Statistical calibration of federal highway administration simplified models for facing tensile forces of soil nail walls

This study first develops a database containing 56 measured facing tensile forces from instrumented soil nail walls during or at completion of construction. Based on the compiled database, the accuracies of both default and modified federal highway administration (FHWA) simplified models for estimation of short-term facing tensile forces are evaluated. Here, accuracy is defined by the model bias computed as the ratio of measured to predicted facing tensile force. The analysis results show that predictions are highly conservative and highly dispersive using the default model, and moderately conservative and medium dispersive using the modified model. Moreover, the prediction accuracy is statistically correlated with magnitudes of the computed facing tensile forces and several model input parameters. An out-of-sample approach is used to develop and validate a recalibrated FHWA model which is then demonstrated to have least empirical constants but best accuracy compared to the default and modified models. The biases for the three models are characterized as lognormal random variables. An example of reliability-based analysis for facing flexure limit state is illustrated to both elaborate the application and highlight the benefit of using the recalibrated model for design practice from the perspective of cost-effectiveness.