Mathematical Modeling for Flow and Transport through Porous Media
Simulation and validation of oilfield reservoir rock properties using 3D-printing and machine learning
Dr. Mohamed S. Jouini
Several developments in Digital Rock Analysis technology showed the ability to better characterize rock properties at pore scale using 3D X-ray Micro CT images in sandstones but faced several limitations in carbonates due to heterogeneity. In this project, we have two main objectives: the first consists of using 3D printing technology to print extracted subsets in larger scale to make laboratory experiments feasible. Producing these 3D prints of rock subsets will allow validating our simulations experimentally at laboratory scale. The second objective of the study is to use texture analysis and machine learning to propose a reproducible workflow for multi scale analysis and to simulate rock properties in complex carbonates rocks. We present an application of our proposed approach on synthetic and real samples.
Data-Driven Reservoir Modeling Using Artificial Intelligence and Deep Learning
Dr. Mohamed S. Jouini
The general frame of the proposed research is the paradigm shift occurring in solving complex oil and gas industry problems using Artificial intelligence (AI). In particular, the research aims to develop reliable data-driven technologies both at reservoir and core plug scales. However, the heterogeneity of the core plug, the presence of pores at macro, micro and nano- scales, makes the characterization very complex. In this context, AI can help in building a reliable reservoir geological model based on mapping of the texture parameters in the core plug to simulated rock properties. The geological model is incorporated in the fluid flow model, which consists of coupled partial-differential equations (PDEs). However, the nonlinearities and heterogeneities within geological model make reservoir simulations a time-consuming task, especially when detailed models are used. In this context, AI and more particularly artificial neural network (ANN) are considered to be useful tools to establish a mapping between input subsurface properties (geological model) and output observed data (recovery factor).