Developing Machine-Learning Models to Predict Airfield Pavement Responses

Abstract

Aviation promotes trade and tourism by connecting regions, people, and countries. Having a functional and efficient airport pavement network is important to improve aviation traffic and to provide safer mobility to almost 800 million passengers travelling in the U.S. per year. The Federal Aviation Administration has initiated and actively been participating in many projects to further advance pavement design and performance to meet user requirements. To accomplish that, quantitative data are needed; such data may be collected from the pavement response to gear and environment loading. In this study, responses from four instrumented taxiway concrete slabs at John F. Kennedy International Airport were analyzed. The collected data were used to develop machine-learning (ML) based prediction models to compute the temperature, curling and bending strains within pavement. The ML models were developed using the support vector machine (SVM) algorithm. The results showed that SVM based ML models can predict pavement responses with a high accuracy and low computation time. Furthermore, in the case of feeding more data from various airports, ML models have proven to be a promising technique for pavement analysis engine for future airport pavement design frameworks. This study also produces recommendations for future data collection projects to have well-designed databases for data-driven models development.

Effects of Different Training Datasets on Machine Learning Models for Pavement Performance Prediction

“…As indicated in previous studies, the most frequently used ML algorithms in pavement performance prediction are support vector regression (SVR), RF, and ANN (8)(9)(10)(11)(12). Therefore, their theoretical bases, characteristics, and calibration hyperparameters are described in the following.…”

“…In addition to the completeness of the database, training data preprocessing has a great impact on a model's predictive performance, is highly dependent on the modeler's experience, and is not typically reported in engineering-related literature (8). In fact, most research focuses on the ML algorithms used and the evaluation metrics, without describing the training database construction and final predictors (8)(9)(10)(11)(12).…”

Multi-depth temperature prediction using machine learning for pavement sections

“…Machine learning (ML) has been utilized for a wide range of applications, such as predicting the temperature within pavement structures. Gungor and Al-Qadi 22 applied ML to calculate the temperature of an HMA layer using a model that included air temperature, direction and speed of the wind, dew point temperature, station pressure, and cloud ceiling. Qiu et al 23 .…”

“…The thermal properties of materials are very important for conventional methods but it is difficult to obtain accurate data (like the thermal conductivity of a material is influenced by its status, such as whether the material is frozen or not), and once the temperature recordings are available, the ML models could be applied to predict the temperature for specific roads without the thermal properties of materials. The ML models 9 , 22 , 23 summarized above show high accuracy in predicting the pavement structure’s temperature, and parameters related to the road materials—i.e., thermal conductivity, specific heat capacity, and so on—are not necessary. This indicates that ML models have potential for improving the temperature prediction at various depths within pavement structures.…”

“…This indicates that ML models have potential for improving the temperature prediction at various depths within pavement structures. Although temperature of the soil layer also influences the pavement’s performance, 5 , 6 previous work 9 , 22 , 23 focused on the temperature of the HMA layer and surface temperature. Thus, it is necessary to extend the application of ML models to different sub-layers.…”