Asphalt-based material low-temperature toughness prediction method based on data driving

Abstract

The invention discloses an asphalt-based material low-temperature toughness prediction method based on data driving, and belongs to the field of asphalt pavement materials. The invention aims to solve the problems of poor low-temperature toughness of the asphalt-based material in the cold region and low efficiency of the traditional performance prediction method. According to the method, a database is constructed by testing performance data of different asphalt and modifier combinations, after cleaning and preprocessing, training and hyper-parameter optimization are performed by using a deep neural network, and an optimal prediction model is screened out; feature engineering analysis is carried out before and after modeling, and key influence factors are identified; and finally, integrating the model to a software interface, realizing low-temperature toughness real-time prediction based on material parameters, and guiding rapid design and performance evaluation of asphalt in a cold region. According to the method, the research and development period of the low-temperature high-toughness asphalt-based material can be remarkably shortened, and a reliable tool with interpretation capability is provided for asphalt material design, so that the durability and service sustainability of the asphalt pavement in the cold region are improved.

Description

Technical Field

[0001] This invention belongs to the field of asphalt pavement materials, and more specifically, relates to a data-driven method for predicting the low-temperature toughness of asphalt-based materials. Background Technology

[0002] my country has a wide distribution of cold regions, and the harsh service environment easily leads to various diseases of asphalt pavements, seriously affecting their performance. Therefore, developing asphalt-based materials with high toughness at low temperatures is of great significance for improving the service capacity of asphalt pavements in cold regions. Currently, improving the low-temperature toughness of asphalt-based materials mainly relies on the addition of modifiers such as rubber powder (CR) and styrene-butadiene-styrene block copolymer (SBS), and their low-temperature performance is usually evaluated based on flexural rheological beam (BBR) tests; however, this method suffers from problems such as long test cycles, unstable results, and low efficiency. In addition, existing performance prediction models for asphalt-based materials mainly include empirical models, mechanical models, and mechanical-empirical combined models, which often fail to reveal the intrinsic mechanisms of performance index changes.

[0003] Machine learning and deep learning technologies offer new approaches to predicting the low-temperature toughness of asphalt-based materials. Machine learning relies on data to train models, enabling computer systems to automatically learn patterns and make predictions or decisions. Deep learning, as a branch of machine learning, utilizes multi-layer neural networks to automatically extract complex features. Currently, machine learning and deep learning have achieved some application in the performance prediction of asphalt-based materials, but most research focuses on predicting rheological properties, identifying complex relationships between variables and achieving high-precision predictions by processing large amounts of data such as rheological parameters, component analysis, and performance tests. Research on low-temperature toughness prediction is relatively lacking. Existing methods, such as predicting the low-temperature performance of asphalt based on dynamic shear rheology (DSR) tests, still struggle to reveal the specific influence of characteristic factors on low-temperature toughness, and cannot effectively support the reverse design of asphalt-based materials with low-temperature toughness as the target. Summary of the Invention This invention aims to address the problems of poor low-temperature toughness of asphalt-based materials in cold regions and the low efficiency of traditional performance prediction methods. It proposes a data-driven method for predicting the low-temperature toughness of asphalt-based materials, significantly shortening the development cycle of low-temperature high-toughness asphalt-based materials. This provides a reliable and interpretable tool for asphalt material design, improving the durability and sustainability of asphalt pavements in cold regions.

