Supervised Learning to Improve Software for Crude Oil Analysis Using Low-Field NMR Relaxometry

Abstract

An important challenge in the petroleum industry is finding efficient methods to determine the physicochemical characteristics of crude oils, including but not limited to viscosity, density, and sulfur content. The conventional American Society for Testing and Materials (ASTM) methods applied for petroleum characterization are labor-intensive and involve toxic chemicals. These drawbacks have prompted researchers to seek alternative approaches. Among these, the low-field nuclear magnetic resonance (LF-NMR) method has gained significant attention. Despite NMR technology’s long-standing use in the petroleum industry for over 60 years, LF-NMR has recently been adopted due to its cost-effectiveness, ease of operation, and minimal sample preparation requirements. In this contribution, we improved our previously developed software, based on 24 crude oils, in terms of accuracy and precision with 87 crude oil samples. Additionally, we now integrate new features with the supervised learning approach to enhance the fast and reliable identification of crude oils to provide solutions in handling crude oils at different stages, such as production and refinery.