A Step Towards Quantifying the Uncertainty of the Soil Mechanical Response Through the Use of Genetic Algorithms

This article proposes a methodology based on genetic algorithms to quantify the uncertainty of soil mechanical response through an optimization tool for the automatic parameter calibration of constitutive models. The optimization tool was developed in order to improve not only the efficiency of the parameter calibration process, but also to establish an objective framework for quantifying material uncertainty by even considering the natural variability resulting from laboratory investigations. This tool was validated for the Hypoplastic constitutive model using Karlsruhe fine sand and considering odeometer tests and triaxial tests under both drained and undrained conditions. The “best fitting” or global parameters are calibrated with the tool based on the aggregate of the experimental data and a methodology is proposed whereby an uncertainty based approach is utilised for the determination of parameter bounds. In addition, a range of different parameter “sets” or local parameters can be determined, where the model parameters are calibrated separately for each of the various ‘planes’ or loading paths of geotechnical interest, e.g. ε1-q. The potential of the tool is demonstrated by a comparison of finite-element simulations of a braced excavation using the Hypoplastic constitutive model, performed using the global and local calibrated parameter sets.