Method and device for evaluating anti-skid performance of asphalt pavement in aeolian sand state
By using 3D texture point cloud separation and Transformer model to evaluate the skid resistance performance of asphalt pavement, the problem of insufficient evaluation accuracy in aeolian sandy environments is solved, achieving efficient and accurate skid resistance performance evaluation and supporting road safety and maintenance decisions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHANGAN UNIV
- Filing Date
- 2026-02-09
- Publication Date
- 2026-06-02
AI Technical Summary
Existing technologies lack accuracy and generalizability in assessing the skid resistance of asphalt pavements in aeolian sand environments. They also lack analysis of the nonlinear effects of macro- and micro-texture-related features on pavement, and lack data-driven machine learning models.
Environmental data is acquired using 3D textured point clouds. Macro and micro texture features are separated by spline region filters. Important features are selected by combining random forest algorithm. A sample attention mechanism model based on Transformer is constructed for evaluation.
It improves the accuracy and efficiency of asphalt pavement skid resistance assessment, enhances the assessment capability in aeolian sandy environments, and provides data support for road driving safety and maintenance decisions.
Smart Images
Figure CN122135147A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of large language model technology, specifically relating to a method and device for evaluating the skid resistance performance of asphalt pavement under aeolian sand conditions. Background Technology
[0002] Road surface skid resistance is a crucial factor in ensuring road driving safety and a key focus of road maintenance. In real-world environments, road surface texture is inevitably affected by external environmental factors such as rain, snowfall, and icing. These factors alter the frictional characteristics between the tire and the road surface, thus impacting skid resistance. Particularly in desert regions, the presence of aeolian sand poses a serious threat to the skid resistance of asphalt pavements. Aeolian sand not only reduces the effective contact area between the tire and the road surface but also lowers the coefficient of friction, leading to a significant decrease in skid resistance. Therefore, accurately and efficiently assessing road surface skid resistance under sandy conditions is of paramount importance.
[0003] Current methods for assessing the skid resistance of pavements still suffer from relatively insufficient accuracy and generalizability in evaluating the skid resistance of asphalt pavements in aeolian sandy environments. Existing methods lack analysis of the nonlinear influence of pavement macro- and micro-texture-related features on skid resistance, leading to feature redundancy. Furthermore, there is a lack of data-driven machine learning models to characterize the variation patterns of skid resistance in sandy pavements, resulting in insufficient accuracy and generalizability of existing assessment methods. Summary of the Invention
[0004] To address the aforementioned problems in the prior art, this invention provides a method and apparatus for evaluating the skid resistance of asphalt pavement under aeolian sand conditions.
[0005] The technical problem to be solved by this invention is achieved through the following technical solution: In a first aspect, the present invention provides a method for evaluating the skid resistance of asphalt pavement under aeolian sand conditions, comprising: Acquire environmental data and 3D texture point cloud of the target asphalt pavement area under aeolian sand conditions; Macro- and micro-texture features are extracted from the 3D texture point cloud to obtain the target texture feature data; The environmental data and target texture feature data are input into a pre-trained target skid resistance performance evaluation model, which outputs the skid resistance performance evaluation results corresponding to the target asphalt pavement area.
[0006] Secondly, the present invention provides a device for evaluating the skid resistance performance of asphalt pavement under aeolian sand conditions, comprising: The acquisition module is used to acquire environmental data and three-dimensional texture point cloud of the target asphalt pavement area under aeolian sand conditions; The data preprocessing module is used to extract macroscopic and microscopic texture features from the 3D texture point cloud to obtain target texture feature data; The evaluation result prediction module is used to input environmental data and target texture feature data into a pre-trained target skid resistance performance evaluation model and output the skid resistance performance evaluation results corresponding to the target asphalt pavement area.
[0007] This invention provides a method and apparatus for evaluating the skid resistance performance of asphalt pavement under aeolian sand conditions. By analyzing the nonlinear influence of pavement texture-related features on skid resistance performance, the method has stronger characterization capabilities than the features used in existing methods. Furthermore, this invention employs a target skid resistance performance evaluation model, overcoming the problem of existing methods lacking data-driven machine learning models, thereby improving the efficiency and accuracy of pavement skid resistance performance evaluation.
