Prediction of pavement stress using deflection data from falling weight deflectometer tests

Vertical stress is a key parameter for assessing subgrade performance and permanent deformation in a pavement structure. Traditionally, vertical stress is estimated by back-calculating pavement layer moduli from falling weight deflectometer (FWD) deflections, followed by analysis using layered elastic analysis (LEA) or finite element analysis (FEA). However, these methods are often time-consuming and sensitive to layer moduli inputs. This study explores a data-driven approach to directly predict vertical stress under moving load from FWD deflection data, thus eliminating the need for back-calculation of layer moduli. A shallow artificial neural network (ANN) model was trained and tested using field-measured FWD deflections and vertical pressure data collected from 11 full-scale pavement test lanes at the Turner-Fairbank Highway Research Center (TFHRC). The predictions using the ANN model showed better agreement with measured vertical stresses than a fitted multi-linear regression model and conventional LEA-based estimates using back-calculated moduli. The findings suggest that ANN-based models can serve as an effective alternative to conventional mechanistic approaches for predicting vertical stresses in flexible pavement structures using FWD deflection data.