모델링 오차를 고려한 아스팔트 포장구조체의 구조건전성 평가

Abstract

The structural integrity assessment technique for the asphalt pavement system is studied considering the modeling errors introduced by material uncertainties. To this end, the artificial neural network is utilized to estimate the elastic modulus of soil layers using the measured deflection data from the FWD (Falling Weight Deflectometer) test. A wave analysis program for a multi-layered pavement system is developed based on the spectral element method for more accurate and faster calculation. The developed program is applied for the numerical simulation of the FWD tests, more specifically the reliability analysis and the gen-eration of training and testing patterns for the neural network. The effects of uncertainties in the material properties for modeling a given pavement system such as Poisson ratio and layer thickness are intensively investigated using Monte Carlo Simulation. It is observed that the amplitude of impact load is most significant and then the layer thickness and the Poisson ra-tio are significant on the maximum deflections than other parameters. The evaluation capa-bility of neural network is also investigated when the input data is corrupted by the modeling errors, and it is found that the estimation results can be significantly deviated due to the mod-eling errors. To reduce the effect of the modeling error, (in other words to improve the ro-bustness of the algorithm), we proposed an alternative scheme to generate the training patterns considering modeling errors, and it is finally concluded that the estimation results can be improved by using the proposed training patterns from extensive numerical simulation study.