Development of asphalt dynamic modulus master curve using falling weight deflectometer measurements.

PDFTech ReportIHRB Project TR-659InTrans Project 13-471Asphalt concreteDynamic modulus of elasticityMechanistic-empirical pavement designCurves (Geometry)Falling weight deflectometersNeural networksViscoelasticityCoefficientsPavement layersUnited StatesIowa State University. Institute for TransportationGopalakrishnan, KasthuriranganKim, SunghwanCeylan, HalilKaya, OrhanIowa State University. Institute for TransportationIowa. Dept. of TransportationIowa. Highway Research BoardUS Transportation CollectionThe asphalt concrete (AC) dynamic modulus (|E*|) is a key design parameter in mechanistic-based pavement designmethodologies such as the American Association of State Highway and Transportation Officials (AASHTO) MEPDG/Pavement-ME Design. The objective of this feasibility study was to develop frameworks for predicting the AC |E*| master curve fromfalling weight deflectometer (FWD) deflection-time history data collected by the Iowa Department of Transportation (IowaDOT). A neural networks (NN) methodology was developed based on a synthetically generated viscoelastic forward solutionsdatabase to predict AC relaxation modulus (E(t)) master curve coefficients from FWD deflection-time history data. According tothe theory of viscoelasticity, if AC relaxation modulus, E(t), is known, |E*| can be calculated (and vice versa) through numericalinter-conversion procedures. Several case studies focusing on full-depth AC pavements were conducted to isolate potentialbackcalculation issues that are only related to the modulus master curve of the AC layer. For the proof-of-concept demonstration,a comprehensive full-depth AC analysis was carried out through 10,000 batch simulations using a viscoelastic forward analysisprogram. Anomalies were detected in the comprehensive raw synthetic database and were eliminated through imposition ofcertain constraints involving the sigmoid master curve coefficients.The surrogate forward modeling results showed that NNs are able to predict deflection-time histories from E(t) master curvecoefficients and other layer properties very well. The NN inverse modeling results demonstrated the potential of NNs tobackcalculate the E(t) master curve coefficients from single-drop FWD deflection-time history data, although the currentprediction accuracies are not sufficient to recommend these models for practical implementation. Considering the complex natureof the problem investigated with many uncertainties involved, including the possible presence of dynamics during FWD testing(related to the presence and depth of stiff layer, inertial and wave propagation effects, etc.), the limitations of current FWDtechnology (integration errors, truncation issues, etc.), and the need for a rapid and simplified approach for routineimplementation, future research recommendations have been provided making a strong case for an expanded research study