Genetic Algorithm-Based Relationships among Outputs from TSDDs and FWD

Abstract: The traffic speed deflectometer devices (TSDDs) have emerged as a viable tool for the network level evaluation of the structural capacity of roads. Several recent studies have shown that TSDDs meet a minimum set of specifications, including precision and accuracy, required for network level evaluation. However, not many rigorous algorithms have been developed for the more quantitative analysis of the pavements. For that reason, TSDD-measured deflection parameters are used for the most part for segmentation of network into the less stiff to more stiff areas. Some recent more quantitative studies have utilized the well-established analysis algorithms developed for falling weight deflectometer (FWD) data, assuming that the TSDD is acting as a “moving FWD.” Even though pragmatic, this process may not be accurate due to the different loading regimes and testing configurations.
In this study, an attempt has been made to develop transfer functions among the FWD- and TSDD-measured deflection parameters considering the differences in their operational parameters and the characteristics of the pavement structure being tested. A symbolic regression approach was employed to establish a relationship for deflections from 10,000 three-layer flexible pavement sections (comprising of AC, base, and subgrade) with different randomly-distributed layer thicknesses and moduli. The 3D-MOVE software was used to simulate and analyze the responses of the pavement sections under the moving load of TSDD and the static load of FWD. The major parameters controlling the transfer functions are the thickness of the hot mix asphalt (HMA) layer and the modulus of subgrade. While the HMA thickness can be obtained from ground penetrating radar (GPR), which is also a non-destructive technique, the modulus of subgrade is usually unknown. Therefore, an algorithm for estimating subgrade modulus from TSDD measurements is also provided. The symbolic regression approach was used to relate the deflections from the TSDD and FWD devices at the same distances from the load, and for estimating modulus of subgrade. The ability to simultaneously generate both the form and the parameters distinct the symbolic regression from traditional linear or nonlinear regression, which they only optimize the parameters of a predefined equation.