[0008] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description
[0009] Figure 1 This is a flowchart illustrating a method for evaluating the skid resistance of asphalt pavement under aeolian sand conditions, as provided in an embodiment of the present invention. Figure 2A and Figure 2B This is a schematic diagram of three-dimensional texture point cloud separation according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the importance scores for feature selection using a random forest according to an embodiment of the present invention; Figure 4 This is an example image of a wind-blown sand asphalt pavement according to an embodiment of the present invention; Figure 5 This is a test area diagram under different sand accumulation conditions according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the training process of the target anti-skid performance evaluation model according to an embodiment of the present invention; Figures 7A to 7D This is a schematic diagram of the experimental results of an asphalt pavement skid resistance evaluation method under aeolian sand conditions according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of an asphalt pavement skid resistance evaluation device under aeolian sand conditions according to an embodiment of the present invention. Detailed Implementation
[0010] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.
[0011] This invention provides a method for evaluating the skid resistance of asphalt pavement under aeolian sand conditions. See also... Figure 1 The method includes the following steps: S10. Obtain environmental data and three-dimensional texture point cloud of the target asphalt pavement area under aeolian sand conditions.
[0012] Optionally, environmental data may include the amount of sand accumulation in the asphalt pavement area, ambient temperature, ambient humidity, and surface temperature.
[0013] S20. Extract macroscopic and microscopic texture features from the 3D texture point cloud to obtain target texture feature data.
[0014] Optionally, step S20 may specifically include: S201. Use a spline region filter to separate the three-dimensional texture point cloud to obtain macro-texture point cloud and micro-texture point cloud.
[0015] For example, a spline regional filter is a filtering method that combines the smooth fitting characteristics of spline functions with an adaptive region partitioning strategy. It can accurately preserve key details such as edges and textures while denoising and smoothing signals / images. The wavelength range corresponding to macroscopic texture point clouds is 0.5-50 mm, and the wavelength range corresponding to microscopic texture point clouds is 0.001-0.5 mm. Figure 2A The road surface outline is contained within the macroscopic texture point cloud, such as Figure 2B The details of the road surface texture are contained in the micro-texture point cloud.
[0016] S202. Perform feature extraction on the macroscopic texture point cloud and the microscopic texture point cloud respectively to obtain initial texture feature data containing multiple texture features.
[0017] For example, three-dimensional features are extracted from macroscopic and microscopic texture point clouds respectively to obtain initial texture feature data containing multiple texture features. These multiple texture features may specifically include height parameters, function parameters, blending parameters, function parameters (volume), feature parameters, etc., totaling 96 features, as shown in Table 1 below.
[0018] Table 1
[0019] in, for The surface height of the location; It is the projected area of the defined region; and express and Number of sampling points in the direction; It is the contour peak line; and It's the material ratio; It is a constant that converts each square meter into milliliters.
[0020] S203. Select features from the various texture features contained in the initial texture feature data to obtain the target texture feature data.
[0021] Optionally, step S203 may specifically include: S2031. Input the initial texture feature data into the pre-trained random forest model and output the importance scores of each texture feature contained in the initial texture feature data.
[0022] S2032. Texture features with importance scores greater than a preset threshold are identified as target texture feature data.
[0023] For example, a model is built using the random forest algorithm, which uses a pre-trained random forest model to evaluate the importance of each feature of the initial texture feature data to the prediction results.
[0024] The initial texture feature data is used as input to the random forest model. This initial texture feature data contains several texture features, each of which may contribute differently to the random forest. Therefore, the random forest model outputs an importance score for each texture feature. Based on the importance score of each texture feature and a preset threshold, such as 0.025, only texture features with an importance score greater than the preset threshold are retained, resulting in the target texture feature data.
[0025] The processing steps specifically include: 1) Calculating the feature importance within a single tree. For the first... A tree, making For all textured features The set of nodes used for splitting; For nodes The increase in impurity before and after splitting. This relates to texture features. In the The importance score of a tree can be defined as:
[0026] 2) Calculate forest-level normalized importance. For forests... The importance scores of the trees are summed and then normalized to the [0,1] interval. The output importance score is defined as follows:
[0027] in, This represents the total number of texture features.
[0028] 3) Given a preset threshold θ = 0.025, the target texture feature data The definition is as follows:
[0029] Optionally, the target texture feature data includes kurtosis, root mean square surface gradient, development interface area ratio, material volume, ten-point height, five-point peak height, arithmetic mean pit curvature, average height of local peaks, maximum height of local peaks, and standard deviation of local peak heights.
