Identification of non‐linear stress–strain–time relationship of soils using genetic algorithm

Recognition of non‐linear constitutive rock/soil model from experimental results is often multi‐modal in the large parameter space. A genetic evolution algorithm is thus proposed for its recognition, including that of structure of the model and coefficients in the model. The structure of the model can be firstly determined according to mechanical mechanism if the mechanism is clearly understood or searched by using evolutionary algorithm. The coefficients to be determined are then searched in global optional space. With the new evolutionary algorithm, the non‐linear stress–strain–time constitutive law to describe strain softening behaviours of diatomaceous soil under consolidated and undrained state was recognized by learning stress–strain–time behaviour of an intact sample under consolidated pressure of σ_c=0.1 MPa and strain velocity of_a=0.175%/min. This model gave reasonable prediction for diatomaceous soils under varying consolidated pressures (0.1–3.5 MPa) and strain velocities (0.0044–1.75%/min). It indicates that the methodology proposed in this paper is robust enough and strongly attractive for recognition of non‐linear constitutive model of soil and rock materials. Copyright © 2002 John Wiley & Sons, Ltd.