training dataset including raw material training parameters and pavement training density of the historical asphalt pavement; wherein, the raw material training parameters are used to characterize a variety of known parameter combinations associated with the raw materials of the historical asphalt pavement; and the pavement training density is used to characterize the actual density of the historical asphalt pavement. The density prediction model includes a first type of hyperparameters and a second type of hyperparameters; A decision tree model is generated based on the first type of hyperparameters, which are used to determine the structure of the decision tree model; the raw material training parameters and the road surface training dielectric constant are used as input data, and the road surface training density is used as the output target of the density prediction model to iteratively train the node classification rules in the decision tree model; The decision tree model is iteratively optimized based on the XGBoost algorithm according to the second type of hyperparameters to determine the target classification rule corresponding to the maximum gain of the objective function; the objective function includes an error function term, a leaf node term, an L1 regularization term, and an L2 regularization term, and the second type of hyperparameters includes the number of iterations, the learning rate, the leaf node coefficients, the L1 regularization term coefficients, and the L2 regularization term coefficients; The first type of hyperparameters and the second type of hyperparameters are optimized using a Bayesian optimization algorithm.

2. The density detection method for asphalt pavement according to claim 1, characterized in that, The steps for determining the dielectric constant of the asphalt pavement using ground-penetrating radar include: The ground-penetrating radar acquires the first electromagnetic wave signal reflected by the calibration component. The ground-penetrating radar acquires the second electromagnetic wave signal reflected by the asphalt pavement. The road surface dielectric constant is calculated based on the first electromagnetic wave signal and the second electromagnetic wave signal.

3. The density detection method for asphalt pavement according to claim 2, characterized in that, The calibration component is a calibration copper plate.

4. The density detection method for asphalt pavement according to claim 1, characterized in that, The steps for optimizing the first type of hyperparameters and the second type of hyperparameters using the Bayesian optimization algorithm include: Obtain the value range corresponding to each parameter type in the first type of hyperparameters and the second type of hyperparameters; Select the error function and Bayesian iteration settings for the Bayesian optimization algorithm; Based on the obtained parameter types, value ranges, and error function types, Bayesian optimization iterations are performed to obtain multiple current optimal parameter value combinations and multiple error function values ​​corresponding to the current optimal parameter value combinations. The parameter value combination corresponding to the minimum error function value during the Bayesian optimization process is selected as the optimal optimization combination of the first type of hyperparameters and the second type of hyperparameters.

5. The density detection method for asphalt pavement according to claim 4, characterized in that, The steps to obtain the current optimal parameter value combination and the corresponding error function value are as follows: Select multiple initial observation points, set the maximum number of observation points, use a probabilistic surrogate model to predict the function distribution based on the initial observation points, and obtain the current optimal parameter value combination; replace the values ​​of the first type of hyperparameters and the second type of hyperparameters in the density prediction model with the current optimal parameter value combination, and use the training dataset to calculate the error function value in the density prediction model after replacing the values ​​of the first type of hyperparameters and the second type of hyperparameters, and obtain the error function value corresponding to the current optimal parameter value combination.

6. A density testing device for asphalt pavement, comprising a density testing method for asphalt pavement according to any one of claims 1-5, wherein the density of asphalt pavement is tested, characterized in that, include: The material parameter acquisition module is used to acquire the raw material parameters of asphalt pavement. The dielectric constant determination module is used to determine the dielectric constant of the asphalt pavement using ground penetrating radar; the density prediction module is used to input the raw material parameters and the dielectric constant of the pavement into the density prediction model to output the pavement density of the asphalt pavement; wherein, the density prediction model is trained based on the XGBoost algorithm.

Images