The manual interpretation of prolific seismic datasets is both time consuming and expensive, prompting Bhaskar Mandapaka (Data Scientist, Halliburton Landmark) and Dr. Youli Mao (Data Scientist, Halliburton Landmark) to develop a new tool which leverages deep learning and automation, to facilitate and more efficient and cost effective seismic interpretation.
Interpreting geological faults from seismic data is an extremely important step in petroleum exploration and production as faults can provide a special insight in the subsurface structure and reservoir compartmentalization. With the development of acquisition technology, the amount of seismic data to be analyzed has increased tremendously over the years, and the manual interpretation of these prolific seismic datasets within a short timeframe is inefficient expensive in terms of domain-expert time. Although there are several fault interpretation algorithms that have been developed to tackle this challenge, most of these workflows still require significant manual effort.
Currently, many billions of dollars are spent on seismic interpretation in the oil & gas industry every year. The goal of this project was to provide an automatic fault interpretation tool that facilitates the seismic interpretation process. We proposed a deep learning based classification algorithm to identify geological faults from seismic volumes which takes advantage of the latest developments in deep learning as well as some algorithms from image classification.
This tool can efficiently handle the large volume of seismic data and reduce the manual effort across the entire workflow. In addition, it can fully capture all the possible faults and fractures from the seismic volumes. This approach shows great promise in seismic interpretation and can help customers increase productivity and reduce costs. The next step will be to bring the deep learning methods into seismic processing area. Such an approach will help advance the automation of seismic exploration in the future.
The presentation can be found here.