Fault interpretation is an important step in the seismic interpretation process and is critical for understanding challenges such as reservoir compartmentalization, fluid migration, and drilling hazards. Recently, assisted fault interpretation workflows leveraging machine learning techniques have become a promising way to automatically identify faults in seismic. Convolutional neural networks (CNNs) are a popular new method to identify fault attributes in seismic data by analyzing image segmentation and feature extraction. In this abstract, we applied a dual-channel CNN architecture to train seismic data and its discontinuity attribute together to increase the fidelity of the fault prediction process. We first trained a model using synthetic data only, we then trained a second model augmenting it with real data from the study area. We then implemented an unsupervised machine learning clustering approach to analyze the fault probability map and extract fault sticks. This automated workflow took less than an hour to complete compared with over a week for an experienced geoscientist to manually pick approximately 200 faults in the same study area. This experience shows that machine learning-based fault imaging and extraction is a valuable tool for fault interpretation. The automated workflow can be used to provide a quick initial fault interpretation or to identify alternative interpretations and better assess fault uncertainty.

This content is only available via PDF.
You can access this article if you purchase or spend a download.