Automated fault network extraction in complex tectonic regimes: A hybrid machine learning and structural attributes approach

Interpreting seismic faults is crucial for prospect generation, reservoir modeling, and CO₂ storage assessment. However, identifying faults in complex tectonic regimes remains challenging, particularly in regions that have experienced multiple phases of tectonic activity. Despite advancements in structural seismic attributes and machine learning, interpreters often still rely on manual methods to analyze intricate fault systems, such as those found in the Poseidon study area located in the Browse basin, Northwestern Australia, where the fault network is shaped by both extensional and compressional tectonic events. This paper introduces a hybrid approach that combines machine learning with seismic structural attributes to extract complex fault networks from 3D seismic data. The method begins by using pre-trained models to generate a fault probability cube, which is then refined through re-training with manually labeled data to incorporate local structural knowledge. To address false negatives, the model is further retrained using an ant-tracking volume generated from the fault probability cube of the manually trained model as automatically labeled data. The fault probability cube is regenerated from the automatically labeled trained model and further enhanced by post-processing techniques, such as ant-tracking, to improve fault connectivity and streamline the automated fault identification process. This hybrid approach effectively detects and extracts both major and minor discontinuities from 3D seismic data with high accuracy, significantly reducing the time and effort required for interpretation compared to traditional techniques.