[0004] To address the aforementioned technical problems, the present invention adopts the following technical solution: The purpose of this invention is to provide a data-driven method for predicting the low-temperature toughness of asphalt-based materials, specifically including the following steps: Step 1: Test the low-temperature toughness of different types of base asphalt, SBS modified asphalt with different SBS modifier dosages, and CR modified asphalt with different CR modifier dosages at different temperatures and different insulation times, and test their basic physical properties and chemical composition characteristics to build the database required for the model. Step 2: Clean and preprocess the database collected in Step 1, check and handle outliers and missing values, perform one-hot encoding on categorical features, and standardize or normalize numerical features to ensure data scale consistency. Step 3: Divide the preprocessed dataset into a training set and a validation set; Step 4: Determine the input features and output labels of the model. Use the basic physical properties, chemical composition and modification information of asphalt as the input features, and the low-temperature toughness performance index as the output label. Step 5: Pre-modeling feature engineering analysis, constructing composite features, and mining feature combinations strongly related to low-temperature toughness performance; Step 6: Train the initialized DNN model using the training set and optimize the hyperparameters, using root mean square error or mean absolute error as the optimization objective, to find the best combination of hyperparameters for the model. Step 7: Determine the optimal configuration of hidden layers and neurons to balance model prediction performance; Step 8: Train multiple DNN models independently using the fixed architecture and the same hyperparameters from Step 7, using non-fixed random seeds, allowing the DNNs to start with different initial weights; Step 9: Evaluate and select the optimal model; evaluate the generalization performance of all tuned candidate models on the validation set to determine the goodness of fit Rfit. 2 As a standard, the model that performs best on the validation set is selected as the final low-temperature toughness prediction model; Step 10: Post-modeling feature engineering analysis to accurately identify the most important features affecting the low-temperature toughness of asphalt-based materials, quantify the contribution of each feature to the prediction results, and analyze the influence relationship between key features and low-temperature toughness. Step 11: Solidify the trained final model into a model file and integrate it into a user-friendly software interface. Users can predict the low-temperature toughness index of the asphalt-based material in real time by inputting the parameters of the asphalt-based material, which can be used to guide the rapid design and performance evaluation of asphalt materials in cold regions.

[0005] Further specifying, in step 1, the base asphalt is 70# base asphalt or 90# base asphalt.

[0006] Further specifying, in step 1, the SBS modifier dosage is 3%, 4%, 5%, or 6%.

[0007] Further specifying, in step 1, the dosage of the CR modifier is 18%, 20%, 22%, or 24%.

[0008] Further specifying, in step 1, the different temperatures are -12℃, -18℃, and -24℃.

[0009] Further specifying, in step 1, the different heat preservation times are 1h, 6h, 12h, 18h, 24h, 36h, and 48h; Further specifying, in step 1, the low-temperature toughness property is the fracture energy measured by the single-sided notched small beam (SENB) test and the stiffness modulus measured by the bending rheodynamic beam (BBR) test. S .

[0010] Further specifying, in step 1, the basic physical properties are penetration (25℃, 100g, 5s), softening point (ring and ball method), and ductility (5℃, 5cm / min).

[0011] Further specifying, in step 1, the chemical composition characteristic is the content of the four components of asphalt, namely, the content of saturated components, aromatic components, resins, and asphaltenes.

[0012] To further specify, in step 2, the category feature is the type of modifier.

[0013] Further specifying, in step 2, the normalization process is the Z-Score method, as shown in the following formula:

[0014] In the formula, X is the data value, μ is the mean of the dataset, and σ is the standard deviation.

[0015] To further specify, in step 3, the dataset is partitioned randomly, with the ratio of training set to validation set being 8:2.

[0016] Further specifying, in step 4, the input features are asphalt grade, type of modifier, modifier dosage, and content of the four components.

[0017] To further specify, in step 4, the output labels are fracture energy and stiffness modulus.

[0018] Further specifying, in step 5, the feature engineering analysis method before modeling is random forest and analysis of variance test method; To further specify, in step 6, the model uses the Adam method for iteration with an initial learning rate of 0.00001.

[0019] To further specify, in step 7, the model network structure has three hidden layers with 64, 32, and 16 neurons respectively; To further specify, in step 8, there are 6 DNN models.

[0020] Further specifying, in step 9, the R 2 for:

[0021] In the formula: These represent the true values ​​of the low-temperature toughness properties (fracture energy, stiffness modulus) of asphalt-based materials. These are predicted values ​​for the low-temperature toughness properties (fracture energy, stiffness modulus) of asphalt-based materials. It is the average value of the data.

[0022] To further specify, in step 10, the method is the SHAP (Shapley Additive Explanations) method, which is used to interpret the model and assign a specific predicted importance value to each feature.

[0023] Compared with the prior art, the present invention has the following beneficial effects: This invention clarifies the relationship between asphalt type, modifier type, modifier dosage and low-temperature toughness of asphalt-based materials, providing new ideas and methods for the design of cold-region asphalt-based materials guided by low-temperature toughness.

[0024] This invention can significantly shorten the research and development cycle of low-temperature high-toughness asphalt-based materials, provide a reliable and interpretable tool for asphalt material design, and thus improve the durability and service sustainability of asphalt pavements in cold regions.

