The integration of numerical modeling and physical measurements through inverse analysis in geotechnical engineering

Abstract

In many geotechnical problems, boundary measurements of loads and displacements are known while the underlying soil behavior is unknown. Inverse analysis techniques are employed to quantify the underlying material behavior. The most widely used inverse analysis approaches are parameter optimization methods in which unknown properties of a material constitutive model are adjusted until the calculated displacements or forces match the measurements. The main limitation of the inverse analysis parameter optimization methods is the need for pre-defined constitutive models that do not benefit from fundamentally new information on material behavior. This paper describes the application of a new inverse analysis approach called SelfSim for extracting soil constitutive behavior via a neural network (NN) material model. This new type of inverse analysis procedure allows the user to focus on the underlying material behavior instead of a specific material constitutive model. The proposed approach is demonstrated for three geotechnical engineering problems: (1) laboratory triaxial tests with frictional loading plates, (2) deformations around a deep excavation and (3) seismic site response from a downhole array.