NONLINEAR DYNAMIC MODEL FOR ASPHALT PAVEMENT BACK-ANALYSIS

Abstract

The falling weight deflectometer (FWD) is at present the most widely used equipment for nondestructive testing of pavements. Pavement surface deflection data obtained from FWD test are used to derive the resilient properties of pavement layer materials using a back-analysis procedure. In this study, a nonlinear dynamic model for asphalt pavement back-analysis is developed. The finite element method is used for the nonlinear dynamic analysis. For parameter estimation, a nonlinear least squares optimization routine is employed. The developed program was then used to study the major factors affecting the results of asphalt pavement back-analysis. Although it is generally recognized that the nonlinear dynamic model is more realistic for FWD data analysis, most of the current back-analysis methods are based on linear static, nonlinear static or linear dynamic forward analysis models. It was found in Problem Case (1) that the critical strains predicted by these models can differ quite considerably from the ones given by the nonlinear dynamic model. FWD loading is basically an impact load, but most current procedures idealize it as a static load. The effect of neglecting the dynamic loading mode was pronouncedly shown in Problem Case (2). Static models yield different set of resilient parameters for the same hypothetical pavement section when the bedrock depth is varied. The resilient moduli of unbound granular material and subgrade soil are stress-dependent nonlinear elastic materials, but most back-analysis methods at present modelled them as linear elastic materials. Through Problem Case (3), the effect of material nonlinearity is demonstrated. It was found that, when FWD loads of different magnitudes were applied onto a pavement section and parameters were subsequently backcalculated by linear models, the results varied with the FWD load level. The modulus of asphalt concrete is highly temperature dependent. The modulus of a thick asphalt layer may vary with depth following the temperature profile in the layer. It was found in Problem Case (4) that if the temperature profile effect is not taken into account, the predicted critical strain in the asphalt layer can differ quite significantly. Since the deflection time-history data recorded by the FWD sensors contain much more information than the peak values alone, a procedure is proposed to use deflection pulses in dynamic backcalculation. It is demonstrated that the parameter estimation accuracy of deflection time-history method was better than that of deflection basin method. The error decreases with the increase in the number of time instants used for back-analysis. Another advantage of using deflection time-history is better data conditioning. A condition number which is a measure of the overall collinearity of the parameters is found to be smaller for deflection time-history method than deflection basin method.