[0025] For a deeper understanding of the features and technical content of this invention, please refer to the accompanying detailed description and drawings. It should be noted that the drawings are provided for illustrative purposes only and are not intended to limit the scope of the invention. Attached Figure Description

[0026] Figure 1 This is a diagram verifying the fracture energy of asphalt-based materials using the model of this invention; Figure 2 This is a verification diagram of the stiffness modulus of asphalt-based materials using the model of this invention. Detailed Implementation

[0027] The present invention will be described in detail below with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but should not be considered as limiting the present invention. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0028] Example: The data-driven method for predicting the low-temperature toughness of asphalt-based materials in this example is implemented through the following steps: Step 1: Test the low-temperature toughness of 70# and 90# matrix asphalt, SBS modified asphalt with SBS modifier content of 3%, 4%, 5%, and 6%, and CR modified asphalt with CR modifier content of 18%, 20%, 22%, and 24% at different temperatures and different insulation times, and test their basic physical properties and chemical composition characteristics to build the database required for the model. The different temperatures are -12℃, -18℃, and -24℃. Different heat preservation times are 1h, 6h, 12h, 18h, 24h, 36h, and 48h; The low-temperature toughness properties are the fracture energy measured by the single-sided notched small beam (SENB) test and the stiffness modulus measured by the bending rheodynamic beam (BBR) test. S ; The basic physical properties are penetration (25℃, 100g, 5s), softening point (ring and ball method), and ductility (5℃, 5cm / min). The chemical composition characteristics are the content of the four components of asphalt, namely, the content of saturated components, aromatic components, resins, and asphaltenes. Step 2: Clean and preprocess the database collected in Step 1, check and handle outliers and missing values, perform unique thermal coding on the modifier type, and standardize or normalize the numerical features to ensure data scale consistency. The normalization process is the Z-Score method, and the formula is as follows:

[0029] In the formula, X is the data value, μ is the mean of the dataset, and σ is the standard deviation; Step 3: Divide the preprocessed dataset into a training set and a validation set using a random partitioning method, with the ratio of the training set to the validation set being 8:2; Step 4: Determine the input features and output labels of the model. The input features are asphalt grade, type of modifier, modifier dosage, and content of the four components. The output labels are fracture energy and stiffness modulus. The basic physical properties, chemical composition, and modification information of asphalt are used as input features, and the low-temperature toughness performance index is used as output labels. Step 5: Use random forest and analysis of variance to perform feature engineering analysis before modeling, construct composite features, and mine feature combinations that are strongly correlated with low-temperature toughness performance; Step 6: Train the initialized DNN model using the training set. The model is iterated using the Adam method with an initial learning rate of 0.00001. Hyperparameter optimization is performed, with root mean square error or mean absolute error as the optimization objective, to find the optimal combination of hyperparameters for the model. Step 7: Determine the optimal configuration of hidden layers and neurons. The model network structure has three hidden layers with 64, 32, and 16 neurons respectively, to balance the model's predictive performance. Step 8: Train 6 DNN models independently using the fixed architecture and the same hyperparameters from Step 7, and name them DNN-1, DNN-2, DNN-3, DNN-4, DNN-5, and DNN-6 respectively. Use random seeds and allow the DNNs to start with different initial weights. Step 9: Evaluate and select the optimal model. Evaluate the generalization performance of all tuned candidate models on the training and validation sets, using the goodness-of-fit R². 2 Based on prediction error and prediction time as criteria, the best-performing DNN-2 model was selected as the final low-temperature toughness prediction model. The model was then tested using samples from the validation set, and the goodness of fit for fracture energy and stiffness modulus were calculated to be 0.97 and 0.95, respectively. Prediction results for some samples in the validation set are shown below. Figure 1 , Figure 2 As shown, this demonstrates the accuracy of the model; The R 2 for:

[0030] In the formula: These represent the true values ​​of the low-temperature toughness properties (fracture energy, stiffness modulus) of asphalt-based materials. These are predicted values ​​for the low-temperature toughness properties (fracture energy, stiffness modulus) of asphalt-based materials. It is the average value of the data; Step 10: Post-modeling feature engineering analysis to accurately identify the most important features affecting the low-temperature toughness of asphalt-based materials, quantify the contribution of each feature to the prediction results, and analyze the influence relationship between key features and low-temperature toughness. Step 10 uses the SHAP (Shapley Additive Explanations) method to interpret the model and assign a specific predicted importance value to each feature. Step 11: Solidify the trained final model into a model file and integrate it into a user-friendly software interface. Users can predict the low-temperature toughness index of asphalt in real time by inputting the basic parameters of asphalt, which can be used to guide the rapid design and performance evaluation of asphalt-based materials in cold regions.

