Predicting Frost Heave Using FROST Model with Centrifuge Models

The FROST numerical model was used to predict frost heave developing in centrifuge soil models. When uncalibrated predictions of heave using Gardner's coefficients were selected from accompanying FROST documentation based on soil grain size, the predictions were not good. However, when the parameters were calibrated to surface heave developing in one set of models of heave in silt, numerical predictions for other freezing conditions in the same silt showed very good matches to centrifuge model data of heave. These close matches occurred not only in heave development patterns but also in statistical distributions when variations of input soil parameters were considered. The same good fit after a similar calibration exercise was not found in either heave developments or statistical distributions in the case of silty clay, which develops heave following a different pattern. Predicted ultimate depths of frost penetration were considerably less than measured penetrations, and final water contents (frozen and unfrozen) predicted after freezing were reasonably close to centrifuge model results. A sensitivity analysis of seven input soil parameters required for the FROST model was conducted using the Monte Carlo simulation technique to assess the response of the FROST model to their random variation. Based on this analysis, the FROST model was found to be most sensitive to those parameters characterizing movement and retention of water, especially the Gardner's coefficients for unsaturated conditions. These parameters are also the most difficult to measure reliably, and therefore back calculation of these values is common. Data obtained from centrifuge models of soil freezing can therefore provide an attractive design calibration and research tool.