Modeling of Creep Compliance Behavior in Asphalt Mixes
Modeling of Creep Compliance Behavior in Asphalt Mixes Using Multiple Regression and Artificial Neural Networks Prof. Saad A. Abo-Qudais Esra’a I. Alrashydah aboqdais@just. edu. jo aiesraa 13@just. edu. jo Jordan University of Science and Technology – Department of civil engineering Jordan-Irbid
SUMMARY RESEARCH OBJECTIVE: This research aims to provide an appropriate approach to enhance asphalt mixtures creep compliance performance predictions and presents two predictive models, one with multiple regression analysis and the other with feed-forward artificial neural networks (ANN). RESEARCH SIGNIFICANCE: This research provides an attractive alternative for making a better primary decision about selecting asphalt mixtures variables in a quite short time with a very low error rate using feed-forward ANN model.
Research Approach Materials: -Asphalt Cement: The asphalt binder was obtained from Jordan petroleum refinery. Unmodified asphalt binder and asphalt binder modified with 2% Elvaloy by weight of the binder were used. The binders have a performance grade (PG) of 64 -22 and 76 -22, respectively. -Aggregate: Crushed limestone obtained from Al-Huson quarries at the northern part of Jordan was used to prepare the HMA specimens. Mid limits of Superpave aggregate gradation for 12. 5 mm nominal maximum size was used for preparing the HMA specimens. Specimens Preparation and Testing Methods 120 HMA specimens were prepared with different combinations including two asphalt modifications, three air voids levels, two aging conditions. All the prepared HMA specimens were exposed to a dynamic creep test at five different testing temperatures.
Description for Variables of Creep Compliance Model. Variable Description Testing temperature Quantitative variable (5, 15, 25, 45, 60 o. C) Loading time Quantitative variable (seconds) Air voids level Quantitative variable (3. 4, 4. 9, 6. 4 %) HMA modification Categorical variable (modified, unmodified) HMA aging condition Categorical variable (short-term aging, level 2 Long-term aging)
Models development
Results and analysis Creep compliance = 0. 008+0. 007 (Testing Temperature) - 0. 102 (Aging Condition) 0. 102 (Asphalt Modification) + 1. 7 E-5 (Loading Time) + 0. 009 (Air Voids Level) MSE value 1 0, 01 0, 0001 0 2 4 6 8 10 12 14 16 Number of hidden neurons ANN Performance plot. 18 20 22
Creep compliance ANN architecture. Creep compliance ANN prediction model
Relationship between the actual and predicted creep compliance values by ANN for (a). training data, (b). validation data, (c). testing data, and (d). all data.
Prediction performance comparison R 2 and MSE values were used as the basic criteria to accurately evaluate the performance of the two models and to verify how well the models represent the experimental creep compliance data. Statistical parameters Multiple regression model Feed-forward ANN R 2 0. 6180 0. 9915 MSE 0. 014 0. 000111
Conclusions and Recommendations Based on the results of this study, the following outcomes were achieved: - Loading time, testing temperature, asphalt modification, air voids level, and aging conditions have a significant impact on the magnitude of HMA creep compliance behavior. - The feed-forward ANN model has better capability to predict HMA creep compliance compared to multiple regression model as evidenced by the highest R 2 value and the lowest MSE value. - ANN found to be an effective tool to represent the HMA mechanical behavior and can be useful in the design of future studies on this topic for making a primary decision about the use of different variables in asphalt mixtures. - The feed-forward ANN model provides a practical advantage that allows the HMA creep compliance predictions in a quite short time with a very low error rate without performing numerous creep tests which means saving time and cost. Based on the results of this research, it is recommended to: Evaluate the effect of different aggregate gradations and different types of additives on the creep compliance values using feed-forward ANN and multiple regression models.
- Slides: 10