An interactive data-driven assistant for automated petrophysical rock typing

Abstract

Petrophysical Rock Typing (PRT) is vital for reservoir characterization, yet traditional methodsrely on subjective interpretation or complex graphical workflows. The Hydraulic Flow Unit (HFU) method, though conceptually strong, lacks practical adoption due to implementation challenges. This thesis offers a fully automated, reproducible HFU workflow. It uses the Ramer-Douglas-Peucker algorithm to segment the Stratigraphic Modified Lorenz Plot (SMLP) and applies machine learning to classify segments into rock types. A key contribution is an expert-in-the-loop framework that enables iterative refinement using data-driven metrics. Rock types are further characterized with Pore Throat Radius Indicator boundaries to enhance interpretability. The approach was validated on the Gulfaks and Poseidon datasets and implemented as a web-based module combining automation with expert control. It supports both rapid default deployment and advanced customization, offering a scalable, consistent tool for reservoir studies.