GPU-accelerated integral equation method for 3D modelling of induction logs

Drilling a well successfully requires not only reaching the target depth but also ensuring the well is accurately placed within the reservoir. This process, known as geosteering, depends on making real-time adjustments while drilling to keep the borehole within the most productive zone. Effective geosteering brings several advantages: it maximizes hydrocarbon recovery, reduces drilling costs, and lowers the risks associated with deviating into unwanted formations. To achieve this, drilling operations rely on advanced measurement technologies. One of the most important is the tri-axial electromagnetic (EM) induction tool, which records how electromagnetic fields interact with the surrounding formation. By analyzing and inverting this data, we can reconstruct the electrical conductivity distribution around the borehole, which serves as a proxy for geological structure and fluid content. This information provides valuable insights into reservoir boundaries and helps operators make more accurate decisions in real time.

This project focuses on advancing the forward modeling and inversion of tri-axial EM induction data to support such applications. The forward modeling is based on the integral equation method, combined with matrix-free implementations and efficient approximations that significantly reduce computational memory demands and speed up calculations. These improvements allow the methods to scale effectively on modern parallel computing resources such as GPUs. For the inversion, the adjoint method is employed within this matrix-free framework to further enhance efficiency and reduce computational overhead. Together, these innovations are designed to bring real-time 3D inversion of EM induction logs closer to practical use. By enabling faster and more accurate subsurface imaging, the project contributes directly to improved geosteering workflows and supports better well placement decisions during drilling.

People

Project members

Durra Handri Saputera - PhD Candidate

Morten Jakobsen - Supervisor