PARSE: Physical attribute representativity and stationarity evaluator open-source library for 3D images using scalar and vector metrics
Description
The concept of representative volume (REV, or RVE in material sciences) is the cornerstone of continuum scale models – one needs to get the volume big enough so that it could be represented with an averaged value (scalar, vector or tensorial, etc.). While REV should be established in all larger scale simulations, this is rarely done in practice despite widespread adoption of 3D imaging devices in all research areas starting from petroleum engineering, material sciences and spanning to biology. Sometimes REV analysis is performed in the form of very simple procedures, such as a check for convergence of porosity or surface area, which is technically identical to omitting it entirely. The main reason for this to happen is the poor understanding of REV concept in general (mainly its connection to spatial stationarity and necessity for vector metrics with high information content) and unavailability of open-source robust solutions. In this work we present PARSE library that solves exactly this problem – we developed an easy to use and well-documented code based on rigorous research carried out recently in explaining the “dark sides” of representativity. In addition to REV, our code allows spatial stationarity analysis and comparison of samples (with subsequent clusterization into different groups) based on vector metrics – correlation functions, persistence diagrams and pore-network statistics, that altogether possess high information content which is critical in establishing stationarity and REV. We test our library on images produced by known statistical processes, such as Poisson spheres. After verification, we show how to compare different samples and group them depending on their “structural DNA”. All solutions explained in the paper are represented by Jupiter notebooks that can be used to perform similar analysis, moreover, the class structure of PARSE library allows painless modifications to be implemented. We believe that such an open-source library will be useful in numerous fields and will become an invaluable tool for 3D image analysis.