[0030] For example, refer to Figure 3 , Figure 3 A diagram illustrating the selection of scores for random forest features. Target texture feature data includes kurtosis. Root mean square surface gradient Development interface area ratio Material volume 10 o'clock height Five-point peak height Arithmetic mean concave curvature Average height of local peaks Maximum height of local peaks and the standard deviation of the local peak height Features ending with "ma" are extracted from macroscopic texture point clouds, while features ending with "mi" are extracted from microscopic texture point clouds.
[0031] S30. Input the environmental data and target texture feature data into the pre-trained target anti-skid performance evaluation model, and output the anti-skid performance evaluation results corresponding to the target asphalt pavement area.
[0032] Optionally, the pre-training process of the target anti-skid performance evaluation model may specifically include: S1. Obtain environmental data samples, three-dimensional texture point cloud samples, and corresponding pendulum values of the asphalt pavement test area under different sand accumulation conditions.
[0033] For example, refer to Figure 4 , Figure 4 The image shows an example of a sample collection for aeolian sand asphalt pavement. First, a 200mm x 100mm test area was selected on the surface of a 500mm x 500mm specimen (asphalt pavement). Figure 5 To create test area maps under different sediment accumulation conditions, environmental data samples, three-dimensional texture point cloud samples, and corresponding pendulum values were recorded on these test areas under different sediment accumulation conditions to form a sample set. The sample set contains 186 samples collected from four asphalt specimens with different gradations.
[0034] The pendulum value, also known as the pendulum friction coefficient tester value, is measured by a pendulum friction coefficient tester and is measured in units of BPN (Bounce Pendulum Number). It is a key quantitative indicator for measuring the anti-skid ability of a road surface.
[0035] S2. Extract macroscopic and microscopic texture features from the 3D texture point cloud sample to obtain the target texture feature data sample.
[0036] For details on this step, please refer to the specific implementation process of step S20, which will not be elaborated here.
[0037] S3. Using environmental data samples and target texture feature data samples as training samples, and with the pendulum value as the label, supervise the training of the pre-built target Transformer model to obtain the target anti-slip performance evaluation model.
[0038] The target Transformer model is a Transformer model based on the inter-sample attention mechanism.
[0039] Optionally, the target Transformer model includes an embedding layer, a multi-head self-attention submodule, an inter-sample attention submodule, and a feedforward and multilayer perceptron.
[0040] Based on this, step S3 may specifically include: S31. The feature dimensions of the environmental data samples and the target texture feature data samples are unified through the embedding layer, and the first feature representation is output.
[0041] S32. The first feature representation is processed by the multi-head self-attention submodule, and the second feature representation is output.
[0042] S33. The second feature representation is processed by the inter-sample attention submodule to output the third feature representation.
[0043] S34. The third feature representation is mapped through feedforward and multilayer perceptron to output the predicted pendulum value.
[0044] S35. Based on the pendulum value and the predicted pendulum value used as labels, update the parameters of the target Transformer model to obtain the target anti-skid performance evaluation model.
[0045] For example, refer to Figure 6 The training dataset (including environmental data samples and target texture feature data samples) is denoted as... , For the sample size, It is the characteristic number.
[0046] The embedding layer performs a separate learnable linear mapping on each column, unifying the mapping of native features with different dimensions and semantics to... The first feature representation is obtained from the latent space, such as the word embedding matrix. .
[0047] The multi-head self-attention submodule uses the following attention mechanism computation method:
[0048]
[0049] in, , H represents the number of heads, and d represents the feature dimension. This stage only allows interaction between different features within the same row. The first feature represents each row. ,Bundle indivual The vector is fed as a sequence into the multi-head self-attention layer. Q, K, and V all come from the row itself. The output is the updated representation of that row. This stage automatically learns "which feature combinations are more important for the pendulum value within this sample". The updated representation... The nonlinear interaction is further refined by a feedforward neural network, and then the gradient is stabilized by residuals and normalization, finally outputting the second feature representation.
[0050] The inter-sample attention submodule uses the following attention mechanism computation method:
[0051]
[0052]
[0053]
[0054] in, , , This represents the feature set of the training samples. At this stage, the global similarity between rows is also explicitly modeled. Q, K, and V all come from different samples, forming a context-aware sample representation. The context-aware sample representation further refines the nonlinear interactions through a feedforward neural network, and then passes through residuals and normalization to ensure gradient stability, outputting a third feature representation, namely the context representation.
