Integrating 3D neural networks fault detection and implicit fault modeling in a global seismic interpretation workflow

Deep neural networks have significantly improved the accuracy and precision of fault detection. This increase in level of detail in the interpretations has opened the door to a completely fault-constrained interpretation of the entire volume but has also made the fault modeling work more complex. This paper introduces an integrated seismic interpretation workflow combining a deep learning approach for fault detection with an implicit fault modeling method. A 3D convolutional neural network detects faults at different scales using an adaptive seismic resampling approach. Faults are then modeled with a novel level-set method, enabling automated editing operations such as merging, truncation, and extension. Applied to the F3 block in the North Sea, the workflow successfully detects major faults with high continuity and captures small-scale faults through a targeted resampling strategy. Automated merging operations and quality control streamline the fault network construction. Finally, the consistent fault network constrains the horizon propagation within the full volume. This approach enhances the fault interpretation efficiency while preserving structural accuracy and controls.