Prediction of Multi-layered Pavement Moduli Based on Falling Weight Deflectometer Test Using Soft Computing Approaches

The application of supervised machine learning algorithms to provide solutions for various civil engineering problems is an emerging trend. This paper presents the utilization of artificial neural network (ANN) and random forest regression (RFR) for the prediction of the elastic moduli of multi-layered pavement based on the falling weight deflectometer (FWD) test. The establishment of ML models includes data preprocessing, hyperparameter optimization, and performance evaluations. The ML models are created from both ANN and RFR techniques using 122,500 datasets from a theoretical model of the FWD test, generated by employing an exact stiffness matrix method for the analysis of multi-layered flexible pavement. The performance measures of both ML models, developed from the synthetic dataset, indicate that the output variables (the predicted pavement moduli) are precisely explained by the input parameters (the measured surface displacements). Both ML solutions are then compared with the FWD test results performed on the road infrastructures in Thailand, showing good agreement with the predicted moduli from the FWD tests. Between the two ML solutions, RFR displays better accuracy in predicting the pavement moduli from the FWD tests with the R2 values of the predicted elastic moduli exceeding 90%. Besides, a sensitivity analysis is carried out to illustrate the impact of surface deflections recorded at each geophone on the predicted pavement moduli. The present study demonstrates the efficacy of ML techniques in assessing road infrastructures and highlights the significance of sensitivity analysis in enhancing the accuracy of pavement performance prediction.