Implementation of Artificial Neural Network for Prediction of Pavement Structure Strains at Critical Locations

Abstract

Falling Weight Deflectometer (FWD) test is commonly used to evaluate the conditions of a pavement structure. The surface deflections measured by a FWD test are normally used in back-calculation analysis to determine the elastic Young’s modulus of the pavement structure materials, which later on is inputted into a forward calculation, usually by a Linear Elastic Analysis (LEA), to determine the strains mobilised at the critical locations (e_t,ac and e_c,sg) in the pavement structure for evaluation of the remaining life. It is of interest to develop a tool for predicting the values of e_t,ac and e_c,sg directly from the FWD deflections while bypassing the above-mentioned back- and forward calculations, which are highly time-consuming. In this research, artificial neural network (ANN), which is a function built-in MATLAB2020 program, was used as the tool for such a prediction. There are three types of pavement structures investigated, which are: i) cement-modified crushed rock base pavement structure; ii) combined-surface pavement structure; and iii) thin-surface pavement structure. A database consisting of the strains at the critical locations and the FWD deflections for each pavement structure type, which were obtained by data generating with LEA in a previous research, were used. The FWD deflections were transformed to various deflection basin parameters (DBPs), and then used to train ANN to correlate with the strains at the critical locations. By comparing the strains predicted by ANN with the ones from LEA as the input, it is found that, in general, the maximum error is around only 3%. In addition, the results predicted by ANN in the present study are substantially more accurate than the ones predicted by a non-linear regression method with statistical equations of the previous study. Hence, the developed ANN can be used to analyse the FWD deflections to determine the critical location’s strains for evaluating the conditions of a pavement structure.