Validation metrics for non-linear soil models using laboratory and in-situ tests

The article discusses the application of statistical metrics for the validation of comprehensive non-linear soil models. The assessment was carried out on the basis of triaxial, oedometer, consolidation and plate load tests of sandy and clay soils. Validation of non-linear soil models was divided depending on the problem type: strength-type problem and strain-type problem. For a strength-type problem the indicators of failure points should be compared. In the course of strain-type problem the stress-strain curves should be compared. Average ratio of experimental data to calculated and coefficient of variation showed the highest efficiency for standard triaxial and oedometer tests, as they allow taking into account the specificity of the deviation and its variability. Other statistical metrics are less effective in geotechnical engineering. Validation according to consolidation tests is recommended to be performed based on the analysis of the time of 100% primary consolidation and the slope of the consolidation curve during the creep phase. In the course of stress paths analysis (for example, in the course of unloading and further reloading), the advantage should be given to visual assessment. Acceptable values of validation metrics for geotechnical engineering are proposed. The specific values of deviations should be determined by the analyst depending on the required accuracy of calculations, the responsibility of the construction object and the assessment of the risk of an accident.