[0055] Finally, the third feature represents the input to a feedforward multilayer perceptron (MLP) (containing two fully connected layers and random deactivation), and the output scalar. As a prediction pendulum value. During the evaluation of model training, the entire model minimizes the root mean square error loss end-to-end, while L2 regularization is used to prevent overfitting.
[0056] Specifically, the key parameters for defining the target Transformer model structure are shown in Table 1.
[0057] Table 1. Parameter Definitions of Model Structure
[0058] Furthermore, the key parameters for model training are defined as shown in Table 2.
[0059] Table 2 Parameter Definitions for Model Training
[0060] This embodiment provides a method for evaluating the skid resistance performance of asphalt pavement under aeolian sand conditions. By employing the Random Forest ensemble learning algorithm, it explores the nonlinear influence of macro- and micro-texture-related features on skid resistance performance. Feature selection is performed on the initial texture feature data to obtain target texture feature data, which has stronger representational capabilities than features used in existing methods. Simultaneously, by constructing an evaluation model based on an improved Transformer with an inter-sample attention mechanism, it overcomes the problem of existing methods lacking data-driven machine learning models, improving the efficiency and accuracy of pavement skid resistance performance evaluation, and providing a solid data foundation for road safety assurance and pavement maintenance decisions.
[0061] The following experiment further illustrates the method for evaluating the skid resistance of asphalt pavement under aeolian sand conditions provided by this invention.
[0062] This experiment validated the results using measured data from specimens AC13, AC16, OGFC10, and SMA16, achieving an average accuracy rate of over 90%. Experimental results are referenced... Figure 7A (AC13) Figure 7B (AC16) Figure 7C (OGFC10) and Figure 7D (SMA16), where the broken line with black triangles represents the actual measured value of the pendulum, and the broken line with red squares represents the predicted value of the target anti-skid performance evaluation model on the test set samples. It should be noted that the data used here was not involved in the training process of the evaluation model, and is only used to evaluate the evaluation effect of the target anti-skid performance evaluation model on the friction coefficient under different sand accumulation conditions.
[0063] Linear regression models, due to their simple structure and limited number of trainable parameters, cannot adequately model datasets with high feature dimensions. Ensemble learning models such as XGBoost and LightGBM, while offering significant advantages over linear regression, lack the ability to learn the correlations between samples. The target anti-slip performance evaluation model proposed in this invention can effectively learn the potential correlations between high-dimensional features and the correlations between samples. Therefore, as shown in Table 3, it exhibits lower error and higher fitting on the test set.
[0064] Table 3 Comparison of quantitative indicators of anti-slip performance evaluation models on the test set
[0065] Corresponding to the above-described method for evaluating the skid resistance of asphalt pavement under aeolian sand conditions, this invention also provides a device for evaluating the skid resistance of asphalt pavement under aeolian sand conditions; such as Figure 8 As shown, the device may include: The acquisition module 801 is used to acquire environmental data and three-dimensional texture point cloud of the target asphalt pavement area under aeolian sand conditions.
[0066] The data preprocessing module 802 is used to extract macroscopic and microscopic texture features from the 3D texture point cloud to obtain target texture feature data.
[0067] The evaluation result prediction module 803 is used to input environmental data and target texture feature data into a pre-trained target anti-skid performance evaluation model and output the anti-skid performance evaluation result corresponding to the target asphalt pavement area.
[0068] For details regarding the device, please refer to the steps of the method for evaluating the skid resistance of asphalt pavement under aeolian sand conditions provided in the first aspect; these will not be elaborated upon here.
[0069] This embodiment provides a device for evaluating the skid resistance performance of asphalt pavement under aeolian sand conditions. By employing the Random Forest ensemble learning algorithm, it explores the nonlinear influence of macro- and micro-texture-related features on skid resistance performance. Feature selection is performed on the initial texture feature data to obtain target texture feature data, which has stronger representational capabilities than features used in existing methods. Simultaneously, by constructing an evaluation model based on an improved Transformer with an inter-sample attention mechanism, it overcomes the problem of existing methods lacking data-driven machine learning models, improving the efficiency and accuracy of pavement skid resistance performance evaluation, and providing a solid data foundation for road safety assurance and pavement maintenance decisions.
[0070] It should be noted that the device is basically similar to the method embodiment, so the description is relatively simple. For relevant parts, please refer to the description of the method embodiment.
[0071] It should be noted that the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention.
[0072] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.
[0073] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings and the disclosure in carrying out the claimed invention. In the description of the invention, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.