[0031] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0032] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A data-driven based method for predicting low-temperature toughness of asphalt-based materials, characterized in that, The method comprises the following steps: Step 1: test the low-temperature toughness performance of different types of base asphalt, SBS modified asphalt with different SBS modifier contents, and CR modified asphalt with different CR modifier contents at different temperatures and different holding times, and test the basic physical performance and chemical composition characteristics of the asphalts, and build a database required for modeling; Step 2: data cleaning and preprocessing are performed on the database collected in step 1, abnormal values and missing values are checked and processed, category features are one-hot encoded, and numerical features are standardized or normalized to ensure uniform data scale; Step 3: the preprocessed data set is divided into a training set and a validation set; Step 4: the input features and output labels of the model are determined, the basic physical performance, chemical composition and modification information of the asphalt are taken as the input features, and the low-temperature toughness performance index is taken as the output label; Step 5: feature engineering analysis before modeling is performed to construct composite features and mine feature combinations strongly related to the low-temperature toughness performance; Step 6: the initialized DNN model is trained using the training set, and hyperparameter optimization is performed to find the best hyperparameter combination of the model, taking the root mean square error or the mean absolute error as the optimization target; Step 7: the best configuration of the hidden layer and the neuron is determined to balance the prediction performance of the model; Step 8: multiple DNN models are independently trained with the fixed architecture and the same hyperparameters in step 7, a random seed is used, and the DNN is allowed to start with different initial weights; Step 9: Evaluate and select the best model; evaluate the generalization performance of all the tuned candidate models on the validation set to the goodness-of-fit R 2 As a standard, select the model that performs best on the validation set as the final low-temperature toughness prediction model; Step 10: feature engineering analysis after modeling is performed to accurately identify the features most important to the low-temperature toughness performance of the asphalt base material, quantify the contribution of each feature to the prediction result, and analyze the influence law between the key features and the low-temperature toughness; Step 11: the trained final model is solidified into a model file and integrated into a user-friendly software interface, and the user can input the parameters of the asphalt base material to predict its low-temperature toughness index in real time, which is used to guide the rapid design and performance evaluation of asphalt materials in cold regions.

2. The method of claim 1, wherein, In step 1, the base pitch is 70# base pitch, 90# base pitch; the SBS modifier content is 3%, 4%, 5%, 6%; the CR modifier content is 18%, 20%, 22%, 24%; different temperatures are -12℃, -18℃, -24℃; different holding time lengths are 1h, 6h, 12h, 18h, 24h, 36h, 48h; the low-temperature toughness performance is the fracture energy measured by single-edge notched beam (SENB) test, the stiffness modulus measured by bending beam rheological (BBR) test S ; the basic physical performance is penetration (25℃, 100g, 5s), softening point (ring and ball method), ductility (5℃, 5cm / min); the chemical composition feature is the content of saturated component, aromatic component, colloid, asphaltene of pitch.

3. The method of claim 1, wherein, In step 2, the category features are the types of modifiers; the normalization processing is the Z-Score method, and the formula is as follows: In the formula, X is the data value, μ is the average value of the data set, and σ is the standard deviation.

4. The method of claim 1, wherein, In step 3, the data set is divided by random division, and the ratio of the training set to the validation set is 8:

2.

5. The method of claim 1, wherein, In step 4, the input features are the asphalt grade, the type of modifier, the modifier content, and the content of the four components; and the output label is the fracture energy and the stiffness modulus.

6. The method of claim 1, wherein, In step 5, the feature engineering analysis method before modeling is the random forest and the variance analysis test method.

7. The method of claim 1 wherein, In step 6, the model is iterated using the Adam method, and the initial learning rate is 0.00001.

8. The method of claim 1, wherein, In step 7, the model network structure has three hidden layers with 64, 32 and 16 neurons respectively; and in step 8, the DNN model is 6.

9. The method of claim 1 wherein, In step 9, the R 2 is: wherein: is the true value of the low temperature toughness properties (fracture energy, stiffness modulus) of the asphalt-based material, is the predicted value of the low temperature toughness properties (fracture energy, stiffness modulus) of the asphalt-based material, is the average value of the data.

10. The method of claim 1, wherein, In step 10, the method is the SHAP (Shapley Additive Explanations) method, which is used to explain the model and assign importance values to each feature for specific predictions.

Images