Abstract
This paper proposes an evaluation method for the spatial variability of the strength distribution derived from cone penetration test (CPT) results inside an earth-fill dam composed of mixed materials with different particle size distributions. The in-situ data from the studied site contain outlying data which can cause the mis-estimation of the statistical parameters. Therefore, the outliers should be separated from the original data. In order to reduce the effects of outliers in modeling the random field of the studied site, the N-values converted from the CPT results are divided into three groups, namely, high, middle, and low. The root mean squared error (RMSE) in the middle group is calculated from the residual error between the variogram values and the regression function and depends on the threshold values for separating the N-values into the three groups. The threshold values between the high and low groups are determined so that the RMSE is minimized. From among the many candidates of threshold values, a pair of values that corresponds to the minimum case of the RMSE is finally determined as the optimal case. The three groups are modeled and simulated through the use of a geostatistical method. By combining the simulated values of the three groups, the original spatial distribution of strength can be evaluated. To validate the proposed method, part of the data is intentionally removed from the measured data. The remaining data are applied to the simulation. The simulated value and the removed data are compared at the same location to verify the correlation between them. As a result, the random field was simulated reasonably. Thus, it has been verified that the proposed method can be used to reasonably simulate the distribution of in-situ strength including outliers.
