A deep learning-assisted inversion for EM logging tool with tilted-coil antennas in VTI media

An electromagnetic (EM) logging tool with tilted-coil antennas is essential in geoelectrical resistivity measurement for its enhanced capability to discern formation dip, azimuth, and anisotropy, surpassing the capabilities of axial-coil antennas. However, interpreting their measurements in logging-while-drilling is notoriously challenging, often plagued by multi-solution issues and time-consuming processes. As a result, aiming at realizing fast inversion for EM logging tools with titled-coil antennas in a layered vertical transverse isotropy media, this paper proposes a general framework assisted by deep learning. Firstly, a deep neural network (DNN) architecture integrates with adaptive moment estimation and exponential moving averages to enhance the model’s performance. Then, the accuracy of the proposed fast inversion method is validated through several experiments, where different types and levels of noise are introduced to the measurements to test the robustness of the proposed inversion scheme. Finally, the effectiveness of the proposed approach is proved by comparing the inversion scheme with the traditional inversion method in the same scenarios. This study concludes that DNN-assisted inversion can reconstruct subsurface formations in real time and overcomes the limitation of nonlinear iterative algorithms, which typically depend on initial value estimates.