Journal Article

Fault Delineation On Syntetic Seismic 3D Data Using Artificial Intelligence

Febriarta, Bondan, Priyono, Awali

2024

Overview

Fault delineation is a crucial process for exploring stratigraphic structure and reservoir properties from seismic data. It plays a vital role as faults often offer valuable insights into the accumulation and migration paths of geological resources. As the size of seismic data increases, fault picking becomes a laborious task, demanding high accuracy from interpreters. Automation is essential to expedite this process and minimize human subjectivity in fault picking. To automate this task, we employ deep learning algorithms, particularly convolutional neural networks (CNNs). In this research, we utilize a 3D U-Net architecture known as FaultSeg3D, with a focal loss function for the training process. The dataset comprises 220 pairs of synthetic data, including 200 pairs for train/test and 20 pairs for validation. Results from the training process indicate a converging loss function curve, with values of 0.0154 for training and 0.0308 for testing. This convergence signifies the success of the training process. Quantitatively, fault delineation estimates from the CNN model demonstrate favorable values based on performance metrics, including precision, recall, and F-1 score, on validation data—approximately 0.6, 0.9, and 0.75, respectively. Each of these value was obtained from testing on validation data. Qualitatively or visually, the fault delineation estimates on validation data using the CNN model outperform the variance attribute. The resulting fault delineation estimate from the CNN model appears more continuous, with fewer inaccuracies compared to the variance attribute. Considering the excellent performance metrics when applied to synthetic data, this potential could be promising to continue with its application to field data which can be carried out for further research.

• Publisher: IOP Publishing
• Subject: Algorithms / Artificial intelligence / Artificial neural networks / Automation / Business metrics / cnn / Convergence / Deep learning / Delineation / Estimates / fault delineation / Fault lines / geophyscis / Machine learning / Neural networks / Performance measurement / seismic / Stratigraphy / Synthetic data / Training / Variance