No abstract

We illustrate novel application of Machine Learning (ML) assisted fault interpretation in the Middle East, interpreting complex fault structures with subtle throws in a large carbonate field onshore Abu Dhabi. Introduced as part of an integrated multi-disciplinary digital excellence initiative at ADNOC, ML-assisted fault interpretation seeks to overcome historic operational bottlenecks caused by traditional seismic interpretation methods which are slow, labour intensive, repetitive, and subjective. Core objectives for deploying ML-assisted fault interpretation were to reduce evaluation time, improve interpretation accuracy, and ensure integration across an intelligent evaluation ecosystem comprised of various disciplines. Envisaged gains from deploying ML-assisted fault interpretation methodology included effective and efficient utilization of multiple seismic datasets to drive rapid multi-scenario analysis, leading to better subsurface understanding within much shorter time frames. Input data used of the project was standard amplitude volume with minimal user-end conditioning. PSTM time and PSDM depth seismic volumes were used in separate runs to confirm that applied ML technology is domain agnostic. The ML-Assisted workflow included: Generating a fault prediction cube based on user-supplied fault interpretation labels made on 6 training lines (<0.8% of the available lines); Creation of fault planarity and azimuth cubes; Parameterization of automated extraction function; Extraction of segmented 3D fault pointsets; Creation of fault framework and fault sticks that can be integrated into traditional methods in seismic and geological modelling domains. Despite limited fault displacement apparent on the seismic volumes, ML fault predictions were of high quality, closely adhering to the seismic response as guided by user-provided training samples. Advantages envisaged from use of ML-assisted interpretation technology in the project were fully realized as the technology enabled rapid extraction of complicated fault structures within a fraction of the time and effort previously taken using traditional means. Efficiency and precision gains from using ML-assisted fault interpretation presents benefits that single seismic volumes can be evaluated thoroughly, and multiple seismic datasets (e.g. various azimuthal volumes) can be evaluated consistently for multi-scenario analysis to reduce subsurface risk and inform better decisions at all phases of the E&P Asset lifecycle.