Predictive Faulting Models in Jointed Concrete Pavement
Loading...
Date
2020-07-16
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Faulting is a major and commonplace distress in Jointed Concrete Pavement (JCP), which can directly cause pavement roughness and adversely influence the ride quality of a vehicle. Faulting also plays an essential role in concrete pavement design. Notwithstanding the importance of faulting, the accuracy and reasonability of the existing faulting prediction models are still controversial. To enhance the concrete pavement design, this research proposes the novel models to estimate faulting depth at joints in the wheel path in JCP, including mechanistic-empirical models, machine learning model, and probabilistic model.
The mechanistic-empirical models are explicit mathematical equations that are easily used for the concrete pavement design and pavement maintenance. The mechanistic-empirical models were proposed in this research by a comprehensive exploration of the full process of faulting in which two faulting stages are found and an inflection point as a critical faulting depth is discerned to differentiate the two stages of faulting. This research proposes two mechanistic-empirical models to characterize the jointed faulting over the entire service life. 1) One is the complete model which can be used to characterize the faulting depth over time in the full progress of faulting that contains two stages faulting (i.e., the prior-inflection point and posterior-inflection point); 2) the other is the load-related model to determine the axle load distribution on faulting initiation at the stage of prior-inflection point.
Machine learning is believed to be powerful method to explore the knowledge in the LTPP faulting data. Using the LTPP data, a list of the popular machine learning models were trained and validated with the LTPP data. The best model from the machine learning model candidates was selected for the faulting prediction. Through a test of the cross validation, the random forest model was chosen for faulting prediction and proved to be compelling and reliable as it holds the satisfactory accuracy of prediction.
Probabilistic model of the faulting was modeled as a stochastic process. Markov Chain, which is the one of the widely used stochastic models, was adopted to characterize the uncertainty in faulting predictions across the chain. The transition probability matrix in the Markov Chain model was constructed to connect a sequence of Markov chains. The development of Markov chain model requires a sufficient and extensive sample of data. The LTPP data seem inadequate for the model development such that the Monte Carlo simulation was adopted to generate a large amount of data. In a progressive manner, the development of Markov chain model includes 1) preparation of data by conducting the Monte Carlo simulation 2) the repeated construction of the transition probability matrices across chains by counting the occurrences of each state of faulting.
Description
Keywords
Jointed concrete pavement, Joint Faulting, Long-term pavement performance, Prediction model