Evaluating statistical homogeneity of cone penetration test (CPT) data profile using auto-correlation function

Abstract

Cone penetration test (CPT) data profile is often numerically modelled using random field theory and based implicitly on a fundamental assumption: statistical homogeneity of the preprocessed stationary CPT data profile. In classical random field theory, statistical homogeneity evaluation may only be achieved if the CPT data profile is infinite in length. While in engineering practice, it is often the case that a CPT data profile has limited sounding depth. This leads to a critical question that whether the CPT data profile satisfies statistical homogeneity assumption to enable proper random field modelling. This paper proposes a novel non-parametric method for evaluating statistical homogeneity of a single CPT data profile. A novel unified representation of CPT data profile pattern is developed in this study using discrete cosine transform (DCT)-based auto-correlation function (ACF). Statistical homogeneity evaluation of a CPT data profile is reformulated as statistical analysis of pattern similarity between the original CPT data profile and its partial segment profiles. The proposed method is general since it is distribution-free and does not require a parametric correlation function. The proposed method performs well on both simulated and real examples and can identify the possible locations of heterogeneity in a CPT profile.