[0074] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.
Claims
1. A method for evaluating the skid resistance of asphalt pavement under aeolian sand conditions, characterized in that, include: Acquire environmental data and 3D texture point cloud of the target asphalt pavement area under aeolian sand conditions; Macro-texture features and micro-texture features are extracted from the three-dimensional texture point cloud to obtain target texture feature data; The environmental data and the target texture feature data are input into a pre-trained target skid resistance performance evaluation model, and the skid resistance performance evaluation results corresponding to the target asphalt pavement area are output.
2. The method for evaluating the skid resistance of asphalt pavement under aeolian sand conditions according to claim 1, characterized in that, The step of extracting macroscopic and microscopic texture features from the three-dimensional texture point cloud to obtain target texture feature data includes: The three-dimensional texture point cloud is separated by a spline region filter to obtain macroscopic texture point cloud and microscopic texture point cloud; Feature extraction is performed on the macroscopic texture point cloud and the microscopic texture point cloud respectively to obtain initial texture feature data containing multiple texture features; Feature selection is performed on the various texture features contained in the initial texture feature data to obtain the target texture feature data.
3. The method for evaluating the skid resistance of asphalt pavement under aeolian sand conditions according to claim 2, characterized in that, The step of selecting features from the multiple texture features contained in the initial texture feature data to obtain the target texture feature data includes: The initial texture feature data is input into a pre-trained random forest model, which outputs the importance scores of each texture feature contained in the initial texture feature data. Texture features with importance scores greater than a preset threshold are identified as target texture feature data.
4. The method for evaluating the skid resistance of asphalt pavement under aeolian sand conditions according to claim 1, characterized in that, The pre-training process of the target anti-skid performance evaluation model includes: Environmental data samples, three-dimensional texture point cloud samples, and corresponding pendulum values were obtained for the asphalt pavement test area under different sand accumulation conditions. Macro-texture features and micro-texture features are extracted from the three-dimensional texture point cloud sample to obtain target texture feature data sample; Using the environmental data samples and the target texture feature data samples as training samples, and the pendulum value as the label, a pre-constructed target Transformer model is trained in a supervised manner to obtain a target anti-slip performance evaluation model; wherein, the target Transformer model is a Transformer model based on the inter-sample attention mechanism.
5. The method for evaluating the skid resistance of asphalt pavement under aeolian sand conditions according to claim 4, characterized in that, The target Transformer model includes an embedding layer, a multi-head self-attention submodule, an inter-sample attention submodule, and a feedforward and multilayer perceptron. The step of using the environmental data samples and the target texture feature data samples as training samples, and using the pendulum value as the label, to perform supervised training on a pre-constructed target Transformer model to obtain a target anti-slip performance evaluation model includes: The embedding layer unifies the feature dimensions of the environmental data sample and the target texture feature data sample, and outputs a first feature representation. The first feature representation is processed by the multi-head self-attention submodule to output the second feature representation; The second feature representation is processed by the inter-sample attention submodule to output the third feature representation; The third feature representation is mapped using the feedforward and multilayer perceptron to output the predicted pendulum value; Based on the pendulum value and the predicted pendulum value used as labels, the parameters of the target Transformer model are updated to obtain the target anti-skid performance evaluation model.
6. The method for evaluating the skid resistance of asphalt pavement under aeolian sand conditions according to claim 1, characterized in that, The environmental data includes the amount of sand accumulation, ambient temperature, ambient humidity, and surface temperature of the asphalt pavement area.
7. The method for evaluating the skid resistance of asphalt pavement under aeolian sand conditions according to claim 1, characterized in that, The target texture feature data includes kurtosis, root mean square surface gradient, development interface area ratio, material volume, ten-point height, five-point peak height, arithmetic mean pit curvature, average height of local peaks, maximum height of local peaks, and standard deviation of local peak heights.
8. A device for evaluating the skid resistance performance of asphalt pavement under aeolian sand conditions, characterized in that, include: The acquisition module is used to acquire environmental data and three-dimensional texture point cloud of the target asphalt pavement area under aeolian sand conditions; The data preprocessing module is used to extract macroscopic and microscopic texture features from the three-dimensional texture point cloud to obtain target texture feature data. The evaluation result prediction module is used to input the environmental data and the target texture feature data into a pre-trained target anti-skid performance evaluation model, and output the anti-skid performance evaluation result corresponding to the target asphalt pavement area.