Modeling rutting depth on RIOHTrack asphalt pavement using Circle LSTMs

Abstract

In response to the critical need for accurate pavement rutting prediction in pavement design, this study introduces a novel deep-learning model, Circle LSTM, designed for the estimation of rutting depths in asphalt pavements. Leveraging data from a full-scale field accelerated road testing facility in Tongzhou, Beijing, which spans 2.038 kilometres and features 19 distinct asphalt pavements, the Circle LSTM model utilizes variables such as temperature and load axis for its predictions. Methodologically, the Circle LSTM outperforms traditional machine learning approaches by cycling through different LSTM models to enhance short-term dependence modeling, showcasing its strength in addressing nonlinear problems inherent in road engineering. Evaluation of the model on the test dataset yielded promising results, with error metrics showing an average RMSE of 2.4170, an average MAE of 2.9454, an average MAPE of 2.61%, and an average coefficient of determination (average R2) of 0.8253 for 19 different asphalt pavements. These findings underscore the Circle LSTM model’s potential not only for fast and reliable rutting depth prediction but also for broader applications in multivariate time-series forecasting tasks in road engineering and beyond.