نوع مقاله : یادداشت فنی
نویسندگان
دانشکده مهندسی عمران، دانشکدگان فنی، دانشگاه تهران، تهران، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
The measurement of failure characteristics in the semi-circular bending (SCB) test of asphalt mixtures has significantly expanded over the last decade, with laboratory study results being published in numerous studies. The purpose of this article is to develop neural network models to predict the fracture energy and fracture toughness of asphalt mixtures based on data mining principles. For this purpose, 3290 data points from SCB fracture test results of asphalt samples were collected from 102 credible articles. Out of these, 1627 data points are used to predict fracture energy and 1663 data points are used to predict fracture toughness.
The input layer of the neural networks includes data collected on fracture mode, loading rate, test temperature, sample thickness, notch dimension, presence of modifier, maximum nominal aggregate size, air percentage, binder percentage, aging, and binder type of asphalt mixtures. The output layer generates the energy and fracture toughness for each assumed input. The results show that the constructed neural network models can predict fracture energy with 75% accuracy and fracture toughness with 70% accuracy.
The sensitivity analysis reveals that the loading rate, test temperature, airvoid percentage, and binder percentage are the most influential characteristics on the prediction models' results. The integration of data mining principles and neural network algorithms enhances the prediction accuracy of asphalt mixture properties, which can aid in designing more durable pavement materials.
The accuracy of the models was validated using metrics such as Root Mean Squared Error (RMSE) and Mean Squared Error (MSE), with RMSE values of 0.727 for fracture energy and 0.170 for fracture toughness. The regression analysis between actual and predicted values showed R² values of 0.75 for fracture energy and 0.70 for fracture toughness, indicating robust model performance.
In conclusion, the neural network models based on collected SCB test data exhibit acceptable performance in predicting the fracture energy and toughness of asphalt mixtures. The study's findings highlight the importance of considering key input variables such as airvoid percentage, binder percentage, test temperature, and loading rate in developing reliable predictive models for asphalt mixture behavior.
کلیدواژهها [English]
)